Thermal power plant risk early warning method and system based on deep learning

Abstract

The invention discloses a thermal power plant risk early warning method and system based on deep learning, and relates to the technical field of artificial intelligence. The method comprises the following steps: firstly, based on target risk state parameters and historical time sequence data of all monitoring parameters, aiming at all target risk associated monitoring parameters and all deep learning models, applying all positive sample data of corresponding parameters, and optimizing hyper-parameters of corresponding models based on an optimization algorithm; obtaining an optimal search result which corresponds to the hyper-parameter and is used for minimizing an objective function which adopts a product of an output error condition index value and a time length index value required for calculation, and then obtaining a target risk occurrence prediction model of each monitoring parameter based on the optimization result; according to the method, the prediction probability of the target risk event occurring at the future moment is obtained through model estimation, then the final prediction probability is obtained through superposition calculation, and finally when the final prediction probability reaches the preset probability threshold value, the risk early warning action is triggered and executed, so that the risk early warning accuracy and timeliness can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a risk early warning method and system for a thermal power plant based on deep learning. Background Art

[0002] In the operation of modern thermal power plants, production safety is of vital importance. Although thermal power plants are currently equipped with a variety of digital systems for production safety, such as access control, video and inspection systems, these systems have the following obvious shortcomings: (1) They are isolated from each other and cannot capture the dynamic changes of production safety risks in thermal power plants in real time and comprehensively, resulting in a large number of blind spots in risk monitoring; (2) The operating environment of thermal power plants is complex and there are many risk factors, including equipment failure risks, operational error risks, natural disaster risks, and personnel behavior risks. In the absence of effective information integration and intelligent analysis, it is difficult to accurately assess these complex and changing risk scenarios, and thus it is impossible to make risk warnings and correct decisions in a timely manner. The above situation seriously threatens the safe and stable operation of thermal power plants and may lead to major accidents, such as serious damage to equipment, casualties, and power supply interruptions.

[0003] Therefore, how to construct a real-time risk warning solution for thermal power plants in order to more effectively ensure the safe production of thermal power plants is a topic that technical personnel in this field urgently need to study. Summary of the invention

[0004] The purpose of the present invention is to provide a risk warning method and system for thermal power plants based on deep learning, so as to solve the problems that risk warnings in existing thermal power plant safety production digital systems are difficult to trigger accurately and lack timeliness in the absence of effective integrated information and intelligent analysis.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In the first aspect, a risk early warning method for a thermal power plant based on deep learning is provided, including: Collect the historical time series data of all monitoring parameters recorded by all safety production digital devices of the target thermal power plant, and collect the occurrence timestamps of all target risk events that occurred in the target thermal power plant, wherein the historical time series data contains the time series of all the target risk events that are consecutive in time. The unit time and the corresponding parameters are described in The historical parameter value of each unit moment in the unit moments; According to the occurrence timestamps of all target risk events, the unit moments are traversed in the following manner to obtain the historical time series data of the target risk state parameters: Around time If the target risk event occurs within The corresponding target risk state parameter value is set to 1, otherwise it will be the same as the unit moment The corresponding target risk state parameter value is set to 0, where Represents the unit duration between two adjacent unit moments; Performing feature dimension reduction and feature selection processing on the historical time series data of all the monitoring parameters to obtain historical time series data of multiple target risk-associated monitoring parameters, wherein the target risk-associated monitoring parameters refer to monitoring parameters among all the monitoring parameters that are correlated with the target risk state parameters; For each monitoring parameter in the plurality of target risk associated monitoring parameters, first determine the time when each target risk event occurs according to the historical time series data of the target risk state parameter, then extract each previous time series data corresponding to the time when each target risk event occurs from the corresponding historical time series data, and use each previous time series data as the corresponding positive sample data, wherein the previous time series data refers to the data traced back from a fixed time before the corresponding target risk event occurs to The historical time series data corresponding to the time window of the length of time, the fixed time refers to the time when the corresponding target risk event occurs. At a certain moment in time, Indicates the preset first duration. Indicates the preset second duration; Obtain multiple deep learning models with different structures; For each monitoring parameter and each deep learning model in the multiple deep learning models, all the positive sample data of the corresponding parameter are applied, and the hyperparameters of the corresponding model are optimized based on the optimization algorithm to obtain the corresponding hyperparameters and use them to make the objective function Minimize the optimization search results, where the objective function The calculation formula is as follows:

[0006] In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for calculation of the deep learning model; For each monitoring parameter, select the corresponding and minimum objective function from the multiple deep learning models. A certain deep learning model is obtained, and the optimization search result of the hyperparameters of the certain deep learning model and the model parameters obtained in the optimization process and corresponding to the optimization search result are imported into the certain deep learning model to obtain a corresponding target risk occurrence prediction model; For each monitoring parameter, Going back to the All corresponding parameter values ​​received within the time window of the time length are imported into the corresponding target risk occurrence prediction model, and the corresponding and The predicted probability of the occurrence of the target risk event; According to the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and at the future time The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the occurrence of the target risk event; If the final predicted probability reaches a preset probability threshold, a warning action corresponding to the target risk event is triggered.

[0007] Based on the above invention content, a new scheme for risk early warning of thermal power plants based on deep learning models and optimization algorithms is provided, that is, firstly, based on the historical time series data of target risk state parameters and all monitoring parameters, the historical time series data of multiple target risk-related monitoring parameters are obtained through feature dimension reduction and feature selection processing, and then, for each monitoring parameter and each deep learning model, all positive sample data of the corresponding parameter are applied, and the hyperparameters of the corresponding model are optimized based on the optimization algorithm, and the corresponding hyperparameters are obtained. The optimal search result for minimizing the objective function of the product of the output error situation index value and the calculation required time index value is obtained, and then the target risk occurrence prediction model of each monitoring parameter is obtained based on the optimization result, and the model is used to estimate the predicted probability of the target risk event occurring at a future moment, and then, according to the correlation coefficient between parameters, the model output error and the predicted probability, the final predicted probability is obtained by superposition calculation, and finally, when the final predicted probability reaches the preset probability threshold, the target risk early warning action is triggered. In this way, by effectively integrating the monitoring data of all digital systems for safe production to perform intelligent analysis on the risks of thermal power plants, the accuracy and timeliness of risk early warning can be improved, which is beneficial to ensuring the safe production of thermal power plants and convenient for practical application and promotion.

[0008] In one possible design, all monitoring parameters include monitoring temperature, monitoring pressure, monitoring parameters for reflecting vibration conditions, monitoring parameters for reflecting equipment operation conditions, monitoring parameters for reflecting equipment operation conditions, environmental monitoring parameters, monitoring parameters for reflecting access control operation conditions, monitoring parameters for reflecting video AI analysis results, monitoring parameters for reflecting inspection conditions, and / or monitoring parameters for reflecting personnel behavior monitoring conditions; And / or, the target risk events include equipment failure risk events, operational error risk events, natural disaster risk events or personnel behavior risk events.

[0009] In a possible design, feature dimension reduction and feature selection are performed on the historical time series data of all monitoring parameters to obtain historical time series data of multiple target risk-related monitoring parameters, including: Using an improved principal component analysis method to perform feature dimensionality reduction processing on the historical time series data of all the monitoring parameters, to obtain a plurality of historical time series data of the monitoring parameters after preliminary dimensionality reduction processing; Using a local linear embedding method to perform feature dimensionality reduction processing on the historical time series data of the monitoring parameters after the multiple initial dimensionality reduction processing, to obtain the historical time series data of the monitoring parameters after the multiple secondary dimensionality reduction processing; A maximum relevance minimum redundancy algorithm based on mutual information maximization is used to perform feature selection processing on the historical time series data of the multiple monitoring parameters after secondary dimensionality reduction processing to obtain historical time series data of multiple target risk associated monitoring parameters, wherein the target risk associated monitoring parameters refer to monitoring parameters among all the monitoring parameters and that are correlated with the target risk state parameters.

[0010] In a possible design, feature dimension reduction and feature selection are performed on the historical time series data of all monitoring parameters to obtain historical time series data of multiple target risk-related monitoring parameters, including: For each monitoring parameter among all the monitoring parameters, the corresponding historical time series data is first used as the first sample , and taking the historical time series data of the target risk state parameter as the second sample , then for the first sample Perform a normal distribution KS test to calculate the first test statistic value , and also for the second sample Perform the normal distribution KS check to calculate the second check statistic value , and finally in the first check statistic value and the second check statistic value When both are greater than the first preset threshold, the first sample is calculated according to the following formula With the sample The correlation coefficient As the corresponding correlation coefficient:

[0011] In the formula, represents a positive integer, In the first sample The Sample values, In the second sample The Sample values, Indicates the first sample The sample mean of The second sample The sample mean of ; Determine whether the total number of monitoring parameters whose correlation coefficients are greater than the second preset threshold value among all the monitoring parameters exceeds the third preset threshold value ,in, represents a positive integer greater than 2; If so, the monitoring parameters whose correlation coefficients are greater than the second preset threshold value among all the monitoring parameters are used as the target risk associated monitoring parameters; otherwise, all the monitoring parameters are arranged in descending order according to the correlation coefficients to obtain a monitoring parameter sequence, and then the first monitoring parameter is selected from the monitoring parameter sequence. monitoring parameters are respectively used as target risk associated monitoring parameters, wherein the target risk associated monitoring parameters refer to monitoring parameters among all the monitoring parameters and which are correlated with the target risk state parameter.

[0012] In one possible design, the multiple deep learning models include classification and regression trees, perceptrons, multi-layer perceptrons, recurrent neural networks, long short-term memory networks, autoencoders and / or variational autoencoders.

