Artificial Intelligence-Based Safety Monitoring Methods and Systems for Thermal Power Plants

Abstract

This invention discloses an artificial intelligence-based safety monitoring method and system for thermal power plants, comprising: firstly, responding to a thermal power plant safety monitoring start command, acquiring current boiler parameters at preset time intervals; then, calling a pre-trained boiler status assessment model to determine whether the current boiler parameters are in a boiler warning state; and finally, when the current boiler parameters are in a boiler warning state, acquiring and executing a target warning resolution strategy for the current boiler parameters. This design, compared to the prior art's process of reporting first and then processing, can improve the efficiency of safety monitoring and processing for thermal power plants.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring of thermal power plants, and more specifically, to a method and system for safety monitoring of thermal power plants based on artificial intelligence. Background Technology

[0002] Currently, thermal power generation is my country's main source of electricity, and safety monitoring of thermal power plants is a key focus of their safety management. Existing technologies often require manual or sensor-based data collection to identify anomalies and then customize solutions. This approach is no longer sufficient to meet the safety management needs of thermal power plants. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for safety monitoring of thermal power plants based on artificial intelligence.

[0004] In a first aspect, embodiments of the present invention provide an artificial intelligence-based safety monitoring method for thermal power plants, comprising:

[0005] In response to the start command of the safety monitoring system of the thermal power plant, generate periodic safety assessment instructions;

[0006] According to the periodic safety assessment instructions, the current boiler parameters are obtained at preset time intervals.

[0007] Call the pre-trained boiler condition assessment model, input the current boiler parameters into the boiler condition assessment model, and determine whether the current boiler parameters are in a boiler early warning state;

[0008] If so, obtain and execute the target early warning solution strategy for the current boiler parameters;

[0009] If not, generate a boiler operation safety report for the current period and send the report to the preset safety terminal.

[0010] In a second aspect, embodiments of the present invention provide a server system, including a server, which is used to execute the method of the first aspect.

[0011] Compared with existing technologies, the beneficial effects provided by this invention include: the artificial intelligence-based safety monitoring method and system for thermal power plants disclosed in this invention includes: firstly, in response to the start command of the thermal power plant safety monitoring, acquiring the current boiler parameters according to a preset time interval; then, calling a pre-trained boiler status assessment model to determine whether the current boiler parameters are in a boiler warning state; finally, when the current boiler parameters are in a boiler warning state, acquiring and executing the target warning solution strategy for the current boiler parameters. This design, compared with the prior art's reporting and processing process, can improve the efficiency of safety monitoring and processing for thermal power plants. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating the steps of an artificial intelligence-based safety monitoring method for thermal power plants provided in an embodiment of the present invention;

[0014] Figure 2 A schematic block diagram of the server structure provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0016] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating the artificial intelligence-based safety monitoring method for thermal power plants provided in this embodiment. The following is a detailed description of the artificial intelligence-based safety monitoring method for thermal power plants.

[0017] 201. In response to the start command of the safety monitoring of the thermal power plant, generate a periodic safety assessment instruction.

[0018] 202. According to the periodic safety assessment instructions, obtain the current boiler parameters at preset time intervals.

[0019] 203. Call the pre-trained boiler condition assessment model, input the current boiler parameters into the boiler condition assessment model, and determine whether the current boiler parameters are in a boiler early warning state.

[0020] 204. If so, obtain and execute the target early warning solution strategy for the current boiler parameters.

[0021] 205. If not, generate the boiler operation safety report for the current period and send the boiler operation safety report to the preset safety terminal.

[0022] In this embodiment, considering the specific scenario of a thermal power plant, a malfunction or accident risk would significantly impact both on-site personnel and urban electricity users. Therefore, to balance resource consumption and real-time safety, the preset time interval for the periodic safety assessment command can be set to 10 minutes. This interval can be adjusted based on factors such as the size of the power plant and the operating season, without limitation. After acquiring the current boiler parameters collected in the current period, a pre-trained boiler status assessment model can be invoked to evaluate whether the current boiler parameters are in a boiler warning state. If so, a target warning resolution strategy for the current boiler parameters can be obtained and executed to resolve the warning situation and ensure the safety of the power plant. If not, a boiler operation safety report for the current period is generated and sent to a preset safety terminal. The preset safety terminal can be a maintenance personnel terminal or a monitoring center terminal of the power plant. Through the above steps, the boiler operation status of the thermal power plant can be periodically monitored, and the use of a pre-trained boiler status assessment model improves the efficiency of safety assessment processing for the power plant.

