Intelligent monitoring method, device, equipment and storage medium for thermal power unit

Abstract

The present invention discloses a method, device, equipment and storage medium for intelligent monitoring of thermal power units, and relates to the technical field of thermal power equipment failure. The method comprises: constructing a monitoring model of a first thermal power unit according to historical operating condition data of a first thermal power unit; obtaining a first parameter estimation value obtained by predicting a parameter of the first thermal power unit by the monitoring model of the first thermal power unit according to operation and maintenance information of the first thermal power unit; optimizing the monitoring model of the first thermal power unit according to the first parameter estimation value, the first parameter actual value and the parameter matrix value when the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power unit satisfies a preset similarity difference, and obtaining an optimized monitoring model of the first thermal power unit. After the operation and maintenance of the thermal power equipment, the present invention optimizes the monitoring model of the first thermal power unit, which can save a lot of costs and improve monitoring efficiency and adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power equipment failure, and in particular to a method, device, equipment and storage medium for intelligent monitoring of a thermal power unit. Background Art

[0002] As information and digital technologies gradually mature in the field of power production, coupled with the current continued rise in demand for clean, efficient and sustainable energy, smart power plants have gradually become a solution for the intelligent development of thermal power under the background of new power, and thermal power intelligent monitoring has become a typical application of smart power plants.

[0003] In terms of technical route, machine learning algorithms are usually used to establish a prediction model for unit system parameters (i.e., a monitoring and early warning model). However, after normal degradation of thermal power equipment, routine equipment maintenance, and equipment overhaul, the monitoring and early warning model cannot adapt to the operating conditions of the equipment and system, resulting in a large number of false alarms after the model has been running for a period of time. The commonly used method is to collect historical data of the equipment and system operation for one year and re-model it, but the number of intelligent monitoring models for key equipment and systems in power plants is huge. Directly abandoning the existing model and retraining a new model will consume a lot of time and manpower costs, which is not an efficient method. Summary of the invention

[0004] The purpose of the present invention is to provide a method, device, equipment and storage medium for intelligent monitoring of thermal power units, thereby solving the problem that the monitoring and early warning model established in the prior art cannot adapt to the operating conditions of thermal power unit equipment.

[0005] In a first aspect, in order to achieve the above-mentioned purpose, an embodiment of the present invention provides a method for intelligent monitoring of a thermal power unit, comprising:

[0006] According to the historical operating data of the operating conditions of the first thermal power equipment, a monitoring model of the first thermal power unit is constructed;

[0007] According to the operation and maintenance information of the first thermal power equipment, obtaining a first parameter estimation value obtained by predicting the parameter of the first thermal power equipment by the first thermal power unit monitoring model;

[0008] When the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference, the first thermal power unit monitoring model is optimized according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain an optimized first thermal power unit monitoring model;

[0009] The first thermal power unit monitoring model stores the parameters of the first thermal power equipment and parameter matrix values ​​of the parameters.

[0010] Optionally, the method further comprises:

[0011] In the case where the first condition is met, determining that the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference;

[0012] The first condition includes:

[0013] Within a preset time period, the average of the similarities between the first parameter estimation values ​​of all parameters of the first thermal power equipment and the first parameter actual values ​​of the corresponding parameters is less than a first threshold;

[0014] Within a preset time period, the similarity between a first parameter estimation value of any parameter of the first thermal power equipment and a first parameter actual value of the corresponding parameter is less than a second threshold;

[0015] The first threshold is greater than the second threshold.

[0016] Optionally, optimizing the first thermal power unit monitoring model according to the first parameter estimated value, the first parameter actual value and the parameter matrix value to obtain an optimized first thermal power unit monitoring model includes:

[0017] Using the first parameter estimation value as the input of a preset neural network model, using the first parameter actual value as the output of the preset neural network model, training the preset neural network model to obtain a trained neural network model;

[0018] The first thermal power unit monitoring model is optimized according to the trained neural network model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model.

[0019] Optionally, optimizing the first thermal power unit monitoring model according to the trained neural network model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model includes:

[0020] Inputting the parameter matrix value into the trained neural network model to obtain a first parameter prediction value output by the trained neural network model;

[0021] The first thermal power unit monitoring model is optimized according to the first parameter prediction value to obtain the optimized first thermal power unit monitoring model.

[0022] Optionally, when the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference, the first thermal power unit monitoring model is optimized according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain the optimized first thermal power unit monitoring model, including:

[0023] In a case where the similarity between the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter satisfies the preset similarity difference, constructing a fitting model by using a fitting algorithm according to the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter, wherein the ratio between the number of the first target parameters and the total number of parameters of the first thermal power equipment is less than a preset ratio;

[0024] The first thermal power unit monitoring model is optimized according to the fitting model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model.

[0025] Optionally, the fitting algorithm includes at least one of the following:

[0026] Polynomial fitting algorithm;

[0027] Exponential function fitting algorithm.

[0028] Optionally, the method further comprises:

[0029] Acquire a second parameter actual value of a parameter of a second thermal power device, wherein the second thermal power device is of the same type as the first thermal power device;

[0030] Inputting the parameter of the second thermal power equipment into the monitoring model of the first thermal power unit to obtain a second parameter estimation value of the parameter of the second thermal power equipment output by the monitoring model of the first thermal power unit;

[0031] Using the estimated value of the second parameter as an input of a preset neural network model, using the actual value of the second parameter as an output of the preset neural network model, training the preset neural network model to obtain a trained neural network model;

[0032] A second thermal power unit monitoring model of the second thermal power equipment is obtained according to the trained neural network model and the parameter matrix value.

[0033] Optionally, obtaining a second thermal power unit monitoring model of the second thermal power equipment according to the trained neural network model and the parameter matrix value includes:

[0034] Inputting the parameter matrix value into the trained neural network model to obtain a second parameter prediction value output by the trained neural network model;

[0035] A second thermal power unit monitoring model of the second thermal power equipment is obtained according to the predicted value of the second parameter and the parameter of the second thermal power equipment.

[0036] Optionally, the preset neural network model includes at least one of the following:

[0037] Back propagation BP neural network;

[0038] Convolutional Neural Network (CNN);

[0039] Long Short-Term Memory Network LSTM.

[0040] In a second aspect, an embodiment of the present invention further provides an intelligent monitoring device for a thermal power unit, comprising:

[0041] A first processing module is used to construct a monitoring model of the first thermal power unit according to the historical operating condition data of the operating condition of the first thermal power equipment;

[0042] A second processing module is used to obtain, according to the operation and maintenance information of the first thermal power equipment, a first parameter estimation value obtained by predicting the parameters of the first thermal power equipment by the first thermal power unit monitoring model;

[0043] A third processing module is used to optimize the first thermal power unit monitoring model according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain an optimized first thermal power unit monitoring model when the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment meets a preset similarity difference;

[0044] The first thermal power unit monitoring model stores the parameters of the first thermal power equipment and parameter matrix values ​​of the parameters.

[0045] In a third aspect, an embodiment of the present invention further provides an intelligent monitoring device for a thermal power unit, comprising: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps in the intelligent monitoring method for a thermal power unit as described in any one of the first aspects are implemented.

[0046] In a fourth aspect, an embodiment of the present invention further provides a readable storage medium having a program stored thereon, and when the program is executed by a processor, the steps in the intelligent monitoring method for a thermal power unit as described in any one of the first aspects are implemented.

[0047] In a fifth aspect, an embodiment of the present invention further provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the method for intelligent monitoring of a thermal power unit as described in any one of the first aspects.

[0048] The above technical solution of the present invention has at least the following beneficial effects:

[0049] The intelligent monitoring method for a thermal power unit in an embodiment of the present invention obtains historical operating condition data of an operating condition of any thermal power equipment (i.e., a first thermal power equipment), and constructs a first thermal power unit monitoring model for the first thermal power equipment according to parameters of the first thermal power equipment and the historical operating condition data of each parameter. The first thermal power unit monitoring model is used to monitor and warn the operating condition of the first thermal power equipment, obtain operation and maintenance information of the first thermal power equipment, and the operation and maintenance information is used to indicate the operation and maintenance status of the first thermal power equipment. According to the operation and maintenance information, a first parameter estimation value for predicting the parameter of the first thermal power equipment by the first thermal power unit monitoring model is obtained, and the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment is determined. When the similarity reaches a preset similarity difference, the first thermal power unit monitoring model is optimized according to the first parameter estimation value and the first parameter actual value to obtain the optimized first thermal power unit monitoring model. That is, after the operation and maintenance of the thermal power equipment, if the similarity gap between the first parameter estimation value of the parameter of the first thermal power equipment and the first parameter actual value of the parameter of the first thermal power equipment is too large, the monitoring model of the first thermal power unit is optimized to adapt to the operating conditions of the thermal power unit equipment again. Compared with collecting historical operating condition data again and re-training the thermal power unit monitoring model, a lot of costs can be saved and monitoring efficiency and adaptability can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flow chart of a method for intelligent monitoring of a thermal power unit provided by an embodiment of the present invention;

[0051] Figure 2 A flowchart of constructing a monitoring model for the first thermal power unit and performing equipment early warning provided in an embodiment of the present invention;

[0052] Figure 3 A flow chart for optimizing the monitoring model of the first thermal power unit provided in an embodiment of the present invention;

[0053] Figure 4 A schematic diagram of a neural network topology structure provided by an embodiment of the present invention;

[0054] Figure 5 A schematic diagram of a construction process of a second thermal power unit monitoring model of a second thermal power plant provided in an embodiment of the present invention;

[0055] Figure 6The original data of the induced draft fan blade position inversion provided by the embodiment of the present invention and the scatter point comparison diagram after the outliers are removed;

[0056] Figure 7 The original data of the motor current of the induced draft fan provided in the embodiment of the present invention and the scatter point comparison diagram after the abnormal values ​​are eliminated;

[0057] Figure 8 The original data of the induced draft fan front bearing temperature 1 provided in the embodiment of the present invention and the scatter point comparison diagram after the abnormal values ​​are eliminated;

[0058] Fig. 9 The original data of the front bearing temperature 1 of the induced draft fan motor provided in the embodiment of the present invention and the scatter point comparison diagram after the abnormal values ​​are eliminated;

[0059] Fig.10 A scatter plot comparison of normal data of the induced draft fan blade position reversal provided in an embodiment of the present invention and data for constructing a historical memory matrix;

[0060] Fig.11 A scatter plot comparison of normal data of the induced draft fan motor current provided by an embodiment of the present invention and data for constructing a historical memory matrix;

[0061] Fig.12 A scatter plot comparison of normal data of the induced draft fan front bearing temperature 1 provided in an embodiment of the present invention and data for constructing a historical memory matrix;

[0062] Fig.13 A scatter plot comparison of normal data of the induced draft fan motor front bearing temperature 1 and data for constructing a historical memory matrix provided by an embodiment of the present invention;

[0063] Fig.14 A comparison chart of the actual value of the induced draft fan blade position reversal in 2023 and the MEST predicted value provided by an embodiment of the present invention;

[0064] Fig.15 A comparison chart of the actual value of the induced draft fan blade position reversal in 2024 and the MEST predicted value provided by an embodiment of the present invention;

[0065] Fig.16 A comparison chart of the actual value of the induced draft fan motor current in 2023 and the MEST predicted value provided by an embodiment of the present invention;

[0066] Fig.17 A comparison chart of the actual value of the induced draft fan motor current in 2024 and the MEST predicted value provided by an embodiment of the present invention;

[0067] Fig.18 A comparison chart of the actual value of the induced draft fan front bearing temperature 1 in 2023 and the MEST predicted value provided by an embodiment of the present invention;

[0068] Fig.19 A comparison chart of the actual value of the induced draft fan front bearing temperature 1 in 2024 and the MEST predicted value provided by an embodiment of the present invention;