[0013] In one possible design, the optimization algorithm uses a particle swarm optimization algorithm, a Newton optimization algorithm, a genetic optimization algorithm, a gray wolf optimization algorithm, a whale optimization algorithm, or a tuna school optimization algorithm.

[0014] In one possible design, for each monitoring parameter and each deep learning model in the multiple deep learning models, all the positive sample data of the corresponding parameter are applied, and the hyperparameters of the corresponding model are optimized based on the optimization algorithm to obtain the corresponding hyperparameters and use them to make the objective function The minimized optimized search results include the following steps S601 to S616: S601. For any monitoring parameter among all the monitoring parameters and any deep learning model among the multiple deep learning models, initialize the number of search individuals corresponding to and including the search population , maximum number of iterations , the first learning factor , the second learning factor , the third learning factor , the fourth learning factor , the fifth learning factor and the sixth learning factor The optimization algorithm parameters are set, and the initial position of each search individual in the search population is randomly generated, and then step S602 is performed, wherein the initial position of each search individual is randomly generated based on the following formula:

[0015] In the formula, Indicates less than or equal to A positive integer, Indicates less than or equal to A positive integer, represents the number of dimensions of the parameter vector to be optimized, wherein the parameter vector to be optimized contains all the hyperparameters of any of the deep learning models, Indicates the first The initial position of the search individual in the parameter vector to be optimized is The weight in the dimensions, Indicated in the The upper limit of the parameter search space in dimensions, Indicated in the The lower bound of the parameter search space in dimensions, Represents a purely decimal random generator function; S602. For each search individual, the corresponding initial position is imported into any deep learning model as a model hyperparameter to obtain a corresponding first deep learning model, and then all the positive sample data of any monitoring parameter are applied to train and test the first deep learning model to obtain the corresponding first output error situation index value and the first calculation time required index value, and finally the first output error situation index value and the first calculation time required index value are imported into the objective function , and use the output result as the corresponding fitness, and then execute step S603, where the objective function The calculation formula is as follows:

[0016] In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for calculation of the deep learning model; S603. Take the initial position of a search individual with the minimum fitness as the initial global optimal position , and also initialize the current number of iterations , then execute step S604; S604. Calculate the current average position of the search population according to the current position of each search individual, and determine the corresponding first new position for each search individual, and then execute step S605, wherein the first new position of each search individual is determined according to the following formula:

[0017] In the formula, Indicates the The first new position of the search individual is in the The weight in the dimensions, represents the global optimal position In the said The weight in the dimensions, Indicates the current average position of the search population in the The weight in the dimensions, Indicates the The current location of the search individual is The weight in the dimensions; S605. For each search individual, the corresponding first new position is imported as a model hyperparameter into any deep learning model to obtain a corresponding second deep learning model, and then all the positive sample data of any monitoring parameter are applied to train and test the second deep learning model to obtain the corresponding second output error situation index value and the second calculation time required index value, and finally the second output error situation index value and the second calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S606; S606. Determine the Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The first new position of the search individual is determined, and then step S607 is executed; otherwise, step S607 is directly executed; S607. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S608, otherwise directly execute step S608; S608. The current average position of the search population is calculated based on the current position of each search individual, and the corresponding second new position is determined for each search individual, and then step S609 is executed, wherein the second new position of each search individual is determined according to the following formula:

[0018] In the formula, Indicates the The second new position of the search individual is in the The weight in the dimensions, Indicates that The first intermediate variable corresponding to the search individual, Indicates that The second intermediate variable corresponding to the search individual, Indicates less than or equal to A positive integer, represents the remainder function, Indicates the first The current location of the search individual is The weight in the dimensions, Indicates that The third intermediate variable corresponding to the search individual, Indicates that The fourth intermediate variable corresponding to the search individual, represents the maximum value function, Indicates that The fifth intermediate variable corresponding to the search individual, Indicates that The sixth intermediate variable corresponding to the search individual, Indicates 180 degrees; S609. For each search individual, the corresponding second new position is imported into any of the deep learning models as a model hyperparameter to obtain a corresponding third deep learning model, and then all the positive sample data of any of the monitoring parameters are applied to train and test the third deep learning model to obtain the corresponding third output error situation index value and the third calculation time required index value, and finally the third output error situation index value and the third calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S610; S610. Determine the Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The second new position of the search individual is determined, and then step S611 is executed; otherwise, step S611 is executed directly; S611. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S612, otherwise directly execute step S612; S612. The current average position of the search population is calculated and updated according to the current position of each search individual, and the corresponding third new position is determined for each search individual, and then step S613 is executed, wherein the third new position of each search individual is determined according to the following formula:

[0019] In the formula, Indicates the The third new position of the search individual is The weight in the dimensions, Indicates that The seventh intermediate variable corresponding to the search individual, Indicates that The eighth intermediate variable corresponding to the search individual, Indicates that The ninth intermediate variable corresponding to the search individual, Indicates that The tenth intermediate variable corresponding to the search individual, Indicates that The eleventh intermediate variable corresponding to the search individual; S613. For each search individual, the corresponding third new position is imported into any of the deep learning models as a model hyperparameter to obtain a corresponding fourth deep learning model, and then all the positive sample data of any of the monitoring parameters are applied to train and test the fourth deep learning model to obtain the corresponding fourth output error situation index value and fourth calculation time required index value, and finally the fourth output error situation index value and the fourth calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S614; S614. Judgment and the above Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The third new position of the search individual is determined, and then step S615 is executed; otherwise, step S615 is directly executed; S615. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S616, otherwise directly execute step S616; S616. Make the current number of iterations Add 1 to the current number of iterations. Whether the maximum number of iterations has been reached , if so, then the global optimal position As a hyperparameter of any deep learning model corresponding to any monitoring parameter and used to make the objective function Minimize the optimized search results, otherwise return to execute step S604.

[0020] In a possible design, according to the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error indicator value of the target risk occurrence prediction model corresponding to each monitoring parameter and at the future time The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the target risk event occurring includes: According to the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and at the future time The predicted probability of the target risk event occurring at the future time is calculated according to the following formula: The final predicted probability of the target risk event occurring :

[0021] In the formula, represents the total number of parameters of the plurality of target risk associated monitoring parameters, Indicates less than or equal to A positive integer, Indicates the first monitoring parameter associated with the multiple target risks The influence weight coefficient corresponding to each monitoring parameter is: Indicates the The correlation coefficient between the monitoring parameters and the target risk state parameters, Indicates less than or equal to A positive integer, Indicates the first of the multiple target risk associated monitoring parameters The correlation coefficient between the monitoring parameters and the target risk state parameters, Indicates that The output error indicator value of the target risk occurrence prediction model corresponding to the monitoring parameters, Indicates that The output error indicator value of the target risk occurrence prediction model corresponding to the monitoring parameters, Indicates that The monitoring parameters correspond to the future time The predicted probability of the target risk event occurring.

[0022] In the second aspect, a thermal power plant risk early warning system based on deep learning is provided, which includes a monitoring data collection unit, a risk data sorting unit, a monitoring data processing unit, a positive sample data selection unit, a learning model acquisition unit, a model hyperparameter optimization unit, a prediction model generation unit, a prediction model application unit, a prediction probability superposition unit and a warning action triggering unit; The monitoring data collection unit is used to collect the historical time series data of all monitoring parameters recorded by all safety production digital devices of the target thermal power plant, and to collect the occurrence timestamps of all target risk events occurring in the target thermal power plant, wherein the historical time series data contains the time series data that are consecutive in time sequence. The unit time and the corresponding parameters are described in The historical parameter value of each unit moment in the unit moments; The risk data collating unit is communicatively connected to the monitoring data collecting unit, and is used to traverse each unit time according to the occurrence timestamp of all target risk events in the following manner to obtain the historical time series data of the target risk state parameter: if at the currently traversed unit time Around time If the target risk event occurs within The corresponding target risk state parameter value is set to 1, otherwise it will be the same as the unit moment The corresponding target risk state parameter value is set to 0, where Represents the unit duration between two adjacent unit moments; The monitoring data processing unit is respectively connected to the monitoring data collection unit and the risk data sorting unit for performing feature dimension reduction and feature selection processing on the historical time series data of all the monitoring parameters to obtain the historical time series data of multiple target risk-associated monitoring parameters, wherein the target risk-associated monitoring parameter refers to a monitoring parameter among all the monitoring parameters that is correlated with the target risk state parameter; The positive sample data selection unit is respectively connected to the risk data sorting unit and the monitoring data processing unit in communication, and is used to determine the time when each target risk event occurs according to the historical time series data of the target risk state parameter for each monitoring parameter in the multiple target risk associated monitoring parameters, and then extract each previous time series data corresponding to the time when each target risk event occurs from the corresponding historical time series data, and use each previous time series data as the corresponding positive sample data, wherein the previous time series data refers to the time when each target risk event occurs and is traced back to the previous time series data from a fixed time before the corresponding target risk event occurs. The historical time series data corresponding to the time window of the length of time, the fixed time refers to the time when the corresponding target risk event occurs. At a certain moment in time, Indicates the preset first duration. Indicates the preset second duration; The learning model acquisition unit is used to acquire multiple deep learning models with different structures; The model hyperparameter optimization unit is respectively connected to the positive sample data selection unit and the learning model acquisition unit for communication, and is used to apply all the positive sample data of the corresponding parameters to each monitoring parameter and each deep learning model in the multiple deep learning models, and optimize the hyperparameters of the corresponding model based on the optimization algorithm to obtain the corresponding hyperparameters and use them to make the objective function Minimize the optimization search results, where the objective function The calculation formula is as follows:

[0023] In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for calculation of the deep learning model; The prediction model generation unit is communicatively connected to the model hyperparameter optimization unit, and is used to select the corresponding and minimum objective function from the multiple deep learning models for each monitoring parameter. A certain deep learning model is obtained, and the optimization search result of the hyperparameters of the certain deep learning model and the model parameters obtained in the optimization process and corresponding to the optimization search result are imported into the certain deep learning model to obtain a corresponding target risk occurrence prediction model; The prediction model application unit is communicatively connected to the prediction model generation unit, and is used for each monitoring parameter to be Tracing back to the All corresponding parameter values ​​received within the time window of the time length are imported into the corresponding target risk occurrence prediction model, and the corresponding and The predicted probability of the occurrence of the target risk event; The prediction probability superposition unit is respectively connected to the prediction model application unit and the model hyperparameter optimization unit for calculating the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and the output error situation index value of the target risk occurrence prediction model at the future time. The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the occurrence of the target risk event; The early warning action triggering unit is communicatively connected to the predicted probability superposition unit, and is used to trigger the execution of the early warning action corresponding to the target risk event if the final predicted probability reaches a preset probability threshold.

