JP4824518B2 - Gas turbine performance diagnostic system, diagnostic method and display screen - Google Patents

Description

  The present invention relates to a performance diagnosis system and a diagnosis method for a gas turbine. Moreover, it is related with the display screen in a gas turbine performance diagnostic system.

  The performance of a gas turbine varies greatly depending not only on operating conditions such as fuel flow rate but also on atmospheric conditions such as intake air temperature. The measured value of the power generation output is not only greatly waved in the cycle of the year due to the annual fluctuation of the intake air temperature, but also fluctuates due to fluctuations in the load factor during operation and daily fluctuations in the intake air temperature. Therefore, it is impossible to discriminate any abnormality in performance degradation. In order to cope with this, various diagnostic methods have been proposed.

  For example, Patent Document 1 proposes to evaluate actual machine data in a certain common condition in order to exclude performance fluctuations due to operating conditions. Patent Document 2 proposes a method in which several performance factors are modeled in combination with a plurality of specific operating conditions, and fluctuations due to changes in operating conditions are excluded based on the results. Patent Document 3 proposes a moving average processing method for excluding the influence of fluctuations after modeling performance factors and operating conditions in combination.

  On the other hand, there is a Mahalanobis-Taguchi method (hereinafter referred to as MT method) as a method for grouping normality and abnormality of data based on the distribution state of multidimensional information (Non-Patent Documents 1 and 2). The MT method originally defines a normal data group as a reference in advance for multi-dimensional measurement values (this group is called a unit space), and calculates the Mahalanobis distance of the evaluation target data with respect to this normal data group (unit space). This is a method of determining whether the data to be evaluated is normal or abnormal depending on whether the calculated statistical distance is equal to or less than a certain threshold value.

Japanese Patent Laid-Open No. 11-15516 JP 2003-83089 A Japanese Patent No. 3538670 Genichi Taguchi: “Technology Development in MT Systems”, Japanese Standards Association, (2002) Takashi Kamoshita et al .: “Ohanashi MT System”, Japanese Standards Association, (2004)

  The above-described performance diagnosis method has a problem that the data extraction rate (the ratio of data to be extracted) decreases when aiming at high accuracy. The performance of the gas turbine varies greatly depending on the load. Particularly, the characteristics at a partial load are nonlinear and complicated. For this reason, it is important to extract the base load operation data (operation data near the rated load excluding partial loads) with high accuracy in order to build a highly accurate model and evaluate the performance with high accuracy. Become.

However, this decreases the range of the threshold value in order to increase the accuracy of the model, so that the data extraction rate decreases. On the other hand, if the threshold range is widened so that the extraction rate does not decrease, the accuracy of the model decreases. Decreasing the data extraction rate can cause problems as described below.
(1) The operating conditions and the number of data for building the performance model are insufficient, and the accuracy of the performance model is lowered. As a result, the accuracy of performance evaluation is lowered.
(2) Unevenness occurs in the data extraction status, and time that cannot be evaluated is excluded at the extraction stage. Such a time increases, and an operation period in which performance evaluation cannot be performed occurs intermittently in a worm-eaten state, which hinders continuous performance evaluation.
(3) Since the period of actual machine data required to build a performance model becomes longer, even if a performance evaluation system is introduced, the period required to actually build a performance model of the actual machine and start performance evaluation become longer. In addition, the data period necessary to ensure the reliability and confirm the evaluation becomes longer, and the diagnosis becomes slower.
(4) When further analysis is performed using the data of the diagnosis result, the number of original data decreases, so that sufficient accuracy of the detailed analysis cannot be ensured.

  An object of the present invention is to provide a diagnostic system and a diagnostic method that can extract more data while keeping the accuracy sufficiently high in order to enable highly accurate diagnosis.

The gas turbine performance diagnosis system according to the present invention includes characteristic modeling means for characteristic-modeling a performance index as a function including an intake air temperature as an input based on operation data of the gas turbine. In the gas turbine performance diagnosis system for diagnosing the performance of the gas turbine based on the output value obtained by inputting the operation data, the time is determined in advance corresponding to each time of the time-series operation data (hereinafter, the time). The operation data (hereinafter referred to as reference data) at a plurality of times in the past is selected within a predetermined period (hereinafter referred to as reference period) and calculated based on the distribution or actual measurement value of the selected reference data. A measured distance indicating the degree of approach of the index value calculated based on the measured value or the measured value of the time operation data for the distribution of at least one type of index value Then, when the statistical distance is within a predetermined upper and lower limit range, a process of selecting the time operation data as the base load operation data for the characteristic modeling is performed at each time of the time series operation data or predetermined. It is characterized by comprising operation data selection means for repeatedly inputting the base load operation data selected by repetition processing for a plurality of times to the characteristic modeling means .

According to the gas turbine performance diagnosis method of the present invention, at least one performance index among power generation output, power generation efficiency, compressor pressure ratio, fuel flow rate, compressor efficiency, turbine efficiency, and intake air flow rate is based on the operation data of the gas turbine. Is characterized as a function that includes at least the intake air temperature and the intake pressure as inputs, and the performance diagnosis of the gas turbine is performed based on the output value obtained by inputting the operation data of the gas turbine to the obtained characteristic model. In the method, the base load operation data used for the characteristic modeling is selected by determining an actual measurement value or at least one kind of index value calculated based on the actual measurement value. The operation data of a plurality of times in the past is obtained in a predetermined method from the reference period determined corresponding to each time (the time) of the operation data of And calculating a statistical distance of the measured value or index value of the driving data for the time with respect to the distribution of the measured value of the driving data for the plurality of times or the distribution of at least one index value calculated from the measured values. Data of each time series in the time series is determined as to whether or not it is within a predetermined upper and lower limit threshold value range, and if it is within the range, this operation data is selected as the base load operation data. It repeats with respect to, It is characterized by the above-mentioned .

  According to the present invention, it is possible to solve the problem that the data extraction rate decreases when high accuracy is achieved, and it is possible to perform extraction with higher accuracy by effectively combining a plurality of base load extraction methods. As a result, it is possible to evaluate the performance deterioration of the gas turbine or the performance recovery due to maintenance with higher accuracy than the conventional method.

  FIG. 1 shows an example of an embodiment of a gas turbine diagnostic system according to the present invention.

  In this system, time-series operation data collected for the gas turbine 101 at regular time intervals is transmitted from the transmission unit 111 and received by the reception unit 113 via the communication line 112. Then, the received information is processed by the calculation means 102, and the performance evaluation result is output to the output means 103.

  Here, the communication line 112, the transmission unit 111, and the reception unit 113 may be any known information transmission unit and transmission / reception unit thereof. For example, any communication unit such as the Internet or a wired / wireless information line and transmission / reception unit thereof is used. be able to.

  The internal configuration and processing system of the calculation means 102 are as described below.

  The input unit 130 passes the information received from the receiving unit 113 to the base load data extracting unit 131 shown below.

  The base load data extraction unit 131 extracts base load operation data from the operation data of the actual machine input from the input unit 130. The base load operation data refers to operation data in the vicinity of the rated load excluding the partial load of the gas turbine, that is, in the vicinity of the maximum output corresponding to the intake temperature / intake pressure of the gas turbine. However, in the operation plan of the target gas turbine, when the load factor is not 100%, for example, when continuously operating at a load factor of 90%, or when operating at a load factor of 70% at night, these constants It is assumed that the operation data near the load factor is extracted instead of the base load operation data. In the present invention, such partial load data in the vicinity of the constant load factor are also collectively referred to as base load operation data.

  The base load data extraction means 131 is typically referred to as a sample in order to select the base load operation data from the time series operation data delivered from the input means 130 as shown in FIG. A reference period setting unit 134 is provided for determining from which period base load data of a certain period in the past (hereinafter referred to as a reference period) is to be adopted in accordance with a predetermined procedure. Further, reference data selecting means 135 for identifying and selecting reference base load operation data (hereinafter referred to as reference data) from a reference period determined in this manner according to a predetermined method, and the reference data thus selected. A statistical distance calculating means 136 for calculating a statistical distance indicating how close the time-series operation data to be evaluated is statistically close to the group consisting of: If the statistical distance calculated for the driving data is less than or equal to a predetermined threshold, the driving data is selected as base load driving data, and if the threshold is exceeded, the driving data is base loaded. A determination / selection unit 137 that performs a process of determining that the operation data is not selected and not selecting the operation data is provided.

  Examples of specific processing methods in the reference period determination unit 134 to the determination / selection unit 137 will be described in detail with reference to FIGS. The reference period setting means 134 is step 7 in FIG. 2, the reference data selection means 135 is step 21 in FIG. 3, the statistical distance calculation means 136 is step 8 in FIG. 2 and the step 22 in FIG. 3, and the determination / selection means 137 is This corresponds to step 13 in FIG.

  The modeling unit 132 receives the base load operation data output from the base load data extraction unit 131 as input information, and models the performance characteristics of the gas turbine. Specifically, the model constant of the performance characteristic model is determined. Here, the performance characteristic model is a calculation model that calculates and outputs a predicted value of performance with the operating conditions of the gas turbine as input (details will be described later in FIG. 7), and is implemented as a program module. The model constant is a constant specific to the target gas turbine set in these program modules. In the present invention, a model and a model constant refer to a program module that receives a data input, calculates and outputs a result, and a constant set by the program module.

  The performance here refers to, for example, power generation output, power generation efficiency, pressure ratio of the compressor or turbine, intake flow rate or exhaust flow rate, exhaust heat utilization efficiency, fuel flow rate, and the like. These are general indicators whose values change as the driving state changes, and are not limited to those shown here. In addition, the operating conditions include the temperature and pressure of the intake air as a necessary minimum. In addition to this, the operating conditions such as the intake air humidity, the fuel flow rate, the amount of steam and water injected into the combustor, and the overall operating conditions are appropriately selected. Included.

  The performance evaluation unit 133 evaluates the operation data of the gas turbine based on the model and the model constant determined by the modeling unit 132, and quantitatively evaluates the transition of performance degradation over time, performance recovery due to maintenance, and the like. To do. Such evaluation of transition such as deterioration of performance over time and recovery during maintenance is referred to as performance evaluation in the present invention.

  There are two performance evaluation methods: a method for converting to standard conditions and a method for calculating the difference between actual measurement and prediction.

  In the method of converting to standard conditions, the performance measurement value of the base load operation data extracted by the base load data extraction means 131 is converted into an equivalent value (standardized data) when the standard condition is corrected. The standard condition is a fixed operating condition such as a standard atmospheric condition defined by ISO (intake temperature 15 ° C., atmospheric pressure 1 atm, relative humidity 60%). Thereby, the time-series data of the performance of the actual machine is converted into time-series data of the performance when adjusted under a certain condition by correcting the influence of the fluctuation of the operating condition. The amount of time-series decrease in the value of the standardized data generated in this way indicates the amount of performance decrease due to deterioration or abnormality, and the increase amount indicates the amount of performance recovery due to maintenance or the like.

  The advantage of this method is that it is converted to performance under constant operating conditions regardless of annual operating condition fluctuations. It can be regarded as.

  In the method of calculating the difference between the actual measurement and the prediction, the performance of the gas turbine in which the model constant is determined by the modeling means 132 is obtained from the data of the operating conditions included in the base load operation data extracted by the base load data extraction means 131. Input into the characteristic model to predict and calculate the performance of the gas turbine according to the operating conditions, calculate the deviation of the actual measurement value from this prediction result, and from the transition of the deviation in time series, the performance degradation status and maintenance Evaluate the recovery effect. For example, if the deviation is enlarged, the performance degradation amount is evaluated based on the deviation value, and if the deviation is reduced, the performance recovery amount is evaluated based on the reduction amount. The advantage of this method is that it does not require standard condition conversion and is simple.

  In this way, by selecting only the base load operation data by the base load data extraction means 131, the subsequent modeling means 132 avoids a reduction in the accuracy of the model due to the non-linearity of the partial load performance of the gas turbine and increases the accuracy. Modeling is possible, and the performance evaluation means 133 provides a highly accurate evaluation result.