[0069] Fig. 20 A comparison chart of the actual value of the induced draft fan motor front bearing temperature 1 in 2023 and the MEST predicted value provided by an embodiment of the present invention;

[0070] Fig.21 A comparison chart of the actual value of the induced draft fan motor front bearing temperature 1 in 2024 and the MEST predicted value provided by an embodiment of the present invention;

[0071] Fig. 22 A schematic diagram of the inverse similarity of the induced draft fan blade positions in 2023 provided by an embodiment of the present invention;

[0072] Fig.23 A schematic diagram of the inverse similarity of the induced draft fan blade positions in 2024 provided by an embodiment of the present invention;

[0073] Fig.24 A schematic diagram of similarity of bearing temperature 1 in an induced draft fan in 2023 provided by an embodiment of the present invention;

[0074] Fig.25 A schematic diagram of similarity of bearing temperature 1 in an induced draft fan in 2024 provided by an embodiment of the present invention;

[0075] Fig.26 A schematic diagram of the similarity of the induced draft fan motor current in 2023 provided by an embodiment of the present invention;

[0076] Fig. 27 A schematic diagram of the similarity of induced draft fan motor current in 2024 provided by an embodiment of the present invention;

[0077] Fig.28 A comparison chart of the actual value of the induced draft fan blade position reversal provided in an embodiment of the present invention, the predicted value predicted by the MSET algorithm based on the initial memory matrix, and the predicted value predicted by the MSET algorithm based on the memory matrix optimized by the BP neural network;

[0078] Fig.29 A comparison chart of the actual value of the induced draft fan motor current provided in an embodiment of the present invention, the predicted value predicted by the MSET algorithm based on the initial memory matrix, and the predicted value predicted by the MSET algorithm based on the memory matrix optimized by the BP neural network;

[0079] Fig.30A comparison chart of the actual value of the induced draft fan front bearing temperature 1 provided in an embodiment of the present invention, the predicted value predicted by the MSET algorithm based on the initial memory matrix, and the predicted value predicted by the MSET algorithm based on the memory matrix optimized by the BP neural network;

[0080] Fig.31 A comparison chart of the actual value of the front bearing temperature 1 of the induced draft fan motor provided in an embodiment of the present invention, the predicted value predicted by the MSET algorithm based on the initial memory matrix, and the predicted value predicted by the MSET algorithm based on the memory matrix optimized by the BP neural network;

[0081] Fig.32 The original data of the E-mill current provided in the embodiment of the present invention and the scatter comparison diagram after the outliers are removed;

[0082] Fig.33 The original data of the E-mill inlet and outlet differential pressure provided in the embodiment of the present invention and the scatter point comparison diagram after the outliers are removed;

[0083] Fig.34 The original data of the primary air volume 2 at the E mill inlet provided in the embodiment of the present invention and the scatter point comparison diagram after the outliers are removed;

[0084] Fig.35 The original data of the E-mill motor drive end bearing temperature 1 provided in the embodiment of the present invention and the scatter point comparison diagram after the abnormal values ​​are eliminated;

[0085] Fig.36 A scatter plot comparison of normal data of E-mill current provided by an embodiment of the present invention and data for constructing a historical memory matrix;

[0086] Fig.37 A scatter plot comparison of normal data of the E-mill inlet and outlet differential pressure provided by an embodiment of the present invention and data for constructing a historical memory matrix;

[0087] Fig.38 A scatter plot comparison of normal data of the primary air volume 2 at the E mill inlet provided in an embodiment of the present invention and data for constructing a historical memory matrix;

[0088] Fig.39 A scatter plot comparison of normal data of the E-mill motor drive end bearing temperature 1 and data for constructing a historical memory matrix provided in an embodiment of the present invention;

[0089] Fig.40 A comparison chart of the actual value of the E-mill current provided by the embodiment of the present invention and the MEST predicted value;

[0090] Fig.41 A comparison chart of the actual value of the E-mill inlet and outlet differential pressure provided in an embodiment of the present invention and the MEST predicted value;

[0091] Fig.42 A comparison chart of the actual value of the primary air volume 2 at the E mill inlet and the MEST predicted value provided by an embodiment of the present invention;

[0092] Fig.43 A comparison chart of the actual value of the E-mill motor drive end bearing temperature 1 and the MEST predicted value provided in an embodiment of the present invention;

[0093] Fig.44 A comparison chart of the actual value of the A-mill current provided by the embodiment of the present invention and the MEST predicted value;

[0094] Fig.45 A comparison chart of the actual value of the differential pressure between the inlet and outlet of the A mill provided in an embodiment of the present invention and the MEST predicted value;

[0095] Fig.46 A comparison chart of the actual value of the primary air volume 2 at the inlet of mill A and the MEST predicted value provided in an embodiment of the present invention;

[0096] Fig.47 A comparison chart of the actual value of the bearing temperature 1 at the drive end of the A-mill motor provided in an embodiment of the present invention and the MEST predicted value;

[0097] Fig.48 A schematic diagram of the similarity of primary air volume 2 at the E mill inlet provided in an embodiment of the present invention;

[0098] Fig.49 A schematic diagram of the similarity of primary air volume 2 at the inlet of mill A provided in an embodiment of the present invention;

[0099] Fig.50 A schematic diagram of the current similarity of an E-mill coal feeder provided in an embodiment of the present invention;

[0100] Fig.51 A schematic diagram of the current similarity of the coal feeder of the A mill provided in an embodiment of the present invention;

[0101] Fig.52 A schematic diagram of the similarity of the vibration 1 of the driving end bearing of the E-mill motor provided in an embodiment of the present invention;

[0102] Fig.53 A schematic diagram of the similarity of the vibration 1 of the motor drive end bearing of the A mill provided in the embodiment of the present invention;

[0103] Fig.54 A comparison chart of the actual value of the A mill current provided in the embodiment of the present invention, the predicted value predicted by the MSET algorithm based on the memory matrix of the E mill, and the predicted value predicted by the MEST algorithm based on the memory matrix of the A mill after BP neural network optimization;

[0104] Fig.55A comparison chart of the actual value of the inlet and outlet differential pressure of the A mill provided in an embodiment of the present invention, the predicted value predicted by the MSET algorithm based on the memory matrix of the E mill, and the predicted value predicted by the MEST algorithm based on the memory matrix of the A mill after BP neural network optimization;

[0105] Fig.56 A comparison chart of the actual value of the primary air volume 2 at the inlet of the A mill provided in an embodiment of the present invention, the predicted value predicted by the MSET algorithm based on the memory matrix of the E mill, and the predicted value predicted by the MEST algorithm based on the memory matrix of the A mill after BP neural network optimization;

[0106] Fig.57 A comparison chart of the actual value of the bearing temperature 1 of the motor drive end of the A mill provided in an embodiment of the present invention, the predicted value predicted by the MSET algorithm based on the memory matrix of the E mill, and the predicted value predicted by the MEST algorithm based on the memory matrix of the A mill after BP neural network optimization;

[0107] Fig.58 A schematic diagram of the structure of an intelligent monitoring device for a thermal power unit provided by an embodiment of the present invention;

[0108] Fig.59 A schematic diagram of the structure of an intelligent monitoring device for a thermal power unit provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0109] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail in conjunction with the accompanying drawings and specific embodiments. In the following description, specific details such as specific configurations and components are provided only to help fully understand the embodiments of the present invention. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention. In addition, for clarity and brevity, the description of known functions and structures is omitted.

[0110] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present invention. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0111] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0112] In the embodiments provided by the present invention, it should be understood that "B corresponding to A" means that B is associated with A and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, but B can also be determined according to A and / or other information.

[0113] In order to solve the problem that the monitoring and early warning model established in the prior art cannot adapt to the operating conditions of thermal power unit equipment, an embodiment of the present invention provides a thermal power unit intelligent monitoring method, device, equipment and storage medium.

[0114] like Figure 1 As shown, an embodiment of the present invention provides an intelligent monitoring method for a thermal power unit, comprising:

[0115] Step 101: construct a monitoring model of a first thermal power unit according to historical operating condition data of the operating condition of a first thermal power equipment.

[0116] The first thermal power equipment is any equipment in a thermal power unit, such as an induced draft fan of a thermal power unit or a coal mill of a thermal power unit.

[0117] It should also be noted that the application field of the intelligent monitoring method of thermal power units provided in the embodiment of the present invention is not limited to thermal power units, but can also be applied to intelligent monitoring of other power generating units, such as wind power units, hydropower units, nuclear power units, photovoltaic units, etc.

[0118] In this step, the historical operating condition data of each parameter of the first thermal power equipment is first obtained, for example, the parameter value of each parameter within one year before the current time is obtained as the historical operating condition data (or equipment historical data). Specifically, when obtaining the historical operating condition data, it is obtained from the data stored in the power plant safety instrument system (Safety Instrumented System, SIS) or distributed control system (Distributed Control System, DCS).

[0119] After obtaining the historical operating data of each parameter of the first thermal power equipment, according to the historical operating data corresponding to each parameter, principal component analysis (PCA) and fuzzy c-means algorithm (FCM), referred to as PCA-FCM, are used to construct a historical memory matrix, which is the first thermal power unit monitoring model. The first thermal power unit monitoring model is used to monitor the operating status of the first thermal power equipment and conduct monitoring and early warning of the operating status of the thermal power equipment. Afterwards, the multivariate state estimation technique (MEST) algorithm is used to calculate the parameter estimation value of the parameter of the first thermal power equipment predicted in the first thermal power unit monitoring model, and the residual between the parameter estimation value and the actual parameter value of the corresponding parameter is calculated, and whether to generate an alarm message is determined according to the set residual threshold.

[0120] Specifically, when determining whether to generate an alarm message, a sliding window technique is used. First, the residual between the estimated parameter value and the actual parameter value of the corresponding parameter is calculated. A residual sequence is generated based on the residual and the corresponding window. The residual sequence is statistically analyzed to obtain a statistical analysis result. Based on the statistical analysis result, it is determined whether the statistical analysis result satisfies the residual being higher than a preset threshold. If so, the unit intelligent monitoring and early warning is realized, that is, an alarm message is generated.

[0121] Among them, the first thermal power unit monitoring model constructed includes the parameters of the first thermal power equipment and the parameter matrix values ​​of each parameter at different operating times, or the first thermal power unit monitoring model constructed includes the parameters of the first thermal power equipment and the parameter matrix values ​​of each parameter. It can also be understood that the historical memory matrix of the first thermal power unit monitoring model is composed of the parameters of the first thermal power equipment and the parameter matrix values ​​of the parameters, or the historical memory matrix of the first thermal power unit monitoring model is composed of the parameters of the first thermal power equipment and the parameter matrix values ​​of the parameters at different operating times.

[0122] The parameter matrix values ​​in the monitoring model of the first thermal power unit can also be understood as historical operating condition data or processed historical operating condition data, and the processed historical operating condition data is used to generate the monitoring model of the first thermal power unit.

[0123] Step 102: According to the operation and maintenance information of the first thermal power equipment, obtain a first parameter estimation value obtained by predicting the parameters of the first thermal power equipment by the first thermal power unit monitoring model. The operation and maintenance information of the first thermal power equipment includes at least one of the following:

[0124] Normal degradation information of the first thermal power equipment;

[0125] Daily maintenance information of the No.1 thermal power equipment;

[0126] The first thermal power equipment maintenance information.

[0127] After obtaining the operation and maintenance information of the first thermal power equipment, when there is no fault in the thermal power equipment, the monitoring model of the first thermal power unit may fail, resulting in a large deviation between the estimated value of the parameter of the first thermal power equipment predicted by the first thermal power unit monitoring model (i.e., the first estimated value) and the actual value of the parameter of the corresponding parameter, which means that the monitoring and early warning of the first thermal power equipment by the first thermal power unit monitoring model may be inaccurate. Therefore, in this step, based on the operation and maintenance information of the first thermal power equipment, when it is determined that there is no equipment fault in the first thermal power equipment, the first parameter estimated value of the parameter of the first thermal power equipment by the first thermal power unit monitoring model is obtained.