[0024] In a third aspect, the present invention provides a computer system comprising a memory, a processor and a transceiver which are communicatively connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the thermal power plant risk warning method as described in the first aspect or any possible design of the first aspect.

[0025] In a fourth aspect, the present invention provides a computer-readable storage medium having instructions stored thereon, which, when executed on a computer, executes the thermal power plant risk warning method as described in the first aspect or any possible design of the first aspect.

[0026] In a fifth aspect, the present invention provides a computer program product, comprising a computer program or instructions, which, when executed by a computer, implements the thermal power plant risk warning method as described in the first aspect or any possible design of the first aspect.

[0027] Beneficial effects of the above scheme: (1) The present invention provides a new scheme for thermal power plant risk early warning based on deep learning model and optimization algorithm, that is, firstly, based on the historical time series data of target risk state parameters and all monitoring parameters, the historical time series data of multiple target risk-related monitoring parameters are obtained through feature dimension reduction and feature selection processing, and then, for each monitoring parameter and each deep learning model, all positive sample data of the corresponding parameter are applied, and the hyperparameters of the corresponding model are optimized based on the optimization algorithm, so as to obtain the optimal search result for the corresponding hyperparameter and for minimizing the objective function of the product of the output error situation index value and the calculation required time index value, and then, based on the optimization result, the target risk occurrence prediction model of each monitoring parameter is obtained, and the model is used to estimate the predicted probability of the target risk event occurring at a future time, and then, according to the correlation coefficient between parameters, the model output error and the predicted probability, the final predicted probability is obtained by superposition calculation, and finally, when the final predicted probability reaches the preset probability threshold, the target risk early warning action is triggered. In this way, by effectively integrating the monitoring data of all safety production digital systems to perform intelligent analysis on the risks of thermal power plants, the accuracy and timeliness of risk early warning can be improved, which is beneficial to ensuring the safety production of thermal power plants and convenient for practical application and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 A flowchart of a thermal power plant risk warning method based on deep learning provided in an embodiment of the present application.

[0030] Figure 2 A schematic diagram of the structure of a thermal power plant risk warning system based on deep learning provided in an embodiment of the present application.

[0031] Figure 3 A schematic diagram of the structure of a computer system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structures of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other embodiments can be obtained based on these embodiments without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0033] It should be understood that although the terms first and second, etc. may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another object. For example, a first object can be referred to as a second object, and similarly, a second object can be referred to as a first object without departing from the scope of the exemplary embodiments of the present invention.

[0034] It should be understood that the term "and / or" that may appear in this document is merely a description of the association relationship between associated objects, indicating that there may be three relationships. For example, A and / or B can indicate three situations: A exists alone, B exists alone, or A and B exist at the same time. For another example, A, B and / or C can indicate the existence of any one of A, B and C or any combination of them. The term " / and" that may appear in this document describes another type of association object relationship, indicating that there may be two relationships. For example, A / and B can indicate two situations: A exists alone or A and B exist at the same time. In addition, the character " / " that may appear in this document generally indicates that the previous and next associated objects are in an "or" relationship.

[0035] Example like Figure 1 As shown, the thermal power plant risk warning method based on deep learning provided in the first aspect of this embodiment can be executed by, but not limited to, a computer system with certain computing resources, such as an edge device, a cloud server, a personal computer (PC, a multi-purpose computer with a size, price and performance suitable for personal use; desktops, laptops, small laptops, tablets and ultrabooks are all personal computers), smart phones, personal digital assistants (PDA) or wearable devices and other electronic devices. Figure 1 As shown, the thermal power plant risk early warning method may include but is not limited to the following steps S1 to S10.

[0036] S1. Collect the historical time series data of all monitoring parameters recorded by all safety production digital devices of the target thermal power plant, and collect the occurrence timestamps of all target risk events that occurred in the target thermal power plant, wherein the historical time series data contains the time series of all the target risk events that are consecutive in time. The unit time and the corresponding parameters are described in The historical parameter values ​​of each unit moment in each unit moment.

[0037] In step S1, the target thermal power plant is the object for risk warning, and the digital production safety device is an existing system such as access control, video or inspection, which can routinely record and obtain all the monitoring parameters. Specifically, all the monitoring parameters include, but are not limited to, monitoring temperature, monitoring pressure, monitoring parameters for reflecting vibration conditions (such as vibration amplitude and / or vibration frequency, etc.), monitoring parameters for reflecting equipment operation conditions (such as operation instruction execution records, etc.), monitoring parameters for reflecting equipment operation conditions (such as operation parameters, usage time and / or recent maintenance records, etc.), environmental monitoring parameters (such as meteorological parameters and / or geological parameters, etc.), monitoring parameters for reflecting access control operation conditions (such as the number of people entering and exiting and / or the frequency of entry and exit, etc.), monitoring parameters for reflecting video AI (Artificial Intelligence) analysis results (such as the total number of people in the field of view, etc.), monitoring parameters for reflecting inspection conditions (such as inspection records, etc.) and / or monitoring parameters for reflecting personnel behavior monitoring conditions (such as personnel activity trajectories, movement frequency, movement standardization and / or proximity to dangerous areas, etc.); the target risk events specifically include, but are not limited to, equipment failure risk events, operation error risk events, natural disaster risk events or personnel behavior risk events. The unit time may be, but is not limited to, every 10 seconds, every 30 seconds, or every 1 minute, etc., that is, the historical time series data may include, but is not limited to, the monitoring parameters in the most recent The historical parameter values ​​at the minute, The historical parameter values ​​at the last minute, ..., the historical parameter values ​​at the last second minute, and the historical parameter values ​​at the last minute, etc. In addition, after collecting the historical time series data of all the monitoring parameters, it is necessary to perform routine data cleaning on the historical time series data of the monitoring parameters through comparative analysis according to the pre-set data quality standards, so as to remove errors and abnormal values; for example, when the temperature value of a certain device exceeds the historical normal range by 3 times the standard deviation, its validity is judged in combination with the temperature of adjacent devices and the operating status of the device and other related data, and if it is invalid, it is corrected or deleted; for personnel behavior data, if there is an unreasonable action trajectory or behavior pattern (such as abnormal stay in a dangerous area, illegal operation, etc.), it is identified and processed by comparing with standard operating procedures and safety regulations.

[0038] S2. According to the occurrence timestamps of all target risk events, traverse each unit time to obtain the historical time series data of the target risk state parameter in the following manner: if at the currently traversed unit time Around time If the target risk event occurs within The corresponding target risk state parameter value is set to 1, otherwise it will be the same as the unit moment The corresponding target risk state parameter value is set to 0, where Represents the unit duration between two adjacent unit moments.

[0039] In step S2, for example, if the unit time length is 10 minutes, and the occurrence timestamps of all target risk events are 10:56, 11:18, 11:32 and 12:01 respectively, the target risk status parameter value corresponding to the unit time 10:50 can be set to 0, the target risk status parameter value corresponding to the unit time 11:00 can be set to 1, the target risk status parameter value corresponding to the unit time 11:10 can be set to 0, the target risk status parameter value corresponding to the unit time 11:20 can be set to 1, the target risk status parameter value corresponding to the unit time 11:30 can be set to 1, the target risk status parameter value corresponding to the unit time 11:40 can be set to 0, the target risk status parameter value corresponding to the unit time 11:50 can be set to 0, and the target risk status parameter value corresponding to the unit time 12:00 can be set to 1, so that the historical time series data of the target risk status parameter can be obtained as {0,1,0,1,1,0,0,1}.

[0040] S3. Perform feature dimension reduction and feature selection processing on the historical time series data of all the monitoring parameters to obtain historical time series data of multiple target risk-associated monitoring parameters, wherein the target risk-associated monitoring parameters refer to monitoring parameters among all the monitoring parameters that are correlated with the target risk status parameters.

[0041] In step S3, it is considered that different risk events may have different correlations with different monitoring parameters. For example, for equipment failure risk, there are generally strong correlations with equipment operating parameters, usage time and / or recent maintenance records; for operational error risk, there are generally strong correlations with the sequence of operating steps and / or deviation of operating time from standard time; for natural disaster risk, there are generally strong correlations with meteorological parameters and / or geological parameters; for personnel behavior risk, there are generally strong correlations with artificial activity trajectory, action frequency, action standardization and / or approach procedures to dangerous areas. Specifically, feature dimension reduction and feature selection processing are performed on the historical time series data of all monitoring parameters to obtain historical time series data of multiple target risk-related monitoring parameters, including but not limited to the following steps S311 to S313.

