CN113359435B - Correction method for dynamic working condition data of thermal power unit - Google Patents

Abstract

The invention discloses a correction method for dynamic working condition data of thermal power units, relates to the technical field of dynamic data correction of thermal power units, and solves the problems that the number of steady-state samples is relatively small and the distribution of working conditions is unbalanced during the operation of the existing thermal power units. The main point of the technical solution is to obtain historical operation data, go through preliminary preprocessing such as steady-state judgment and working condition division, and then filter out the neighboring samples of each steady-state working condition that are similar to the boundary parameters of the dynamic working condition sample to be corrected through the distance threshold. Samples, the correction coefficients are estimated based on the least squares method and the kernel density weighting method. Finally, the state parameter values and performance index values of the modified dynamic working condition samples are calculated based on the historical nearest steady state working condition with the smallest Minkowski distance. The parameter stability of this method is higher, which meets the actual needs of engineering; and the utilization of dynamic operation data is improved, which can effectively improve the problem of limited data mining accuracy due to unbalanced distribution of samples in working conditions.

Description

Correction method for dynamic working condition data of thermal power unit

technical field

The present disclosure relates to the technical field of dynamic data correction of thermal power units, in particular to a correction method for dynamic working condition data of thermal power units.

Background technique

In the data-driven modeling of thermal process, steady-state condition samples have become the main data source for model training due to their good regularity and less noise. However, in actual operation, affected by peak shaving demand and seasonal changes, the external conditions such as unit load command and environmental factors are constantly changing, resulting in the continuous fluctuation of unit operating parameters. Coupled with the influence of internal factors such as the intervention of the power plant's artificial regulation and the hysteresis characteristics of the thermal system, the frequency and magnitude of field data deviation will be further aggravated. It can be seen that the frequent and violent fluctuation process makes the number of steady-state working conditions in the unit operation history database often much smaller than that of dynamic working conditions.

In fact, a completely steady-state operating condition does not exist during the generating process of the unit. Even the steady-state data obtained through steady-state judgment are only quasi-steady-state data with small fluctuation range and less noise. In order to improve the accuracy and efficiency of data mining, a smaller steady-state threshold is used to screen historical data. Although higher-quality working condition information can be obtained, the number of retained samples will be greatly reduced, and the problem of unbalanced working condition distribution will be reduced. further intensified. The lack of sufficient and comprehensive high-quality samples to participate in model training will inevitably affect the effect of mining the operating rules of the unit, which is obviously fatal to the performance evaluation and diagnosis optimization of the unit under all operating conditions.

For data mining, improving the quality of raw data is always the key. The uneven distribution of the massive data accumulated in the high-dimensional space is a difficulty in the field of data mining. The operating state of the unit is complex and diverse, and the deviation of the working conditions is reflected in the deviation of the operating parameters, which in turn causes the deviation of the system energy consumption. To study the changing law of working conditions under the overall operating range, it is bound to face problems such as high dimension, high coupling, and high error. The operating performance of the system is jointly determined by the boundary parameters and the operating parameters, and the difference in energy consumption of the neighboring samples is also caused by the joint influence of the difference between the boundary parameters and the operating-related parameters. Therefore, how to revise the data in the dynamic process of the unit to obtain high-quality data, so as to provide a more complete and clean data foundation for the mining of operational optimization rules, is an urgent problem to be solved.

SUMMARY OF THE INVENTION

The present disclosure provides a method for correcting dynamic working condition data of thermal power units, and the technical purpose of which is to disclose a dynamic working condition for the relatively small number of steady-state samples and unbalanced working condition distribution during the operation of existing thermal power units. The correction method of process data to obtain a large number of steady-state samples, so as to solve problems such as unbalanced distribution of operating conditions.