  The functions of the base load data extraction unit 131, the modeling unit 132, and the performance evaluation unit 133 described above can be performed at any place on the information path from the data collection unit attached to the gas turbine 101 to the final output unit 103. It may be dispersed. For example, the base load data extraction means 131 for extracting the base load operation data can be provided in a stage before the information is transmitted by the transmission means 111. In this way, the information transmitted through the communication line 112 is only the base load data after extraction, so the amount of information communication can be reduced, and the execution of the base load data extraction means 131 on the calculation means side becomes unnecessary. Therefore, the calculation load can be reduced.

  The result processed by the calculation means 102 as described above is transmitted to the output means 103 via the information line 114 and output. Specifically, the output means 103 may be any known means such as a display device such as a computer display, electronic data storage means such as a hard disk or memory, and information printing / drawing means such as a printing machine. The output data in the output means 103 is output as a trend graph indicating time series transition of standardized data or time series record data.

  In this example, the positional relationship between the installation locations of the gas turbine 101, the calculation means 102, and the output means 103 is assumed to be in a location different from the installation site of the gas turbine 101.

  In this way, the deterioration status of a large number of facilities can be intensively monitored and grasped at a remote monitoring center or the like. In addition, the calculation means 102 can be operated in its own management, and the calculation means can be updated to new equipment and the calculation method can be improved without changing the information management system of the actual machine site. It can be operated flexibly and efficiently.

  However, this is only an example, and the positional relationship between the installation locations is not limited to this, and may be arbitrary. For example, the output unit 103 may be installed at the installation site of the gas turbine 101. This is when a business operator (referred to as a site business operator here) where the gas turbine 101 is installed receives a performance diagnosis service for the gas turbine 101 from another business operator (referred to as a diagnostic business operator). This is a preferred form. The site provider side can obtain only the information of the diagnosis result of deterioration that is finally necessary without installing and maintaining the equipment such as the calculation means 102 by itself.

  Alternatively, for example, the output unit 103 may be arranged at a different location from the installation site of the gas turbine 101 and the installation site of the calculation unit 102, for example, at the head office of the installation company (site company) of the gas turbine 101. This is suitable when the department that develops and examines the maintenance plan based on the diagnosis result information of the deterioration state of the gas turbine 101 is located at a site different from the site such as the head office of the site operator.

  FIG. 2 shows an internal processing procedure in the base load data extraction unit 131 of FIG. Broadly speaking, a statistical distance calculation step 4 for calculating the statistical distance that the data at each time to be evaluated has with respect to the distribution of data in the reference period, and a threshold at which the calculated statistical distance at each time is constant The threshold value determination / data selection step 5 is performed for determining whether or not the data is within the range and selecting data within the range. Each is described below.

  The statistical distance calculation process 4 is an iterative process starting from the process 6 for the data at each time in the evaluation target period, and the repetitive process refers to the past referred to in order to calculate the statistical distance of the data at each time. A reference period setting step 7 for setting the data period (reference period), and a step 8 for calculating a statistical distance of the data at the time for the past data group selected in the reference period setting step 7; In this way, step 9 for determining whether or not the calculation of the statistical distance performed in steps 7 and 8 has been completed for all times of the operation data in the evaluation target period, and if not completed, the process proceeds to the data at the next time. It consists of process 10.

  The reference period setting method in the reference period setting step 7 may be a predetermined method, and a plurality of data in a certain period before the time with respect to data at an arbitrary time in the evaluation target period. What is necessary is just to be able to specify. As an example, there is a method of setting past data for a certain number of days or hours as a reference period corresponding to each time. In this case, the length of the reference period is preferably about one week. The operation of gas turbines may be lighter on weekends when power generation demand decreases, and one week is a measure of the operation cycle. It is preferable. Alternatively, in the same way, when considering from the guideline of the operation cycle, one day may be set as the length of the reference period in which the intake air temperature and the operating conditions are completed. Alternatively, a method of extracting data in which the intake air temperature and the power generation output are within a predetermined range from data in a certain period in the past may be used. In addition, for example, the demand for electric power is reduced at night, and the load factor of the gas turbine may also be reduced. In consideration of such diurnal characteristics, the night is excluded, or only a certain time of day is used. A method may be used in which certain selection conditions are configured by combining various selection criteria, such as extracting or excluding weekend data, thereby specifying and selecting data.

  As a statistical distance calculation method in the step 8 (statistical distance calculation step of data at each time), at least one of reflecting the load state of the gas turbine with respect to the past data group selected from the reference period. One or more of the above indices, for example, power generation output, power generation efficiency, pressure ratio, fuel flow rate, exhaust gas temperature, intake air flow rate, or a load index calculated by the method of FIGS. A statistical distance, for example, Mahalanobis distance, of the data at the time is calculated with respect to a distribution in a multi-dimensional space composed of indices (for example, a four-dimensional space if there are four indices).

  If the determination result in Step 9 is false (False, incomplete) and the process proceeds to data at the next time in Step 10, Steps 7 and 8 are repeated again. When this process is repeated and the process for the entire evaluation period is completed, the determination in step 9 becomes true, and the process proceeds to process 5 in the latter half.

  The threshold determination / data selection step 5 is an iterative process starting from step 11 for the data at each time in the evaluation target period. In the iterative process, it is determined whether or not the statistical distance with respect to the reference period data of the data at the time calculated in the statistical distance calculation step 4 is within the upper and lower limits of the threshold set based on a predetermined method. A threshold determination step 12, and a selection step 13 for selecting data at the time if the result of the threshold determination step 12 is true (TRUE) and not selecting it if false (FALSE); A determination step 14 for determining whether or not such threshold determination and selection according to the determination result are completed for all times of the operation data in the evaluation target period, and the determination result is false (FALSE, incomplete) In this case, the process 15 advances to data at the next time.

  If the determination result of the determination step 14 is false (FALSE, incomplete) and the process proceeds to the data at the next time in step 15, select the determination of the threshold determination step 12 and the determination result is true The selection step 13 to be repeated is repeated for all time data. When this is completed for all times, the determination result of the determination step 14 becomes true, and the data extraction process ends. The group of data at the time selected in the selection step 13 in the above-described threshold determination / data selection step 5 repetition process becomes the data extracted as the base load operation data.

  As a threshold setting method in the threshold determination step 12, the upper and lower limits of the statistical distance are set to preset fixed values, or are sequentially calculated as a function of data variation. It is also possible to use a method that is set as follows.

  In this way, in this way, the threshold determination is performed for the statistical distance with respect to the past reference period for the multidimensional evaluation index of the driving data. By using the statistical distance of the multidimensional index in this way, for example, when one specific index is a noise or a sudden value, the correlation with the other index is statistically considered. The If there is a correlation such that the fluctuation or sudden change of the indicator occurs simultaneously in other indicators, the statistical distance becomes relatively small, and only the indicator that does not occur in other indicators. In the case of noise, the statistical distance becomes relatively large, and thus the proximity to the center of the distribution is quantified in consideration of the variation of multiple indices. For this reason, in the threshold value determination step 12 and the selection step 13, the base load operation data can be selected from the vicinity of the center of the fluctuation by excluding a statistically large portion. Therefore, the accuracy of the performance model constructed based on these data becomes high, and the evaluation accuracy for performance degradation and recovery becomes high.

  In the above description, the timing for determining the past data reference period in the reference period setting step 7 is executed every time the processes 6 to 9 are repeated, but the reference period is not updated every time as described above. For example, even if the update timing is determined depending on whether the calculated value of a predetermined function that evaluates changes in operating conditions or environments every certain number of hours or days exceeds a preset threshold value Good. In this way, the amount of calculation of the reference data update process is reduced, so that the diagnosis calculation process can be speeded up.

  FIG. 3 shows an example of a statistical distance calculation flow for the data at each time. This is a detailed procedure example of the step 8 (or the reference data selection means 135 and the statistical distance calculation means 136 in FIG. 1) for calculating the statistical distance at each time in FIG. The whole is a step of selecting past data at a plurality of times from a reference period (unit space setting step 21), a step of calculating a statistical distance of the time data and a group of data selected by the step (reference of each time data) It comprises a statistical distance calculation step 22) from the period to the selected data. Step 21 corresponds to the reference data selection means 135 of FIG. Step 22 corresponds to the statistical distance calculation means 136 of FIG. 1 and step 8 of FIG. Each is described below.

  The step of selecting past data at a plurality of times from the reference period (unit space setting step 21) starts from the step 23 for the data at each time in the reference period selected in the reference period setting step 7 (FIG. 2). It is an iterative process. The repetitive processing is performed in steps 24 to 30 described below. First, for the target data at each time, at least one index reflecting the load state of the gas turbine, that is, the power generation output, power generation efficiency, pressure ratio, fuel flow rate, exhaust gas temperature, intake air flow rate set in step 8 of FIG. Alternatively, for each of at least one of the load factor indicators calculated by the methods of FIGS. 4 and 5 (to be described later) (hereinafter collectively referred to as load factor indicators), upper and lower limit ranges predetermined for each indicator. It is determined whether it is within the range (index-specific threshold determination steps 24-26). Based on the threshold value determination result of each index executed in the threshold value determination steps 24 to 26 for each index, whether the determination result of all these indexes is true or not (at least one false index If it is false, true / false is determined (logical index determination step 27 for all indices). If the determination result of the logical product determination step 27 for all indexes is true, the operation data at that time in the reference period is selected (selection step 28). Determining whether the above-described steps 24 to 28 (when the result of step 27 is false, step 28 is not performed) have been executed for all data in the past reference period (that is, all data of step 7 described above) (Completion determination step 29), and when the result of the completion determination is false (FALSE, incomplete), the process proceeds to data at the next time (step 30).

  If the result of the completion determination step 29 is false (not completed for all data) and the process proceeds to the next time data in the reference period in step 30, the steps 24 to 28 (or 27) are performed. The process is repeated until the determination in step 29 becomes true (until it is completed for all times of operation data in the reference period). When the determination in step 29 becomes true in this way, the process proceeds to the next step 22.

  Note that the threshold value determination in the index-specific threshold value determination steps 22 to 24 shows the determination example of three indexes in FIG. 3, but in the case where the number of indexes is n, similarly n index values Threshold determination processing is executed for.

  The distribution of the data group selected in the selection step 28 corresponds to a unit space of the MT method (Mahalanobis Taguchi method). In this respect, the data of this group is hereinafter referred to as a unit space as appropriate.

  In the subsequent step 22, the group of data selected in the step 28 in the repetition process of the above steps 23 to 29 is referred to, and the statistical distance between this group and the data of the evaluation target time is calculated. As a method for calculating the statistical distance, a known Mahalanobis distance or the like may be used. Among them, a Mahalanobis distance D based on Taguchi definition represented by the following equation (1) is preferable.

本方法には以下の効果がある。

This method has the following effects.

  (A) By performing determination using viewpoints from a plurality of indices as in steps 24 to 26, it is possible to extract unit space data serving as a reference for base load determination with high accuracy. In addition, in the logical product determination step 27 for all indexes, only data for which the determination result for all indexes is true is selected as unit space data. Therefore, the data included in the unit space data is extremely outlier in statistical terms. Less. As a result, the distance calculated in step 22 becomes a highly accurate evaluation value of the distance to the base load operation area, and the accuracy of base load data extraction in the subsequent threshold determination / data selection step 5 (FIG. 2). Becomes higher. As a result, the performance characteristic model of the subsequent modeling unit 132 (FIG. 1) becomes highly accurate, and the performance degradation and recovery evaluation accuracy in the final performance evaluation unit 133 (FIG. 1) becomes high.

