JP2005301582A - Process management device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently discriminate and estimate a normal operation state and a defective state from limited sensor signals, and to estimate the factor when any failure is generated, and to efficiently execute the performance verification, failure detection, failure prediction and failure diagnosis of this system, and to reduce energy consumption and maintenance costs. <P>SOLUTION: A statistical analyzing means 3 uses process variable connection weighting parameters "a<SB>11</SB>, a<SB>12</SB>to a<SB>mp</SB>" to perform the principal component analysis of process result data, and prepares a principal component score "Z<SB>m</SB>" by reducing the number of variables. A preliminarily learnt artificial intelligence means 9 analyzes the main component scores "Z<SB>m</SB>", and decides the presence/absence of abnormality, and estimates a defective factor. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a process management apparatus that predicts in advance an abnormality of various types of industrial and consumer equipment or a situation where operation is impossible, and estimates the cause when a failure occurs, and there are many BEMS apparatuses in particular. Of the functions, the present invention relates to a process management apparatus that performs failure detection, failure prediction, and failure diagnosis.

  Currently, almost the same can be said for general processes, but in recent years, it has become difficult to secure skilled engineers in building equipment operation management, and as a result, building scale has increased, resulting in building equipment such as air conditioning systems. The detection and treatment of abnormalities occurring in the country are slowing down, and the possibility of causing deterioration of the indoor environment and energy waste is increasing. As a system that solves these problems and is expected to have a high energy-saving effect, a building, environment, and energy management system (BEMS) apparatus is now attracting attention.

BEMS equipment manages all building management items or multiple building management items from building management to energy management in a lump, and is important as an energy-saving measure for commercial buildings such as offices and stores. It has become.
JP-A-6-208568

  By the way, there are various building facilities. Today, air conditioning systems for heating and cooling are indispensable for buildings. Several air conditioning systems are required for each floor of the building, and the number of large air conditioning systems increases. Moreover, since this air conditioning system is composed of a plurality of devices, problems often occur.

  If a malfunction occurs, the air conditioning system does not operate normally, and a comfortable air conditioning environment cannot be provided to the residents. Further, even if energy saving control or the like is introduced, usually only sensors necessary for the control are installed, and those for detecting a failure and its factor are not installed.

  For this reason, even if a malfunction occurs in the equipment, it is not discovered and it is operated as it is, and the state in which the equipment efficiency is greatly reduced is continued and energy is often wasted.

  Therefore, in recent years, in order to quickly find defects in each facility installed in a building, many buildings and the center are connected by a communication line, and the facilities in each building are intensively monitored by an observer placed in the center. Even when no management device is installed on each building side, a monitoring system is often introduced that allows the sensor side to monitor the presence or absence of abnormality of each facility provided on each building side.

  However, in such a monitoring system, a huge amount of data is concentrated on the center side, so there is a limit to the detection of malfunctions by manual monitoring, and a huge amount of data is automatically analyzed and a failure is detected. There has been a strong demand for the development of an apparatus that can determine whether or not there is any.

  In view of the above circumstances, the present invention is capable of efficiently discriminating and estimating a normal operation state and a failure state from a limited sensor signal, and estimating the factor when a failure occurs to verify system performance, failure detection, and failure prediction. The main object of the present invention is to provide a process management apparatus that can accurately perform failure diagnosis, reduce energy consumption, and reduce maintenance costs.

  In order to achieve the above object, the present invention provides a statistical analysis means for statistically analyzing multivariables obtained by measuring a process using analysis parameters, reducing the number of variables and generating statistical analysis data, and a learning process. And an artificial intelligence means for analyzing the statistical analysis data output from the statistical analysis means, determining the presence or absence of an abnormality, and estimating a failure factor.

  In the process management apparatus according to the present invention, it is possible to efficiently discriminate and estimate the normal operation state and the failure state from the limited sensor signal, and when the failure occurs, the cause is estimated, and the system performance verification, failure detection, failure prediction, failure Diagnosis can be performed with high accuracy, energy consumption can be reduced, and maintenance costs can be reduced.