[0042] S311. Use an improved principal component analysis method to perform feature dimensionality reduction processing on the historical time series data of all the monitoring parameters, and obtain multiple historical time series data of the monitoring parameters after preliminary dimensionality reduction processing.

[0043] In step S311, the improved principal component analysis method is one of the dimensionality reduction processing technologies improved based on the existing principal component analysis (PCA), and its specific principle is as follows: Assume that the original data matrix is ,in, represents the number of samples, Represents the number of features; calculates the covariance matrix , for the matrix Perform eigenvalue decomposition to obtain eigenvalues and the eigenvector ; Sort by eigenvalue and select the top ones whose cumulative contribution rate reaches a certain threshold (e.g. 90%) The eigenvectors form the projection matrix , get the initial dimension reduction data .

[0044] S312. Use the local linear embedding method to perform feature dimensionality reduction processing on the historical time series data of the monitoring parameters after the multiple initial dimensionality reduction processing, and obtain the historical time series data of the monitoring parameters after the multiple secondary dimensionality reduction processing.

[0045] In step S312, the Locally Linear Embedding (LLE) method is one of the commonly used dimensionality reduction processing technologies. The specific principle is as follows: for each data point , find the reconstruction weight in the local area formed by its neighboring points (determined by the K nearest neighbor algorithm) , so that , express The set of neighboring points of ; By solving the problem of minimizing the reconstruction error To get the weight matrix ; Then calculate the low-dimensional embedding coordinates , so that ,in, Represents the identity matrix; finally take the smallest The eigenvectors corresponding to the non-zero eigenvalues ​​are , to achieve further dimensionality reduction.

[0046] S313. Use the maximum relevance minimum redundancy algorithm based on mutual information maximization to perform feature selection processing on the historical time series data of the multiple monitoring parameters after secondary dimensionality reduction processing, and obtain historical time series data of multiple target risk-associated monitoring parameters, wherein the target risk-associated monitoring parameters refer to the monitoring parameters among all the monitoring parameters and that are correlated with the target risk state parameters.

[0047] In step S313, the maximum relevance and minimum redundancy algorithm (mRMR) is an existing algorithm mainly used for feature selection. The specific principle is as follows: calculate each feature With target risk variable (i.e., the target risk state parameter in this embodiment) , and calculate the features With other features Mutual information between Then apply the maximum correlation minimum redundancy algorithm to select The biggest feature Add to selected feature set , until a stopping condition is met (such as selecting a certain number of features).

[0048] In the step S3, the feature dimension reduction and feature selection processing can also be performed directly based on the correlation coefficient calculation result, that is, specifically, the historical time series data of all the monitoring parameters are subjected to feature dimension reduction and feature selection processing to obtain the historical time series data of multiple target risk-related monitoring parameters, including but not limited to the following steps S321 to S323.

[0049] S321. For each monitoring parameter among all the monitoring parameters, first use the corresponding historical time series data as the first sample , and taking the historical time series data of the target risk state parameter as the second sample , then for the first sample Perform a normal distribution KS test to calculate the first test statistic value , and also for the second sample Perform the normal distribution KS check to calculate the second check statistic value , and finally in the first check statistic value and the second check statistic value When both are greater than the first preset threshold, the first sample is calculated according to the following formula With the sample The correlation coefficient As the corresponding correlation coefficient:

[0050] In the formula, represents a positive integer, In the first sample The Sample values, In the second sample The Sample values, Indicates the first sample The sample mean of The second sample The sample mean of .

[0051] In the step S321, the first preset threshold may be 0.05, for example.

[0052] S322. Determine whether the total number of monitoring parameters whose correlation coefficients are greater than the second preset threshold value among all monitoring parameters exceeds the third preset threshold value. ,in, Represents a positive integer greater than 2.

[0053] In step S322, the second preset threshold may be 0.3, and the third preset threshold may be An example is 10.

[0054] S323. If yes, then the monitoring parameters with correlation coefficients greater than the second preset threshold value among all the monitoring parameters are used as the target risk associated monitoring parameters; otherwise, all the monitoring parameters are first arranged in descending order according to the correlation coefficients to obtain a monitoring parameter sequence, and then the first monitoring parameter is selected from the monitoring parameter sequence. monitoring parameters are respectively used as target risk associated monitoring parameters, wherein the target risk associated monitoring parameters refer to monitoring parameters among all the monitoring parameters and which are correlated with the target risk state parameter.

[0055] Based on the above steps S321 to S323, at least The historical time series data of the target risk associated monitoring parameters.

[0056] S4. For each monitoring parameter among the multiple target risk associated monitoring parameters, first determine the time when each target risk event occurs according to the historical time series data of the target risk state parameter, then extract each previous time series data corresponding to the time when each target risk event occurs from the corresponding historical time series data, and use each previous time series data as the corresponding positive sample data, wherein the previous time series data refers to the data traced back from a fixed time before the corresponding target risk event occurs to The historical time series data corresponding to the time window of the length of time, the fixed time refers to the time when the corresponding target risk event occurs. At a certain moment in time, Indicates the preset first duration. Indicates the preset second duration.

[0057] In step S4, the target risk event occurrence time is a unit time aligned by step S2 and close to the occurrence timestamp of the target risk event, so it can be determined according to the historical time series data of the target risk state parameter. The duration can be determined based on the duration required to respond to the target risk event, and generally needs to be greater than the shortest duration required to respond to the target risk event, such as 10 minutes. The duration can be set based on historical experience, such as 2 hours. For example, if a target risk event occurs at 18:10, the fixed time is 18:00, and the time series data from 16:01 to 18:00 can be extracted from the historical time series data of the monitoring parameter as a positive sample data corresponding to the monitoring parameter.

[0058] S5. Obtain multiple deep learning models with different structures.

[0059] In the step S5, deep learning refers specifically to machine learning based on deep neural network models and methods; it is developed based on algorithm models such as statistical machine learning and artificial neural networks, combined with the development of contemporary big data and large computing power. The most important technical feature of deep learning is the ability to automatically extract features. The extracted features are also called deep features or deep feature representations. Compared with artificially designed features, deep features have stronger and more robust representation capabilities. Specifically, the multiple deep learning models include but are not limited to classification and regression trees (Classification and regression tree, abbreviated as CART, is a learning method for conditional probability distribution of output random variable Y under given input random variable X), perceptron, multi-layer perceptron, recurrent neural network, long short-term memory network, autoencoder and / or variational autoencoder, etc. In addition, the specific method of obtaining the deep learning model is an existing means. Taking the bidirectional LSTM (Long Short-Term Memory) model that has been injected with the self-attention mechanism as an example, it can be obtained according to the following steps: first select the target model building tool (such as pytorch), and introduce the library and module for model building on the target model building tool; then determine the input sequence information, and based on the library and module, construct a self-attention mechanism layer by setting different weights for the input sequence information; then, based on the library and module, construct the original bidirectional LSTM model according to the forward information and reverse information of the input sequence information; finally, integrate the self-attention mechanism layer into the original bidirectional LSTM model to obtain the bidirectional LSTM model that has been injected with the self-attention mechanism.

[0060] S6. For each monitoring parameter and each deep learning model in the multiple deep learning models, apply all the positive sample data of the corresponding parameter, optimize the hyperparameters of the corresponding model based on the optimization algorithm, obtain the corresponding hyperparameters and use them to make the objective function Minimize the optimization search results, where the objective function The calculation formula is as follows:

[0061] In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for deep learning model calculation.

[0062] In step S6, specifically, for each pair of monitoring parameters and deep learning models, all the positive sample data of the corresponding parameters are applied, and the hyperparameters of the corresponding model are optimized based on the optimization algorithm to obtain the corresponding hyperparameters and use them to make the objective function Minimize the optimized search results. Model hyperparameters refer to the parameters pre-set in the learning model. These parameters cannot be directly learned from the data in the standard model training process, but need to be manually set to optimize the performance of the model, such as the number of hidden layer nodes, learning rate and batch size. The output error situation index value is used to reflect the accuracy of model detection. The smaller the value, the higher the model detection accuracy. It specifically includes but is not limited to the average deviation value, variance and / or average error rate of the overall samples, and can be obtained by conventional statistics during the model training and testing process. The calculation time required index value is used to reflect the model detection speed. The shorter the value, the faster the model detection speed, which can be obtained by conventional timing during the model testing process. The objective function It is used as a comprehensive indicator that takes into account both detection accuracy and detection speed. The smaller the result, the better the trained model can achieve in terms of both detection accuracy and detection speed, so as to select the optimal target risk occurrence prediction model that takes into account both detection accuracy and detection speed. In addition, the optimization algorithm may be, but is not limited to, a particle swarm optimization algorithm, a Newton optimization algorithm, a genetic optimization algorithm, a Grey Wolf Algorithm, a whale optimization algorithm, or a tuna school optimization algorithm.

[0063] In step S6, considering that different optimization algorithms have different advantages and disadvantages, in order to comprehensively integrate the performance of the aforementioned optimization algorithms to achieve the purpose of leveraging strengths and avoiding weaknesses, this embodiment preferably applies all the positive sample data of the corresponding parameters to each monitoring parameter and each deep learning model in the multiple deep learning models, optimizes the hyperparameters of the corresponding model based on the optimization algorithm, obtains the corresponding hyperparameters and uses them to make the objective function The minimized optimized search results include but are not limited to the following steps S601 to S616.