[0128] When obtaining the first parameter estimation value of the first thermal power equipment's parameter by the first thermal power unit monitoring model, the MEST technology is used to obtain the first parameter estimation value obtained by predicting the first thermal power equipment's parameter by the first thermal power unit monitoring model.

[0129] Step 103: When the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment (the first parameter actual value is the actual value of the parameter of the first thermal power equipment after operation and maintenance) meets the preset similarity difference, the first thermal power unit monitoring model is optimized according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain the optimized first thermal power unit monitoring model. The first thermal power unit monitoring model stores the parameters of the first thermal power equipment and the parameter matrix values ​​of the parameters.

[0130] In this step, the similarity between the first parameter estimation value of the parameter of the first thermal power equipment and the actual parameter value of the corresponding parameter is first obtained, wherein the first parameter estimation value of the parameter of the first thermal power equipment and the actual parameter value of the corresponding parameter refer to the parameter estimation value and the actual parameter value of the same parameter of the first thermal power equipment.

[0131] Similarity is a measure of the similarity between two samples, which can reflect the difference between the two samples.

[0132] In this step, the preset similarity difference can be understood as a large difference between the similarity between the first parameter estimation value and the first parameter actual value of the corresponding parameter.

[0133] When the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment meets the preset similarity difference, it means that the first thermal power unit model is invalid and cannot accurately monitor and warn the first thermal power equipment after operation and maintenance. At this time, it is necessary to update and optimize the first thermal power unit model so that the optimized first thermal power unit monitoring model can timely and effectively monitor and warn the first equipment after operation and maintenance. Specifically, in this embodiment, when optimizing the first thermal power unit monitoring model, the first thermal power unit model is directly optimized according to the first parameter estimation value and the first parameter actual value to adapt to the operating conditions of the thermal power unit equipment again. Compared with collecting the historical operating condition data of the parameters of the first thermal power equipment again, training the thermal power unit monitoring model again according to the re-acquired historical operating condition data can save a lot of costs and improve monitoring efficiency and adaptability.

[0134] In an optional embodiment, the method further includes:

[0135] In the case where the first condition is met, determining that the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference;

[0136] The first condition includes:

[0137] Within a preset time length, the average of the similarities between the first parameter estimation values ​​of all parameters of the first thermal power equipment and the first parameter actual values ​​of the corresponding parameters is less than a first threshold value, that is, it can be understood that the average of the similarities between the estimation values ​​and the actual values ​​of all parameters of the first thermal power equipment is less than the first threshold value within the preset time length;

[0138] Within a preset time length, the similarity between the first parameter estimation value of any parameter of the first thermal power equipment and the first parameter actual value of the corresponding parameter is less than a second threshold value, that is, it can be understood that the similarity between the estimation value and the actual value of any parameter of the first thermal power equipment is less than the second threshold value within the preset time length;

[0139] The first threshold is greater than the second threshold. The first threshold and the second threshold are both values ​​less than 1. Optionally, the first threshold is 95% and the second threshold is 90%.

[0140] Optionally, the preset duration may be between 10 minutes and 30 minutes.

[0141] Among them, similarity is a measure of the similarity between two samples, which can reflect the difference between the two samples. According to the estimated value and actual value of each predicted parameter, the similarity between the estimated value and the actual value of the parameter can be calculated (or the health value score). The calculation formula of similarity is as follows:

[0142]

[0143] in, represents the similarity between a parameter estimation value (first parameter estimation value) and an actual value of the parameter (first parameter actual value) of a parameter of the first thermal power equipment, represents the actual value of a parameter (also called observed value, parameter observation value), represents the parameter estimate of the parameter, n Indicates the first device n parameters.

[0144] The following is a detailed description of the process of building the monitoring model of the first thermal power unit:

[0145] The step of constructing a monitoring model of the first thermal power unit according to the historical operating condition data of the operating condition of the first thermal power equipment comprises:

[0146] Acquire historical operating condition data of parameters of the first thermal power equipment under operating conditions of the first thermal power equipment;

[0147] Preprocessing the historical operating condition data to obtain first operating condition data;

[0148] According to the first operating condition data, a monitoring model of the first thermal power unit is constructed.

[0149] Optionally, preprocessing the historical operating condition data to obtain first operating condition data includes:

[0150] An abnormal data cutoff value is set in the historical operating condition data, and first abnormal data is deleted in the historical operating condition data according to the abnormal data cutoff value to obtain second operating condition data, wherein the first abnormal data includes start and stop data and equipment maintenance data of the first thermal power equipment;

[0151] Using a local outlier factor (LOF) algorithm, second abnormal data in the second operating condition data is detected, and the second abnormal data is deleted from the second operating condition data to obtain third operating condition data;

[0152] The third operating condition data is normalized to obtain the first operating condition data.

[0153] Optionally, constructing the first thermal power unit monitoring model according to the first operating condition data includes:

[0154] Performing dimensionality reduction processing on the first operating condition data by using a PCA algorithm to obtain fourth operating condition data;

[0155] Performing clustering processing on the fourth operating condition data by using an FCM algorithm to obtain first clustering data corresponding to each parameter of the first thermal power equipment;

[0156] Performing group sampling in the first cluster data to obtain sampling data of each parameter;

[0157] According to the parameters of the first thermal power equipment and the corresponding sampling data, a historical memory matrix is ​​constructed, and the historical memory matrix is ​​the monitoring model of the first thermal power unit.

[0158] The following is a detailed description of the principle of using the LOF algorithm to detect abnormal data:

[0159] Data quality plays a crucial role in model performance. Therefore, in the initial data cleaning stage, a strategy based on setting thresholds was adopted to remove obvious outliers in the data set. However, in order to capture potential minor anomalies more accurately, the local outlier factor LOF algorithm was introduced. The LOF algorithm is a density-based anomaly detection method, and its core is to identify outliers by comparing the local density difference between a data point and its neighboring data points. Unlike traditional methods that rely on global statistical models, the LOF algorithm makes judgments based on the unique characteristics of the local neighborhood around each data point, so it is particularly suitable for processing data sets with nonlinear and non-uniform distribution characteristics. The basic principle is that normal data points are usually located in high-density areas, while outliers are more likely to appear in low-density areas or areas that are significantly isolated from other points.

[0160] For a given data sample set D Any sample point in p ,The calculation process of LOF algorithm is described as follows.

[0161] 1. Calculate k-neighbor distance:

[0162] Define sample points p With point o The direct distance between For the sample point p ,That k - The proximity distance is from p To its k The distance to the nearest neighbor is denoted by .

[0163] 2. Build k -Distance Neighborhood:

[0164] Sample point p is the center of the circle, k The neighborhood with distance radius is defined as k -Distance neighborhood, denoted as . This neighborhood contains allp The distance is no greater than The points:

[0165]

[0166] 3. Calculation k- Reachable distance:

[0167] Sample points p About sample points o of k The reachable distance is the k Distance or sample point p With this point o The larger value of the distance between samples. p About the sample o of k The reachable distance is expressed as:

[0168]

[0169] 4. Calculation k - Locally accessible density:

[0170] Sample points p of k The local reachability density is its k The distance from all points in the neighborhood k The reciprocal of the sum of the reachable distances, divided by k The number of points in the distance neighborhood, which is the inverse of the average reachable distance. The formula for local reachability density is:

[0171]

[0172] From the above formula, we can see that the smaller the average reachable distance, the greater the local reachable density of the sample, and it is likely to belong to the same cluster; otherwise, it may be an outlier.

[0173] 5. Calculation k - Local abnormal factors:

[0174] Sample points p of k The local abnormal factor is k The average local reachability density in the distance neighborhood is equal to the point k The ratio of local reachable density. The formula is:

[0175]

[0176] When the value is close to 1, the sample p Its neighboring points belong to the same cluster; when it is less than 1, p is a relatively dense point; when it is greater than 1, pThe larger the value, the higher the abnormality.

[0177] The following is a detailed description of the principle of the PCA algorithm:

[0178] Principal component analysis (PCA) is an efficient data dimensionality reduction technique that aims to project high-dimensional data into a lower-dimensional subspace through linear transformation while maximizing the retention of key data information, which is mainly reflected in the variance characteristics of the data. PCA deeply analyzes the original features and reconstructs mutually orthogonal (independent) principal components. These principal components, as linear combinations of the original data in various dimensions, achieve effective data compression and information retention. The PCA algorithm searches for a series of orthogonal coordinate axes in descending order of variance, so that most of the variance is concentrated in the front k The variance of the subsequent coordinate axes is small and is often ignored. In practical applications, the previous coordinate axes with larger variance are selectively retained. k The coordinate axis not only simplifies the data model and reduces the computational complexity, but also enhances the interpretability and visualization of the data.

[0179] The following is the calculation process of the PCA algorithm, for a data set containing M N-dimensional samples ,in Representative i samples, and the goal is to reduce the data dimension to N' dimensions.

[0180] Step 1: Data preprocessing — decentralization:

[0181] To simplify subsequent calculations, we first need to decentralize the sample data to ensure that the mean of the data in each dimension is 0. The specific operations of decentralization are as follows:

[0182]

[0183] in, is the mean vector of all samples:

[0184]

[0185] Step 2: Construct the covariance matrix:

[0186] Covariance matrix S Describes the correlation between the dimensions in the sample. For the decentralized sample matrix , the covariance matrix S It can be expressed as:

[0187]

[0188] in, is a M ×N A matrix where each row is a decentralized sample vector.

[0189] Step 3: Eigendecomposition covariance matrix:

[0190] Covariance matrix S Perform eigendecomposition to extract its principal components. The purpose of eigendecomposition is to find the eigenvalues ​​and eigenvectors of the covariance matrix. Let the eigenvalues ​​be , the corresponding eigenvector is , then:

[0191]

[0192] Eigenvalue Indicates that the data is The variance in a direction reflects the information content in that direction; and the eigenvector are the unit vectors in these directions.

[0193] Step 4: Select the principal components and construct the projection matrix:

[0194] According to the dimension after dimension reduction , select the largest The eigenvalues ​​corresponding to the eigenvectors These eigenvectors form the projection matrix W :

[0195]

[0196] Step 5: Project the data into low-dimensional space:

[0197] Using the projection matrix W , project the original data into a low-dimensional space to obtain the reduced-dimensional data. To project:

[0198]

[0199] in, yes dimensional vector, representing the sample after dimensionality reduction.

[0200] Finally, the sample data set after dimensionality reduction is formed .

[0201] The following is a detailed description of the principle of the fuzzy C-means clustering algorithm:

[0202] Fuzzy C-means clustering (FCM), proposed by Bezdek in 1973, is unique in that it allows data points to belong to multiple clusters at the same time. The core of the algorithm is to construct a membership matrix, which aims to maximize the similarity within the cluster and minimize the similarity between clusters, so as to determine the degree of belonging of each data point to each cluster. In the FCM algorithm, the membership of each data point to each cluster is a real number between 0 and 1, which represents the degree or probability of the data point belonging to the cluster.

[0203] The core goal of the fuzzy C-means clustering algorithm is to solve the membership matrix U and cluster centers V , to minimize the objective function J ( U , V ), which measures the weighted sum of the squared distances from a sample to its cluster center:

[0204]

[0205] Where: is the membership matrix, satisfying ; is the cluster center, where ; Representation sample To cluster center The distance ; m is the fuzzy weighted index.

[0206] The specific steps of the fuzzy C-means clustering algorithm are as follows.

[0207] 1. Initialization:

[0208] Set the number of clusters , fuzzy weighted index , Iteration termination tolerance , maximum iteration step Initialize the membership moment ,make Indicates the first iteration.