  (B) Since the Mahalanobis distance characteristic is used, where the distance is small when the correlation between multidimensional indices is high and the distance is large when the correlation is low, the data is excluded in the judgment of a single index. However, when the correlation with other indexes is high, the Mahalanobis distance falls within the threshold range and can be extracted as base load data. The method of expanding the extraction range is based on the distribution characteristics of data in a multidimensional space, that is, whether it is dense, scattered, or in what direction the density is biased, From the vicinity of the center of gravity where the data is concentrated toward the peripheral direction, the portions having the same density are sequentially selected in order from the portion having the higher distribution density to the portion having the lower distribution density. Therefore, it is possible to secure a range in which the data extraction rate can be increased while keeping the extraction accuracy substantially the same without drastically reducing.

  (C) As described in this figure, the unit space is sequentially updated for each data of the evaluation target, so the influence of seasonal variations in operating conditions and performance characteristics due to deterioration over time Can be considered. That is, since the unit space is set following these state changes, the base load data can be extracted with high accuracy in consideration of the influence of seasonal variation and aging deterioration.

  As described above, the present invention can solve the problem that the data extraction rate decreases when the accuracy is increased. In addition, it is possible to provide a method for extracting more accurately by effectively combining a plurality of base load extraction methods.

  FIG. 4 shows an example of a calculation process flow of the load factor index value determined in step 24 (or steps 25 and 26) in FIG. Hereinafter, the calculation method based on this processing procedure is referred to as an output regression method. The procedure includes an output characteristic model identification step 41 for identifying an output characteristic model representing a change in power generation output according to the intake air temperature and the intake pressure based on actual machine time-series operation data, and the intake air temperature and the intake air pressure in the output characteristic model. The load factor calculation step 42 for calculating the value of the load factor based on the calculated value of the power generation output that is output by substituting the actual measurement value and the actual value of the power generation output, and the load calculated in the load factor calculation step 42 It comprises a load factor calculation step 43 for converting so as to suppress fluctuations in the index value. Details of each step will be described below.

The output characteristic model in the output characteristic model identification step 41 is a function represented by the following equation that includes the intake air temperature Tci and the intake air pressure Pci as inputs and outputs the power generation output We.
We = f_We (Tci, Pci) (2-1)
Specifically, this function f_We is preferably expressed as follows, for example.
_{We = a e × (Tci-} Tci 0) + b e × (Pci-Pci 0) + α e × t + We 0 ... (2-2)
Here, t represents the operating time, and Tci ₀ and Pci ₀ are standard condition values of the intake air temperature and the intake air pressure. The operating time t only needs to be indicative of the passage of time. For example, the operating time of the gas turbine is considered taking into account factors that affect deterioration, such as the equivalent operating time converted in consideration of the effect on the life due to combustion temperature and start / stop. You may use the corrected time. In the following description, it is assumed that the operation time t includes such conversion time. The term of the operating time t can be omitted when the influence of deterioration over time is considered to be negligible in the time-series operation data period to be evaluated. It is desirable to include this term if it cannot be ignored that performance has deteriorated.

By the influence of deterioration over time is separated from the section to the coefficients alpha _e, it can be avoided to include bias due time degradation to claim a _e, b _e operating conditions characteristic of the output, the high output characteristics Can be modeled with accuracy.

In the formula (2-2), a _e , b _e , α _e and We ₀ indicate model constants. Model identification in the output characteristic model identification step 41 is to determine these model constant values. a _e and b _e are coefficients for correcting the power generation output by the deviation of the intake air temperature and pressure from the standard conditions. α _e indicates a rate of decrease in the power generation output with time t (hereinafter referred to as a deterioration coefficient). The deterioration coefficient α _e can be omitted when it is considered that there is almost no performance deterioration during the period of the actual machine time series operation data used in the output characteristic model identification step 41. We ₀ indicates the power generation output in the initial stage (before deterioration during the period starts) when Tci ₀ and Pci ₀ are standard condition values.

In addition, as an item calculated by Expression (2-2), in the case of a gas turbine that injects a refrigerant such as steam (hereinafter referred to as a refrigerant injection type, a steam injection type, etc.), the standard condition of the refrigerant injection amount Gs is used. The term C _e × (Gs−Gs ₀ ) to be corrected in accordance with the deviation from the value Gs ₀ may be added. Here C _e is a model constant. In this way, when the change in the flow rate or temperature of the refrigerant particularly injected in the refrigerant injection type gas turbine is large, the load factor can be quantified with high accuracy in consideration of the influence.

As a method for determining these model constants a _e , b _e , α _e , We ₀ , the power generation output calculated by substituting the operating condition included in the time series operation data of the actual machine into the right side of the equation (2-2) It is preferable to determine such that the error between the time-series calculation result and the time-series data of the power generation output included in the actual machine time-series operation data is minimized throughout the data period. This can be obtained by a known method such as a least square method. An example of such a method will be described later with reference to FIG.

Alternatively, if there are design values, plan specification values, and experience values (hereinafter collectively referred to as a priori values) of the model constants a _e , b _e , α _e , We ₀ , these may be used. This method may be more accurate than inaccurate model regression when there are few actual machine data shortly after the start of operation, or when there are many partial load operations and the number of base load data cannot be secured sufficiently. Many and effective. In addition, such an a priori value is set in the initial stage after the start of operation with a small number of data, and after that, when a certain amount of operation data is accumulated, it is updated with a value that is regressed from the data. It is also effective to perform a weighted weighted average according to the number and gradually shift to the regression value from the actual machine data. In the following description, identification of various model constants will be described using a known method such as the least square method as an example, but if there is an a priori value as in the description here, it may be used.

In the output characteristic model identification step 41, actual machine data collected by variously changing operating conditions such as intake air temperature may be used instead of the actual machine time-series operation data. In this case, if the data collection period is short, the deterioration coefficient α _e can be omitted as described above.

The calculation of the value of the load factor index in the step 42 (load factor calculation step A) is an output characteristic model (for example, a model constant determined in the output characteristic model identification step 41, as shown in equation (3-1)). For the predicted value We _{calc of the} power generation output that is output by inputting the time series data of the actual machine operating conditions (measured values of the intake air temperature and pressure in the above example) in the evaluation target period in the equation (2-2)), This is to calculate the ratio of the actual power generation output measured value We _act of the actual machine (referred to as load factor Load _A ).
Load _A = We _act / We _calc (3-1)
The calculation of the value of the load factor index in the step 43 (load factor calculation step B) includes the value of the load factor Load _A at a certain time as shown in Expression (3-2). The ratio (load factor Load _B ) divided by the moving average value or the specific percentile value Load _Amav during a certain period is calculated. The moving average value here refers to the above-described simple moving average value or weighted moving average value of a certain period including the time point. Further, the specific percentile value indicates a cumulative percentage value in the distribution of values during the period (for example, a 50% value (median value), a cumulative 70% value from the lower order, etc.). Hereinafter, in other descriptions, these moving average values and percentile values are collectively referred to as moving average values. By taking the ratio to the moving average value in this way, the error fluctuation range of the load factor index is made sufficiently small, excluding the effects of deterioration of plant performance over time and seasonal performance fluctuations, with a constant value at the center. Can be converted to vertical fluctuations.
Load _B = Load _A / Load _Aav (3-2)
In this case, as a method of taking the moving average period, for example, it is preferable to take one week immediately before the target time, or a data period of about 100 to 200 points. There are the following reasons for this. For example, when driving data is stored at intervals of 1 hour as a typical measurement record, moving average over a period of 1 to 2 days (24 to 48 points), for example, Saturdays and Sundays of weekend holidays When the power demand decreases and the load factor falls, the moving average period is included in this period, and the calculation result of the load factor in Equation (2-2) is overestimated. There is sex. In this case, the load factor actually decreases on the weekend, but the load factor does not decrease in calculation, and the load factor cannot be accurately evaluated. On the other hand, if the moving average is performed over a period of one month or longer, there is a risk that the deterioration in performance in units of one week cannot be captured because the period is too long. Considering these, when processing based on data at one hour intervals, moving average over the period or number of data points as described above may reflect the effects of various operating condition fluctuations and deterioration over time. It is possible and suitable.

In this way, among Load _A calculated in Step 42 and Load _B calculated in Step 43, Load _B calculated in Step 43 is normally determined as a load factor threshold value determination (Step of FIG. 3). 24 to 26). However, the step 43 can be omitted when the fluctuation range of the calculation result of the load factor Load _A according to the equation (2-1) in the step 42 is sufficiently small or the value of the load factor Load _A is sufficiently close to 1. is there. In this case, the value of the load factor Load _A may be used as the load factor index value.

When the load factor index value calculated by the output regression method is determined as a threshold value in the above-described step 24 (or steps 25 and 26), the threshold value may be set as follows, for example. . That is, the upper and lower limit values of the load factor Load _A or Load _B are set to 100 ± P% (in this case, the value of the load factor Load _A is 100-P% or more and 100 + P% or less is true in the determination) The value of P should be about 1 to 3% for industrial gas turbines with a relatively large base load operating time, and about 10% at maximum for power generation gas turbines with relatively large load factor fluctuations. . Alternatively, the upper and lower limit values may not be set symmetrically above and below 100%, and may be set separately, for example, the lower limit is 97% and the upper limit is 102%.

  It should be noted that the period of the actual machine time series data when calculating the load factor in the steps 42 to 43 may be different from the period of the actual machine time series data used when determining the model constant in the step 41 described above. For example, in step 41, constants of the output characteristic model (formula (2-1), formula (2-2)) are determined in a certain period of the actual machine time series data, and in steps 42 to 43, the model constants are determined. Can be applied to the entire period for which the load factor is to be evaluated. In this way, the model constant can be determined efficiently with a small amount of data without using all the data in the target period for degradation evaluation.

  In the method shown by the equations (2-1) to (3-2) of the output regression method, the load factor is calculated using only three pieces of information such as intake air temperature, pressure, and power generation output. Further, in Formula (2-2), the output operating condition correction and the correction due to deterioration are separated. Therefore, it is possible to efficiently calculate the load factor taking into account fluctuations in operating conditions and aging deterioration with a small amount of information and calculation amount. In the output regression method, the load factor is directly evaluated from the operation results as the ratio of the actual output to the power generation output at the expected full load (so-called maximum expected output). It has the feature that it can be quantified from the viewpoint of the output result, not from the viewpoint of the plant output command or plan.

  FIG. 5 shows another example of the calculation process flow of the load factor index value determined in step 24 (or steps 25 and 26) in FIG. Hereinafter, the calculation method based on this processing procedure is referred to as an exhaust gas temperature method. In the exhaust temperature method, in order to calculate the index value of the load factor, the exhaust temperature at the full load state of the gas turbine or the intermediate stage gas temperature of the turbine, or the upper limit value of the combustion temperature management index for control is used. Hereinafter, these are collectively referred to as an exhaust gas temperature index value Tmx. The calculation procedure of the load factor index by the exhaust gas temperature method includes an exhaust gas temperature index model identification step 44 for determining a constant of a characteristic model representing the relationship between the exhaust gas temperature index value and the intake air temperature based on time series operation data of the actual machine, A load factor index calculating step 45 for calculating an index value of the load factor from the predicted value and the actual value of the exhaust temperature index value output by inputting the actual intake air temperature to the exhaust temperature index model identifying step 44; The load factor indicator calculation step 46 for converting so as to suppress the fluctuation of the load factor indicator value calculated in step (b). Details of each step will be described below.

The exhaust gas temperature index characteristic model in the exhaust gas temperature index model identification step 44, which is the exhaust gas temperature characteristic model identification step, receives the intake air temperature Tci and inputs the exhaust gas temperature index value Tmx at this time. It is calculated and output according to the function relationship.
Tmx = f_Tmx (Tci) (4-1)
As the characteristic model of the exhaust gas temperature index, that is, the function f_Tmx in the equation (4-1), a relational expression incorporated in the operation control logic of the gas turbine can be generally used. Gas turbine operation control is generally managed to protect the maximum cycle temperature from exceeding the heat-resistant temperature of the high temperature material when given several operating conditions such as the intake air temperature and the amount of refrigerant injected into the combustor. There is an index value to do. In this index value, the upper limit temperature that should be observed regardless of the operating condition is set as a function of the operating condition and the measured value. During operation, the amount of fuel input is controlled so that the index value is less than or equal to this upper limit value. Therefore, how close the actual index value is to the upper limit temperature is how close the equipment is to the maximum operating temperature, ie how close to full load conditions, in other words how close to the base load operating range. It becomes a guide. Therefore, when this relational expression is used as a model expression of Expression (4-1), the actual control state is reflected as it is, and the index value at the time of base load operation can be accurately predicted. As a result, base load operation data can be extracted with higher accuracy, and degradation diagnosis can be finally performed with higher accuracy.