The above-mentioned technical purpose of the present disclosure is achieved through the following technical solutions:

A method for correcting dynamic working condition data of thermal power units, comprising:

Preliminarily screen the first characteristic parameter related to the system operation performance through mechanism analysis, and select the second characteristic parameter related to the system operation performance index from the first characteristic parameter through the grey correlation degree algorithm;

Calculate the statistic of the second characteristic parameter to obtain a steady-state factor describing the stability of the working condition, compare the steady-state factor with the steady-state threshold value, and if the steady-state factor is smaller than the steady-state threshold value, it is regarded as a steady-state working condition sample;

Calculate the Minkowski distance between the steady-state working condition sample and the dynamic working condition sample S with known boundary conditions, if the distance between the first steady-state working condition in the steady-state working condition sample and the dynamic working condition sample S If the Minkowski distance is less than the distance threshold d _ε , the first steady state condition is a neighbor condition, and the Minkowski distance is used to screen out the neighbor condition samples from the steady state condition samples {w ₁ ,w ₂ ,...,w _K ,w _N }, then there are (K+1) neighboring working conditions in the neighbor working condition sample, and w _N represents the neighboring working condition with the closest distance to the dynamic working condition sample S;

Calculate the kernel density distribution of the neighbor working case samples {w ₁ ,w ₂ ,...,w _K ,w _N };

According to the kernel density distribution, the least squares estimation is performed on the energy consumption evaluation index of the dynamic working condition sample S and the correction coefficient of the related parameters to obtain the final correction coefficient;

Correcting the dynamic working condition sample S according to the final correction coefficient to obtain a corrected quasi-steady-state working condition sample S';

I^S表示动态工况下系统的能耗评价指标，

表示动态工况下系统的边界参数，

表示动态工况下系统的相关参数，其中，u就表示边界参数，r表示相关参数，m、n分别表示边界参数和相关参数的个数。Among them, the dynamic working condition sample S is expressed as

^IS represents the energy consumption evaluation index of the system under dynamic conditions,

represents the boundary parameters of the system under dynamic conditions,

Represents the relevant parameters of the system under dynamic conditions, where u represents the boundary parameters, r represents the relevant parameters, and m and n represent the number of boundary parameters and relevant parameters, respectively.

The beneficial effect of the present disclosure is that: by obtaining historical operation data in the power plant operation database, after preliminary preprocessing such as steady-state judgment, working condition division, etc., and then by a certain distance threshold, the samples of each steady-state working condition that are to be corrected are screened out. For the adjacent samples with similar boundary parameters of the samples in the dynamic condition, the correction coefficients are estimated based on the least squares method and the kernel density weighting method. Finally, the state parameter values and performance index values of the modified dynamic working condition samples are calculated based on the historical nearest steady state working condition with the smallest Minkowski distance.

The present application is a steady-state data supplement method, which is faster than the traditional modeling supplement method, has higher parameter stability, and meets the actual needs of the project; and improves the utilization of dynamic operation data, which can effectively improve the efficiency of engineering Due to the unbalanced sample distribution, the data mining accuracy is limited. In addition, the present application does not require complicated hardware equipment, and the price is low.

Description of drawings

Fig. 1 is the flow chart of the method described in this application;

Figure 2 is a schematic diagram of the scatter of the turbine heat consumption with the load distribution under different working conditions;

Figure 3 is a schematic diagram of the distribution of the mean value of the steam turbine heat consumption rate in different load intervals.

Detailed ways

The technical solutions of the present disclosure will be described in detail below with reference to the accompanying drawings. In the description of this application, it should be understood that the terms "first" and "second" are only used for description purposes, and cannot be interpreted as indicating or implying relative importance or indicating the number of technical features indicated, It is only used to distinguish different technical characteristics.

Fig. 1 is the flow chart of the modification method described in the application for thermal power unit dynamic working condition data, as shown in Fig. 1, the method includes:

S1: Preliminarily screen first characteristic parameters related to system operation performance through mechanism analysis, and select second characteristic parameters related to system operation performance indicators from the first characteristic parameters through a grey correlation degree algorithm.

Specifically, the relevant parameters of the system operation are preliminarily screened by the mechanism analysis, and the characteristic variables with greater correlation with the performance indicators of the unit are selected by the grey correlation algorithm, and reduced by a certain correlation threshold to obtain the energy consumption characteristic parameters, denoted by X ={X _u , X _r } is the energy consumption characteristic parameter of a certain system, and X _r (related parameter) is composed of the uncontrollable characteristic parameter X _u (boundary parameter) and the controllable characteristic parameter. Here it is necessary to select an appropriate grey relational degree reduction threshold.