[0064] S601. For any monitoring parameter among all the monitoring parameters and any deep learning model among the multiple deep learning models, initialize the number of search individuals corresponding to and including the search population , maximum number of iterations , the first learning factor , the second learning factor , the third learning factor , the fourth learning factor , the fifth learning factor and the sixth learning factor The optimization algorithm parameters are set, and the initial position of each search individual in the search population is randomly generated, and then step S602 is performed, wherein the initial position of each search individual is randomly generated based on the following formula:

[0065] In the formula, Indicates less than or equal to A positive integer, Indicates less than or equal to A positive integer, represents the number of dimensions of the parameter vector to be optimized, wherein the parameter vector to be optimized contains all the hyperparameters of any of the deep learning models, Indicates the first The initial position of the search individual in the parameter vector to be optimized is The weight in the dimensions, Indicated in the The upper limit of the parameter search space in dimensions, Indicated in the The lower bound of the parameter search space in dimensions, Represents a purely decimal random generator.

[0066] In step S601, specifically, the corresponding optimization algorithm parameters are initialized for each pair of monitoring parameters and deep learning models. In addition, considering that different deep learning models may have different model hyperparameters, the parameter vector to be optimized may be different for different deep learning models.

[0067] S602. For each search individual, the corresponding initial position is imported into any deep learning model as a model hyperparameter to obtain a corresponding first deep learning model, and then all the positive sample data of any monitoring parameter are applied to train and test the first deep learning model to obtain the corresponding first output error situation index value and the first calculation time required index value, and finally the first output error situation index value and the first calculation time required index value are imported into the objective function , and use the output result as the corresponding fitness, and then execute step S603, where the objective function The calculation formula is as follows:

[0068] In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for deep learning model calculation.

[0069] In the step S602, considering that some hyperparameters need to be integers (such as the number of hidden layer nodes), it is necessary to round the corresponding parameter values ​​before importing the initial position into the model as a model hyperparameter, such as rounding the hidden layer node values. The aforementioned application of all the positive sample data of any monitoring parameter to train and test the first deep learning model is an existing conventional calibration verification method and will not be repeated here. In addition, when obtaining the first output error situation index value and the first calculation time required index value, it is also necessary to record the model parameters corresponding to the initial position and obtained after model training, so as to use them in the subsequent step S7, and it is also necessary to record the first output error situation index value, so as to use it in the subsequent step S9.

[0070] S603. Take the initial position of a search individual with the minimum fitness as the initial global optimal position , and also initialize the current number of iterations , then execute step S604.

[0071] S604. Calculate the current average position of the search population according to the current position of each search individual, and determine the corresponding first new position for each search individual, and then execute step S605, wherein the first new position of each search individual is determined according to the following formula:

[0072] In the formula, Indicates the The first new position of the search individual is in the The weight in the dimensions, represents the global optimal position In the said The weight in the dimensions, Indicates the current average position of the search population in the The weight in the dimensions, Indicates the The current location of the search individual is The weight in a dimension.

[0073] S605. For each search individual, the corresponding first new position is imported as a model hyperparameter into any deep learning model to obtain a corresponding second deep learning model, and then all the positive sample data of any monitoring parameter are applied to train and test the second deep learning model to obtain the corresponding second output error situation index value and the second calculation time required index value, and finally the second output error situation index value and the second calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S606.

[0074] In the step S605, it is also considered that some hyperparameters need to be integers (such as the number of hidden layer nodes). Therefore, before the first new position is imported into the model as a model hyperparameter, it is necessary to round the corresponding parameter values, such as rounding the hidden layer node values. The aforementioned application of all the positive sample data of any monitoring parameter to train and test the second deep learning model is also an existing conventional calibration verification method and will not be repeated here. In addition, when obtaining the second output error situation index value and the second calculation time required index value, it is also necessary to record the model parameters corresponding to the first new position and obtained after model training, so as to use them in the subsequent step S7, and it is also necessary to record the second output error situation index value, so as to use it in the subsequent step S9.

[0075] S606. Determine the Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The first new position of the search individual is determined, and then step S607 is executed; otherwise, step S607 is directly executed.

[0076] S607. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S608, otherwise directly execute step S608.

[0077] S608. The current average position of the search population is calculated based on the current position of each search individual, and the corresponding second new position is determined for each search individual, and then step S609 is executed, wherein the second new position of each search individual is determined according to the following formula:

[0078] In the formula, Indicates the The second new position of the search individual is in the The weight in the dimensions, Indicates that The first intermediate variable corresponding to the search individual, Indicates that The second intermediate variable corresponding to the search individual, Indicates less than or equal to A positive integer, represents the remainder function, Indicates the first The current location of the search individual is The weight in the dimensions, Indicates that The third intermediate variable corresponding to the search individual, Indicates that The fourth intermediate variable corresponding to the search individual, represents the maximum value function, Indicates that The fifth intermediate variable corresponding to the search individual, Indicates that The sixth intermediate variable corresponding to the search individual, Indicates 180 degrees.

[0079] S609. For each search individual, the corresponding second new position is imported into any of the deep learning models as a model hyperparameter to obtain a corresponding third deep learning model, and then all the positive sample data of any of the monitoring parameters are applied to train and test the third deep learning model to obtain the corresponding third output error situation index value and the third calculation time required index value, and finally the third output error situation index value and the third calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S610.

[0080] In the step S609, it is also considered that some hyperparameters need to be integers (such as the number of hidden layer nodes). Therefore, before the second new position is imported into the model as a model hyperparameter, it is necessary to round the corresponding parameter values, such as rounding the hidden layer node values. The aforementioned application of all the positive sample data of any monitoring parameter to train and test the third deep learning model is also an existing conventional calibration verification method, which will not be repeated here. In addition, when obtaining the third output error situation index value and the third calculation time required index value, it is also necessary to record the model parameters corresponding to the second new position and obtained after model training, so as to use them in the subsequent step S7, and it is also necessary to record the third output error situation index value, so as to use it in the subsequent step S9.

[0081] S610. Determine the Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The second new position of the search individual is determined, and then step S611 is executed; otherwise, step S611 is directly executed.

[0082] S611. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S612, otherwise directly execute step S612.

[0083] S612. The current average position of the search population is calculated based on the current position of each search individual, and the corresponding third new position is determined for each search individual, and then step S613 is executed, wherein the third new position of each search individual is determined according to the following formula:

[0084] In the formula, Indicates the The third new position of the search individual is The weight in the dimensions, Indicates that The seventh intermediate variable corresponding to the search individual, Indicates that The eighth intermediate variable corresponding to the search individual, Indicates that The ninth intermediate variable corresponding to the search individual, Indicates that The tenth intermediate variable corresponding to the search individual, Indicates that The eleventh intermediate variable corresponding to the search individual.

[0085] S613. For each search individual, the corresponding third new position is imported into any of the deep learning models as a model hyperparameter to obtain a corresponding fourth deep learning model, and then all the positive sample data of any of the monitoring parameters are applied to train and test the fourth deep learning model to obtain the corresponding fourth output error situation index value and fourth calculation time required index value, and finally the fourth output error situation index value and the fourth calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S614.

[0086] In the step S613, it is also considered that some hyperparameters need to be integers (such as the number of hidden layer nodes). Therefore, before the third new position is imported into the model as a model hyperparameter, it is necessary to round the corresponding parameter values, such as rounding the hidden layer node values. The aforementioned application of all the positive sample data of any monitoring parameter to train and test the fourth deep learning model is also an existing conventional calibration verification method, which will not be repeated here. In addition, when obtaining the fourth output error situation index value and the fourth calculation time required index value, it is also necessary to record the model parameters corresponding to the third new position and obtained after model training, so as to use them in the subsequent step S7, and it is also necessary to record the fourth output error situation index value, so as to use it in the subsequent step S9.

[0087] S614. Judgment and the above Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The third new position of the search individual is determined, and then step S615 is executed; otherwise, step S615 is directly executed.

[0088] S615. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S616, otherwise directly execute step S616.

[0089] S616. Make the current number of iterations Add 1 to the current number of iterations. Whether the maximum number of iterations has been reached , if so, then the global optimal position As a hyperparameter of any deep learning model corresponding to any monitoring parameter and used to make the objective function Minimize the optimized search results, otherwise return to execute step S604.

[0090] Based on the above steps S601 to S616, since each search individual performs three fitness calculations with the help of spirals and group centers in each iteration, it can be regarded as a fusion of the particle swarm optimization algorithm and the whale optimization algorithm. Therefore, under all the same conditions, its optimization performance is better than other optimization algorithms, which is conducive to quickly and accurately obtaining the hyperparameters and using them to make the objective function Minimize the optimization of search results.

[0091] S7. For each monitoring parameter, select the corresponding and minimum objective function from the multiple deep learning models. A certain deep learning model is obtained, and the optimization search result of the hyperparameters of the certain deep learning model and the model parameters obtained during the optimization process and corresponding to the optimization search result are imported into the certain deep learning model to obtain the corresponding target risk occurrence prediction model.

[0092] In step S7, specifically, according to the global optimal position of each deep learning model The corresponding fitness, if the global optimal position of a model is found The corresponding fitness is the minimum value, then the model is regarded as the one with the minimum objective function The optimal search result of the hyperparameters of the deep learning model is the global optimal position of the deep learning model. , also considering that some hyperparameters need to be integers (such as the number of hidden layer nodes), it is necessary to round the corresponding parameter values ​​before importing the optimization search results of the hyperparameters of the deep learning model as model hyperparameters into the model, such as rounding the hidden layer node values. In addition, the model parameters obtained in the optimization process and corresponding to the optimization search results are the model parameters recorded in the aforementioned S602, S604, S609 or S613 and obtained after model training.