If the control logic cannot be used as a characteristic model for the exhaust gas temperature index, simply represent the following equation and model the exhaust gas temperature based on actual machine operation data and planned specification values. It may be replaced.
_{Tto = a to × (Tci-} Tci 0) + b to × (Gs-Gs 0) + Tto 0 ... (4-2)
Here, Gs is the refrigerant injection amount to the combustor, and Gs ₀ is a value under the standard conditions (temperature 15 ° C., 1 atm, relative humidity 60% as intake conditions). However, if the gas turbine is of a type that does not inject refrigerant, or if the influence of refrigerant injection is relatively small compared to the intake air temperature, the second term on the right side is omitted. If there are operating conditions other than the intake air temperature that have a strong influence on the exhaust temperature, use this instead of Gs, or change the term from the standard conditions similar to the first and second terms. May be added as a term to be corrected by the model constant. Examples of such conditions include compressor outlet temperature and turbine intermediate stage gas temperature. Also, a _to , b _to , and Tto _o in the above formula are model constants, and in particular, Tto _o is the exhaust temperature under the standard conditions. These model constants can be determined by fitting based on a plurality of actual machine operation data, plan specification values, test operation data, or the like. That is, the error between the exhaust temperature value calculated by substituting the intake air temperature and the refrigerant injection amount into the right side and the exhaust temperature of the actual data using multiple sets of exhaust temperature data according to the intake air temperature and the refrigerant injection amount The values of these model constants should be determined so that the sum of squares is minimized. For this, a known method such as a least square method can be used. According to this method, even if there is no information on the exhaust temperature control logic, it is possible to easily model if there is information such as operation data and planned specification values about changes in the exhaust temperature according to the intake air temperature and the refrigerant injection amount. Is possible.

Note that the equation (4-2) may be further simplified as described above, omitting the term of the refrigerant injection amount, and the following equation (4-3). Even with a steam-injected gas turbine, with a control-type facility that maintains a constant ratio between the fuel flow rate and the steam-injection flow rate, the relationship between the exhaust temperature and the intake air temperature can be modeled with sufficiently high accuracy even with this method. The advantage in this case is that the exhaust temperature can be estimated easily with fewer variables.
_{Tto = a to × (Tci-} Tci 0) + Tto o ... (4-3)
The calculation of the load factor index value in the step 45 (load factor index calculation step) is a function (formula (4-1)) that includes the intake air temperature Tci specified in the step 44 and outputs the exhaust gas temperature index value Tmx. )), The estimated value Tmx _calc of the exhaust gas temperature index value that is output by substituting the operation data of the intake air temperature Tci at each time point in the evaluation target period, and the deviation ΔTmx _A of the actual measurement value Tmx _act with respect to this, It is a calculation. This ΔTmx _A is an index value of the load factor. When this value is 0, the load factor is 100%, when the value is positive, the load factor exceeds 100%, and when the value is negative, the load factor is less than 100. .
_{ΔTmx A = Tmx act - Tmx calc} ... (5-1)
The calculation of the load factor index value in the load factor index calculation step 46 includes the value at each time point in the evaluation target period for the load factor index value ΔTmx _A calculated in the previous step 45 and the time point. This is a calculation based on the following equation for the deviation from the moving average value or specific percentile value ΔTmx _Amak for a certain period. This ΔTmx _B is obtained by taking a deviation from the moving average, and when ΔTmx _A has a periodic fluctuation or the like, this fluctuation is excluded and converted into a vertical fluctuation around a constant value. Note that the moving average and percentile value are set in the same manner as the procedure of the output regression method.
ΔTmx _B = ΔTmx _A −ΔTmx _Amak (5-2)
Of the ΔTmx _A calculated in the step 45 and the ΔTmx _B calculated in the step 46, the ΔTmx _B calculated in the step 46 is usually used as a load index threshold determination (FIG. 3). Used in Steps 24-26). However, for the process 46, when the fluctuation range of the calculation result of the load factor index ΔTmx _A by the expression (5-1) of the process 45 is sufficiently small or the value is sufficiently close to 0 (for example, when the fluctuation is within 1 ° C.) ) Etc. can be omitted. In this case, the value of the load factor index ΔTmx _A may be the load factor index value.

When the threshold value is determined in the aforementioned step 24 (or steps 25 and 26) using the load factor index of the exhaust gas temperature method thus calculated, the threshold value is, for example, ΔTmx _A or ΔTmx _B. Is determined to be true if the value is within the range of −P ° C. or higher and + P ° C. or lower, and false if the value is out of the range. A typical value of P is typically about 1 to 3 ° C. However, if the partial load operation time is relatively long, or depending on the accuracy of the measurement data and the type of the gas turbine cycle / combustor, it may be expanded up to about 10 ° C. In particular, when the actually measured value of the load factor index is lower than the predicted value, it is preferable to further narrow the allowable range of deviation. In such a case, the operating state is on the partial load state side due to a decrease in output or a decrease in the amount of heat generated by the fuel. If data on the partial load side is eliminated as much as possible in this way, operation data can be extracted with higher accuracy near a load factor of 100%, and deterioration can be diagnosed with higher accuracy in the end.

Further, the moving average value ΔTmx _Amak used in the step 46 may be a fixed value for each operation period sandwiched by maintenance instead. Such a fixed value for each period is often included in the setting information at the time of trial operation adjustment at the time of plant maintenance. If you can not get takes a measured value Tmx distribution _act of exhaust temperature index value of each period, it is preferable to select the highest value of the distribution density of within this. In the case of an actual machine with a long base load operation time, the standard of the value is around the cumulative distribution 50% value. When the load is reduced at night or on weekends, for example, it may be adjusted within a range of, for example, a 70% value or a maximum value of about 90%.

  As a feature of this method (exhaust temperature method), first, a temperature management index for protecting a high-temperature part of a gas turbine is used as it is, so that the accuracy is relatively high. Further, it is not necessary to model a complicated characteristic of the load factor from actual machine data unlike the output regression method described in FIG. In other words, the calculation model for the load factor based on the output regression method is that the 100% load output of the gas turbine is changed depending on conditions such as the intake air temperature, the fuel flow rate, and the refrigerant injection amount. On the other hand, in the exhaust temperature method, the relationship between the exhaust temperature and the intake air temperature usually has a control logic and a simple relational expression used for control. In step 44, these existing relational expressions can be used without building a special model independently. Therefore, it is possible to build a model in a labor-saving manner and a highly accurate model from the beginning (because it is actually used for control, it is normal that the operation data of the actual machine is controlled and operated according to its value) Can be used. This also allows a diagnostic system to be quickly constructed and operated efficiently.

  Moreover, the information used by the method shown by Formula (4-1)-Formula (5-2) is only intake air temperature and exhaust temperature, or also refrigerant | coolant injection amount, and 2-3 pieces. In this method, it is possible to determine whether or not the load factor is in the vicinity of the steady load by representing how various differences in various operating conditions affect the load factor by using these several conditions. Therefore, the extraction of the base load operation data taking into account the difference in the operation conditions can be effectively performed with a small amount of information and a calculation amount separately from the case of the output regression method.

  In the exhaust gas temperature method, the base load operation data can be extracted effectively with such a small amount of variables because the exhaust gas temperature has the following characteristics. In other words, the exhaust temperature is determined thermodynamically mainly by three factors, that is, the intake flow rate, the fuel flow rate, and the refrigerant injection flow rate. These three factors are determined in common from the result of the exhaust gas temperature control of the gas turbine. And exhaust gas temperature control is controlled so that the load factor of a gas turbine becomes near the maximum. Therefore, it can be said that the relationship between the intake air temperature and the exhaust gas temperature, which is the main input / output of the exhaust gas temperature control, is an input / output relationship that summarizes the effects of the above three factors that affect the load factor. The exhaust gas temperature method can reduce the input variables for determining the load factor to the above-mentioned 2-3 items by paying attention to this characteristic. As a result, the exhaust gas temperature method can quantify the load factor state of the gas turbine relatively stably and with high accuracy.

  The exhaust gas temperature method evaluates the load factor from the viewpoint of the control side, that is, the exhaust gas temperature management index (exhaust temperature or control index). This is because the load factor is evaluated from the viewpoint of how the plant is controlled based on the operation data that reflects these influences under the influence of operating conditions such as deterioration and seasonal conditions. Will be. At this time, the control setting value of the exhaust gas temperature index (or a function for calculating the control command value from the measured value) is generally kept constant regardless of operating conditions such as deterioration and intake air temperature. is there. For this reason, the load factor index value according to this method is not affected by fluctuations in operating conditions such as deterioration and seasonal conditions, and is evaluated as vertical fluctuations around a certain value. There is an advantage that can be.

  FIG. 6 shows another example of the calculation process flow of the load factor index value determined in step 24 (or steps 25 and 26) in FIG. Hereinafter, the calculation method based on this processing procedure is referred to as a heat input regression method. The procedure includes a heat input characteristic model identification step 47 for identifying a heat input characteristic model representing a change in the amount of heat input to the gas turbine according to the intake air temperature and the intake pressure based on the actual time series operation data, and a heat input characteristic model. A load factor calculation step 48 for calculating a load index value based on a predicted value of the heat input amount output by substituting the intake air temperature and the intake pressure of the actual machine and the actual heat input amount, and a load index calculated in the step 48 The load factor calculation process 49 which converts so that the fluctuation | variation of a value may be suppressed.

  Here, the heat input is the energy of fuel input to the gas turbine. However, when the refrigerant such as steam is injected into the gas turbine, the total heat input including the heat quantity of the refrigerant may be included. Alternatively, when the calorific value or density of the fuel is considered to be substantially constant, the flow rate of the fuel may be used.

  Details of the steps 47 to 49 will be described below.

The heat input characteristic model in the heat input characteristic model identification step 47 is a function as shown by the following expression that includes the intake air temperature Tci and the intake air pressure Pci and outputs the heat input GF.
GF = f_GF (Tci, Pci) (6-1)
Specifically, the function f_GF is preferably expressed by the following equation, for example.
_{GF = a F × (Tci-} Tci 0) + b F × (Pci-Pci 0) + α F × t + GF 0 ... (6-2)
Here, t represents the operating time, and Tci ₀ and Pci ₀ are standard condition values of the intake air temperature and the intake air pressure. The term of the operating time t can be omitted when the influence of deterioration over time is considered to be negligible in the time-series operation data period to be evaluated. It is desirable to include this term if it cannot be ignored that performance has deteriorated. By the influence of deterioration over time is separated from the section to the coefficients alpha _F, section a _e operating conditions characteristic of the _output, to b _e can avoid to include bias caused by aging, the heat input characteristic Model with high accuracy.

Further, in the formula _{(6-2), a F, b} F, α F, GF 0 indicates the model constants. The model identification in this step 47 is to determine these model constant values. a _F and b _F are coefficients for correcting the power generation output based on deviations from the standard conditions of the intake air temperature and pressure, and α _F indicates the rate of decrease in power generation output over time (hereinafter referred to as a degradation coefficient). . The deterioration coefficient α _F can be omitted when it is considered that there is almost no performance deterioration during the period of the actual machine time-series operation data used in this step 47. GF ₀ indicates an initial heat input amount (before deterioration begins during the period) when Tci ₀ and Pci ₀ are standard condition values.