  Embodiments of the present invention will be described below with reference to the drawings. It is obvious that the present invention can be applied to a general process. However, in the embodiment, as a specific example, the intake-side damper 102 that controls the intake amount of outside air and the exhaust amount are controlled as shown in FIG. Using the exhaust-side damper 103, the circulation-side damper 104 that controls the amount of air circulation, and cold water, hot water, steam, etc., the temperature and humidity of outside air taken in via the intake-side damper 102 are adjusted. An air conditioner 105, an air blower 107 that takes in air whose temperature and humidity have been adjusted by the air conditioner 105 and sends the air into the room 106, an air fan 108 that takes air from inside the room 106, exhausts and circulates, and a communication line 109. The control unit for controlling the air conditioner 105 based on the control command supplied from the central CPU via the central control unit, the measurement result obtained by the thermometer 110 and the hygrometer 111 in the room 106, and the like. Using the air-conditioning process 101 that includes a roller 112, via a network, the conditioning process 101 Bill environment and energy management for remote control (BEMS) for fault detection, prediction and diagnosis function of the apparatus will be described.

<< Configuration Description of Embodiment >>
FIG. 1 is a block diagram showing the overall configuration of a process management apparatus according to the present invention applied to such an air conditioning process.

  The process management apparatus 1 shown in this figure includes a process result data storage unit 2 that collects process data, a statistical analysis unit 3 that analyzes process result data using a principal component analysis (PCA), and an analysis Analytical data storage means 4 for storing results, and artificial intelligence means 5 for performing process performance verification, failure detection, failure prediction, failure diagnosis and the like based on the stored analysis results. And the energy consumption of each air conditioner installed in each building and other process values by each sensor constituting each measurement monitoring control means 10 installed in the target building side (office building, department store, hospital, etc.) While detecting (outside air temperature, room temperature, indoor humidity, etc.), the detection result (process performance data) is taken in via the network 11, and the failure detection of the air-conditioning process 101, failure prediction, failure diagnosis failure detection, etc. are performed.

  The process result data storage means 2 is used to measure the energy consumption of the air conditioner, other process values (outside temperature, room temperature, indoor humidity, etc.) from the measurement monitoring control means 10 installed on the target building side such as an office building, department store, or hospital. ) Is output via the network 11 and stored separately in normal process performance data and failure process performance data.

  Further, the statistical analysis means 3 is obtained by the process simulation means 6 and the process simulation means 6 which perform the simulation of the normal state of the process and the simulation of the various fault states in advance using each parameter such as various buildings and load conditions. Simulation data accumulating means 7 for accumulating the obtained simulation results, and using the principal component analysis, the simulation results accumulated in the simulation data accumulating means 7 and the normal process results accumulated in the process performance data accumulating means 2 Data and process performance data at the time of failure occurrence are statistically analyzed, and the main component that performs the pre-stage processing necessary to quickly and easily extract the difference between the normal state characteristic value and the abnormal state characteristic value And analyzing means 8.

  Then, before operating the process management apparatus 1, a simulation is performed by performing a simulation when the process is in a normal state and a simulation when there are various failure states using various parameters such as various buildings and load conditions. The results are accumulated, and the simulation results are subjected to principal component analysis to obtain process variable connection weight parameters necessary to quickly and easily extract the difference between the normal state characteristic value and the abnormal state characteristic value. Using this process variable coupling weight parameter, the principal component analysis is performed on the simulation result when the process is in a normal state and the simulation result when the various failure states occur, and each main analysis score is stored in the analysis data storage means 4. accumulate. Thereafter, when the process management apparatus 1 is operating, the process result data supplied via the process result data storage means 2 is principally analyzed using the process variable coupling weight parameter, and the main analysis score is obtained. The artificial intelligence means 5 is supplied. Then, when the operation period of the process management apparatus 1 becomes long and the actual performance data and the actual performance data at the time of failure have been sufficiently accumulated in the process performance data storage means 2, process variables are used to The connection weight parameter is updated, and each main analysis score when the process stored in the analysis data storage unit 4 is normal and each main analysis score when the process is faulty are updated.

  Also, the analysis data storage means 4 is a new processing result obtained by performing principal component analysis on the simulation result in the normal state from the statistical analysis means 3 from the process management apparatus 1 and a new principal component analysis on the simulation result in various failure states. Each time a processing result is output, it is fetched, and the latest processing result is supplied to the artificial intelligence means 5 while updating the previous processing result.

  The artificial intelligence means 5 updates each main analysis score (each main analysis score when the process is normal, each main analysis score when the process is faulty) stored in the analysis data storage means 4. In addition, based on these main analysis scores, while updating the weight parameters of the neural network, the principal component score of the process performance data output from the statistical analysis means 3 is analyzed to detect failure of the air conditioning process 111, failure prediction, failure Neural network means 9 for performing diagnostic failure detection, energy saving control, etc. is provided, and learning is performed based on each principal component score stored in the analysis data storage means 4 before the process management apparatus 1 is operated. Determine the weight parameters of the neural network. Thereafter, when the process management apparatus 1 is in operation, the principal component score of the process performance data output from the statistical analysis unit 3 is analyzed, and failure detection, failure prediction, failure diagnosis, failure detection, and the like of the air conditioning process 111 are performed. .