[0093] S8. For each monitoring parameter, the Going back to the All corresponding parameter values ​​received within the time window of the time length are imported into the corresponding target risk occurrence prediction model, and the corresponding and The predicted probability of the target risk event occurring.

[0094] In step S8, for example, if the monitoring parameter is the monitoring temperature, the current moment At 20:00, The duration is 2 hours. If the duration is 10 minutes, the temperature parameter values ​​received from 18:01 to 20:00 can be imported into the target risk occurrence prediction model corresponding to the monitored temperature, and the predicted probability of the target risk event occurring at 20:10 corresponding to the monitored temperature can be output.

[0095] S9. Based on the correlation coefficients between the monitoring parameters and the target risk state parameters and the output error indicator values ​​of the target risk occurrence prediction model corresponding to the monitoring parameters and the output error indicator values ​​of the target risk occurrence prediction model at the future time The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the target risk event occurring.

[0096] In the step S9, the correlation coefficients between the various monitoring parameters and the target risk state parameters can be specifically calculated based on the aforementioned step S321. The output error situation index value of the target risk occurrence prediction model corresponding to the various monitoring parameters is the output error situation index value recorded in the aforementioned S602, S604, S609 or S613 and obtained through model verification. Since the correlation coefficient can be positively correlated with the influence weight coefficient of the detection parameter on the target risk state parameter (that is, the larger the correlation coefficient, the larger the influence weight coefficient), and the output error situation index value is negatively correlated with the influence weight coefficient of the detection parameter on the target risk state parameter (that is, the lower the output error situation index value, the larger the influence weight coefficient), it is possible to calculate the output error situation index value of the target risk occurrence prediction model corresponding to the various monitoring parameters based on the correlation coefficients between the various monitoring parameters and the target risk state parameters and the output error situation index value of the target risk occurrence prediction model corresponding to the various monitoring parameters and at the future moment. The predicted probability of the target risk event occurring is realized at the future moment. Specifically, according to the correlation coefficients between the monitoring parameters and the target risk state parameters and the output error index values ​​of the target risk occurrence prediction model corresponding to the monitoring parameters and the target risk occurrence prediction model at the future time, The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the target risk event occurring includes, but is not limited to: the output error indicator value of the target risk occurrence prediction model corresponding to each monitoring parameter and the output error indicator value of the target risk occurrence prediction model corresponding to each monitoring parameter at the future time. The predicted probability of the target risk event occurring at the future time is calculated according to the following formula: The final predicted probability of the target risk event occurring :

[0097] In the formula, represents the total number of parameters of the plurality of target risk associated monitoring parameters, Indicates less than or equal to A positive integer, Indicates the first monitoring parameter associated with the multiple target risks The influence weight coefficient corresponding to each monitoring parameter is: Indicates the The correlation coefficient between the monitoring parameters and the target risk state parameters, Indicates less than or equal to A positive integer, Indicates the first of the multiple target risk associated monitoring parameters The correlation coefficient between the monitoring parameters and the target risk state parameters, Indicates that The output error indicator value of the target risk occurrence prediction model corresponding to the monitoring parameters, Indicates that The output error indicator value of the target risk occurrence prediction model corresponding to the monitoring parameters, Indicates that The monitoring parameters correspond to the future time The predicted probability of the target risk event occurring.

[0098] S10. If the final predicted probability reaches a preset probability threshold, a warning action corresponding to the target risk event is triggered.

[0099] In the step S10, the preset probability threshold may be 85% for example. The warning action may be, but not limited to, an audible and visual alarm action, and may be triggered and executed based on conventional techniques.

[0100] Therefore, based on the thermal power plant risk warning method described in the aforementioned steps S1 to S10, a new scheme for thermal power plant risk warning based on deep learning model and optimization algorithm is provided, that is, firstly, based on the historical time series data of the target risk state parameters and all monitoring parameters, the historical time series data of multiple target risk-related monitoring parameters are obtained through feature dimension reduction and feature selection processing, and then, for each monitoring parameter and each deep learning model, all positive sample data of the corresponding parameter are applied, and the hyperparameters of the corresponding model are optimized based on the optimization algorithm, and the corresponding hyperparameters and the optimal search result for minimizing the objective function of the product of the output error situation index value and the calculation required time index value are obtained, and then the target risk occurrence prediction model of each monitoring parameter is obtained based on the optimization result, and the model is used to estimate the predicted probability of the target risk event occurring at a future moment, and then the final predicted probability is obtained by superimposing the calculation according to the correlation coefficient between the parameters, the model output error and the predicted probability, and finally, when the final predicted probability reaches the preset probability threshold, the target risk warning action is triggered. In this way, by effectively integrating the monitoring data of all digital systems of safe production to perform intelligent analysis on the risks of thermal power plants, the accuracy and timeliness of risk warning can be improved, which is beneficial to ensuring the safe production of thermal power plants and convenient for practical application and promotion.

[0101] like Figure 2 As shown, the second aspect of this embodiment provides a virtual system for implementing the thermal power plant risk warning method described in the first aspect, including but not limited to a monitoring data collection unit, a risk data sorting unit, a monitoring data processing unit, a positive sample data selection unit, a learning model acquisition unit, a model hyperparameter optimization unit, a prediction model generation unit, a prediction model application unit, a prediction probability superposition unit and a warning action triggering unit; The monitoring data collection unit is used to collect the historical time series data of all monitoring parameters recorded by all safety production digital devices of the target thermal power plant, and to collect the occurrence timestamps of all target risk events occurring in the target thermal power plant, wherein the historical time series data contains the time series data that are consecutive in time sequence. The unit time and the corresponding parameters are described in The historical parameter value of each unit moment in the unit moments; The risk data collating unit is communicatively connected to the monitoring data collecting unit, and is used to traverse each unit time according to the occurrence timestamp of all target risk events in the following manner to obtain the historical time series data of the target risk state parameter: if at the currently traversed unit time Around time If the target risk event occurs within The corresponding target risk state parameter value is set to 1, otherwise it will be the same as the unit moment The corresponding target risk state parameter value is set to 0, where Represents the unit duration between two adjacent unit moments; The monitoring data processing unit is respectively connected to the monitoring data collection unit and the risk data sorting unit for performing feature dimension reduction and feature selection processing on the historical time series data of all the monitoring parameters to obtain the historical time series data of multiple target risk-associated monitoring parameters, wherein the target risk-associated monitoring parameter refers to a monitoring parameter among all the monitoring parameters that is correlated with the target risk state parameter; The positive sample data selection unit is respectively connected to the risk data sorting unit and the monitoring data processing unit in communication, and is used to determine the time when each target risk event occurs according to the historical time series data of the target risk state parameter for each monitoring parameter in the multiple target risk associated monitoring parameters, and then extract each previous time series data corresponding to the time when each target risk event occurs from the corresponding historical time series data, and use each previous time series data as the corresponding positive sample data, wherein the previous time series data refers to the time when each target risk event occurs and is traced back to the previous time series data from a fixed time before the corresponding target risk event occurs. The historical time series data corresponding to the time window of the length of time, the fixed time refers to the time when the corresponding target risk event occurs. At a certain moment in time, Indicates the preset first duration. Indicates the preset second duration; The learning model acquisition unit is used to acquire multiple deep learning models with different structures; The model hyperparameter optimization unit is respectively connected to the positive sample data selection unit and the learning model acquisition unit for communication, and is used to apply all the positive sample data of the corresponding parameters to each monitoring parameter and each deep learning model in the multiple deep learning models, and optimize the hyperparameters of the corresponding model based on the optimization algorithm to obtain the corresponding hyperparameters and use them to make the objective function Minimize the optimization search results, where the objective function The calculation formula is as follows:

[0102] In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for calculation of the deep learning model; The prediction model generation unit is communicatively connected to the model hyperparameter optimization unit, and is used to select the corresponding and minimum objective function from the multiple deep learning models for each monitoring parameter. A certain deep learning model is obtained, and the optimization search result of the hyperparameters of the certain deep learning model and the model parameters obtained in the optimization process and corresponding to the optimization search result are imported into the certain deep learning model to obtain a corresponding target risk occurrence prediction model; The prediction model application unit is communicatively connected to the prediction model generation unit, and is used for each monitoring parameter to be Going back to the All corresponding parameter values ​​received within the time window of the time length are imported into the corresponding target risk occurrence prediction model, and the corresponding and The predicted probability of the occurrence of the target risk event; The prediction probability superposition unit is respectively connected to the prediction model application unit and the model hyperparameter optimization unit for calculating the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and the output error situation index value of the target risk occurrence prediction model at the future time. The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the occurrence of the target risk event; The early warning action triggering unit is communicatively connected to the predicted probability superposition unit, and is used to trigger the execution of the early warning action corresponding to the target risk event if the final predicted probability reaches a preset probability threshold.

[0103] The working process, working details and technical effects of the aforementioned system provided in the second aspect of this embodiment can be referred to the risk warning method for a thermal power plant described in the first aspect, and will not be described in detail here.