S2: Calculate the statistic of the second characteristic parameter to obtain a steady state factor describing the stability of the working condition, compare the steady state factor with the steady state threshold value, and consider the steady state working condition sample if the steady state factor is smaller than the steady state threshold value. This process can be achieved by the R test method.

S3: Calculate the Minkowski distance between the steady-state working condition sample and the dynamic working condition sample S with known boundary conditions, if the first steady-state working condition in the steady-state working condition sample is the same as the dynamic working condition sample The Minkowski distance of S is smaller than the distance threshold d _ε , then the first steady state condition is a neighbor condition, and the Minkowski distance is used to screen out the neighbor condition samples from the steady state condition samples {w ₁ , w ₂ ,...,w _K ,w _N }, then there are (K+1) neighboring working conditions in the neighboring working condition sample, and w _N represents the neighboring working condition with the closest distance to the dynamic working condition sample S .

When screening the neighboring working conditions, the samples in the adjacent grids can be preferentially selected to calculate the distance, which is similar to the function of narrowing the search range, that is, the working conditions are divided. The working conditions are generally divided by the equal width method.

The distance threshold d _ε will be determined according to the actual system operation.

其中，d(A,B)表示m维空间内任意两点A(a₁,a₂,...a_m)与B(b₁,b₂,...b_m)之间的闵可夫斯基距离，A(a₁,a₂,...a_m)表示所述动态工况样本S中的任意一个动态工况，B(b₁,b₂,...b_m)表示所述稳态工况样本中的任意一个稳定工况，所述p表示变参数，且p＝2。Specifically, the Minkowski distance is expressed as:

Among them, d(A,B) represents the Minkoffs between any two points A(a ₁ ,a ₂ ,...am ) and B(b ₁ ,b ₂ ,...b _m ) in the _m -dimensional space Base distance, A(a ₁ ,a ₂ ,... _am ) represents any dynamic condition in the dynamic condition sample S, and B(b ₁ ,b ₂ ,...b _m ) means the For any stable working condition in the steady-state working condition sample, the p represents a variable parameter, and p=2.

M represents not only the number of boundary parameters, but also the number of dimensions for dividing the working conditions. Each operating point corresponds to m boundary parameters. When calculating the distance in the high-dimensional space, it is necessary to calculate the difference of the boundary parameters between the two points in each dimension and then sum them up to obtain the final distance.

The Minkowski distance from any one of the neighboring working condition samples to the dynamic working condition sample S needs to be calculated separately from the neighboring working condition to each working condition in the dynamic working condition sample S. A maximum Minkowski distance of a working condition in the dynamic working condition sample S is the Minkowski distance between the neighboring working condition and the dynamic working condition sample S.

Before calculating the Minkowski distance, the boundary parameters of the working conditions need to be normalized.

S4: Calculate the kernel density distribution of the neighbor working condition samples {w ₁ , w ₂ ,...,w _K ,w _N }, and a Gaussian kernel function can be used for estimation.

其中，f_h(d)表示任意近邻工况与最近近邻工况w_N的距离为d时对应的概率密度，d_k表示表示近邻工况w_k与w_N的距离，k＝1,2,...,K，h表示带宽，h＝d_ε/10，g(.)表示核函数。Specifically, the kernel density distribution of the samples {w ₁ ,w ₂ ,...,w _K ,w _N } of the nearest neighbors includes:

Among them, f _h (d) represents the corresponding probability density when the distance between any nearest neighbor working condition and the nearest neighbor working condition w _N is d, d _k represents the distance between the nearest neighbor working condition w _k and w _N , k=1,2, ..., K, h represents the bandwidth, h=d _ε /10, g(.) represents the kernel function.

S5: Perform a least square estimation on the energy consumption evaluation index of the dynamic working condition sample S and the correction coefficient of the related parameters according to the kernel density distribution, to obtain a final correction coefficient.