Further, as an item calculated by the equation (6-2), in the case of a gas turbine that injects a refrigerant such as steam (hereinafter referred to as a refrigerant injection type, a steam injection type, etc.), the standard condition of the refrigerant injection amount Gs is used. The term C _F × (Gs−Gs ₀ ) to be corrected in accordance with the deviation from the value Gs ₀ may be added. Here, C _F is a model constant. In this way, when the change in the flow rate or temperature of the refrigerant particularly injected in the refrigerant injection type gas turbine is large, the load factor can be quantified with high accuracy in consideration of the influence.

As a method for determining these model constants a _F , b _F , α _F , and GF ₀ , the power generation output calculated by substituting the operation conditions included in the time series operation data of the actual machine into the right side of the equation (6-2) It is preferable to determine such that the error between the time-series calculation result and the time-series data of the power generation output included in the actual machine time-series operation data is minimized throughout the data period. This can be obtained by a known method such as a least square method. An example of such a method will be described later with reference to FIG. Alternatively, when there are design values, plan specification values, and experience values (hereinafter collectively referred to as a priori values) of model constants a _F , b _F , α _F , and GF ₀ , these may be used. This method may be more accurate than inaccurate model regression when there are few actual machine data shortly after the start of operation, or when there are many partial load operations and the number of base load data cannot be secured sufficiently. Many and effective. In addition, such an a priori value is set in the initial stage after the start of operation with a small number of data, and after that, when a certain amount of operation data is accumulated, it is updated with a value that is regressed from the data. It is also effective to perform a weighted weighted average according to the number and gradually shift to the regression value from the actual machine data.

In the step 47, actual machine data collected by variously changing operating conditions such as intake air temperature may be used instead of the actual machine time-series operation data. In this case, if the data collection period is short, the deterioration coefficient α _F can be omitted as described above.

Calculation of the value of the load factor index in the step 48 (load factor calculation step A) is a heat input characteristic model in which the model constant is determined in the heat input characteristic model identification step 47 as shown in equation (7-1). (Equation (6-1) or Equation (6-2)) is input the time series data of actual machine operating conditions (in the above example, measured values of intake air temperature and pressure) during the evaluation target period, This is to calculate the ratio of the heat input GF _act calculated from the measured value of the actual machine to the predicted value GF _calc (referred to as load factor Load _A ).
Load _A = GF _act / GF _calc (7-1)
The calculation of the value of the load factor index in the step 48 (load factor calculation step B) includes the value of the load factor Load _A at a certain time as shown in the equation (7-2). The ratio (load factor Load _B ) divided by the moving average value or the specific percentile value Load _Amav during a certain period is calculated. The moving average value here refers to a simple moving average value or a weighted moving average value for a certain period including that time, as described in the load factor calculation step 43 of FIG. The specific percentile value indicates a cumulative percentage value in the value distribution during the period. By taking the ratio to the moving average value in this way, the error fluctuation range of the load factor index is made sufficiently small, excluding the effects of deterioration of plant performance over time and seasonal performance fluctuations, with a constant value at the center. Can be converted to vertical fluctuations.
Load _B = Load _A / Load _Amav (7-2)
In this case, the moving average period can be determined in the same manner as described in the load factor calculation step 43 in FIG. 4, for example, when processing is performed on the basis of data at one hour intervals, up to the target time. Taking the data period of one week immediately before or about 100-200 points is preferable because the influence of various operating condition fluctuations and deterioration over time can be taken into account.

In this way, among Load _A calculated in Step 48 and Load _B calculated in Step 49, Load _B calculated in Step 49 is usually used as a load index threshold determination (Step 24 in FIG. 3). To 26). However, for step 49, when the fluctuation range of the calculation result of the load factor index GF according to formula (6-1) or formula (6-2) in step 48 is sufficiently small or the value is sufficiently close to 1 (for example, 0) .95 or more) can be omitted. In this case, the value of the load factor index Load _A may be set as the load factor index value.

  The heat input regression method for calculating the load factor index in this way can efficiently evaluate the load state of the gas turbine based on relatively few data items as in the output regression method. The load regression evaluation by the output regression method was an evaluation from the viewpoint of the output side (operation result side) as the power generation output, whereas the heat input regression method is the input side (so-called heat input to the plant). The evaluation is based on the input energy from the viewpoint of the operation command side. For this reason, the load factor evaluation by the heat input regression method has a feature that the load factor can be evaluated from the viewpoint of the transition of the input energy state to the plant. In the event of an abnormality in the gas turbine, the operation command and the actual operation state may differ, so that it is possible to utilize the viewpoint of the input heat input method separately from the output regression method on the output side in this way. Useful.

  FIG. 17 shows another example of the calculation processing flow inside the base load data extraction means 131 in FIG. Hereinafter, the calculation method based on this processing procedure is referred to as a coexistence method. The procedure is an iterative process for each time of the time-series data to be evaluated, and the iterative process has at least one index reflecting the load state of the gas turbine for each time data, that is, At least one of the power generation output, power generation efficiency, pressure ratio, fuel flow rate, exhaust gas temperature, intake air flow rate, or load factor index calculated by the method shown in FIGS. For each of the above indices (load factor indices), an index-specific threshold value determination process 24-26 for determining whether or not each index is within a predetermined upper and lower limit range, and the determination processes 24-26 Based on the threshold value determination result of each index executed in step 1, the determination result of all these indices is true or not (false even if one is a false index) AND determination step 152 in FIG. If the determination result of the step 152 is true, the selection step 153 for selecting the operation data at that time, and the above steps 24-26, 152, 153 (if the determination of the step 152 is false, 153 is none) and the completion determination step 154 for determining whether or not the processing of the time series data to be evaluated has been executed for all times, and the result of the completion determination is false (FALSE, incomplete) Step 155 proceeds to data at the next time.

  If the determination result of the completion determination step 154 is false (FALSE, incomplete) and the process proceeds to data at the next time of the time series data in step 155, the process 152 is true through the steps 24-26 to 152. In this case, the process from the selection step 153 to the completion determination step 154 is executed again. This series of processing is repeated until the determination in the completion determination step 154 becomes true (TRUE, completed for all times of time-series data).

  Thus, when the determination in the completion determination step 154 becomes true (TRUE), the operation data selected in the selection step 153 of the above repeated processing becomes the extracted data based on the base load.

  This method has the following effects.

  (A) Since multiple indicators are used as in steps 24 to 26 and only data for which the determination result of all indicators is true is selected as base load data, the accuracy of base load data extraction increases, and this continues. The performance characteristic model of the modeling unit 132 (FIG. 1) becomes highly accurate, and the performance degradation accuracy and recovery evaluation accuracy in the final performance evaluation unit 133 (FIG. 1) become high.

  (A) Base load data extraction accuracy equivalent to or better than the most accurate method among the output regression method, exhaust temperature method, and heat input regression method can be obtained. This is a coexistence method that takes AND conditions (logical product of whether all the conditions are satisfied) of multiple indicators such as output regression method, exhaust temperature method, heat input regression method, etc. This is because the ranges extracted by each index are overlapped and the extracted ranges are narrowed down to the overlapping portions (a specific example is shown later in FIG. 11).

  (C) Preliminary information on which method of the output regression method, exhaust temperature method, and heat input regression method is suitable, and the effort to confirm this are not required, so high-precision results can be obtained quickly and efficiently. It is done. For example, among the output regression method, exhaust temperature method, and heat input regression method, which method of base load extraction accuracy is high depends on the accuracy of each measurement item on the site and the model type of the gas turbine. If you do not know in advance which base load extraction method is suitable in this way, you usually perform all the output regression, exhaust temperature, and heat input regression methods, compare the accuracy, and then It is necessary to select a method. However, in the compatibility method, these multiple methods are implemented together and the results are automatically incorporated to obtain an extraction accuracy equivalent to or higher than the most accurate method among these methods as described above. It is because it can do.

  Hereinafter, an example of a processing procedure in the above-described modeling unit 132 in FIG. 1 will be described. As described above, the modeling unit 132 determines the model constant of the performance characteristic model of the gas turbine based on the base load operation data extracted by the base load data extraction unit 131.

  At this time, if the target gas turbine is used for the energy service business or is an aged machine, the owner of the equipment and the operator of the operation and maintenance have the performance model and model constants of the target machine. It is common to be unknown. Model constants, in particular, are manufacturer's design information and are generally not disclosed. Even if they are disclosed, performance characteristics may differ from the original design due to repeated maintenance and parts replacement on aged machines. There are many.

  In such a case, the performance characteristic model needs to be constructed from the operation data of the actual machine. For this purpose, it is necessary to collect data so that the range of the intake air temperature, that is, the atmospheric temperature, which is a typical operation condition, includes the entire operation range. For example, in order to include the range of annual maximum and minimum temperatures, it is desirable that a period of about half a year is required as a period between at least the maximum temperature and the minimum temperature. In the case of a continuous operation facility, this corresponds to an operation time of about 4000 to 8000 hours.

  However, the performance of the gas turbine deteriorates even during such a period. For example, it has been reported as a typical example that the power generation output decreases by about 3% in 5000 hours after the start of operation (Harry G. Stall, Creating Owner's Competitive Advantage through Structural Services, GE Power Systems, ER 4208). , (2001)). The problem with this is that the influence of deterioration is included in the data collected to construct a model in a normal state before the equipment deteriorates. Therefore, there is a need for a method capable of identifying a performance characteristic model before deterioration even from actual machine operation data including the influence of deterioration over time.

  In order to cope with this, after modeling the performance characteristics of actual machine data taking into account the effects of deterioration, a method for identifying the initial performance characteristic model before the progress of deterioration is excluded by excluding the calculation of deterioration from this model. It is shown below.

  FIG. 7 is a schematic diagram of a performance characteristic model used for identification of an initial performance characteristic model. In this performance characteristic model, an operating condition 64 such as atmospheric conditions and a model constant 62 specific to the target machine determined by design conditions are input to the performance characteristic model 61, and a performance calculation result (performance calculation value 65) is output. At this time, a model constant for reflecting the progress of deterioration in the performance calculation (hereinafter referred to as the deterioration coefficient 63 in particular) is added to the input. First, the operating condition data included in the time series data of the base load operation of the actual machine extracted by the base load data extracting means 131 (FIG. 1) is input to the operating condition 64. The known performance parameter determination method such as the least square method is used so that the output performance calculation value 65 is within the predetermined allowable range with respect to the performance value included in the time series data of the same base load operation. The model constant 62 and the deterioration coefficient 63 are identified. The degradation coefficient 63 and the performance calculation portion using the degradation coefficient 63 are removed from the performance characteristic model 61 thus identified, and this is used as the initial performance characteristic model before performance degradation.

Specific examples are shown below. Expressions (8-1), (8-2), and (8-3) are examples of modeling methods provided by the present invention (hereinafter, these are collectively referred to as Expression (8)). Expression (8-1) calculates the performance Y by adding the performance correction amount ΔY due to the deviation from the standard condition to the performance Y ₀ when the operating condition is the standard condition. At this time, the equation (8-2) calculates the correction amount ΔY as a function f_opr (hereinafter referred to as an operation condition correction function) of at least one operation condition x _i (i = 1, 2,...). To do. The above-described model constant 62 in FIG. 7 is included in the coefficients and constant terms used in the operating condition correction function f_opr. Further, the equation (8-3) calculates the performance Y ₀ under the standard conditions by a function f_dgr (hereinafter referred to as a time-decreasing function) so as to change according to the progress of deterioration. The time-decreasing function f_dgr includes the operating time t as a variable. In addition to this, the initial performance value (a value under standard conditions and a constant) before the deterioration occurs, and the progress of the performance deterioration due to the elapse of the operating time t. Includes at least one constant coefficient (deterioration coefficient 63 in FIG. 7) as a model constant.