[0104] like Figure 3As shown, the third aspect of this embodiment provides a computer system for executing the thermal power plant risk warning method as described in the first aspect, including a memory, a processor and a transceiver that are sequentially connected in communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the thermal power plant risk warning method as described in the first aspect. For example, the memory may include, but is not limited to, a random access memory (Random-Access Memory, RAM), a read-only memory (Read-Only Memory, ROM), a flash memory (Flash Memory), a first-in-first-out memory (First Input First Output, FIFO) and / or a first-in-last-out memory (First Input Last Output, FILO), etc.; the processor may include, but is not limited to, a microprocessor of the STM32F105 series. In addition, the computer device may also include, but is not limited to, a power module, a display screen and other necessary components.

[0105] The working process, working details and technical effects of the aforementioned computer system provided in the third aspect of this embodiment can be referred to the risk warning method for a thermal power plant as described in the first aspect, and will not be described in detail here.

[0106] The fourth aspect of this embodiment provides a computer-readable storage medium storing instructions including the thermal power plant risk warning method as described in the first aspect, that is, the computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the thermal power plant risk warning method as described in the first aspect is executed. The computer-readable storage medium refers to a carrier for storing data, which may include but is not limited to computer-readable storage media such as floppy disks, optical disks, hard disks, flash memories, USB flash drives, and / or memory sticks, and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0107] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment can be referred to the risk warning method for a thermal power plant as described in the first aspect, and will not be described in detail here.

[0108] In a fifth aspect of this embodiment, a computer program product is provided, including a computer program or an instruction, wherein when the computer program or the instruction is executed by a computer, the thermal power plant risk warning method as described in the first aspect is implemented. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0109] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A thermal power plant risk early warning method based on deep learning, characterized in that: include: Collect the historical time series data of all monitoring parameters recorded by all safety production digital devices of the target thermal power plant, and collect the occurrence timestamps of all target risk events that occurred in the target thermal power plant, wherein the historical time series data contains the time series of all the target risk events that are consecutive in time. The unit time and the corresponding parameters are described in The historical parameter value of each unit moment in the unit moments; According to the occurrence timestamps of all target risk events, the unit moments are traversed in the following manner to obtain the historical time series data of the target risk state parameters: Around time If the target risk event occurs within The corresponding target risk state parameter value is set to 1, otherwise it will be the same as the unit moment The corresponding target risk state parameter value is set to 0, where Represents the unit duration between two adjacent unit moments; Performing feature dimension reduction and feature selection processing on the historical time series data of all the monitoring parameters to obtain historical time series data of multiple target risk-associated monitoring parameters, wherein the target risk-associated monitoring parameters refer to monitoring parameters among all the monitoring parameters that are correlated with the target risk state parameters; For each monitoring parameter in the plurality of target risk associated monitoring parameters, first determine the time when each target risk event occurs according to the historical time series data of the target risk state parameter, then extract each previous time series data corresponding to the time when each target risk event occurs from the corresponding historical time series data, and use each previous time series data as the corresponding positive sample data, wherein the previous time series data refers to the data traced back from a fixed time before the corresponding target risk event occurs to The historical time series data corresponding to the time window of the length of time, the fixed time refers to the time when the corresponding target risk event occurs. At a certain moment in time, Indicates the preset first duration. Indicates the preset second duration; Obtain multiple deep learning models with different structures; For each monitoring parameter and each deep learning model in the multiple deep learning models, all the positive sample data of the corresponding parameter are applied, and the hyperparameters of the corresponding model are optimized based on the optimization algorithm to obtain the corresponding hyperparameters and use them to make the objective function Minimize the optimization search results, where the objective function The calculation formula is as follows: In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for calculation of the deep learning model; For each monitoring parameter, select the corresponding and minimum objective function from the multiple deep learning models. A certain deep learning model is obtained, and the optimization search result of the hyperparameters of the certain deep learning model and the model parameters obtained in the optimization process and corresponding to the optimization search result are imported into the certain deep learning model to obtain a corresponding target risk occurrence prediction model; For each monitoring parameter, Going back to the All corresponding parameter values ​​received within the time window of the time length are imported into the corresponding target risk occurrence prediction model, and the corresponding and The predicted probability of the occurrence of the target risk event; According to the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and at the future time The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the occurrence of the target risk event; If the final predicted probability reaches a preset probability threshold, a warning action corresponding to the target risk event is triggered.

2. The thermal power plant risk early warning method according to claim 1, characterized in that: All monitoring parameters include monitoring temperature, monitoring pressure, monitoring parameters for reflecting vibration conditions, monitoring parameters for reflecting equipment operation conditions, monitoring parameters for reflecting equipment operation conditions, environmental monitoring parameters, monitoring parameters for reflecting access control operation conditions, monitoring parameters for reflecting video AI analysis results, monitoring parameters for reflecting inspection conditions, and / or monitoring parameters for reflecting personnel behavior monitoring conditions; And / or, the target risk events include equipment failure risk events, operational error risk events, natural disaster risk events or personnel behavior risk events.

3. The thermal power plant risk early warning method according to claim 1, characterized in that: The historical time series data of all the monitoring parameters are subjected to feature dimension reduction and feature selection processing to obtain the historical time series data of multiple target risk-related monitoring parameters, including: Using an improved principal component analysis method to perform feature dimensionality reduction processing on the historical time series data of all the monitoring parameters, to obtain a plurality of historical time series data of the monitoring parameters after preliminary dimensionality reduction processing; Using a local linear embedding method to perform feature dimensionality reduction processing on the historical time series data of the monitoring parameters after the multiple initial dimensionality reduction processing, to obtain the historical time series data of the monitoring parameters after the multiple secondary dimensionality reduction processing; A maximum relevance minimum redundancy algorithm based on mutual information maximization is used to perform feature selection processing on the historical time series data of the multiple monitoring parameters after secondary dimensionality reduction processing to obtain historical time series data of multiple target risk associated monitoring parameters, wherein the target risk associated monitoring parameters refer to monitoring parameters among all the monitoring parameters and that are correlated with the target risk state parameters.

4. The thermal power plant risk early warning method according to claim 1, characterized in that: The historical time series data of all the monitoring parameters are subjected to feature dimension reduction and feature selection processing to obtain the historical time series data of multiple target risk-related monitoring parameters, including: For each monitoring parameter among all the monitoring parameters, the corresponding historical time series data is first used as the first sample , and taking the historical time series data of the target risk state parameter as the second sample , then for the first sample Perform a normal distribution KS test to calculate the first test statistic value , and also for the second sample Perform the normal distribution KS check to calculate the second check statistic value , and finally in the first check statistic value and the second check statistic value When both are greater than the first preset threshold, the first sample is calculated according to the following formula With the sample The correlation coefficient As the corresponding correlation coefficient: In the formula, represents a positive integer, In the first sample The Sample values, In the second sample The Sample values, Indicates the first sample The sample mean of The second sample The sample mean of ; Determine whether the total number of monitoring parameters whose correlation coefficients are greater than the second preset threshold value among all the monitoring parameters exceeds the third preset threshold value ,in, represents a positive integer greater than 2; If so, the monitoring parameters whose correlation coefficients are greater than the second preset threshold value among all the monitoring parameters are used as the target risk associated monitoring parameters; otherwise, all the monitoring parameters are arranged in descending order according to the correlation coefficients to obtain a monitoring parameter sequence, and then the first monitoring parameter is selected from the monitoring parameter sequence. monitoring parameters are respectively used as target risk associated monitoring parameters, wherein the target risk associated monitoring parameters refer to monitoring parameters among all the monitoring parameters and which are correlated with the target risk state parameter.

5. The thermal power plant risk early warning method according to claim 1, characterized in that: The multiple deep learning models include classification and regression trees, perceptrons, multi-layer perceptrons, recurrent neural networks, long short-term memory networks, autoencoders and / or variational autoencoders.

6. The thermal power plant risk early warning method according to claim 1, characterized in that: The optimization algorithm adopts a particle swarm optimization algorithm, a Newton optimization algorithm, a genetic optimization algorithm, a gray wolf optimization algorithm, a whale optimization algorithm or a tuna swarm optimization algorithm.