[0209] 2. Update cluster centers:

[0210] Calculate the The cluster center of the step :

[0211]

[0212] 3. Update the membership matrix:

[0213] Modified membership matrix , calculate the objective function value :

[0214] ;

[0215]

[0216] in, .

[0217] 4. Iteration judgment:

[0218] The iteration stops if any of the following conditions is met:

[0219] (membership termination tolerance);

[0220] and (Objective function termination tolerance);

[0221] (maximum iteration step size);

[0222] otherwise , and return to step 2 to update the cluster center.

[0223] 5. Final Result:

[0224] Get the final membership matrix and cluster centers , so that the objective function The value of reaches the minimum. According to the final membership matrix The values ​​of the elements in can determine the attribution of all samples. When kind.

[0225] Optionally, after constructing the first thermal power unit monitoring model, the method further includes:

[0226] Using the MEST technology, obtaining a first parameter estimation value obtained by predicting the parameter of the first thermal power equipment by the first thermal power unit monitoring model;

[0227] Obtaining an actual value of a first parameter of a parameter of the first thermal power equipment;

[0228] Using the MEST technique, a residual value between the estimated value of the first parameter and the actual value of the first parameter is obtained, and a residual sequence is generated according to the residual value;

[0229] Using the sliding window technology, the residual sequence is statistically analyzed according to the preset sliding window to obtain the statistical analysis results;

[0230] Based on the statistical analysis results, when it is determined that the residual value is higher than a preset threshold, a fault warning is issued for the first thermal power equipment. Based on the statistical analysis results, when it is determined that the residual value is lower than or equal to the preset threshold, it is determined that the first thermal power equipment is normal.

[0231] The principle of multivariate state estimation technology (MSET) is described in detail below:

[0232] Multivariate state estimation technology (MSET) is a nonlinear multivariate predictive diagnostic technology proposed by Singer et al. This technology builds a historical memory matrix based on the health data of the equipment during normal operation. The matrix reflects the complex relationship between the various parameters of the equipment under normal working conditions. During actual operation, MSET receives the parameters monitored by the equipment in real time, and compares and analyzes them with the health data in the historical memory matrix. It evaluates the equipment status by calculating the deviation between the real-time status and the estimated status.

[0233] MSET uses a weight vector to measure the similarity between actual data and healthy data to reflect the closeness of the current state of the device to the normal state. When the device fails or is abnormal, the deviation between the real-time data and the healthy data increases and the similarity decreases. The advantage of MSET lies in its nonlinearity and multivariate nature. It can handle the complex relationship between device status and parameters and consider the influence of multiple parameters at the same time, so as to more accurately evaluate the device status.

[0234] For a device, if there is n interrelated monitoring variables (one variable corresponds to one parameter), then at time t i When monitoring, this n variables can be combined into an observation vector. , expressed as follows:

[0235] Formula 1

[0236] By analyzing the historical operating data and extracting features of the equipment, a historical memory matrix of the equipment operation is constructed. n Measurement points (one measurement point corresponds to one parameter of the device) m The historical memory matrix of states (a state can be understood as the parameter values ​​of all parameters at a certain moment) is denoted as D , expressed as follows:

[0237] Formula 2

[0238] This matrix contains n The measuring points are m Data in each state. m represents the number of observed states,n Represents the number of monitoring parameters, Indicates n The variables in The value of the moment.

[0239] Observation vector of real-time operation of the device X obs (It can also be understood as the actual values ​​of multiple parameters) as the input parameters of the model (historical memory matrix), the observation vector X obs and stored in the memory matrix D By using similarity theory and linearly weighting each state in the memory matrix, an estimate of the observation vector under normal operating conditions is obtained. X est The estimated value can be used as a reference for normal operation and to determine whether the current observation vector is similar to the normal operation state or has an anomaly. In this way, the state estimation and anomaly detection of the equipment can be achieved.

[0240] Assume the observation vector is X obs , the estimated vector obtained by the MSET algorithm is X est , the formula is as follows:

[0241] Formula 3

[0242] Among them, the vector W Represents the weight vector corresponding to each state in the observation vector. It reflects the similarity between the current state of the observation vector and the memory matrix. Weight vector W The calculation of is based on evaluating the similarity between the observation vector and each state in the memory matrix, and assigning different weights to each state according to the degree of similarity.

[0243] To determine the weight W , the residual can be minimized, and the residual of the observed vector and the estimated vector is:

[0244] Formula 4

[0245] Using the least squares principle, minimizing the residual can determine the weight vector W , the formula is as follows:

[0246] Formula 5

[0247] Then we get the estimated value of the observation vector X est The calculation formula is as follows:

[0248] Formula 6

[0249] Since the correlation between the process memory matrix data may cause the matrix to be irreversible, formula 6 cannot be calculated normally. In actual application scenarios, nonlinear operators are used instead. To replace the conventional matrix multiplication, in order to prevent the occurrence of matrix irreversibility problem. and , for:

[0250] Formula 7

[0251] in, Represents the row vector of matrix A and the column vector of matrix B The Euclidean distance between is calculated as follows:

[0252] Formula 8

[0253] Among them, when the vector X and Y When the two vectors are highly consistent or similar, the result of the nonlinear operation approaches 0; and as the difference between the two vectors increases, the result of the nonlinear operation will increase accordingly. Based on this characteristic, the original matrix multiplication operation in Formula 6 is replaced with a nonlinear operator operation to derive a new expression for the estimated value of the observation vector:

[0254] Formula 9

[0255] in C is a constant matrix. In the actual calculation process, the matrix can be calculated in advance C The results are stored so that they can be directly used in subsequent calculations, thus saving calculation time.

[0256] Combine the following Figure 2 , specifically explain the process of building the monitoring model of the first thermal power unit and conducting equipment early warning:

[0257] Obtain historical operating data of the first thermal power equipment through the SIS system or the DCS system; perform feature selection on the historical operating data based on expert experience or the Pearson correlation analysis algorithm; perform data preprocessing on the historical operating data after feature selection to obtain first operating data, wherein the data preprocessing process includes: performing outlier truncation screening, detecting outliers using the LOF algorithm, and performing normalization processing; construct a historical memory matrix based on the first operating data, wherein the process of constructing the historical memory matrix includes: performing dimensionality reduction processing using the PCA algorithm, performing clustering processing using the FCM clustering algorithm, and performing group sampling; use the MEST model to obtain the residual value between the actual parameter value and the parameter estimation value predicted by the historical memory matrix, and use the sliding window technology to determine whether the residual value is higher than the residual threshold, and if so, issue a fault warning, otherwise determine that the equipment is normal.

[0258] In an optional embodiment, the first thermal power unit monitoring model is optimized according to the first parameter estimated value, the first parameter actual value and the parameter matrix value to obtain the optimized first thermal power unit monitoring model, including:

[0259] Taking the first parameter estimation value as the input of a preset neural network model, taking the first parameter actual value (the parameter actual value of the parameter of the first thermal power equipment after operation and maintenance, or the parameter observation value of the parameter of the first thermal power equipment after operation and maintenance) as the output of the preset neural network model, training the preset neural network model to obtain a trained neural network model;

[0260] The first thermal power unit monitoring model is optimized according to the trained neural network model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model.

[0261] The preset neural network model includes at least one of the following:

[0262] Back Propagation (BP) neural network;

[0263] Convolutional Neural Networks (CNN);

[0264] Long Short-Term Memory (LSTM).

[0265] In this optional embodiment, the preset neural network model is taken as an example of BP neural network.

[0266] A BP neural network is adopted, with the estimated value of the first parameter as input and the actual value of the first parameter (actual observed value) as output, to train the BP neural network and obtain a trained BP neural network model, which is a multivariate regression model. Afterwards, the trained BP neural network model is used to optimize the monitoring model of the first thermal power unit and obtain an optimized monitoring model of the first thermal power unit.

[0267] The process of using CNN and LSTM for training and optimizing the monitoring model of the first thermal power unit is basically the same as the process of using BP neural network for training and optimizing the monitoring model of the first thermal power unit mentioned above, and will not be repeated here.

[0268] Specifically, the first thermal power unit monitoring model is optimized according to the trained neural network model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model, including:

[0269] Inputting the parameter matrix value into the trained neural network model to obtain a first parameter prediction value output by the trained neural network model, that is, the first parameter prediction value can be understood as the parameter matrix value optimized by the trained neural network model;

[0270] The first thermal power unit monitoring model is optimized according to the first parameter prediction value to obtain the optimized first thermal power unit monitoring model. That is, after obtaining the first parameter prediction value output by the trained neural network model, the optimized first thermal power unit monitoring model is generated according to the first parameter prediction value and the parameters of the first thermal power equipment.

[0271] It can also be understood that the first parameter prediction value is a parameter matrix value in the optimized monitoring model of the first thermal power unit.

[0272] Combine the following Figure 3 , specifically explain the process of optimizing the monitoring model of the first thermal power unit:

[0273] Obtain historical operating data of the first thermal power equipment through the SIS system or the DCS system; perform data preprocessing on the historical operating data to obtain first operating data, wherein the data preprocessing process includes: performing outlier truncation screening, detecting outliers using the LOF algorithm, and performing normalization processing; construct a historical memory matrix (i.e., the first thermal power unit monitoring model) according to the first operating data; use the MEST algorithm to obtain a first parameter estimation value output by the historical memory matrix; use the first parameter estimation value as the input for training the BP neural network model, use the first parameter actual value of the first thermal power equipment after operation and maintenance as the output of the training BP neural network model, train the BP neural network model, and obtain a trained BP neural network model; use the parameter matrix value in the historical memory matrix as the model prediction input of the trained BP neural network model, and obtain the first parameter prediction value of the trained BP neural network model prediction output; generate an optimized historical memory matrix (i.e., the optimized first thermal power unit monitoring model) according to the first parameter prediction value and the parameters of the first thermal power equipment.

[0274] The following is a detailed explanation of the idea of ​​model optimization:

[0275] When the thermal power equipment operates normally, since the operating parameters of the thermal power equipment (the actual value of the first parameter) are similar to the parameter matrix values ​​in the historical memory matrix, the MEST algorithm is used to estimate the parameter values ​​predicted by the historical memory matrix. The actual value of the parameter Basically consistent with the historical memory matrix D A piece of operating condition data in the set (i.e. parameter matrix value) Similar ones are:

[0276]

[0277] After thermal power equipment deteriorates or undergoes normal maintenance, the operating data after deterioration or maintenance may not be included in the historical memory matrix. D In the same working condition as before the overhaul, the MSET algorithm is used to obtain the predicted value (the first parameter estimate) and the observed value after maintenance (actual value of the first parameter) The similarity of is larger than that of the previous one, but it is also different from a certain working condition data (i.e. parameter matrix value) in the historical memory matrix D set. Similar ones are:

[0278]

[0279] Build a neural network model:

[0280]

[0281] Equivalent to establishing a historical memory matrix D The mapping relationship between the sample working conditions in and , therefore, the history matrix D Input the trained model to obtain the optimized historical memory matrix (i.e. the optimized monitoring model of the first thermal power unit) satisfy:

[0282]

[0283] This completes the model update and realizes the predicted value after maintenance. With observed value Consistent.

[0284] The historical memory matrix is ​​updated and optimized through the back-propagation BP neural network model. The back-propagation BP neural network model can gradually adjust the elements in the memory matrix by iteratively optimizing the weights, so that it can better adapt to the new working condition data after maintenance. This update mechanism not only improves the prediction accuracy of the model, but also enhances the generalization ability of the model, enabling it to more effectively handle unknown or changing data.