Here, the performance Y is, for example, power generation output, power generation efficiency, compressor pressure ratio, fuel flow rate, and the like. Y is not limited to these, and any operation data whose value changes when performance deteriorates is a target. The operating conditions indicate the environmental conditions such as the atmospheric temperature and pressure described above, and the operating conditions such as the fuel flow rate. The standard conditions are, for example, the standard atmospheric conditions determined by ISO and the operating conditions corresponding to the atmospheric conditions. Indicates the value of.
Y = ΔY + Y ₀ (8-1)
ΔY = f_opr (x _i ) (8-2)
Y ₀ = f_dgr (t) (8-3)
The characteristic of these equations (8) is that the influence of the change in performance characteristics due to deterioration is a term with time deterioration indicating the performance under standard conditions (second term on the right side of equation (8-1), equation (8-3)). It is to have been consolidated. Strictly speaking, the change in the performance characteristics due to the deterioration extends to the operating condition correction function f_opr (formula (8-2)) of the correction term ΔY (the first term on the right side of the formula (8-1)). However, the inventors have found that analysis of the performance characteristics of the gas turbine, the effect of the correction term [Delta] Y, as compared with the effect on the time course of standard conditions in terms performance Y _0, was found to be relatively small. Therefore, the deterioration over time of the performance is modeled as a time-decreasing function f_dgr that changes the value according to the deterioration with respect to the performance value Y ₀ under the standard conditions as described above. A modeling method has been devised in which the operating condition correction function f_opr for correcting is constant regardless of deterioration. This method models the complicated effects of gas turbine performance degradation, focusing only on the aging of the output value under standard conditions, as described above, and omits the effect on performance correction due to differences in operating conditions. Yes. With this simplification, the present method can separate and evaluate the time-dependent degradation of gas turbine performance and the complex characteristics of performance changes due to operating conditions in a labor-saving and efficient manner.

A specific example of the model formula of the performance characteristic model 61 of the formula (8) is as the following formula (9-1). In this equation, x _i0 represents a value under a standard condition of at least one operating condition x _i (i = 1, 2,..., N). The symbol Σ indicates that the term a _i × (x _i −x _i0 ) is added to i = 1, 2,... N (the description of Σ is the same hereinafter). The coefficient a _i (i = 1, 2,...) Is a correction coefficient for the performance Y according to the deviation of the i-th operating condition x _{i from} the standard condition x _i0 . α is a deterioration coefficient of the performance Y according to the operating time. Constant term Y ₀₀ represents an initial value prior to the Performance values Y ₀ at standard conditions, the performance is degraded. In the present invention, the initial value before performance deterioration refers to the initial value during the target period.
Y = Σa _i × (x _i −x _i0 ) + (α × t + Y ₀₀ ) (9-1)
_{Y = Σa i × (x i} -x i0) + Y 00 ... (9-2)
In the equation (9-1), the coefficient a _i (i = 1, 2,...) And the constant term Y ₀₀ correspond to the model constant 62 in FIG. As a method for determining the values of a _i , α, and Y ₀₀ , calculation is performed by the known least square method of multiple regression analysis using the base load actual machine time-series operation data extracted by the base load data extraction unit 132. Good.

  When the model constant is determined in this way in equation (9-1), a model equation indicating performance characteristics before the progress of deterioration can be obtained by removing the term of deterioration with time as shown in equation (9-2). In this method, the performance deterioration due to the change in the operating condition of the gas turbine and the change in the performance due to the deterioration are separated by separating the term of the deterioration of the performance with time and the term of the change in the operating condition in a very simple form shown in the equation (9-1). It can be easily simulated.

Table 1 shows a preferred example of the correspondence variables x _i of performance metrics Y and operating conditions for modeling the performance characteristics by the formula (9-1). In Table 1, double circles are particularly desirable as operating conditions for modeling target performance indicators, and single circles should be additionally used when the accuracy of the model is insufficient. It can be done.

表１に示すように、性能指標Ｙのうち、発電出力、発電効率、燃料流量は次式のように表すことで十分高精度にモデル化できる。圧縮機圧力比は、変数ｘ_ｉに吸気圧力を使わずに吸気温度のみの関数とすることでよいが、吸気温度と湿度の関数にするとさらに特性モデルを高精度化できる。
Ｙ ＝ ａ_１×（Ｔｃｉ−Ｔｃｉ_０） ＋ ａ_２×（Ｐｃｉ−Ｐｃｉ_０） ＋ α×ｔ ＋ Ｙ_００ …（９−３
なお、表１の一重丸印の変数の使い方としては、上述のようにモデル化の式に組み込むのではなく、前述のベースロードデータ抽出手段１３１のデータ抽出条件として組み合わせてもよい。すなわち、前述したベースロードデータ抽出手段１３１においてデータを抽出するかどうか判定する工程２４〜２６（図３）の判定は、これら一重丸印の変数の値についての上下限許容範囲の条件がさらに付け加えられたものであってもよい。このようにすると、性能指標の特性モデルに影響は及ぼすものの、支配的に影響はしないこれらの変数の変動による影響を、特性モデルを構築前のデータ抽出の段階（ベースロードデータ抽出手段１３１）で予め除外することができる。したがって、本モデル化手段１３２で性能特性をモデル化する際に、主要な影響を及ぼす運転条件に絞って、高精度にモデル化できる。

As shown in Table 1, among the performance index Y, the power generation output, the power generation efficiency, and the fuel flow rate can be modeled with sufficiently high accuracy by being expressed as follows. Compressor pressure ratio may be to a function of only the intake air temperature without the intake pressure to the variable x _i, can accurate the more characteristic model when the function of intake air temperature and humidity.
_{Y = a 1 × (Tci-} Tci 0) + a 2 × (Pci-Pci 0) + α × t + Y 00 ... (9-3
The usage of the single circle variable in Table 1 may be combined as a data extraction condition of the base load data extraction unit 131 described above, instead of being incorporated in the modeling equation as described above. That is, the determination in the steps 24 to 26 (FIG. 3) for determining whether or not to extract data in the base load data extraction means 131 described above further adds the conditions of the upper and lower limit allowable ranges for the values of these single circle variables. It may be what was made. In this way, the influence of the fluctuation of these variables, which have an influence on the characteristic model of the performance index but do not have a dominant influence, is detected at the data extraction stage (base load data extraction means 131) before the characteristic model is constructed. It can be excluded in advance. Therefore, when the performance characteristics are modeled by the modeling unit 132, it is possible to model with high accuracy by focusing on the operating conditions that have the main influence.

  When the performance characteristic model formula is not linear as in formulas (9-1) to (9-3), the model constant may be determined by the procedure shown in FIG. That is, initial values of the model constant 62 (FIG. 7) and the deterioration coefficient 63 (FIG. 7) are set in the model constant and deterioration coefficient initial value setting step 71, and then the initial values are set. A performance calculation value 65 (FIG. 7) calculated by inputting the operating condition 64 (FIG. 7) into the performance characteristic model 61 (FIG. 7) is output in the performance calculation step 72, and the performance calculation value 65 is an actually measured value. In the error determination step 73, whether the error is within a predetermined error range determined in advance is true / false. If false, the model constant 62 and the deterioration coefficient 63 are corrected in a set value correction step 74 to calculate the performance. Step 72 may be executed again, and this may be repeated until the determination result of error determination step 73 falls within the allowable error range. The values of the model constant 62 and the deterioration coefficient 63 when the end condition of the error determination step 73 is satisfied are values finally determined. In such a case, it is preferable to use an a priori value such as a design value, a planned specification value, and an experience value as the initial value of the step 71 as described above. This not only speeds up the convergence of the iterative calculation, but also reduces the risk of searching for a solution with an impossible value that may occur in the mechanical convergence calculation.

  The processing contents in the performance evaluation unit 133 of FIG. 1 will be described with reference to FIG. Here, a case where performance time-series data is converted into values under standard conditions will be described as an example.

  As shown in FIG. 9, the model that performs this (hereinafter referred to as the standard condition conversion model 82) receives the input of the model constant 62 and the operation data 81, and calculates the performance value of the actual machine that fluctuates depending on the operation conditions. A converted value (hereinafter referred to as standardized data 83 or standard condition converted value) when the ISO standard condition is corrected is output.

  As the model constant 62, a value (FIGS. 7 and 8) determined by the modeling unit 132 described above can be used. The operation data 81 is data composed of actual machine operating conditions 64 (FIG. 7) and actual measured values of actual machine performance Y at this time.

The model equation of the standard conditions in terms model 82 becomes as shown in Equation (10-1), the normalized data _{Y 0,} the operating conditions described in the equation (8-2) on the performance value Y and the previous correction function f_opr _{(X i} ) To calculate. Specifically, for example, if the form corresponding to the above-described expression (9-1) is as shown in the expression (10-2), the form corresponding to the expression (9-3) is as shown in the expression (10-3). Good. In these cases, as the model constant a _i , the one previously obtained by the equation (9-1) can be used. In this way, the standard condition conversion can be efficiently performed using the model constant 12 determined by the modeling means 132 in the previous stage without creating a new standard condition conversion model from the beginning.
_{Y 0 = Y - f_opr (X} i) ... (10-1)
_{Y 0 = Y - Σa i ×} (x i -x i0) ... (10-2)
_{Y 0 = Y - a 1 ×} (Tci-Tci 0) - a 2 × (Pci-Pci 0) ... (10-3)
Note that the period of the actual machine time series data used for determining the model constant in the modeling unit 132 described above may be different from the period of the actual machine time series data in which the actual machine data is converted into the standard conditions by the performance evaluation unit 133. . For example, the modeling unit 132 determines a model constant of the performance characteristic model in a part of the actual time series data, and the performance evaluation unit 133 uses this model constant to evaluate the tendency of performance recovery due to performance degradation or maintenance. It can be applied to the whole. In this way, the model constant can be determined efficiently with a small amount of data without using all the data in the target period for degradation evaluation. This is particularly essential when evaluating a period that includes when performance is restored by maintenance. For example, when maintenance is performed at a facility whose performance has deteriorated over time, and performance has recovered significantly at once, performance characteristics can be determined based on actual machine operation data including performance discontinuities. Since it will be modeled, the accuracy of a model will worsen. In such a case, for example, when the model constant of the performance characteristic model is determined in the period before (or after) the maintenance is performed (modeling unit 132), and the performance is converted into the standard condition (performance evaluation unit 133). By calculating the standardized data of performance using the same value without changing the value of this model constant before and after maintenance, the performance recovery effect of maintenance can be quantified with high accuracy.

  Hereinafter, with reference to FIGS. 10 to 14, performance degradation evaluation is performed by the output regression method, the exhaust temperature method, the compatibility method, and the MT method for the actual operation data shown in FIG. To state.

  FIG. 10 shows model prediction accuracy and data extraction when performance degradation evaluation is performed by the four methods of the output regression method, the exhaust gas temperature method, the compatibility method, and the MT method for the operation data shown in FIG. It is the example which compared the relationship of the rate. Here, the MT method is the method described with reference to FIGS. 2 and 3, and the calculation of the statistical distance is based on the formula (1) of the Mahalanobis-Taguchi method (MT method) described above.

  Here, the threshold value of each base load extraction method was set as follows. In the output regression method and the exhaust temperature method, a threshold value is set so that the accuracy is equal when the extraction rate is in the range of about 60-70%. In the coexistence method, both threshold values are used as they are, and the data that is true in both determinations is determined as the base load data. In the MT method, the selection data of the compatibility method is set as a unit space, and the Mahalanobis distance of the target data with respect to the distribution of the unit space data is determined by the threshold value 3. Note that the unit space of the MT method is sequentially updated according to the progress of time with the past week as a reference period.

  The results show that the compatibility method has the same accuracy as the exhaust temperature method, which is the more accurate of the output regression method and the exhaust temperature method. In this actual machine data example, the exhaust gas temperature method is more accurate than the output regression method, but which of the two methods is higher depends on the accuracy of the measurement items on the site and the model type. In such a situation, the compatibility method can automatically obtain an accuracy equal to or higher than that of the higher accuracy method without checking and comparing the accuracy of both of these methods individually. This is because, as described above with reference to FIG. 17, the determination of the extraction range by the compatibility method is narrowing down to the overlapping range of the extraction ranges of the exhaust temperature method and the output regression method.