<< Description of Operation of Embodiment >>
Next, with reference to the block diagram shown in FIG. 1, the flowchart shown in FIG. 2, and the schematic diagram shown in FIG. 3, the process management device 1 is divided into before and when the process management device 1 is operating. The operation of the process management apparatus 1 will be described.

<Before the process management device 1 is operated>
First, before the process management apparatus 1 operates, the simulation means 6 of the statistical analysis means 3 organizes each piece of data relating to the building that has been input in advance, each load condition data, etc. for each time.

  Thereafter, as shown in the flowchart of FIG. 2, the simulation means 6 reads each piece of information relating to the building side inputted in advance, and the size, heat capacity, wall area, and wall size of the room provided on the building side. After data such as the heat transmissibility is set (step S1), and based on each load condition data, etc., the outside air temperature, the outside air humidity, and the amount of solar radiation, which are meteorological factors related to the air conditioning load, are set in advance (step S2). ) The amount of heat acquired from the window glass surface, the amount of heat generated by the person, the amount of heat generated by the person, OA, and the amount of heat generated by the lighting are calculated in advance for the simulation time corresponding to the change in the occupancy rate of the person. The heat load data is set (step S3).

  Thereafter, the simulation means 6 checks whether the sampling control period is an integer multiple of the simulator calculation period, and when it is the sampling period (step S4), the room temperature is set by the DDC (Direct Digita1 Contro11er) control algorithm method. A PID parameter (speed type) is calculated based on the deviation between the value and room temperature, the deviation between the humidity setting value and the room humidity, etc. (step S5).

  Next, the simulation unit 6 uses the theoretical calculation formula of the heat exchanger and the calculation formula of the air diagram in the order of the cooling coil, the heating coil, and the humidifier, so that each state point of air in the air conditioner 105 ( Temperature, humidity, enthalpy, etc.) are calculated (step S6), and the wall of the room in the building is divided into the outer wall (with windows), the floor, the inner wall, and the ceiling, and the heat transfer equation (heat transfer model) The heat balance is described (step S7), the dynamic characteristic change between the room temperature and the wall temperature is calculated by the differential decomposition method, and the air state point at the air conditioner inlet is obtained by the mixing calculation of the recirculated air and the introduced outside air. (Step S8).

  Hereinafter, the simulation unit 6 performs the above-described calculation until the simulation end time is reached while the time data is updated, and the simulation when the process is in a normal state and the simulation when in various fault states are performed. As a result, the process result data when the process is in a normal state and the process result data when various processes are in failure are stored in the simulation data storage unit 8 (steps S4 to S10).

  In parallel with this operation, before operating the process management apparatus 1, the process result data when the process accumulated in the simulation data accumulation means 7 is in a normal state by the principal component analysis means 8 and various The process performance data in the failure state is analyzed, and a process variable connection weight parameter necessary to easily and quickly extract the difference between the characteristic value in the normal state and the characteristic value in the abnormal state is obtained.

In this case, for example, the process performance data is p multivariable data with the number of variables “X ₁ , X ₂ ,..., X _p ” and the number of observation objects n, and using this multivariate data, When creating _m new variables “Z ₁ , Z ₂ ,..., Z _m ” less than p, process variable connection weight parameters “a ₁₁ , a _{12 ,.} "Is calculated.

[Equation 1]
Var (Z ₁ ) ≧ Var (Z ₂ ) ≧… ≧ Var (Z _m ) (1)
a _i1 ² + a _i2 ² + ... 10 a _ip ² = 1 (2)
Where i: i = 1, 2,..., M
Var (Z _i ): Variance At this time, the new variable “Z ₁ ” is made independent of the new variable “Z ₂ ”. The new variable “Z ₃ ” is independent of the new variable “Z ₁ ” and the new variable “Z ₂ ”. The same applies from the new variable “Z ₄ ” to the new variable “Z _m ”.

Under such conditions, the process variable connection weight parameter _{"a 11, a 12, ...} , a mp" to seek the original variables _{"X 1, X 2, ...} , X p" covariance matrix of The process variable coupling weight parameters “a ₁₁ , a ₁₂ ,..., A _mp ” are none other than finding the eigenvectors.