Specifically include:

表示边界参数

对能耗评价指标I^S的修正系数，

表示边界参数

对第j个相关参数

的修正系数，j∈[1,n]；

表示近邻工况w_k的能耗评价指标与w_N的能耗评价指标的差值；

表示近邻工况w_k的边界参数与w_N的边界参数的差值集合；

表示近邻工况w_k的第j个相关参数与w_N的第j个边界参数的差值；f_h(d_k)表示任意近邻工况与w_N的距离为d_k时对应的概率密度；θ1、θ2表示使argmin(.)结果最小的参数。

represents the boundary parameter

Correction coefficient for the energy consumption evaluation index ^IS ,

represents the boundary parameter

For the jth relevant parameter

The correction coefficient of , j∈[1,n];

represents the difference between the energy consumption evaluation index of the neighboring working condition w _k and the energy consumption evaluation index of w _N ;

Represents the set of difference values between the boundary parameters of the neighbor case w _k and the boundary parameters of w _N ;

Represents the difference between the jth related parameter of the nearest neighbor working condition w _k and the jth boundary parameter of w _N ; f _h (d _k ) represents the corresponding probability density when the distance between any nearest neighbor operating condition and w _N is d _k ; θ1, θ2 represent the parameters that minimize the result of argmin(.).

S6: Correct the dynamic working condition sample S according to the final correction coefficient to obtain a corrected quasi-steady-state working condition sample S'.

I^S表示动态工况下系统的能耗评价指标，

表示动态工况下系统的边界参数，

表示动态工况下系统的相关参数，其中，u就表示边界参数，r表示相关参数，m、n分别表示边界参数和相关参数的个数。The dynamic condition sample S is expressed as

^IS represents the energy consumption evaluation index of the system under dynamic conditions,

represents the boundary parameters of the system under dynamic conditions,

Represents the relevant parameters of the system under dynamic conditions, where u represents the boundary parameters, r represents the relevant parameters, and m and n represent the number of boundary parameters and relevant parameters, respectively.

The correction process includes:

表示动态工况样本S的边界参数与w_N的边界参数的差值集合；

表示w_N的能耗评价指标；

表示w_N的第j个相关参数；最终得到准稳态工况样本

Represents the set of differences between the boundary parameters of the dynamic working condition sample S and the boundary parameters of w _N ;

represents the energy consumption evaluation index of w _N ;

Represents the jth related parameter of w _N ; finally, the quasi-steady state sample is obtained

As a specific example, the on-site DCS sampling data is stored in the historical database of the factory-level monitoring information system (SIS). Combined with the specific system to be analyzed, the on-site data is first subjected to data cleaning, steady-state screening and other preprocessing to obtain steady-state operation. The operating condition database is reserved for subsequent analysis. Then, the characteristic variables of energy consumption during the operation of the turbine are selected through the grey correlation algorithm, and the relatively uncontrollable characteristic parameters are used as boundary conditions to divide the steady-state data, and the remaining parameters are operation-related parameters. For the dynamic working condition samples to be corrected, its boundary parameters can be matched in the steady-state operating condition library, and the steady-state samples within the boundary similar working conditions can be screened out, and the difference between these steady-state samples and the working conditions to be corrected can be calculated separately. The Minkowski distance between them. Set the distance threshold d _ε of the Minkowski distance to 0.5, filter out a certain number of samples of neighboring working conditions, and calculate the kernel density probability of the Minkowski distance of each sample as the weight, and use the least squares method to correct the relevant parameters. The coefficients are fitted, and the parameter results of the modified working conditions are finally given.

The invention analyzes the practicability of the method of the invention in combination with the turbine system of a 600MW subcritical air-cooling unit of a power plant in Inner Mongolia. Taking the heat consumption of the steam turbine as an example, the original data comes from the PI database of the SIS system of the unit. The time is from August 1st to August 15th, 2020. The sampling interval is 1min, and there are a total of 21,600 sets of data. Taking the steady-state threshold as 1.5, a total of 8765 sets of steady-state samples were obtained, and the remaining 12,835 samples were dynamic samples.