[0023] To more clearly describe the solution provided in the embodiments of this application, the aforementioned step 203 can be implemented through the following detailed steps.

[0024] 101. Obtain the current boiler parameters, which include boiler parameters in at least one category.

[0025] The current boiler parameters refer to the boiler parameters to be tested for safety. The current boiler parameters can refer to boiler parameters collected in real time.

[0026] The current boiler parameters may include boiler temperature, segmented wall tube temperature, boiler pressure, residue amount, steam-water system feedback parameters, fuel system feedback parameters, ventilation equipment feedback parameters, pulverizing equipment feedback parameters, water supply equipment feedback parameters, etc. This embodiment does not limit these parameters.

[0027] 102. Perform feature vector extraction on the boiler parameters of each category to obtain the boiler state vector of the current boiler parameters.

[0028] This involves extracting feature vectors from the boiler parameters across different categories within the current boiler parameters. Specifically, this can be done by constructing features for each category of boiler parameters. Through feature construction, the boiler state vector corresponding to the current boiler parameters can be obtained. The boiler state vector corresponding to the boiler parameters may contain feature vectors of statistical types corresponding to the boiler parameters, such as count features, density features, and deviation features, etc.

[0029] For example, feature construction can generate new fields by transforming or combining existing data feature fields (i.e., boiler parameters in each category of the current boiler parameters) through operations such as addition, subtraction, multiplication, and division. After the original data has been appropriately transformed or combined, it can have strong information value.

[0030] Specifically, feature vector extraction is performed on boiler parameters of various categories in the current boiler parameters. This can be done through a neural network model or by manually constructing features. This embodiment does not impose any restrictions on this method.

[0031] 103. Using the boiler condition assessment model, the boiler condition assessment operation is performed on the current boiler parameters based on the boiler condition vector to obtain the condition confidence of the current boiler parameters. The condition confidence represents the probability that the current boiler parameters are in a boiler warning state. The boiler condition assessment model includes at least one condition assessment sub-network, where each condition assessment sub-network corresponds to at least one condition assessment strategy.

[0032] The boiler early warning status safety warning information refers to the boiler status that requires a warning. After the corresponding warning is issued, maintenance personnel quickly check to determine if there is a problem, or perform maintenance according to the preset warning response strategy to avoid boiler and boiler-linked equipment failure, which could threaten the safety of the entire thermal power plant. The status confidence level can be regarded as the initial prediction result of the current boiler parameters. Optionally, in this embodiment, based on the status confidence level, combined with the evaluation strategy group and the controllable evaluation strategy group, the current boiler parameters can be further tested for safety.

[0033] Specifically, the boiler state assessment operation is performed on the current boiler parameters based on the boiler state vector, which means predicting the probability that the current boiler parameters are in a boiler warning state based on the boiler state vector.

[0034] Among them, the status assessment strategy is a strategy used to detect whether there is a safety risk status in the boiler parameters. For example, the status assessment strategy can be abnormal boiler temperature, abnormal segment wall tube temperature, abnormal boiler pressure, abnormal residue amount, abnormal steam and water content, abnormal fuel combustion efficiency, abnormal ventilation status, abnormal pulverizing status, abnormal feedwater status, etc. This embodiment does not limit this.

[0035] The boiler condition assessment model may include a neural network model constructed using a neural network algorithm, a tree model constructed using a decision tree algorithm, a support vector machine, etc., and this embodiment does not impose any restrictions on this. The boiler condition assessment model includes at least one condition assessment sub-network. When the boiler condition assessment model is a tree model, it may consist of at least one decision tree, and each condition assessment sub-network may be a decision tree.

[0036] Optionally, in this embodiment, the step "using the boiler condition assessment model to perform boiler condition assessment on the current boiler parameters based on the boiler condition vector, and obtaining the condition confidence of the current boiler parameters" may include:

[0037] For each state assessment subnetwork in the boiler condition assessment model, the boiler condition assessment operation is performed on the current boiler parameters based on the boiler condition vector through the state assessment subnetwork to obtain the undetermined state confidence of the current boiler parameters.

[0038] The confidence scores of the undetermined states corresponding to each state assessment subnetwork are integrated to obtain the state confidence scores of the current boiler parameters.

[0039] Optionally, in this embodiment, the step "integrating the confidence scores of the undetermined states corresponding to each state evaluation sub-network to obtain the state confidence scores of the current boiler parameters" may include:

[0040] Determine the contribution coefficient of the confidence of the undetermined state corresponding to each state evaluation subnetwork;

[0041] Based on the contribution coefficient, the confidence scores of the undetermined states corresponding to each state evaluation subnetwork are weighted and averaged to obtain the state confidence scores of the current boiler parameters.