However, this time, the original variable _{"X 1, X 2, ...} , X p" and the units of the new variable _{"Z 1, Z 2, ......} , Z m" from that and units of different, principal component analysis , The original variables “X ₁ , X ₂ ,..., X _p ” are normalized (standardized).

That is, when the average value of all target data of the kth variable is “M _k ” and the standard deviation is “S _k ”, each variable “X _kj ” is normalized as follows.

[Equation 2]
X _kj → (X _kj −M _k ) / S _k (3)
However, j: j = 1 to n (n is the number of data)
Next, the process variable combination weight parameters “a ₁₁ , a ₁₂ ,..., A _mp ” are used by the principal component analysis means 8 to calculate the process performance data (or , Process actual data at the time of various fault conditions) are combined with original variables “X ₁ , X ₂ ,..., X _p ” to form new variables “Z ₁ , Z ₂ , ..., Z _m ″ is created.

[Equation 3]
_{Z 1 (t) = a 11} · X 1 (t) + a 12 · X 2 (t) + ... Ten _{a 1p · X p (t)} ... (4)
_{Z 2 (t) = a 21} · X 1 (t) + a 22 · X 2 (t) + ... Ten _{a 2p · X p (t)} ... (5)
...
...
Z _m (t) = a _m1 · X ₁ (t) + a _m2 · X ₂ (t) + ... 10 _amp · X _p (t) (6)
However, _amp : process variable connection weight parameter t: time In this case, it is assumed that the new variable (principal component score) “Z ₁ , Z ₂ ,..., Z _m ” has the following properties.

The new variable “Z ₁ ” allows the information of the original variables “X _{1 to} X _p ” to be aggregated to the maximum extent. Further, the new variable “Z ₂ ” is configured so that the information of the original variables “X _{1 to} X _p ” is aggregated as much as possible after the new variable “Z ₁ ”. Hereinafter, the same applies from the new variable “Z ₃ ” to the new variable “Z _m ”.

Then, the principal component analysis means 8 performs principal component analysis processing on all process result data when the process stored in the simulation data storage means 7 is in a normal state, and applies to all process result data when various failure states occur. when the principal component analysis process and was complete the principal component scores "Z _m" of normal that obtained by principal component analysis, and the principal component scores "Z _m" at the time of failure is transferred to the analysis data storage means 4 Accumulated.

In parallel with the above-described operation, the normal principal component score “Z _m ” stored in the analysis data storage unit 4 by the neural network unit 9 of the artificial intelligence unit 5 before the process management apparatus 1 is operated. Then, learning is performed based on the principal component score “Z _m ” at the time of failure, and the weight parameter of the neural network is determined.

  As a result, the room temperature, the operation valve opening, the room humidity, the outside air temperature, and the outside air humidity measured in the air conditioning process 101 are used as variables of the process performance data obtained by the simulation process of the simulation means 6, and the neural network means 9 is divided into cooling coil failure, cooling fan failure, cooling water operation valve failure, duct pressure loss, heat source machine failure, and when the failure factor is estimated, according to the weight parameter of the neural network, as shown in FIG. Each neuron element in the input layer, each neuron element in the intermediate layer, and each neuron element in the output layer constituting the neural network means 9 are coupled.

<When the process management device 1 is operating>
Further, when the process management apparatus 1 is operated, each process result data output from the measurement monitoring control means 10 installed on the target building side such as an office building, department store, or hospital by the process result data storage means 2, for example, each The energy consumption of the air-conditioning process 101 and other process values (outside temperature, room temperature, indoor humidity, etc.) are captured and output to the statistical analysis means 3, and based on the determination result of the artificial intelligence means 5, the normal process The result data and the process result data at the time of the failure are divided and accumulated.

In parallel with this operation, the principal component analysis unit 8 of the statistical analysis unit 3 uses the process variable coupling weight parameters “a ₁₁ , a _{12 ,.} The calculation shown in the equations (3) to (6) is performed on the process result data output from the result data storage unit 2 to obtain the principal component score “Z _m ”, which is supplied to the artificial intelligence unit 5 Is done.

As a result, the neural network means 9 of the artificial intelligence means 5 performs neural network processing on the principal component score “Z _m ” to determine whether the air conditioning process 101 on the target building side such as an office building, department store, or hospital is out of order. When it is estimated that a failure has occurred, a failure factor is output, and the process result data stored in the process result data storage means 2 is a process at the time of failure until the restoration of the air conditioning process 101 is completed. Accumulated as actual data.