[0285] BP neural network is a network algorithm based on error back propagation training proposed by McClelland and Rumelhart in 1986. It is also a multi-layer feedforward network including hidden layers. It is one of the most mature and widely used neural network models. In 1989, Hecht-Nielsen proved that a single hidden layer BP network can approximate any continuous function in any closed interval with arbitrary accuracy. It is currently widely used in nonlinear modeling, pattern recognition, system identification, prediction and control. The topological structure of a neural network with a single hidden layer and a single node output layer is as follows: Figure 4 As shown in Figure 1, the network structure includes input layer, hidden layer and output layer. Linear weight connections are used between the input layer and the hidden layer, and between the hidden layer and the output layer. Activation functions are used to transfer input and output between the hidden layer and the output layer.

[0286] The principle of BP algorithm is as follows:

[0287] The BP neural network algorithm includes two parts: signal forward transmission and error reverse propagation. Assume that the number of neuron nodes in the input layer is n , the number of neuron nodes in the hidden layer is s , then the input value of the hidden layer is:

[0288]

[0289] In the above formula: i =1,2,…, n ; j =1,2,…, s ; The input layer nodei The value of The input layer node i To the hidden layer nodes j The connection weight of is the hidden layer node j The threshold value.

[0290] The activation function between hidden layer outputs is selected as sigmoid or tanh function, that is:

[0291] or

[0292] Hidden layer nodes j The output value is:

[0293]

[0294] A linear weight connection is used between the hidden layer and the output layer, and the input value of the output layer h for:

[0295]

[0296] in: j =1,2,…, s ; is the hidden layer node j The connection weights to the output layer; b is the threshold of the output layer nodes.

[0297] The transfer function of the output layer selects purelin or tanh function, that is:

[0298] or

[0299] Output layer output value y for:

[0300]

[0301] After calculating the output layer result based on the input sample, the error between the network output value and the actual output sample is further calculated, and then the error is propagated back to the neurons in the hidden layer to adjust the connection weights and thresholds until the error reaches the accuracy requirement or reaches certain stopping conditions. The main training methods of BP neural network include gradient descent method, gradient descent method with momentum factor, Newton algorithm, Gauss-Newton algorithm, LM algorithm, etc.

[0302] In an optional embodiment, when the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference, the first thermal power unit monitoring model is optimized according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain an optimized first thermal power unit monitoring model, including:

[0303] In a case where the similarity between the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter satisfies the preset similarity difference, constructing a fitting model by using a fitting algorithm according to the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter, wherein the ratio between the number of the first target parameters and the total number of parameters of the first thermal power equipment is less than a preset ratio;

[0304] Optionally, the preset ratio is a value less than 1. The preset ratio can take a smaller value, and its specific value is not limited in this embodiment.

[0305] In this optional embodiment, among the parameters of the first thermal power equipment, only the similarity between the first parameter estimation value of the first target parameter (the number of the first target parameter is one or more of the parameters of the first thermal power equipment) and the first parameter actual value of the first target parameter satisfies the above-mentioned preset similarity difference, then a fitting algorithm can be used to construct a fitting model (or called a fitting function) to update and optimize the historical memory matrix.

[0306] The first thermal power unit monitoring model is optimized using the fitting model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model.

[0307] Specifically, the parameter matrix value is input into the fitting model to obtain the fitting parameter value output by the fitting model;

[0308] Generate the optimized first thermal power unit monitoring model according to the first target parameter and the fitting parameter value, and the second target parameter and the parameter matrix value corresponding to the second target parameter;

[0309] The second target parameter is a parameter of the first thermal power equipment other than the first target parameter.

[0310] Wherein, the fitting algorithm includes at least one of the following:

[0311] Polynomial fitting algorithm;

[0312] Exponential function fitting algorithm.

[0313] In an optional embodiment, when the thermal power unit monitoring model of the first thermal power equipment is used to establish a thermal power unit monitoring model for equipment of the same type, the method further includes:

[0314] Obtain an actual value of a second parameter of a parameter of a second thermal power device, wherein the second thermal power device is of the same type as the first thermal power device. For example, the thermal power unit system includes multiple coal mills in the same unit, the first thermal power device may be one of the coal mills, and the second thermal power device may be another coal mill.

[0315] In the case where the type of the first device is the same as the type of the second device, it can be considered that the parameters of the first thermal power device and the parameters of the second thermal power device are the same.

[0316] The parameters of the second thermal power equipment are input into the first thermal power unit monitoring model to obtain a second parameter estimation value of the parameters of the second thermal power equipment output by the first thermal power unit monitoring model.

[0317] Specifically, the MEST algorithm is used to preliminarily predict the parameters of the second thermal power equipment through the first thermal power unit monitoring model to obtain second parameter estimation values ​​of the parameters of the second thermal power equipment.

[0318] Using the estimated value of the second parameter as an input of a preset neural network model, using the actual value of the second parameter as an output of the preset neural network model, training the preset neural network model to obtain a trained neural network model;

[0319] Wherein, the preset neural network model includes at least one of the following:

[0320] Back propagation BP neural network;

[0321] Convolutional Neural Network (CNN);

[0322] Long Short-Term Memory Network LSTM.

[0323] In this optional embodiment, the preset neural network model is also taken as an example to illustrate the BP neural network model.

[0324] A BP neural network is adopted, with the estimated value of the second parameter as input and the actual value (actual observed value) of the second parameter as output, to train the BP neural network to obtain a trained BP neural network model, which is a multivariate regression model. Then, according to the trained neural network model and the parameter matrix value in the first thermal power unit monitoring model, the second thermal power unit monitoring model of the second thermal power equipment is obtained.

[0325] Specifically, according to the trained neural network model and the parameter matrix value, a second thermal power unit monitoring model of the second thermal power equipment is obtained, including:

[0326] Inputting the parameter matrix value into the trained neural network model to obtain a second parameter prediction value output by the trained neural network model;

[0327] According to the second parameter prediction value and the parameters of the second thermal power equipment, a second thermal power unit monitoring model of the second thermal power equipment is obtained. That is, after obtaining the second parameter prediction value output by the trained neural network model, the second thermal power unit monitoring model of the second thermal power equipment is generated according to the second parameter prediction value and the parameters of the second thermal power equipment.

[0328] Combine the following Figure 5 , specifically explain the construction process of the monitoring model of the second thermal power unit of the second thermal power equipment:

[0329] Obtain historical operating data of the first thermal power equipment through the SIS system or the DCS system; perform data preprocessing on the historical operating data to obtain first operating data, wherein the data preprocessing process includes: performing outlier truncation screening, detecting outliers using the LOF algorithm, and performing normalization processing; constructing a historical memory matrix of the first thermal power equipment (i.e., a first thermal power unit monitoring model) according to the first operating data; using the MEST algorithm to obtain a first parameter estimation value output by the historical memory matrix; using the first parameter estimation value as an input for training a BP neural network model, using the second parameter actual value of the second thermal power equipment as an output of the training BP neural network model, training the BP neural network model, and obtaining a trained BP neural network model; using the parameter matrix value in the historical memory matrix as a model prediction input of the trained BP neural network model, and obtaining a second parameter prediction value output by the trained BP neural network model; generating a historical memory matrix of the second thermal power equipment (i.e., a second thermal power unit monitoring model of the second thermal power equipment) according to the second parameter prediction value and the parameters of the second thermal power equipment.

[0330] The following two specific examples illustrate the specific process of optimizing the thermal power unit monitoring model provided by the embodiment of the present invention:

[0331] Example 1:

[0332] Take the first thermal power equipment, the induced draft fan of a 660MW thermal power unit, as an example.

[0333] First, establish the fault warning model of the induced draft fan (i.e. the monitoring model of the first thermal power unit):

[0334] The core of the fault warning method based on MSET lies in constructing a historical memory matrix. The specific construction process is as follows: First, the principal component analysis (PCA) technology is used to reduce the dimension of the original high-dimensional data (historical operating condition data) and map it to N-dimensional space; then, the fuzzy C-means clustering algorithm is used to perform cluster analysis on the reduced-dimensional data, and M cluster clusters are set; then, according to the data volume of each cluster and the ratio of the total target data volume to the total original data volume, a data extraction strategy is formulated to extract data from each cluster in proportion. The specific operation is to sort and group the samples in the cluster, and then select the minimum value, maximum value and data closest to the median of each group to ensure the diversity and representativeness of the data; finally, the extracted data is integrated to form a historical memory matrix containing the total target data volume. This method effectively reduces the complexity of data processing.

[0335] Specifically, the core link of the fault warning method based on MSET is to construct a historical memory matrix. The construction process is described as follows: First, the principal component analysis (PCA) technology is used to reduce the dimension of the original high-dimensional data (historical operating condition data) and map it to a n dimensional space; then, the fuzzy C-means clustering algorithm is used to perform cluster analysis on the reduced-dimensional data, and the preset k clusters. On this basis, a data extraction strategy is designed according to the data volume of each cluster and the ratio of the total target data to the total original data. Specifically, data is extracted from each cluster in proportion. The operation process includes sorting and grouping the samples in the cluster, and then selecting the minimum value, maximum value and the data closest to the median of the group from each group to ensure the diversity and representativeness of the extracted data. Finally, these extracted data are integrated to form a historical memory matrix containing the total target data. This method not only significantly reduces the complexity of data processing, but also improves the efficiency and quality of data processing.

[0336] Parameter selection of induced draft fan:

[0337] In the process of building the induced draft fan monitoring and fault warning model (MSET), selecting appropriate modeling variables is a key step to ensure model performance. This requires an in-depth understanding of the actual monitoring points (i.e. parameters) on site, aiming to select key measurement points that can capture the dynamic characteristics of the induced draft fan and are highly sensitive to major faults from the complex monitoring parameters. To this end, a combination of business expert analysis and algorithms was adopted, and Pearson correlation analysis technology was introduced to determine and select the following core measurement points as the basis for modeling analysis. The selected core measurement points are specifically listed in Table 1 below:

[0338] Table 1 Input parameters of induced draft fan fault warning model

[0339]

[0340] Abnormal data cleaning of induced draft fan:

[0341] The historical operating data of the induced draft fan from June 2023 to November 2023 are collected from the DCS system. The data set is sampled at intervals of 5 minutes, with a total of 48,001 data. In the preliminary analysis stage, the scatter plot was used to find that the data of some measuring points were abnormally deviated from the normal range, so the outlier truncation method was used to set the screening criteria of generator power greater than 220, induced draft fan inlet flue gas temperature greater than 90, induced draft fan motor front bearing temperature 1 greater than 45.6, and induced draft fan motor current greater than 140. In order to completely eliminate potential abnormal data, the LOF method was further applied for secondary screening, and finally 33,390 valid records were screened out from the original data. In view of the fact that the induced draft fan equipment contains many measuring points, several representative key measuring points are selected, including induced draft fan blades, current, and bearing temperature, and the original data and cleaned data of these measuring points are plotted and displayed. The specific display content is as follows: Figures 6 to 9 As shown, Figure 6 The original data (historical operating data) of the induced draft fan blade position reversal and the scatter plot comparison chart after outliers are removed. Figure 7 The original data of the induced draft fan motor current and the scatter plot comparison chart after the outliers are removed. Figure 8 The original data of the induced draft fan front bearing temperature 1 and the scatter plot comparison chart after the outliers are removed. Fig. 9 The original data of the induced draft fan motor front bearing temperature 1 and the scatter plot comparison chart after outliers are removed. Figures 6 to 9 In the figure, the gray dots represent the original data, the red dots represent the data after the original data is truncated and filtered, and the blue dots represent the normal values ​​after the LOF algorithm is filtered.