The gray correlation degree analysis was carried out on the samples under steady state conditions, and the results are shown in Table 1.

Table 1

Taking the gray correlation degree reduction threshold as 0.75, for the turbine operating conditions, the main steam pressure, main steam temperature, reheat steam temperature, vacuum degree, load, main steam flow, regulating stage pressure, and regulating stage temperature are screened as Turbine energy consumption characteristic variables. Among them, the main steam pressure, main steam temperature, reheat steam temperature, vacuum degree and load are the boundary parameters of the turbine operating conditions, and the main steam flow, regulating stage pressure and regulating stage temperature are the relevant parameters of the turbine operating conditions. According to the historical fluctuation range of the parameters, the interval for dividing the working condition interval can be artificially determined. The specific dividing parameters are shown in Table 2.

Main steam pressure/MPa Main steam temperature/℃ Reheat steam temperature/℃ Vacuum degree/kPa Parameter variation range 13-17 520-570 500-570 7-18 interval 0.5 5 5 1

Table 2

For each piece of dynamic process data to be corrected, the 40 working conditions closest to the Minkowski distance are taken as reference samples. Based on the least squares estimation method, the correction coefficients of the operation-related parameters such as the load, the main steam flow, and the exhaust steam pressure of the high-pressure cylinder are estimated by the difference of the main steam pressure, the main steam temperature, the reheat steam temperature, and the vacuum degree. The parameter distributions of some of the original dynamic conditions, the corrected conditions and the nearest neighbor conditions are shown in Table 3. Figure 2 is the scatter plot of the turbine heat consumption distribution with load under different conditions. , the standard deviation and deviation rate of turbine heat consumption corresponding to different working conditions. Figure 3 shows the mean distribution of turbine heat consumption rate in different load intervals. It can be seen from Figure 2 that the projected area of the heat consumption rate is significantly reduced for the operating conditions after interpolation and correction of the neighboring operating conditions. From Table 4 and Figure 3, it can be seen that the corrected heat consumption deviation rate is greatly reduced, and the fluctuation is suppressed. Judging from the average heat consumption of each load interval, the trend of the revised turbine heat consumption gradually decreasing with the load is more obvious, and the value is closer to the result in the test report, which proves that the effect of dynamic sample correction is ideal, and it can meet engineering needs.

table 3

Table 4

In order to further verify the reproducibility of the modified operating parameters, the steady-state heat consumption prediction model was used to verify the coupling between the energy consumption characteristic parameters. The input parameters of the neural network model are the energy consumption characteristic parameters. A total of 500 steady-state working condition samples are screened. The first 400 steady-state working condition samples train the steady-state heat consumption prediction model based on the BP neural network algorithm, and the next 100 steady-state working condition samples are used to train the steady-state heat consumption prediction model. The working condition samples and 100 dynamically corrected working condition samples are the errors of the test set evaluation model. The evaluation indicators of the errors are the mean relative error (MRE) and the root mean square error (RMSE). , the errors of training set and test set are shown in Table 5. It can be seen from Table 5 that the error of the modified operating parameters in predicting the heat consumption of the turbine is slightly larger than that of the steady state sample, but the MRE is also within 1.5%, which proves that the coupling relationship between the relevant parameters is still guaranteed. , which is still instructive in practical operation optimization.

table 5

The invention makes full use of the historical steady-state data of the unit to correct the complex dynamic working conditions. Compared with the traditional modeling supplementary method, the speed is faster, the calculation amount is smaller, and the parameter stability is higher, and it is suitable for thermal power unit data. Mining and running optimization areas, data sparse areas and sample supplementation.

The above are exemplary embodiments of the present application, and the protection scope of the present application is defined by the claims and their equivalents.