Further, when the operation period of the process management apparatus 1 becomes long and the process result data storage unit 2 sufficiently stores the process result data at the normal time and the process result data at the time of failure, the process result data is obtained by the principal component analysis unit 8. The normal process result data and the process result data at the time of failure stored in the storage means 2 are analyzed, and new process variable coupling weight parameters “a ₁₁ , a ₁₂ ,..., A _mp ” are obtained, The process variable combination weight parameters “a ₁₁ , a ₁₂ ,..., A _mp ” are updated, and the new process variable combination weight parameters “a ₁₁ , a _{12 ,.} With respect to normal process performance data and failure process performance data stored in the storage means 2, the above equations (3) to (6) are used. And performs the operation shown in equation a principal component scores "Z _m" new normal obtained by principal component analysis, principal component scores during new fault "Z _m" and is the analysis data storage means 4 are transferred, the principal component scores "Z _m" of normal, and the principal component scores "Z _m" at the time of fault is updated.

Then, a new normal principal component score “Z _m ” and a new failure principal component score “Z _m ” stored in the analysis data storage unit 4 by the neural network unit 9 of the artificial intelligence unit 5 are obtained. Based on this, learning is performed to obtain a weight parameter of a new neural network, and the connection contents of each neuron element in the input layer, each neuron element in the intermediate layer, and each neuron element in the output layer are updated.

<< Simulation result of embodiment >>
Next, simulation contents when the normality / abnormality determination capability of the process management apparatus 1 is confirmed will be described with reference to FIGS.

First, the air conditioning process 101 shown in FIG. 9 is used as the target process, and the following five variables (P = 5) that can be normally measured as process performance data variables, that is, the actual air conditioning process that is currently operating. Variables “X ₁ , X ₂ , X ₃ , X ₄ , X ₅ ” that can collect process performance data without adding a new sensor are used for 101 and the like.

X ₁ : Room temperature (with measurement noise) (7)
X ₂ : Cold water operation valve MV (8)
X ₃ : Indoor humidity (9)
X ₄ : outside air temperature (10)
X _5: ambient humidity ... (11)
<Air supply fan failure simulation>
And the size of the room is “2,009 m ³ ”, there are windows in the south and west, and the operating time of the air conditioning process 101 is from “8 o'clock” to “19 o'clock (calculation cycle: 1 minute)” As shown in the table of FIG. 4, the simulator means 3 performs a normal simulation of the cooling time while changing the air conditioning load conditions such as the outside air temperature and the indoor heat load, and obtains the principal component score “Z _i ”. As for the failure simulation, it is assumed that the supply air fan has failed and the supply amount of 7.0 [kg / s] is reduced to “4.5 [kg / s]” under normal conditions. When the score “Z _i ” was obtained, each principal component score time series display (trend graph) shown in FIG. 5 could be obtained.

Among the principal component scores “Z _i ” shown in the trend graph shown in FIG. 5, the normal data portion and the abnormal data portion of the principal component score “Z _i ” indicated by “No. 4” are respectively expanded. as can be seen from the trend graph shown in FIG. 6 described, can be the one threshold, to separate the principal component scores "Z _i" of normal, the principal component scores "Z _i" of the abnormality.

  Next, a method for obtaining a threshold value (adjustment parameter) that can discriminate between a normal state and an abnormal state will be described.

Now, assuming that normal / abnormal can be separated with the “i-th” principal component score “Z _i ”, the design parameter is “n”, and the “i-th” principal component score “Z _i ” in the normal state. When a standard deviation of “σ _i ” is “σ _i ”, a threshold value for separating normal / abnormal is set to “nσ _i ”.

In this case, among the normal principal component scores “Z _i ”, the number of data that does not satisfy the threshold “nσ _i ”, that is, the number of data determined to be abnormal is “Ne”, and the principal component score “ When the total number of data of Z _i ”is“ N _a ”, if the threshold parameter“ n ”is determined as shown in the following formula, the principal component score“ Z _i ”at normal time and the principal component score at abnormal time The error rate when separating “Z _i ”, that is, the error rate at which normal data is determined to be abnormal can be reduced to “5%” or less.

[Equation 4]
n = 100 (Ne / Na) (12)
n <5 (13)
Then, when separating whether the air supply fan is normal or abnormal using the “No. 4” principal component score “Z _i ”, when the threshold value “nσ _i ” shown below is used It was confirmed that the discovery probability of abnormal data could be over “90.3%”.