7. The thermal power plant risk early warning method according to claim 1, characterized in that: For each monitoring parameter and each deep learning model in the multiple deep learning models, all the positive sample data of the corresponding parameter are applied, and the hyperparameters of the corresponding model are optimized based on the optimization algorithm to obtain the corresponding hyperparameters and use them to make the objective function The minimized optimized search results include the following steps S601 to S616: S601. For any monitoring parameter among all the monitoring parameters and any deep learning model among the multiple deep learning models, initialize the number of search individuals corresponding to and including the search population , maximum number of iterations , the first learning factor , the second learning factor , the third learning factor , the fourth learning factor , the fifth learning factor and the sixth learning factor The optimization algorithm parameters are set, and the initial position of each search individual in the search population is randomly generated, and then step S602 is performed, wherein the initial position of each search individual is randomly generated based on the following formula: In the formula, Indicates less than or equal to A positive integer, Indicates less than or equal to A positive integer, represents the number of dimensions of the parameter vector to be optimized, wherein the parameter vector to be optimized contains all the hyperparameters of any of the deep learning models, Indicates the first The initial position of the search individual in the parameter vector to be optimized is The weight in the dimensions, Indicated in the The upper limit of the parameter search space in dimensions, Indicated in the The lower bound of the parameter search space in dimensions, Represents a purely decimal random generator function; S602. For each search individual, the corresponding initial position is imported into any deep learning model as a model hyperparameter to obtain a corresponding first deep learning model, and then all the positive sample data of any monitoring parameter are applied to train and test the first deep learning model to obtain the corresponding first output error situation index value and the first calculation time required index value, and finally the first output error situation index value and the first calculation time required index value are imported into the objective function , and use the output result as the corresponding fitness, and then execute step S603, where the objective function The calculation formula is as follows: In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for calculation of the deep learning model; S603. Take the initial position of a search individual with the minimum fitness as the initial global optimal position , and also initialize the current number of iterations , then execute step S604; S604. Calculate the current average position of the search population according to the current position of each search individual, and determine the corresponding first new position for each search individual, and then execute step S605, wherein the first new position of each search individual is determined according to the following formula: In the formula, Indicates the The first new position of the search individual is in the The weight in the dimensions, represents the global optimal position In the said The weight in the dimensions, Indicates the current average position of the search population in the The weight in the dimensions, Indicates the The current location of the search individual is The weight in the dimensions; S605. For each search individual, the corresponding first new position is imported as a model hyperparameter into any deep learning model to obtain a corresponding second deep learning model, and then all the positive sample data of any monitoring parameter are applied to train and test the second deep learning model to obtain the corresponding second output error situation index value and the second calculation time required index value, and finally the second output error situation index value and the second calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S606; S606. Determine the Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The first new position of the search individual is determined, and then step S607 is executed; otherwise, step S607 is directly executed; S607. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S608, otherwise directly execute step S608; S608. The current average position of the search population is calculated based on the current position of each search individual, and the corresponding second new position is determined for each search individual, and then step S609 is executed, wherein the second new position of each search individual is determined according to the following formula: In the formula, Indicates the The second new position of the search individual is in the The weight in the dimensions, Indicates that The first intermediate variable corresponding to the search individual, Indicates that The second intermediate variable corresponding to the search individual, Indicates less than or equal to A positive integer, represents the remainder function, Indicates the first The current location of the search individual is The weight in the dimensions, Indicates that The third intermediate variable corresponding to the search individual, Indicates that The fourth intermediate variable corresponding to the search individual, represents the maximum value function, Indicates that The fifth intermediate variable corresponding to the search individual, Indicates that The sixth intermediate variable corresponding to the search individual, Indicates 180 degrees; S609. For each search individual, the corresponding second new position is imported into any of the deep learning models as a model hyperparameter to obtain a corresponding third deep learning model, and then all the positive sample data of any of the monitoring parameters are applied to train and test the third deep learning model to obtain the corresponding third output error situation index value and the third calculation time required index value, and finally the third output error situation index value and the third calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S610; S610. Determine the Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The second new position of the search individual is determined, and then step S611 is executed; otherwise, step S611 is executed directly; S611. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S612, otherwise directly execute step S612; S612. The current average position of the search population is calculated based on the current position of each search individual, and the corresponding third new position is determined for each search individual, and then step S613 is executed, wherein the third new position of each search individual is determined according to the following formula: In the formula, Indicates the The third new position of the search individual is The weight in the dimensions, Indicates that The seventh intermediate variable corresponding to the search individual, Indicates that The eighth intermediate variable corresponding to the search individual, Indicates that The ninth intermediate variable corresponding to the search individual, Indicates that The tenth intermediate variable corresponding to the search individual, Indicates that The eleventh intermediate variable corresponding to the search individual; S613. For each search individual, the corresponding third new position is imported into any of the deep learning models as a model hyperparameter to obtain a corresponding fourth deep learning model, and then all the positive sample data of any of the monitoring parameters are applied to train and test the fourth deep learning model to obtain the corresponding fourth output error situation index value and fourth calculation time required index value, and finally the fourth output error situation index value and the fourth calculation time required index value are imported into the objective function , and use the output result as the corresponding and new fitness, and then execute step S614; S614. Determine the Is the fitness corresponding to the current position of the search individual greater than that of the The current fitness of the search individual, if so, then the The current position of the search individual is updated to the The third new position of the search individual is determined, and then step S615 is executed; otherwise, step S615 is directly executed; S615. Determine the global optimal position Is the corresponding fitness greater than the minimum value among the current fitness of each search individual? If so, the global optimal position Update to the current position of any search individual with the minimum value, and then execute step S616, otherwise directly execute step S616; S616. Make the current number of iterations Add 1 to the current number of iterations. Whether the maximum number of iterations has been reached , if so, then the global optimal position As a hyperparameter of any deep learning model corresponding to any monitoring parameter and used to make the objective function Minimize the optimized search results, otherwise return to execute step S604.

8. The thermal power plant risk early warning method according to claim 1, characterized in that: According to the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and at the future time The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the target risk event occurring includes: According to the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and at the future time The predicted probability of the target risk event occurring at the future time is calculated according to the following formula: The final predicted probability of the target risk event occurring : In the formula, represents the total number of parameters of the plurality of target risk associated monitoring parameters, Indicates less than or equal to A positive integer, Indicates the first monitoring parameter associated with the multiple target risks The influence weight coefficient corresponding to each monitoring parameter is: Indicates the The correlation coefficient between the monitoring parameters and the target risk state parameters, Indicates less than or equal to A positive integer, Indicates the first of the multiple target risk associated monitoring parameters The correlation coefficient between the monitoring parameters and the target risk state parameters, Indicates that The output error indicator value of the target risk occurrence prediction model corresponding to the monitoring parameters, Indicates that The output error indicator value of the target risk occurrence prediction model corresponding to the monitoring parameters, Indicates that The monitoring parameters correspond to the future time The predicted probability of the target risk event occurring.

9. A thermal power plant risk early warning system based on deep learning, characterized in that: It includes a monitoring data collection unit, a risk data collation unit, a monitoring data processing unit, a positive sample data selection unit, a learning model acquisition unit, a model hyperparameter optimization unit, a prediction model generation unit, a prediction model application unit, a prediction probability superposition unit and a warning action triggering unit; The monitoring data collection unit is used to collect the historical time series data of all monitoring parameters recorded by all safety production digital devices of the target thermal power plant, and to collect the occurrence timestamps of all target risk events occurring in the target thermal power plant, wherein the historical time series data contains the time series data that are consecutive in time sequence. The unit time and the corresponding parameters are described in The historical parameter value of each unit moment in the unit moments; The risk data collating unit is communicatively connected to the monitoring data collecting unit, and is used to traverse each unit time according to the occurrence timestamp of all target risk events in the following manner to obtain the historical time series data of the target risk state parameter: if at the currently traversed unit time Around time If the target risk event occurs within The corresponding target risk state parameter value is set to 1, otherwise it will be the same as the unit moment The corresponding target risk state parameter value is set to 0, where Represents the unit duration between two adjacent unit moments; The monitoring data processing unit is respectively connected to the monitoring data collection unit and the risk data sorting unit for performing feature dimension reduction and feature selection processing on the historical time series data of all the monitoring parameters to obtain the historical time series data of multiple target risk-associated monitoring parameters, wherein the target risk-associated monitoring parameter refers to a monitoring parameter among all the monitoring parameters that is correlated with the target risk state parameter; The positive sample data selection unit is respectively connected to the risk data sorting unit and the monitoring data processing unit in communication, and is used to determine the time when each target risk event occurs according to the historical time series data of the target risk state parameter for each monitoring parameter in the multiple target risk associated monitoring parameters, and then extract each previous time series data corresponding to the time when each target risk event occurs from the corresponding historical time series data, and use each previous time series data as the corresponding positive sample data, wherein the previous time series data refers to the time when each target risk event occurs and is traced back to the previous time series data from a fixed time before the corresponding target risk event occurs. The historical time series data corresponding to the time window of the length of time, the fixed time refers to the time when the corresponding target risk event occurs. At a certain moment in time, Indicates the preset first duration. Indicates the preset second duration; The learning model acquisition unit is used to acquire multiple deep learning models with different structures; The model hyperparameter optimization unit is respectively connected to the positive sample data selection unit and the learning model acquisition unit for communicating with each other, and is used to apply all the positive sample data of the corresponding parameters to each monitoring parameter and each deep learning model in the multiple deep learning models, and optimize the hyperparameters of the corresponding model based on the optimization algorithm to obtain the corresponding hyperparameters and use them to make the objective function Minimize the optimization search results, where the objective function The calculation formula is as follows: In the formula, Indicates the output error indicator value of the deep learning model. Indicates the time required for calculation of the deep learning model; The prediction model generation unit is communicatively connected to the model hyperparameter optimization unit, and is used to select the corresponding and minimum objective function from the multiple deep learning models for each monitoring parameter. A certain deep learning model is obtained, and the optimization search result of the hyperparameters of the certain deep learning model and the model parameters obtained in the optimization process and corresponding to the optimization search result are imported into the certain deep learning model to obtain a corresponding target risk occurrence prediction model; The prediction model application unit is communicatively connected to the prediction model generation unit, and is used for each monitoring parameter to be Tracing back to the All corresponding parameter values ​​received within the time window of the time length are imported into the corresponding target risk occurrence prediction model, and the corresponding and The predicted probability of the occurrence of the target risk event; The prediction probability superposition unit is respectively connected to the prediction model application unit and the model hyperparameter optimization unit for calculating the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and the correlation coefficient between each monitoring parameter and the target risk state parameter and the output error situation index value of the target risk occurrence prediction model corresponding to each monitoring parameter and the output error situation index value of the target risk occurrence prediction model at the future time. The predicted probability of the target risk event occurring is calculated by superimposing the probability of the target risk event occurring at the future time. The final predicted probability of the occurrence of the target risk event; The early warning action triggering unit is communicatively connected to the predicted probability superposition unit, and is used to trigger the execution of the early warning action corresponding to the target risk event if the final predicted probability reaches a preset probability threshold.

10. A computer system, characterized in that: The method comprises a memory, a processor and a transceiver which are communicatively connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the thermal power plant risk warning method as described in any one of claims 1 to 8.

Images