[0342] Construction of the historical memory matrix of the induced draft fan:

[0343] By applying the principal component analysis (PCA) technique, the dimension of the normal data set after the induced draft fan data cleaning was reduced. This step aims to reduce the complexity of the data while retaining as much original information as possible for subsequent analysis. Subsequently, the fuzzy C-means (FCM) clustering algorithm was used to perform cluster analysis on the reduced-dimensional data, aiming to divide the data into several groups with similar characteristics. After the cluster analysis was completed, group sampling was performed, that is, representative data samples were selected from each cluster, and a historical memory matrix was constructed based on this. Several important measurement points were selected, and the distribution of these measurement points in the memory matrix and the original data was compared and displayed. The visualization results are shown in Figure 2. Figures 10 to 13 As shown, Fig.10 This is a scatter plot comparison of the normal data of the induced draft fan blade position reversal and the data used to construct the historical memory matrix. Fig.11This is a scatter plot comparison of the normal data of the induced draft fan motor current and the data used to construct the historical memory matrix. Fig.12 This is a scatter plot comparison of the normal data of the induced draft fan front bearing temperature 1 and the data for constructing the historical memory matrix. Fig.13 It is a scatter plot comparison of the normal data of the induced draft fan motor front bearing temperature 1 and the data used to construct the historical memory matrix. Figures 10 to 13 The red dots represent the normal data after removing outliers from the original data (historical operating condition data), and the blue dots represent the data used to construct the historical memory matrix.

[0344] After the above process, the historical memory matrix of the induced draft fan has been constructed. In order to verify the performance of the historical memory matrix, the data from the second half of 2023 and the first half of 2024 were used for testing. By applying the MSET algorithm, the predicted values ​​of the test data were obtained and compared with the actual values. The comparison results are shown in Figure 2. Figures 14 to 21 As shown, Fig.14 This is a comparison chart of the actual value of the induced draft fan blade position in 2023 and the MEST predicted value. Fig.15 This is a comparison chart of the actual value of the induced draft fan blade position in 2024 and the MEST predicted value. Fig.16 This is a comparison chart of the actual value of the induced draft fan motor current in 2023 and the MEST predicted value. Fig.17 This is a comparison chart of the actual value of the induced draft fan motor current in 2024 and the MEST predicted value. Fig.18 This is a comparison chart of the actual value of the induced draft fan front bearing temperature 1 in 2023 and the MEST predicted value. Fig.19 This is a comparison chart of the actual value of the induced draft fan front bearing temperature 1 in 2024 and the MEST predicted value. Fig. 20 This is a comparison chart of the actual value of the front bearing temperature 1 of the induced draft fan motor in 2023 and the MEST predicted value. Fig.21 This is a comparison chart of the actual value of the front bearing temperature 1 of the induced draft fan motor in 2024 and the MEST predicted value. Figures 14 to 21 In the figure, the red line represents the actual value of the parameter, and the green dotted line represents the estimated value of the parameter predicted by the MEST algorithm.

[0345] By drawing the similarity graph of parameters, we found that the similarity of many parameters is around 95%. Here we select several parameters with a similarity close to 90% in 2024 for display, such as Figure 22 to Figure 27 As shown, Fig. 22 This is a schematic diagram of the inverse similarity of the induced wind turbine blade position in 2023. Fig.23 This is a schematic diagram of the inverse similarity of the induced wind turbine blade position in 2024. Fig.24 This is a similarity diagram of the bearing temperature 1 in the induced draft fan in 2023. Fig.25 This is a similarity diagram of the bearing temperature 1 in the induced draft fan in 2024. Fig.26This is a schematic diagram of the similarity of the induced draft fan motor current in 2023. Fig. 27 This is a schematic diagram of the similarity of the induced draft fan motor current in 2024.

[0346] It can be seen from the above prediction results that the induced draft fan blades were overhauled at the end of 2023, so the data in 2024 has changed significantly compared to before the overhaul. When using the traditional MSET model, the operating conditions of the equipment often change significantly after the overhaul. Due to these changes, the original memory matrix may not accurately reflect the new operating conditions after the overhaul, resulting in poor prediction results. As a core component of the MSET model, the construction of the memory matrix relies on the statistical characteristics of historical data. However, equipment maintenance often introduces new operating conditions, performance parameters or failure modes, which may not be fully considered in the original memory matrix.

[0347] It is proposed to use the back propagation (BP) algorithm to update the memory matrix. The BP algorithm can gradually adjust the elements in the memory matrix by iteratively optimizing the weights, so that it can better adapt to the new working condition data after maintenance. This update mechanism not only improves the prediction accuracy of the model, but also enhances the generalization ability of the model, enabling it to more effectively handle unknown or changing data.

[0348] Application of induced draft fan BP-MSET model (i.e., the model after optimizing the memory matrix using the BP neural network model):

[0349] In practical applications, first, an initial memory matrix is ​​constructed based on the data before the induced draft fan overhaul, and based on the memory matrix, the MSET model is used to predict the data after the overhaul. However, as shown in the figure above, the preliminary prediction results have large errors. In order to optimize the prediction performance, the BP algorithm is introduced to iteratively update the memory matrix of MSET. The specific steps are as follows: Thousands of data from 2024 are randomly selected as test sets, and the memory matrix of 2023 is used to predict through the MSET model; then, the prediction results of the test set are used as the input of the BP model, and the actual test data is used as the output to train the BP model; then, the memory matrix data of 2023 is input into the trained BP model to obtain the predicted value of the memory matrix data, and this is used as the new memory matrix to update the memory matrix. After this improvement, the accuracy of the prediction results has been significantly improved, and the error has been significantly reduced. The improved prediction results are as follows. Figure 28 to Figure 31 shown. Fig.28 The figure is a comparison chart of the actual value of the induced draft fan blade position, the predicted value predicted by the MSET algorithm based on the initial memory matrix, and the predicted value predicted by the MSET algorithm based on the memory matrix optimized by the BP neural network. Fig.29The figure is a comparison chart of the actual value of the induced draft fan motor current, the predicted value predicted by the MSET algorithm based on the initial memory matrix, and the predicted value predicted by the MSET algorithm based on the memory matrix optimized by the BP neural network. Fig.30 This is a comparison chart of the actual value of the induced draft fan front bearing temperature 1, the predicted value predicted by the MSET algorithm based on the initial memory matrix, and the predicted value predicted by the MSET algorithm based on the memory matrix optimized by the BP neural network. Fig.31 This is a comparison chart of the actual value of the front bearing temperature 1 of the induced draft fan motor, the predicted value predicted by the MSET algorithm based on the initial memory matrix, and the predicted value predicted by the MSET algorithm based on the memory matrix optimized by the BP neural network. Figure 28 to Figure 31 In the figure, the red line represents the actual value of the parameter, the green dotted line represents the estimated value of the parameter predicted by the MEST algorithm based on the initial memory matrix, and the blue dotted line represents the estimated value of the parameter predicted by the MEST algorithm based on the memory matrix optimized by the BP neural network.

[0350] For the memory matrix model predicted by the MEST algorithm (referred to as the MEST model) and the memory matrix model optimized by the BP neural network predicted by the MEST algorithm (referred to as the BP-MEST model), a series of regression model indicators can be used to evaluate the model, mainly including the following indicators, and the calculation formulas are as follows.

[0351] Mean Square Error:

[0352]

[0353] in, is the actual value of the sample, is the sample estimate, n is the sample size.

[0354] Root mean square deviation:

[0355]

[0356] in, is the actual value of the sample, is the sample estimate, n is the sample size.

[0357] Mean absolute error:

[0358]

[0359] Mean absolute percentage error:

[0360]

[0361] in, is the actual value of the sample, is the sample estimate, n is the sample size.

[0362] R-squared value:

[0363]

[0364] in, is the actual value of the sample, is the sample estimate, is the sample mean, n is the sample size.

[0365] The comparison of the indicators of the MEST model and the BP-MEST model is shown in Table 2 below:

[0366] Table 2 Evaluation index data of each model of induced draft fan

[0367]

[0368] The following is a specific example to illustrate the process of building a thermal power unit monitoring model for the same type of equipment:

[0369] Example 2:

[0370] Take the coal mill of a 660MW thermal power unit as an example.

[0371] The concept of BP-MSET is extended to more complex model application scenarios, and the BP-MSET method is used to achieve model optimization between the same type of equipment in the same unit. This method only requires building a basic model for the same type of equipment, thereby reducing the time and resources required for data analysis and modeling. To this end, from the seven coal mills in a power plant, coal mill E with the best historical data performance was selected as the modeling object, and the model of coal mill E was used to predict the data of coal mill A.

[0372] Coal Mill E Modeling:

[0373] Parameter selection for coal mill E modeling:

[0374] The analysis of business experts and algorithms were combined, and Pearson correlation analysis technology was introduced to select the following core measurement points as the basis for coal mill modeling analysis. The selected core measurement points are listed in Table 3 below:

[0375] Table 3 Input parameters of coal mill fault warning model

[0376]

[0377] Coal mill E abnormal data cleaning:

[0378] The historical data of coal mill E from June 2023 to April 2024 are collected from the DCS system. The data set is sampled at 5-minute intervals, with a total of 92,001 data. In the preliminary analysis stage, the outlier truncation method is used to set the screening criteria that the instantaneous coal quantity data of the coal feeder is between 40 and 90, the mill hydraulic oil station loading force valve instruction is less than 65, and the mill hydraulic oil station loading force valve instruction is between 35 and 60. The LOF method is used for secondary screening, and finally 66,416 valid records are screened out from the original data. Several representative key measuring points (parameters) are selected, including E mill current, E mill inlet and outlet differential pressure, E mill inlet primary air volume 2, E mill motor drive end bearing temperature 1, etc., and the original data and cleaned data of these measuring points are plotted and displayed. The specific display content is as follows: Figure 32 to Figure 35 As shown, Fig.32 The original data of the E mill (coal mill E) current and the scatter plot comparison after the outliers are removed are shown below. Fig.33 The original data of the differential pressure of the E mill in and out and the scatter plot comparison chart after the outliers are removed. Fig.34 The original data of the primary air volume 2 at the E mill inlet and the scatter plot comparison chart after outliers are removed. Fig.35 The original data of the bearing temperature 1 at the drive end of the E-mill motor and the scatter plot comparison chart after the outliers are removed are shown in Figure 1. Figure 32 to Figure 35 In the figure, the gray dots represent the original data, the red dots represent the data after the original data is truncated and filtered, and the blue dots represent the normal values ​​after the LOF algorithm is filtered.

[0379] The principal component analysis (PCA) technique was first applied to the normal data of coal mill E for dimensionality reduction processing to reduce the dimension of the data and retain key information. Subsequently, the fuzzy C-means (FCM) clustering algorithm was used to perform cluster analysis on the reduced-dimensional data, and data with similar characteristics were grouped into one category. On this basis, group sampling was performed, and representative data were selected from each cluster to construct a memory matrix. In order to more intuitively display the data characteristics, several key measurement points were selected, and the distribution of these measurement points in the memory matrix was compared with the original data. The visualization results are shown in the figure below. Figure 36 to Figure 39 shown. Fig.36 This is a scatter plot comparison of the normal data of E-mill current and the data used to construct the historical memory matrix. Fig.37 This is a scatter plot comparison of the normal data of the differential pressure of the E mill and the data used to construct the historical memory matrix. Fig.38 This is a scatter plot comparison of the normal data of the primary air volume 2 at the E mill inlet and the data used to construct the historical memory matrix. Fig.39 This is a scatter plot comparison of the normal data of the E-mill motor drive end bearing temperature 1 and the data used to construct the historical memory matrix. Figure 36 to Figure 39The red dots represent the normal data after removing outliers from the original data (historical operating condition data), and the blue dots represent the data used to construct the historical memory matrix.

[0380] From the continuous normal operation data of coal mill E, a section of historical data is selected as the test set. After inputting these data into the MSET model, the prediction value corresponding to the test set can be obtained. By comparing and analyzing the predicted value with the actual observed value, the results show that the model performs well in predicting the data of coal mill E itself, and the error between the predicted value and the actual observed value is very small. Figure 40 to Figure 43 The results of this test are shown. Fig.40 This is a comparison chart of the actual value of E mill current and the predicted value of MEST. Fig.41 This is a comparison chart of the actual value of the differential pressure at the inlet and outlet of the E mill and the value predicted by MEST. Fig.42 This is a comparison chart of the actual value of the primary air volume 2 at the E mill inlet and the MEST predicted value. Fig.43 This is a comparison chart of the actual value of the E-mill motor drive end bearing temperature 1 and the MEST predicted value. Figure 40 to Figure 43 In the figure, the red line represents the actual value of the parameter, and the green dotted line represents the estimated value of the parameter predicted by the MEST algorithm.