[Equation 5]
Threshold = nσ _i
= −1.03 · σ ₄
= -0.2923 (14)
However, σ ₄ : σ ₄ = 0.2741
<Failure simulation of cooling water operation valve>
In addition, the size of the room is “2,009 m ³ ”, there are windows in the south and west, and the operation time of the air conditioning process 101 is from “8 o'clock” to “19 o'clock (calculation cycle: 1 minute)” As shown in FIG. 4, the simulator means 3 performs a normal simulation of the cooling time while changing the air conditioning load conditions such as the outside air temperature and the indoor heat load, and obtains the principal component score “Z _i ”. As for the simulation, when the principal component score “Z _i ” is calculated on the assumption that the cooling water operation valve has failed and a delay of “5 minutes” has occurred with respect to the operation output of the controller, FIG. It was possible to obtain a time series display (trend graph) of each principal component score shown.

The trend graph shown in FIG. 8 in which the normal data portion and the abnormal data portion of the “No. 2” main component score are enlarged among the principal component scores “Z _i ” shown in the trend graph shown in FIG. as it can be seen from the can by one threshold, to separate the principal component scores "Z _i" of normal, the principal component scores "Z _i" of the abnormality.

  Next, a method for obtaining a threshold value (adjustment parameter) that can discriminate between a normal state and an abnormal state will be described.

Now, assuming that the normal / abnormal can be separated by the “i” -th principal component score “Z _i ”, the design parameter is “n”, and the “i” -th principal component score “Z _i in the normal state”. When a standard deviation of “σ _i ” is “σ _i ”, a threshold value for separating normal / abnormal is set to “nσ _i ”.

In this case, among the normal principal component scores “Z _i ”, the number of data that does not satisfy the threshold “nσ _i ”, that is, the number of data determined to be abnormal is “Ne”, and the principal component score “ When the total number of data of Z _i ”is“ N _a ”, if the threshold parameter“ n ”is determined as shown in the following formula, the principal component score“ Z _i ”at normal time and the principal component score at abnormal time The error rate when separating “Z _i ”, that is, the error rate at which normal data is determined to be abnormal can be reduced to “5%” or less.

[Equation 6]
n = 100 (Ne / Na) (15)
n <5 (16)
Then, when the “No. 2” principal component score “Z _i ” is used to separate whether the cooling water operation valve is normal or abnormal, the threshold value “nσ _i ” shown below is used. It was confirmed that the probability of finding abnormal data could be “90.8%” or higher.

[Equation 7]
Threshold = nσ _i
= ± 1.8 · σ ₂
= 1.8016 (17)
However, σ ₄ : σ ₄ = 1.0009
As can be seen from these simulation results, by monitoring each principal component score “Z _i ” of the principal component analysis, not only normal and abnormal can be separated, but also the principal component score “Z” that appears prominently due to the failure factor. _i ”changes and an abnormality occurs on the air-conditioning process 101 side by setting a threshold value“ nσ _i ”or learning to the neural network means 9 for each principal component score“ Z _i ” This can be detected and an alarm can be issued.

<< Effects of the Embodiment >>
As described above, in this embodiment, the statistical analysis means 3 performs principal component analysis on the process performance data using the process variable coupling weight parameters “a ₁₁ , a ₁₂ ,..., A _mp ”, and calculates the number of variables. The reduced principal component score “Z _m ” is created, and the principal component score “Z _m ” is analyzed by the artificial intelligence means 9 learned in advance to determine whether there is an abnormality and to estimate the failure factor. Therefore, it is possible to efficiently discriminate and estimate the normal operation state and the failure state from the limited sensor signal, and when the failure occurs, the cause is estimated and the system performance verification, failure detection, failure prediction, and failure diagnosis are executed accurately. Energy consumption can be reduced and maintenance costs can be reduced.

In this embodiment, the principal component analysis means 8 is arranged in the statistical analysis means 3, and the process results are obtained using the process variable connection weight parameters “a ₁₁ , a ₁₂ ,..., A _mp ” of the principal component analysis. Principal component analysis is performed on the process result data supplied via the data storage means 2 and the principal component score “Z _m ” is obtained. Therefore, the number of sensors installed on each building side is reduced, and the sensor It is possible to efficiently discriminate and estimate the normal operation status and failure status from limited sensor signals while estimating the system performance verification, failure detection, failure prediction, and failure diagnosis. Can be performed with high accuracy, energy consumption can be reduced, and maintenance costs can be reduced.