Claims (6)

1. A correction method for dynamic working condition data of a thermal power generating unit is characterized by comprising the following steps:

primarily screening first characteristic parameters related to system operation performance through mechanism analysis, and selecting second characteristic parameters related to system operation performance indexes from the first characteristic parameters through a grey correlation algorithm;

calculating the statistic of the second characteristic parameter to obtain a steady-state factor describing the stability of the working condition, comparing the steady-state factor with a steady-state threshold value, and considering the steady-state factor smaller than the steady-state threshold value as a steady-state working condition sample;

calculating the Minkowski distance between the steady-state operating condition sample and the dynamic operating condition sample S of known boundary conditions, if the Minkowski distance between the first steady-state operating condition sample and the dynamic operating condition sample S is less than a distance threshold d _ε If the first steady-state working condition is a neighboring working condition, screening out samples { w ] of the neighboring working condition from the samples of the steady-state working condition through Minkowski distance ₁ ,w ₂ ,...,w _K ,w _N And f, obtaining (K +1) neighbor working conditions in the neighbor working condition samples, w _N Representing the nearest neighbor working condition to the dynamic working condition sample S;

calculating the neighbor condition samples { w ₁ ,w ₂ ,...,w _K ,w _N Nuclear density distribution of };

performing least square estimation on the energy consumption evaluation index of the dynamic working condition sample S and the correction coefficient of the related parameter according to the nuclear density distribution to obtain a final correction coefficient;

correcting the dynamic working condition sample S according to the final correction coefficient to obtain a corrected quasi-steady-state working condition sample S';

wherein the dynamic condition sample S is represented as

I ^S The energy consumption evaluation index of the system under the dynamic working condition is shown,

representing the boundary parameters of the system under dynamic conditions,

and expressing the relevant parameters of the system under the dynamic working condition, wherein u expresses the boundary parameters, r expresses the relevant parameters, and m and n respectively express the number of the boundary parameters and the relevant parameters.

2. A method as claimed in claim 1 wherein the Minkowski distance is expressed as:

wherein d (A, B) represents any two points A (a) in m-dimensional space ₁ ,a ₂ ,...a _m ) And B (B) ₁ ,b ₂ ,...b _m ) Minkowski distance, A (a) ₁ ,a ₂ ,...a _m ) Represents any one of the dynamic conditions, B (B) ₁ ,b ₂ ,...b _m ) And representing any stable working condition in the steady-state working condition samples, wherein p represents a variable parameter.

3. The method of claim 2, wherein the computing the neighbor condition samples { w } ₁ ,w ₂ ,...,w _K ,w _N A nuclear density distribution of { includes:

wherein f is _h (d) Representing any near neighbor working condition and nearest neighbor working condition w _N When the distance of (d) is the corresponding probability density, d _k Indicating neighbor condition w _k And w _N K1, 2, K, h denotes the bandwidth, h d _ε The/10, g (.) represents the kernel function.

4. The method according to claim 3, wherein the least square estimation is performed on the energy consumption evaluation index of the dynamic working condition sample S and the correction coefficient of the related parameter according to the nuclear density distribution to obtain a final correction coefficient,

the method comprises the following steps:

wherein,

representing boundary parameters

Energy consumption evaluation index I ^S The correction coefficient of (a) is determined,

representing boundary parameters

For j relevant parameter

J is within [1, n ]]；

Indicating neighbor condition w _k Energy consumption evaluation index of (1) and w _N The difference of the energy consumption evaluation indexes;

indicating neighbor condition w _k Boundary parameter and w _N The set of difference values of the boundary parameters of (a);

indicating neighbor condition w _k J-th correlation parameter of (1) and w _N The difference of the jth boundary parameter of (a); f. of _h (d _k ) Represents any neighbor operating condition and w _N A distance of d _k Probability density of temporal correspondence; θ 1, θ 2 represent parameters that minimize the argmin (.) result.

5. The method according to claim 1, wherein the modifying the dynamic condition sample S according to the final modification coefficient to obtain a modified quasi-steady state condition sample S' comprises:

wherein,

boundary parameter and w representing dynamic condition sample S _N A set of difference values of the boundary parameters of (a);

denotes w _N The energy consumption evaluation index of (1);

denotes w _N The j-th correlation parameter of (1); finally obtaining a quasi-steady state working condition sample

6. The method of claim 2, wherein p is 2.

Images