[0381] Coal Mill A - MSET Model Prediction:

[0382] Given that different coal mills are usually equipped with similar measurement points, although there may be slight differences in the data range, an idea is to directly apply the mature model of coal mill E to coal mill A. This strategy aims to simplify the modeling process of coal mill A and avoid repeated complex and time-consuming steps such as data cleaning and memory matrix construction, thereby effectively saving resources and time.

[0383] A continuous historical data of coal mill A under normal operating conditions was selected as the test set, and the MSET model of coal mill E was used to predict these data. After the prediction was completed, the predicted values ​​were compared with the actual observed values. The results show that although the overall trend of the model in predicting the coal mill A data is relatively consistent with the actual situation, there is a large error between the predicted value and the actual observed value. This shows that although the model has demonstrated the ability to generalize to different equipment to a certain extent, it still needs further optimization to improve the accuracy of the prediction. The specific presentation of the test results is as follows Figures 44 to 47 shown. Fig.44 This is a comparison chart of the actual current of mill A (coal mill A) and the MEST predicted value. Fig.45 This is a comparison chart of the actual value of the differential pressure between the inlet and outlet of mill A and the value predicted by MEST. Fig.46 This is a comparison chart of the actual value of the primary air volume 2 at the inlet of mill A and the MEST predicted value. Fig.47This is a comparison chart of the actual value of the bearing temperature 1 at the drive end of the A mill motor and the MEST predicted value. Figures 44 to 47 In the figure, the red line represents the actual value of the parameters of coal mill A, and the green dotted line represents the estimated value of the parameters predicted by the MEST algorithm and the memory matrix of coal mill E.

[0384] By drawing the similarity graph of parameters, we found that the similarity of many parameters is around 95%. Here we select several parameters with similarity close to 90% for display, such as Figures 48 to 53 shown. Fig.48 This is the similarity diagram of primary air volume 2 at the E mill inlet. Fig.49 This is a similarity diagram of primary air volume 2 at the inlet of mill A. Fig.50 This is the current similarity diagram of the E-mill coal feeder. Fig.51 This is the current similarity diagram of the coal feeder of mill A. Fig.52 53 is a similarity diagram of the vibration of the bearing of the driving end of the motor of the E mill, and 54 is a similarity diagram of the vibration of the bearing of the driving end of the motor of the A mill.

[0385] Coal mill BP-MSET model (using BP neural network model to generate the memory matrix of coal mill A according to the memory matrix of coal mill E) application:

[0386] In order to improve the prediction ability of coal mill E model for coal mill A data, the BP algorithm is used to optimize the memory matrix of coal mill E. The specific optimization steps are as follows:

[0387] First, thousands of samples are randomly selected from the normal historical data of coal mill A to form a test set. Then, the current memory matrix of coal mill E is used to predict the data in the test set through the MSET model.

[0388] Then, the prediction results of these test sets are used as the input of the BP model, and the corresponding actual test data are used as the output of the BP model to train the BP model.

[0389] After the training is completed, the memory matrix data of coal mill E is input into the trained BP model to obtain the predicted values ​​of the memory matrix data. These predicted values ​​are used to replace the original memory matrix to form a new, optimized memory matrix.

[0390] By updating the memory matrix with the BP algorithm, it is found that the accuracy of the prediction results has been significantly improved and the error has been significantly reduced. The prediction results before and after the model optimization are as follows: Figure 54 to Figure 57 shown. Fig.54 The figure is a comparison of the actual value of the current of mill A, the predicted value predicted by the MSET algorithm based on the memory matrix of mill E, and the predicted value predicted by the MEST algorithm based on the memory matrix of mill A after BP neural network optimization. Fig.55This is a comparison chart of the actual value of the inlet and outlet differential pressure of mill A, the predicted value predicted by the MSET algorithm based on the memory matrix of mill E, and the predicted value predicted by the MEST algorithm based on the memory matrix of mill A after BP neural network optimization. Fig.56 This is a comparison chart of the actual value of the primary air volume 2 at the inlet of mill A, the predicted value predicted by the MSET algorithm based on the memory matrix of mill E, and the predicted value predicted by the MEST algorithm based on the memory matrix of mill A after BP neural network optimization. Fig.57 This is a comparison chart of the actual value of the bearing temperature 1 at the drive end of the motor of mill A, the predicted value predicted by the MSET algorithm based on the memory matrix of mill E, and the predicted value predicted by the MEST algorithm based on the memory matrix of mill A after BP neural network optimization. Figure 54 to Figure 57 In the figure, the red line represents the actual parameter value, the green dotted line represents the parameter estimation value predicted by the MEST algorithm based on the memory matrix of coal mill E, and the blue dotted line represents the parameter estimation value predicted by the MEST algorithm based on the memory matrix of coal mill A obtained by BP neural network optimization.

[0391] The comparison of the indicators of the MEST model (memory matrix of coal mill E) and the BP-MEST model (memory matrix of coal mill A) is shown in Table 4 below:

[0392] Table 4 Comparison of evaluation indexes of various models for coal mill E prediction of coal mill A data

[0393]

[0394] The intelligent monitoring method for thermal power units provided in the embodiment of the present invention aims at the problem of model incompatibility due to normal equipment degradation and after-maintenance. Through incremental learning, the model parameters are automatically corrected using a small amount of data without retraining the model, and the existing model is updated and iterated, so that the performance of the failed warning model can be restored, which greatly reduces the operation and maintenance cost after the model is put into use. At the same time, for modeling of the same type of equipment in a unit, by transfer learning, the model parameters can be corrected without remodeling, and the modeling workload of the same type of equipment can be greatly reduced. The problem of a large number of false alarms in the existing intelligent monitoring system of thermal power units is solved due to the normal degradation of equipment, daily maintenance of equipment, and after equipment maintenance, the monitoring and warning model cannot adapt to the operating conditions of the equipment and system; the problem of model incompatibility of the same type of equipment in the same thermal power unit is solved; the problem of model incompatibility of the same type of equipment between different thermal power units is solved.

[0395] like Fig.58 As shown, the embodiment of the present invention also provides an intelligent monitoring device for a thermal power unit, comprising:

[0396] The first processing module 5801 is used to construct a monitoring model of the first thermal power unit according to the historical operating condition data of the operating condition of the first thermal power equipment;

[0397] The second processing module 5802 is used to obtain a first parameter estimation value obtained by predicting the parameters of the first thermal power equipment by the first thermal power unit monitoring model according to the operation and maintenance information of the first thermal power equipment;

[0398] The third processing module 5803 is used to optimize the first thermal power unit monitoring model according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain an optimized first thermal power unit monitoring model when the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment meets a preset similarity difference;

[0399] The first thermal power unit monitoring model stores the parameters of the first thermal power equipment and parameter matrix values ​​of the parameters.

[0400] Optionally, the device further comprises:

[0401] A fourth processing module, configured to determine, when the first condition is met, that a similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference;

[0402] The first condition includes:

[0403] Within a preset time period, the average of the similarities between the first parameter estimation values ​​of all parameters of the first thermal power equipment and the first parameter actual values ​​of the corresponding parameters is less than a first threshold;

[0404] Within a preset time period, the similarity between a first parameter estimation value of any parameter of the first thermal power equipment and a first parameter actual value of the corresponding parameter is less than a second threshold;

[0405] The first threshold is greater than the second threshold.

[0406] Optionally, the third processing module 5803 includes:

[0407] A first processing unit, configured to use the first parameter estimation value as an input of a preset neural network model, use the first parameter actual value as an output of the preset neural network model, train the preset neural network model, and obtain a trained neural network model;

[0408] The second processing unit is used to optimize the first thermal power unit monitoring model according to the trained neural network model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model.

[0409] Optionally, the second processing unit is specifically configured to:

[0410] Inputting the parameter matrix value into the trained neural network model to obtain a first parameter prediction value output by the trained neural network model;

[0411] The first thermal power unit monitoring model is optimized according to the first parameter prediction value to obtain the optimized first thermal power unit monitoring model.

[0412] Optionally, the third processing module 5803 includes:

[0413] a third processing unit, configured to construct a fitting model using a fitting algorithm according to the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter, when the similarity between the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter satisfies the preset similarity difference, wherein a ratio between the number of the first target parameters and the total number of parameters of the first thermal power equipment is less than a preset ratio;

[0414] The fourth processing unit is used to optimize the first thermal power unit monitoring model according to the fitting model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model.

[0415] Optionally, the fitting algorithm includes at least one of the following:

[0416] Polynomial fitting algorithm;

[0417] Exponential function fitting algorithm.

[0418] Optionally, the device further comprises:

[0419] A fifth processing module, configured to obtain an actual value of a second parameter of a parameter of a second thermal power device, wherein the second thermal power device is of the same type as the first thermal power device;

[0420] A sixth processing module, configured to input the parameter of the second thermal power equipment into the monitoring model of the first thermal power unit, and obtain a second parameter estimation value of the parameter of the second thermal power equipment output by the monitoring model of the first thermal power unit;

[0421] a seventh processing module, configured to use the estimated value of the second parameter as an input of a preset neural network model, use the actual value of the second parameter as an output of the preset neural network model, train the preset neural network model, and obtain a trained neural network model;

[0422] An eighth processing module is used to obtain a second thermal power unit monitoring model of the second thermal power equipment according to the trained neural network model and the parameter matrix value.

[0423] Optionally, the eighth processing module includes:

[0424] A fifth processing unit, configured to input the parameter matrix value into the trained neural network model to obtain a second parameter prediction value output by the trained neural network model;

[0425] The sixth processing unit is used to obtain a second thermal power unit monitoring model of the second thermal power equipment according to the second parameter prediction value and the parameters of the second thermal power equipment.

[0426] Optionally, the preset neural network model includes at least one of the following:

[0427] Back propagation BP neural network;

[0428] Convolutional Neural Network (CNN);

[0429] Long Short-Term Memory Network LSTM.

[0430] It should be noted that the intelligent monitoring device for thermal power units provided in the embodiment of the present invention is a device capable of executing the above-mentioned intelligent monitoring method for thermal power units. All embodiments of the above-mentioned intelligent monitoring method for thermal power units are applicable to the device and can achieve the same or similar technical effects.

[0431] like Fig.59 As shown, an embodiment of the present invention also provides an intelligent monitoring device for a thermal power unit, including: a processor 5901; and a memory 5903 connected to the processor 5901 via a bus interface 5902, wherein the memory 5903 is used to store programs and data used by the processor 5901 when performing operations, and the processor 5901 calls and executes the programs and data stored in the memory 5903.

[0432] The transceiver 5904 is connected to the bus interface 5902 and is used to receive and send data under the control of the processor 5901. Specifically, the processor 5901 is used to read the program in the memory 5903. The processor 5901 performs the following process:

[0433] According to the historical operating data of the operating conditions of the first thermal power equipment, a monitoring model of the first thermal power unit is constructed;

[0434] According to the operation and maintenance information of the first thermal power equipment, obtaining a first parameter estimation value obtained by predicting the parameter of the first thermal power equipment by the first thermal power unit monitoring model;

[0435] When the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference, the first thermal power unit monitoring model is optimized according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain an optimized first thermal power unit monitoring model;

[0436] The first thermal power unit monitoring model stores the parameters of the first thermal power equipment and parameter matrix values ​​of the parameters.