In this embodiment, the neural network means 9 is used as the artificial intelligence means 5 and before the process management device 1 is operated, the relationship between each principal component score “Z _m ” and each failure factor is learned. When the process management apparatus 1 is in operation, the principal component score “Z _m ” corresponding to the process performance data on the air conditioning process 101 side is analyzed to determine whether there is an abnormality and to estimate the failure factor. So, without using a complicated normal / abnormal judgment program, you can efficiently discriminate and estimate the normal operation state and the failure state with only a simple operation of learning. System performance verification, failure detection, failure prediction, and failure diagnosis can be performed with high accuracy, energy consumption can be reduced, and maintenance costs can be reduced.

In this embodiment, before the process management apparatus 1 is operated, the simulation means 3 performs simulation processing to calculate process result data in a normal state and process result data in an abnormal state. The process variable data “a ₁₁ , a ₁₂ ,..., A _mp ” is obtained in the principal component analysis means 8 by using the process result data in the normal state and the process result data in the abnormal state, and the neural network Since the weight parameter of the neural network is obtained by the means 9, before the process management apparatus 1 is operated, a simulation is performed, and the process variable coupling weight parameters “a ₁₁ , a ₁₂ ,..., A _mp ”, The neural network weight parameters and the like are obtained, and the process management apparatus 1 When the operation was, these process variables binding weight parameter _{"a 11, a 12, ...} , a mp", by using a weighting parameter of the neural network to analyze the process performance data, efficiently normal operating condition and a fault condition It can discriminate and estimate the cause when a failure occurs, and can perform system performance verification, failure detection, failure prediction, and failure diagnosis with high accuracy, reduce energy consumption, and reduce maintenance costs. it can.

Further, in this embodiment, when the operation period of the process management apparatus 1 becomes long and the process result data storage unit 2 has sufficiently accumulated normal process result data and failure process result data, the principal component analysis unit 8, the process result data at the normal time and the process result data at the time of failure stored in the process result data storage means 2 are analyzed, and the process variable combination weight parameters “a ₁₁ , a ₁₂ ,..., A _mp ”, normal time of the main component _"Z m", updates the main component _{"Z m"} at the time of failure, as a main component _{"Z m"} at the time of a new normal, based on the main component _{"Z m"} at the time of new fault Since the neural network means 9 is trained to update the weight parameters of the neural network, the operation period of the process management device 1 is long. , Aging, due to such a change in control contents, when the process performance data has changed, in response to this, the process variable connection weight parameter _{"a 11, a 12, ...} , a mp", the normal efficiency principal component scores "Z _m", principal component scores of failure "Z _m", update the weighting parameters of the neural network, aging, while addressing the like change in control contents, a fault condition to the normal operation state It can be well discriminated and estimated, and when the failure occurs, the cause can be estimated, and system performance verification, failure detection, failure prediction and failure diagnosis can be performed automatically and accurately, reducing energy consumption and maintenance costs. .

  In this embodiment, when the air-conditioning process 101 is selected as a process to be managed, room temperature (with measurement noise) that can be measured with a normal measuring instrument, cold water (hot water) operation valve output, Since only the humidity, the outside air temperature, and the outside air humidity are used to determine whether or not the principal component score “” of the air conditioning process 101 side is broken, the normal air conditioning process 101 does not use a special sensor. It is possible to efficiently discriminate and estimate most of the failures that occur, and to estimate the causes when failures occur, to perform system performance verification, failure detection, failure prediction, and failure diagnosis with high accuracy, and to reduce energy consumption, Maintenance costs can be reduced.

  In this embodiment, when the air-conditioning process 101 is designated as a process to be managed, the artificial intelligence means 9 is roughly divided into five systems, a water system, an air system, a heat exchanger system, a heat source system, Since the failure factor is estimated, most failures occurring in the normal air conditioning process 101 can be discriminated and estimated efficiently. When a failure occurs, the cause is estimated to verify system performance, failure detection, failure prediction, failure Diagnosis can be performed with high accuracy, energy consumption can be reduced, and maintenance costs can be reduced.

<< Other embodiments >>
In addition, this invention is not limited to the said embodiment, It can implement variously in the range which does not deviate from the summary.

  Further, in the above-described embodiment, the failure detection, failure prediction, and failure diagnosis functions of the BEMS device that controls the air conditioning process of each building are used as specific examples. However, it is obvious that the present invention can be applied to general processes. is there.