  In addition, the MT method has the highest extraction rate while maintaining the same or higher accuracy as the other three methods. In particular, in comparison with the coexistence method, the extraction rate of the coexistence method is 51%, whereas the MT method extracts 73%, which is about 20% more data, and the accuracy is equivalent. Compared with the exhaust temperature method, the MT method has the same accuracy and the extraction rate is 4% higher. As described above, in the MT method, more data can be extracted with the same accuracy than when the output regression method and the exhaust gas temperature method are performed alone or simply combined as in the compatible method.

FIG. 11 shows how base load data is extracted by each method in the example of FIG. 10 described above. In the figure, the load factor Load _B , which is a load factor indicator of the output regression method, and the load factor indicator ΔTmx _B of the exhaust temperature method are plotted on the horizontal axis and the vertical axis, respectively, and the extraction range by the MT method is compared with the other three methods. In contrast.

  In the output regression method, the threshold value is 100 ± 2%, and data in a region sandwiched between two solid lines (vertical lines) of these threshold values shown in the figure is determined as base load data.

  In the exhaust temperature method, the threshold value is 0 ± 1 ° C., and data in a region sandwiched between two solid lines (horizontal lines) of these threshold values shown in the figure is determined as base load data.

  In the compatibility method, data in a region surrounded by a rectangle that satisfies both the output regression method and the exhaust temperature method is determined as base load data. In the compatibility method, extraction is performed by narrowing down to the overlapping area of the extraction range of the output regression method and exhaust temperature method in this way, so regardless of which accuracy is higher in the original output regression method or exhaust temperature method Even if there is no such prior information, an accuracy equivalent to or higher than that of the more accurate method can be realized.

  In the MT method, data in the compatible method region is defined as a unit space, and a range that extends from here to the periphery of the distribution is determined as base load data. From the figure, it can be seen that the MT method reflects the correlation between the output regression method and the exhaust gas temperature method, and is equally and equally extracted from the center of the distribution toward the direction of lower density according to the distribution status. As a result, the MT method can extract more data while maintaining the same accuracy as the other three methods (extraction rate is high).

  In this example, the output regression method is less accurate than the exhaust temperature method, but the exhaust temperature method may be less accurate than the output regression method. In this way, the quality of the single method depends on the model of the actual machine and the measurement accuracy of each item, but it is the compatible method and the MT method that can obtain highly accurate results regardless of this. The reason why the accuracy of the coexistence method is high is as described above. On the other hand, the MT method can reinforce the less accurate method among the individual methods (output regression method and exhaust temperature method), and can obtain better results than the case of each method alone. . That is, it is possible to stably obtain the result with the highest accuracy and the highest extraction rate by taking advantage of the goodness of both types. This is due to the characteristics of the method using the MT method in which extraction is performed from a location close to the statistical distance in consideration of the correlation of distributions for a plurality of load factor indexes.

  FIG. 12 compares the change in extraction rate and model prediction accuracy when the threshold value for base load determination of each method is changed with respect to FIG. 11 (prediction accuracy is indicated by an average error). ing). In this figure, the threshold is a parameter, so it is not displayed directly on the graph. The graph shows the relationship between extraction rate and model prediction accuracy.

Before describing the results, the standardization of threshold values for comparing each method and the setting of each method are described below. The unit of threshold value differs between the output regression method and the exhaust temperature method. For this reason, for each load factor index value, the peripheral variation of the average value was normalized as a constant multiple of the standard deviation σ. That is, when changing the threshold value, n normalized by the standard deviation was changed.
Load factor 100 ± nσ [%]
Exhaust temperature control index value deviation 0 ± nσ [%]
The threshold value of the compatibility method depends on the threshold values of both the output method and the exhaust temperature method. As a method of changing these simultaneously, the threshold value of each method was set by changing n normalized by the standard deviation as described above. In the MT method, data determined as a base load by the compatibility method was used for the unit space. The threshold values of the output regression method and exhaust temperature method, which are the basis of the MT method, are set at points where the extraction rate is about 68%, and the threshold value of the MT method is the Mahalanobis distance calculated by the equation (1) The threshold was judged.

  The results show that the MT method has a lower average error rate than the other three methods in any extraction rate range. In particular, the error rate is consistently lower than that of the other three methods even in the region where the extraction rate is 80% or more where the error rate starts to increase rapidly as the extraction rate increases. Thus, the MT method can stably obtain high accuracy regardless of the range of the extraction rate.

  As described above, this method, which applies the MT method proposed in the present invention, takes into account the correlation between a plurality of indices, that is, the temporary load factor and the exhaust temperature management index deviation, and is statistically calculated from the center of gravity of these distributions. This is because the threshold value is determined so that the degree of variation is uniform. Since the threshold value is set in this way, even when the threshold value is changed, the MT method preferentially selects a value closer to the center of the distribution along the correlation of the load factor index. become. As a result, even if the extraction rate is comparable, it is more suitable from the viewpoint of multiple indicators than a single method such as the output regression method or exhaust temperature method, or a combined method that simply combines these methods. It can be said that the data considered to be selected.

  FIG. 13 individually shows changes in the extraction rate and the model prediction accuracy corresponding to the changes in the actual threshold value of each method with respect to FIG. 11 described above.

  The results show that in the three methods of the output regression method, the exhaust temperature method, and the coexistence method, the extraction rate changes to about 80 to 100% in the range of the normalized threshold value of 1 to 10, and 95 at the same time. The% error rate has rapidly increased from about 1% to 11%. Therefore, in these three methods, the sensitivity of the influence of the threshold setting on the model prediction accuracy is large, and care must be taken in setting the threshold.

  On the other hand, in the MT method, even if the threshold value is changed from 1 to 10, the 95% error rate remains approximately 1%, and there is no significant change. The rate increases to about 3%, but the change is gradual.

  This is because, in the MT method, the threshold value is converted from the original scale of the load factor index into the Mahalanobis distance, which is a statistical variation scale. In this conversion, the statistical distance is close in the vicinity where the values are dense, and the distance increases as the density decreases. As a result, in the method using the original value scale such as the output regression method, exhaust temperature method, and compatibility method, the transition from the dense part of the distribution to the thin part changes rapidly in a short interval. On the other hand, in the MT method, since the threshold value changes according to the change in the degree of statistical variation, the slope of the extraction rate with respect to the change in threshold value is gentler than in the other three methods. As described above, the MT method uses the statistical distance as a scale, so that the section where the extraction rate changes greatly can be expanded and grasped gently. Thereby, the influence on the accuracy by setting the threshold value is moderately suppressed rather than abruptly.

  As described above, the MT method has a low sensitivity to the influence on accuracy due to the threshold setting. Therefore, the threshold value can be set stably in the operation of base load extraction.

  FIG. 14 shows the results of evaluating the deterioration tendency using the four methods for the operation data shown in FIG.

  Here, unlike the above-described FIGS. 10 and 11, the threshold values of the respective methods are set as described below.

  That is, in FIGS. 10 and 11, the threshold values of the output regression method and the exhaust temperature method are used as they are for the threshold value of the compatible method and the unit space of the MT method. However, considering the relationship between the extraction rate and accuracy of each method as shown in FIG. 12, it is necessary to unify the criteria for setting the most appropriate threshold value for each method. For this reason, looking at the trade-off trend, in the example of FIG. 12, the accuracy is almost constant when the extraction rate is about 0% to about 80%, but when it exceeds about 80%, the accuracy suddenly increases. It turns out that it falls. In this example, in the region where the extraction rate is about 0 to 80%, the operation data selection range is the base load operation region, and the selection range is expanded from the center of the distribution toward the peripheral direction. However, on the other hand, when the extraction rate exceeds about 80%, it indicates that the data of the partial load area is included.

  Therefore, it is desirable to avoid an area where partial load data is included and accuracy is lowered while taking an extraction rate as large as possible as a threshold selection criterion. On such a graph as shown in FIG. 12, the accuracy is almost constant and the error is kept at a low level, and the level of the output is increased while the extraction rate is increased and the error starts to increase from a certain point. This is the point just before. Such a point is hereinafter referred to as an error rate inflection point. Here, the average error rate is shown as a result of evaluating the degree of deterioration by setting a threshold value at the inflection point of each method. In addition, the vertical line in a figure has shown the period when maintenance, such as replacement | exchange of an engine or components, was implemented.

  The results show that, with either method, the performance declines over time during the operation period between maintenance, but the tendency of performance recovery every time maintenance is performed can be clarified. . In particular, (2) and (5) in the figure are replacement maintenance of the entire core engine including the compressor, the combustor, and the turbine. In such overall replacement, the turbine blade of (1) in the figure is used. It can also be confirmed that the recovery range of performance is large compared to partial maintenance such as replacement, fuel control valve replacement in (3), and thermocouple replacement in (4). (Note that (6) in the figure is not a maintenance, but only an inspection, so there is no performance fluctuation. Also, since (6) and after, a special load distribution operation was performed on the actual machine, the performance has changed). In addition, since the vertical fluctuation width of the standardized data from time to time is uniform, it can be said that the performance deterioration with time can be identified in the same manner as the example of performance recovery at the time of maintenance. In the time-series transition of the original power generation output measurement value shown above (FIG. 15), there is always a large vertical fluctuation range due to the annual fluctuation of the intake air temperature, the fluctuation of the load factor during operation, and the daily fluctuation of the intake air temperature. The recovery effect due to maintenance could not be determined at all.However, by extracting the base load data using the method of the present invention, modeling it, and converting it to standard conditions, the performance transitions according to the operation and maintenance situation You can see clearly.

  Furthermore, when the results of these methods are compared in detail, the accuracy of the output regression method is relatively low in this example, and values that deviate from the vicinity of the center of the distribution are seen in comparison with the exhaust temperature method. On the other hand, in the coexistence method, the value out of such a distribution is reduced, and in the MT method, it is further reduced.

  This is because, as described above with reference to FIG. 11, in the coexistence method, extraction is performed by narrowing down to the overlapping region of the extraction range of the output regression method and the exhaust temperature method, and the non-overlapping region is excluded. This is because outliers from these two-dimensional distributions are excluded in consideration of the correlation between the output regression method and the exhaust gas temperature method. In this example, for the data selected once by the output regression method, false data (outside the predetermined upper and lower limits) is excluded by the threshold determination of the exhaust gas temperature index by the compatible method, and by the MT method the exhaust gas is exhausted. It can be said that such a result that the deviation value is smaller is obtained by excluding the data having the deviation value in view of the correlation with the temperature index from the periphery of the distribution.

  FIG. 16 shows an example of a display screen that is preferably provided in the gas turbine performance diagnosis system of the present invention.

  The display screen 140 is an example of one screen of the display device showing the output means 103 (FIG. 1). The display screen 140 includes an original data graph 41 that displays a time series change of operation data (the original data) during the diagnosis target period for the performance index, buttons or display menus arranged on the display screen, and the like. An object (hereinafter referred to as an extraction execution button 142) whose base load operation data extraction means starts extracting base load operation data from the original data when clicked or selected on the screen. , Instead of the original data graph and the extracted data graph, the extracted data graph 143 displaying the time-series change of the base load operation data (hereinafter referred to as extracted data) extracted by the base load data extracting means 131 Change the plot symbol and color so that the original data can be distinguished from the extracted data. Graph displayed over the rough (hereinafter, extracted superimposed graph) includes a or a further graph that displays the time-series change of the normalized data outputted from the standard conditions conversion means (normalized data graph 144).

  Although the number of extraction execution buttons 142 is one in the figure, the number is not limited to this, and a plurality of extraction load buttons 142 corresponding to various base load operation data extraction methods such as the output regression method, the exhaust gas temperature method, and the compatibility method described above are applicable. It is also possible to arrange the buttons so that the extraction method can be selected. At this time, a plurality of buttons are allowed to be selected at the same time, and a screen for inputting and setting each threshold value according to a plurality of selected extraction methods is also displayed. Shall be. Also, it is assumed that an option button is provided that allows the user to specify whether or not to combine multiple AND conditions into a single result for a plurality of base load extraction results.