[0437] Optionally, the processor 5901 is further configured to:

[0438] In the case where the first condition is met, determining that the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference;

[0439] The first condition includes:

[0440] Within a preset time period, the average of the similarities between the first parameter estimation values ​​of all parameters of the first thermal power equipment and the first parameter actual values ​​of the corresponding parameters is less than a first threshold;

[0441] Within a preset time period, the similarity between a first parameter estimation value of any parameter of the first thermal power equipment and a first parameter actual value of the corresponding parameter is less than a second threshold;

[0442] The first threshold is greater than the second threshold.

[0443] Optionally, the processor 5901 is configured to:

[0444] Using the first parameter estimation value as the input of a preset neural network model, using the first parameter actual value as the output of the preset neural network model, training the preset neural network model to obtain a trained neural network model;

[0445] The first thermal power unit monitoring model is optimized according to the trained neural network model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model.

[0446] Optionally, the processor 5901 is specifically configured to:

[0447] Inputting the parameter matrix value into the trained neural network model to obtain a first parameter prediction value output by the trained neural network model;

[0448] The first thermal power unit monitoring model is optimized according to the first parameter prediction value to obtain the optimized first thermal power unit monitoring model.

[0449] Optionally, the processor 5901 is configured to:

[0450] In a case where the similarity between the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter satisfies the preset similarity difference, constructing a fitting model by using a fitting algorithm according to the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter, wherein the ratio between the number of the first target parameters and the total number of parameters of the first thermal power equipment is less than a preset ratio;

[0451] The first thermal power unit monitoring model is optimized according to the fitting model and the parameter matrix value to obtain the optimized first thermal power unit monitoring model.

[0452] Optionally, the fitting algorithm includes at least one of the following:

[0453] Polynomial fitting algorithm;

[0454] Exponential function fitting algorithm.

[0455] Optionally, the processor 5901 is further configured to:

[0456] Acquire a second parameter actual value of a parameter of a second thermal power device, wherein the second thermal power device is of the same type as the first thermal power device;

[0457] Inputting the parameter of the second thermal power equipment into the monitoring model of the first thermal power unit to obtain a second parameter estimation value of the parameter of the second thermal power equipment output by the monitoring model of the first thermal power unit;

[0458] Using the estimated value of the second parameter as an input of a preset neural network model, using the actual value of the second parameter as an output of the preset neural network model, training the preset neural network model to obtain a trained neural network model;

[0459] A second thermal power unit monitoring model of the second thermal power equipment is obtained according to the trained neural network model and the parameter matrix value.

[0460] Optionally, the processor 5901 is specifically configured to:

[0461] Inputting the parameter matrix value into the trained neural network model to obtain a second parameter prediction value output by the trained neural network model;

[0462] A second thermal power unit monitoring model of the second thermal power equipment is obtained according to the predicted value of the second parameter and the parameter of the second thermal power equipment.

[0463] Optionally, the preset neural network model includes at least one of the following:

[0464] Back propagation BP neural network;

[0465] Convolutional Neural Network (CNN);

[0466] Long Short-Term Memory Network LSTM.

[0467] Among them, Fig.59 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 5901 and memory represented by memory 5903. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides a user interface 5905. The transceiver 5904 may be a plurality of components, i.e., including a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium. The processor 5901 is responsible for managing the bus architecture and general processing, and the memory 5903 may store data used by the processor 5901 when performing operations.

[0468] Those skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by instructing relevant hardware through a computer program, wherein the computer program includes instructions for executing part or all of the steps of the above method; and the computer program may be stored in a readable storage medium, and the storage medium may be any form of storage medium.

[0469] In addition, the embodiment of the present invention further provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, each process of the above-mentioned thermal power unit intelligent monitoring method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0470] In addition, it should be pointed out that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above-mentioned series of processing can naturally be performed in the order of description or in chronological order, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understandable that all or any steps or components of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or a network of computing devices in hardware, firmware, software or a combination thereof, which can be achieved by those of ordinary skill in the art using their basic programming skills after reading the description of the present invention.

[0471] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general-purpose device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code that implements the method or device. In other words, such a program product also constitutes the present invention, and a storage medium storing such a program product can also constitute the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future.

[0472] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0473] A specific embodiment of the present invention further provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0474] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for intelligent monitoring of thermal power units, characterized in that: include: According to the historical operating data of the operating conditions of the first thermal power equipment, a monitoring model of the first thermal power unit is constructed; According to the operation and maintenance information of the first thermal power equipment, when it is determined that the first thermal power equipment has no equipment failure, obtaining a first parameter estimation value of a parameter of the first thermal power unit monitoring model for the first thermal power equipment; wherein the operation and maintenance information of the first thermal power equipment includes at least one of the following: normal degradation information of the first thermal power equipment, daily maintenance information of the first thermal power equipment, and maintenance information of the first thermal power equipment; In the case where the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment meets the preset similarity difference, the first thermal power unit monitoring model is optimized according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain the optimized first thermal power unit monitoring model; wherein the first thermal power unit monitoring model is a multivariate state estimation technology MSET model; the first thermal power unit monitoring model stores the parameters of the first thermal power equipment and the parameter matrix values ​​of the parameters; Obtaining an actual value of a second parameter of a parameter of a second thermal power device, wherein the second thermal power device is of the same type as the first thermal power device; and the first thermal power device and the second thermal power device are devices of the same type between different thermal power units; Inputting the parameter of the second thermal power equipment into the monitoring model of the first thermal power unit to obtain a second parameter estimation value of the parameter of the second thermal power equipment output by the monitoring model of the first thermal power unit; Using the estimated value of the second parameter as an input of a preset neural network model, using the actual value of the second parameter as an output of the preset neural network model, training the preset neural network model to obtain a trained neural network model; Inputting the parameter matrix value into the trained neural network model to obtain a second parameter prediction value output by the trained neural network model; According to the predicted value of the second parameter and the parameter of the second thermal power equipment, a monitoring model of a second thermal power unit of the second thermal power equipment is obtained; According to the first parameter estimation value, the first parameter actual value and the parameter matrix value, the first thermal power unit monitoring model is optimized to obtain the optimized first thermal power unit monitoring model, including: Using the first parameter estimation value as the input of a preset neural network model, using the first parameter actual value as the output of the preset neural network model, training the preset neural network model to obtain a trained neural network model; Inputting the parameter matrix value into the trained neural network model to obtain a first parameter prediction value output by the trained neural network model; Optimizing the first thermal power unit monitoring model according to the first parameter prediction value to obtain the optimized first thermal power unit monitoring model; Wherein, when the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference, the first thermal power unit monitoring model is optimized according to the first parameter estimation value, the first parameter actual value and the parameter matrix value to obtain the optimized first thermal power unit monitoring model, including: In a case where the similarity between the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter satisfies the preset similarity difference, constructing a fitting model by using a fitting algorithm according to the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter, wherein the ratio between the number of the first target parameters and the total number of parameters of the first thermal power equipment is less than a preset ratio; Inputting the parameter matrix value into the fitting model to obtain the fitting parameter value output by the fitting model; The optimized first thermal power unit monitoring model is generated according to the first target parameter and the fitting parameter value, as well as the second target parameter and the parameter matrix value corresponding to the second target parameter; wherein the second target parameter is a parameter of the first thermal power equipment other than the first target parameter.

2. The method according to claim 1, characterized in that The method further comprises: In the case where the first condition is met, determining that the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment satisfies a preset similarity difference; The first condition includes: Within a preset time period, the average of the similarities between the first parameter estimation values ​​of all parameters of the first thermal power equipment and the first parameter actual values ​​of the corresponding parameters is less than a first threshold; Within a preset time period, the similarity between a first parameter estimation value of any parameter of the first thermal power equipment and a first parameter actual value of the corresponding parameter is less than a second threshold; The first threshold is greater than the second threshold.

3. The method according to claim 1, characterized in that The fitting algorithm includes at least one of the following: Polynomial fitting algorithm; Exponential function fitting algorithm.

4. The method according to claim 1, characterized in that: The preset neural network model includes at least one of the following: Back propagation BP neural network; Convolutional Neural Network (CNN); Long Short-Term Memory Network LSTM.

5. An intelligent monitoring device for a thermal power unit, characterized in that: include: A first processing module is used to construct a monitoring model of the first thermal power unit according to the historical operating condition data of the operating condition of the first thermal power equipment; A second processing module is used to obtain, based on the operation and maintenance information of the first thermal power equipment, a first parameter estimation value of a parameter of the first thermal power equipment for a monitoring model of a first thermal power unit, when it is determined that the first thermal power equipment has no equipment failure; wherein the operation and maintenance information of the first thermal power equipment includes at least one of the following: normal degradation information of the first thermal power equipment, daily maintenance information of the first thermal power equipment, and maintenance information of the first thermal power equipment; The third processing module is used to optimize the first thermal power unit monitoring model according to the first parameter estimation value, the first parameter actual value and the parameter matrix value when the similarity between the first parameter estimation value and the first parameter actual value of the parameter of the first thermal power equipment meets the preset similarity difference, so as to obtain the optimized first thermal power unit monitoring model; wherein the first thermal power unit monitoring model is a multivariate state estimation technology MSET model; the first thermal power unit monitoring model stores the parameters of the first thermal power equipment and the parameter matrix values ​​of the parameters; A fifth processing module is used to obtain a second parameter actual value of a parameter of a second thermal power device, wherein the second thermal power device is of the same type as the first thermal power device; and the first thermal power device and the second thermal power device are devices of the same type between different thermal power units; A sixth processing module, configured to input the parameter of the second thermal power equipment into the monitoring model of the first thermal power unit, and obtain a second parameter estimation value of the parameter of the second thermal power equipment output by the monitoring model of the first thermal power unit; a seventh processing module, configured to use the estimated value of the second parameter as an input of a preset neural network model, use the actual value of the second parameter as an output of the preset neural network model, train the preset neural network model, and obtain a trained neural network model; An eighth processing module, comprising: a fifth processing unit and a sixth processing unit; The fifth processing unit is used to input the parameter matrix value into the trained neural network model to obtain a second parameter prediction value output by the trained neural network model; The sixth processing unit is used to obtain a second thermal power unit monitoring model of the second thermal power equipment according to the second parameter prediction value and the parameter of the second thermal power equipment; Wherein, the third processing module includes: A first processing unit, configured to use the first parameter estimation value as an input of a preset neural network model, use the first parameter actual value as an output of the preset neural network model, train the preset neural network model, and obtain a trained neural network model; The second processing unit is used to input the parameter matrix value into the trained neural network model to obtain a first parameter prediction value output by the trained neural network model; optimize the first thermal power unit monitoring model according to the first parameter prediction value to obtain the optimized first thermal power unit monitoring model; a third processing unit, configured to construct a fitting model using a fitting algorithm according to the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter, when the similarity between the first parameter estimation value of the first target parameter and the first parameter actual value of the first target parameter satisfies the preset similarity difference, wherein a ratio between the number of the first target parameters and the total number of parameters of the first thermal power equipment is less than a preset ratio; The fourth processing unit is used to input the parameter matrix value into the fitting model to obtain the fitting parameter value output by the fitting model; generate the optimized first thermal power unit monitoring model according to the first target parameter and the fitting parameter value, and the second target parameter and the parameter matrix value corresponding to the second target parameter; wherein the second target parameter is a parameter of the first thermal power equipment other than the first target parameter.

6. An intelligent monitoring device for a thermal power unit, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the steps in the method for intelligent monitoring of a thermal power unit as described in any one of claims 1 to 4 are implemented.

7. A readable storage medium, characterized in that: The readable storage medium stores a program, and when the program is executed by the processor, the steps in the intelligent monitoring method for a thermal power unit as described in any one of claims 1 to 4 are implemented.

8. A computer program product, characterized in that It comprises computer instructions, which, when executed by a processor, implement the steps in the intelligent monitoring method for a thermal power unit as described in any one of claims 1 to 4.

Images