  In addition, as a normal / abnormal determination algorithm, an algorithm that is determined based on whether or not a preset threshold value is deviated is used. An algorithm for determining normal / abnormal using an instantaneous value is used. It may be used. However, if such a judgment algorithm is used, there is a possibility of misjudgment due to the influence of data fluctuation due to “Noise Factor”. It is also possible to reduce the risk of misjudgment caused by noise by combining with an algorithm that evaluates the degree of normality and abnormality.

  In the above-described embodiment, the present invention has been described by taking as an example a BEMS device that remotely monitors and controls each air-conditioning facility installed in a plurality of buildings. You may carry out with. In this case, the process management apparatus according to the present invention is installed in the building, not remote.

The block diagram which shows the example of whole structure of the process management apparatus by this invention. The flowchart which shows the operation example of the process simulation means shown in FIG. The schematic diagram which shows the operation example of the neural network means shown in FIG. The table used when the process management apparatus shown in FIG. 1 processed normal data and air supply fan abnormality data to check normality / abnormality judgment capability. The trend graph which shows the example of a main component score when the data at the time of normality and the data at the time of an air supply fan abnormality are processed into the process management apparatus shown in FIG. In the trend graph shown in FIG. No. 4 corresponding to the main component score portion of No. 4, the data when the air supply fan is abnormal. The trend graph which expanded each principal component score part of 4. The trend graph which shows the example of a principal component score when the data at the time of normality and the data at the time of cooling water operation valve abnormality are processed into the process management apparatus shown in FIG. In the trend graph shown in FIG. No. 2 corresponding to the principal component score part 2 and data when the air supply fan is abnormal. The trend graph which expanded each principal component score part of 2. The schematic block diagram which shows the example of an air-conditioning process managed by the process management apparatus by this invention.

Explanation of symbols

  1: process management device, 2: process performance data storage means, 3: statistical analysis means, 4: analysis data storage means, 5: artificial intelligence means, 6: process simulation means, 7: simulation data storage means, 8: principal component Analysis means, 9: Neural network means, 10: Measurement monitoring control means, 11: Network, 101: Air conditioning process, 102: Damper on the intake side, 103: Damper on the exhaust side, 104: Damper on the circulation side, 105: Air conditioner 106: Indoor, 107: Blower, 108: Refrigerant fan, 109: Communication line, 110: Thermometer, 111: Hygrometer, 112: Controller.

Claims (7)

Statistical analysis means for statistically analyzing multiple variables obtained by measuring processes using analysis parameters, reducing the number of variables and generating statistical analysis data,
Using the determination parameters obtained by the learning process, analyzing the statistical analysis data output from the statistical analysis means, determining the presence or absence of an abnormality, and artificial intelligence means for estimating a failure factor;
A process management apparatus comprising: The process management apparatus according to claim 1, wherein
The statistical analysis means uses a process variable coupling weight parameter as the analysis parameter, and converts the multi-variable into the statistical analysis data by principal component analysis using the process variable coupling weight parameter.
A process management apparatus characterized by that. In the process management device according to any one of claims 1 and 2,
The artificial intelligence means uses a weight parameter of a neural network as the determination parameter, determines the presence or absence of abnormality by the neural network, and estimates a failure factor.
A process management apparatus characterized by that. In the process management device according to any one of claims 1 to 3,
A process simulation means for simulating the normal state of the process and various failure states is provided,
Before the operation of the process, based on the simulation data obtained by the process simulation means, determine the analysis parameter or the determination parameter,
A process management apparatus characterized by that. In the process management device according to any one of claims 1 to 4,
A process performance data storage means for storing process performance data at normal times and process performance data at the time of failure is provided.
When the process operation period becomes longer and the process performance data at the normal time and the process performance data at the time of the failure are sufficiently accumulated, the process performance data at the normal time and the process performance data at the time of the failure are used to Correcting variable coupling weight parameters or weight parameters of the neural network;
A process management apparatus characterized by that. The process management apparatus according to any one of claims 1 to 5,
When the air conditioning process is specified as the process, as a variable, at least one of room temperature, cold water (hot water) operation valve operation output, room humidity, outside air temperature, outside air humidity is measured.
A process management apparatus characterized by that. The process management apparatus according to any one of claims 1 to 6,
When an air conditioning process is designated as the process, the artificial intelligence means roughly divides the water system, the air system, the heat exchanger system, the heat source system, and other five factors to estimate the failure factor.
A process management apparatus characterized by that.

Images