  The extraction of the base load operation data is extremely important because it greatly affects the accuracy of the performance diagnosis, and this is as described in detail in the example in the prior invention. Since the display screen 140 includes the extraction execution button 142 and the original data graph 141, the extracted data graph 143, or the extracted overlapping graph as described above, how to extract base load operation data important for diagnosis is described. When the user wants to check whether it has been done, it can efficiently display the result and confirm the result or try a new condition.

  Further, when a standardized data graph 144 converted into standard conditions is also provided, the user can see the base load extraction status in comparison with the performance transition status shown in the standard condition conversion graph. This helps the user comprehensively determine the performance degradation status and the base load extraction status that affects the performance, and perform the diagnostic work intuitively and with high accuracy.

  If a plurality of extraction execution buttons 142 are provided according to the extraction method as described above, the user of this system determines that it is not preferable, for example, by observing the extraction status of the base load data on the screen. In this case, the extraction method and combination can be changed. Since the extraction result and the degradation diagnosis result according to this change are immediately reflected in the extracted data graph 143 and the standardized data graph 144, the user can easily try various extraction methods and their combinations on the screen and display them. The results (extracted graph, standardized graph) can be compared. As described above, the present system can interactively display the base load extraction status corresponding to the extraction method according to the user input, and therefore can efficiently assist the user in selecting the most accurate extraction method.

  In the method for diagnosing performance degradation of the gas turbine described in all the above embodiments, the accuracy of diagnosis can be improved by further performing the following. That is, in the processing by the base load data extracting unit 131, it is preferable to perform the above extraction after narrowing down to data in which the operating condition is within a certain range in advance (under the AND condition). Such operating conditions include, for example, the injection flow rate of refrigerant such as steam or water sprayed on the combustor to reduce the nitrogen oxide concentration in the exhaust gas or increase the power generation output, or the temperature, pressure, and heat amount thereof, Or the conditions regarding the input of the process of a gas turbine, such as temperature, flow volume, and pressure conditions of a reheat cycle, correspond.

  Similarly, in the calculation of the output regression method in the base load data extraction unit 131 and the determination of the performance characteristic model in the modeling unit 132, it is also effective to divide the modeling into a plurality of intake air temperature condition ranges. Is the method. The output of gas turbines improves when the intake air temperature decreases, but in practice, there are many operations that suppress the output in the region where the intake air temperature is low in accordance with the capacity of the generator. Performance characteristics may change in the intake air temperature range. This method can flexibly cope with such characteristic changes.

The block diagram of an example of the diagnostic system using the performance degradation diagnostic apparatus by this invention. The flowchart which shows an example of the selection method of the data of the reference period for calculation of statistical distance in the base load driving | operation data extraction method of this invention. The flowchart which shows an example of the calculation method of the statistical distance with the data of a reference period in the base load driving | operation data extraction method of this invention. The flowchart of the output regression method which shows an example of the extraction method of the base load driving | operation data in the performance degradation diagnosis method of this invention. The flowchart of the exhaust temperature method which shows an example of the extraction method of the base load operation data in the performance degradation diagnostic method of this invention. The flow chart of the heat input regression method which shows an example of the extraction method of the base load driving | operation data in the performance degradation diagnosis method of this invention. The figure which shows an example of the input / output of a performance characteristic model in the performance degradation diagnostic method of this invention. The flowchart which shows an example of the identification procedure of the model constant of a performance characteristic model in the performance degradation diagnostic method of this invention. The figure which shows an example of the input / output of the standard condition conversion model in the performance degradation diagnostic method of this invention. Compare the relationship between model prediction accuracy and data extraction rate when performance degradation is evaluated by applying output regression method, exhaust temperature method, compatibility method, and MT method base load extraction method for actual machine Example. The figure which shows the extraction range of the base load data by each system of an output regression method, an exhaust temperature method, a coexistence method, and MT method. The comparison figure of the change of an extraction rate and a model prediction accuracy when changing the threshold of base load judgment of each method of an output regression method, an exhaust temperature method, a coexistence method, and an MT method. The figure which shows the change of the extraction rate and model prediction accuracy corresponding to the change of the actual threshold value in each method of an output regression method, an exhaust temperature method, a coexistence method, and an MT method. The trend figure which shows the result of having converted the performance characteristic model into the base condition data extracted using each method of output regression, exhaust temperature method, compatibility method, and MT method, and converting the time series output data to standard conditions. The trend figure which shows the raw data of a time-dependent change of gas turbine performance (output). The figure which shows the example of a display screen of the diagnostic system using this invention. The flowchart of the coexistence method which shows an example of the extraction method of the base load driving | operation data in the performance degradation diagnosis method of this invention.

Explanation of symbols

4 ... Statistical distance calculation step, 5 ... Threshold determination / data selection step, 21 ... Unit space setting step, 22 ... Statistical distance calculation step from the reference period of each time data to selected data, 24-26 ... Threshold determination process for each index, 27 ... AND determination process for all indices, 41 ... Output characteristic model identification process, 42 ... Load factor calculation process, 43 ... Load factor calculation process, 44 ... Exhaust temperature index model identification process, 45 ... load factor index calculation process, 46 ... load factor index calculation process, 47 ... heat input characteristic model identification process, 48 ... load factor calculation process, 49 ... load factor calculation process, 61 ... performance characteristic model, 62 ... model constant, 63 Deterioration coefficient, 64: Operating conditions, 65: Performance calculation value, 71: Model constant and deterioration coefficient initial value setting process, 72: Performance calculation process, 73: Error determination process, 74: Set value correction process, 81: Operation data, 2 ... Standard condition conversion model, 83 ... Standardized data, 101 ... Gas turbine, 102 ... Calculation means, 103 ... Output means, 131 ... Base load data extraction means, 132 ... Modeling means, 133 ... Performance evaluation means, 134 ... see Period setting means, 135 ... reference data selection means, 136 ... statistical distance calculation means, 137 ... determination / selection means, 140 ... display screen, 141 ... original data graph, 142 ... extraction execution button, 143 ... extraction data graph, 144 ... Standardized data graph, 152 ... logical product judgment process of all indexes, 153 ... selection process.

Claims (8)

Based on the operation data of the gas turbine, equipped with characteristic modeling means for characteristic modeling of the performance index as a function including the intake air temperature as an input, and the output value obtained by inputting the operational data of the gas turbine to the obtained characteristic model In the gas turbine performance diagnosis system for diagnosing gas turbine performance based on
Operation data (hereinafter referred to as reference data) at a plurality of times in the past by a predetermined method within a predetermined period (hereinafter referred to as reference period) corresponding to each time of the time-series operation data (hereinafter referred to as the time). And the index value calculated based on the measured value or measured value of the time operation data with respect to the distribution of the measured value of the selected reference data or the distribution of at least one index value calculated based on the measured value Calculating the statistical distance indicating the degree of approach, and selecting the time operation data as the base load operation data for the characteristic modeling when the statistical distance is within a predetermined upper and lower limit range, The operation data is repeatedly input for each time of the operation data or a plurality of predetermined times, and the base load operation data selected by the repetitive processing is input to the characteristic modeling means. Performance Diagnosis system for a gas turbine comprising the data selection means.   In Claim 1, the selection of the reference data in the operation data selecting means is the actual value of each time operation data during the reference period or the value of the index value calculated based on the actual value is set to the upper and lower limits set in advance. It is determined whether or not it is within the range of the threshold value, and when the determination results of all the indexes are true, the operation data of the time is selected as the reference data. A performance diagnosis system for a gas turbine, which is performed by repeating a plurality of predetermined times. In claim 1, the means for calculating the index value in the operation data selection means,
(1) Substitute the operation data of the intake air temperature and intake air pressure at each time into the function identified based on the time series operation data that includes the intake air temperature and intake air pressure as input and outputs the power generation output. Means for calculating the ratio of the measured value to the predicted value of the power generation output at the time, or
(2) Means for calculating a ratio between each time value of the ratio calculated in (1) above and a moving average value of a certain period including the time or a specific percentile value used in statistics. A gas turbine performance diagnosis system characterized by comprising: In claim 1, the means for calculating the index value in the operation data selection means,
(1) The intake air temperature is included in the input, and at this time, the exhaust temperature at the maximum output of the gas turbine or the intermediate stage gas temperature of the turbine, or the upper limit value of the combustion temperature management index for control (hereinafter collectively referred to as the exhaust temperature index value) A means for calculating the deviation of the measured value from the predicted value at the maximum output of the exhaust gas temperature index value, which is output by substituting the operation data of the intake air temperature at each time into the function that outputs
(2) Means for calculating a deviation between each time value of the deviation calculated in (1) and a moving average value or a specific percentile value used in statistics for a certain period including the time. A gas turbine performance diagnosis system characterized by comprising: In claim 1, the means for calculating the index value in the operation data selection means,
(1) A function that includes the intake air temperature as an input and outputs the input fuel energy amount at the maximum output of the gas turbine or the total amount of input energy to the combustor of the gas turbine (hereinafter referred to as heat input amount) at each time Means for calculating the ratio of the measured value to the predicted value of the heat input at the maximum output, which is output by substituting the operation data of the intake air temperature of, or
(2) A gas turbine comprising: means for calculating a ratio between the value of each time and a moving average value of a certain period including the time or a specific percentile value used in statistics Performance diagnostic system . Based on the operation data of the gas turbine, at least one performance index of power generation output, power generation efficiency, compressor pressure ratio, fuel flow rate, compressor efficiency, turbine efficiency, intake air flow rate, and at least intake air temperature and intake air pressure are input. In the performance diagnosis method of the gas turbine, which is characterized by including a characteristic model as a function including, and diagnosing the performance of the gas turbine based on an output value obtained by inputting operation data of the gas turbine to the obtained characteristic model
Selecting base load operation data used for the characteristic modeling by determining at least one kind of index value calculated based on an actual measurement value or the actual measurement value;
The determination method is to select operation data at a plurality of past times by a predetermined method from a reference period determined corresponding to each time of the time-series operation data (the time), and the operation data at the plurality of times. The statistical distance of the actual measured value or index value of the time operation data for the distribution of at least one type of index value calculated from the actual measured value or the measured value is calculated, and the statistical distance is a predetermined upper and lower limit threshold. It is determined whether it is within the range of values, and if it is within the range, the operation of selecting this operation data as the base load operation data is repeated for the data at each time in the time series. A characteristic gas turbine performance diagnosis method . A display screen provided in the gas turbine performance diagnosis system according to claim 1,
A graph (hereinafter referred to as the original graph) that displays a time series change of the operation data (hereinafter referred to as the original data) at a plurality of time points during the diagnosis target period,
By clicking or selecting on the screen, extraction of base load operation data from the original data is started by the operation data selection means, a button or a display menu object,
A graph (hereinafter referred to as an extraction graph) displaying a time-series change of base load operation data (hereinafter referred to as extraction data) extracted by the operation data selection means, or the original data and the original graph instead of the original graph and the extraction graph A graph that is superimposed on a single graph with different plot symbols and colors so that the extracted data can be distinguished,
A display screen for a gas turbine performance diagnosis system, comprising a graph for displaying an output value obtained as a result of inputting the base load operation data to the characteristic modeling means . 2. The characteristic modeling means according to claim 1, wherein the characteristic modeling means includes an operation condition correction function for correcting an influence caused by a deviation of the operation condition of the gas turbine from a predetermined standard condition value, and a performance index value of the gas turbine at the standard condition value. The operation condition correction function and the time decrease function of the characteristic model expressed as the sum of the time decrease function representing the time decrease corresponding to the accumulated operation time or equivalent operation time are used as the operation data selected by the operation data selection means. A gas turbine performance diagnosis system characterized by comprising means for identification based on multiple regression analysis .

Images