[0042] It should be noted that the boiler condition assessment model is trained by multiple labeled sample training sets. In this embodiment, the sample training set includes training boiler parameters and preset labels corresponding to the training boiler parameters. The label refers to the prediction state confidence level corresponding to the training boiler parameters. The boiler condition assessment model can be trained by other equipment and then provided to the boiler risk assessment device, or it can be trained by the boiler risk assessment device itself.

[0043] If the boiler risk assessment device is trained automatically, then before the step "using the boiler condition assessment model to perform boiler condition assessment on the current boiler parameters based on the boiler condition vector, and obtain the condition confidence of the current boiler parameters", the following may also be included:

[0044] Obtain a sample training set, which includes training boiler parameters and the prediction state confidence corresponding to the training boiler parameters. The training boiler parameters include boiler parameters in at least one category.

[0045] Feature vector extraction is performed on the boiler parameters of each category in the training boiler parameters to obtain the boiler state vector of the training boiler parameters;

[0046] Using the initial boiler condition assessment model, the boiler condition assessment operation is performed on the training boiler parameters according to the boiler condition vector to obtain the true state confidence of the training boiler parameters as the boiler early warning state.

[0047] Based on the confidence levels of the actual and predicted states of the trained boiler parameters, the parameters of the initial boiler condition assessment model are adjusted to obtain the trained boiler condition assessment model.

[0048] The training process involves first calculating the true state confidence of the training boiler parameters, then using the backpropagation algorithm to adjust the parameters of the initial boiler condition assessment model. Based on the true state confidence and predicted state confidence of the training boiler parameters, the parameters of the initial boiler condition assessment model are optimized so that the true state confidence of the training boiler parameters approaches the predicted state confidence, thus obtaining the trained boiler condition assessment model.

[0049] Specifically, a prediction state confidence level of 0 indicates that there is no safety warning information for the training boiler parameter, while a prediction state confidence level of 1 indicates that there is a safety warning information for the training boiler parameter. The prediction state confidence level can be set according to the actual situation.

[0050] 104. When the state confidence exceeds the initial state confidence threshold, execute the command to call the evaluation strategy to obtain the evaluation strategy group. The evaluation strategy group includes at least one evaluation strategy, which is selected from the state evaluation strategies of the boiler state evaluation model.

[0051] The initial state confidence threshold can be set according to actual conditions, and this embodiment does not impose any restrictions on it. The evaluation strategy can be used to determine whether the boiler parameters are in a boiler warning state.

[0052] Optionally, in some embodiments, the current boiler parameters whose state confidence does not exceed the initial state confidence threshold can be determined as the boiler controllable state.

[0053] 105. When the current boiler parameters match the evaluation strategy, determine that the current boiler parameters are in a boiler early warning state.

[0054] Specifically, the current boiler parameter matching and evaluation strategy can be a preset abnormal condition corresponding to the current boiler parameter matching and evaluation strategy group. This preset abnormal condition can be set according to actual conditions, and this embodiment does not impose any restrictions on it. For example, the preset abnormal condition can be the number of evaluation strategies in the current boiler parameter matching and evaluation strategy group.

[0055] For example, if the current boiler parameters correspond to the boiler state vector A=6, B=9, C=3, and the evaluation strategy in the evaluation strategy group is A>10, B≤5, C≤10, then the current boiler parameters match the evaluation strategy C≤10. If the preset abnormal condition is that the boiler parameters are in a boiler warning state when they match more than two evaluation strategies in the evaluation strategy group, then the current boiler parameters only match one, and therefore the boiler is in a controllable state.

[0056] Optionally, the following examples are also provided in this embodiment:

[0057] When the state confidence exceeds the advanced state confidence threshold, the command to call the controllable evaluation strategy is executed to obtain a controllable evaluation strategy group. The controllable evaluation strategy group includes at least one controllable evaluation strategy, which is selected from the state evaluation strategies of the boiler state evaluation model.

[0058] When the current boiler parameters match the controllable evaluation strategy, the current boiler parameters are determined to be in a controllable state.

[0059] The confidence threshold for the advanced state can be set according to actual conditions, and this embodiment does not impose any restrictions on it. Specifically, the confidence threshold for the advanced state can generally be set larger than the confidence threshold for the initial state. The confidence thresholds for the initial state and the advanced state mainly depend on the actual performance of the model during testing, and it is necessary to observe the model's accuracy, recall, and other evaluation indicators under different threshold settings. The controllable evaluation strategy can be used to determine whether the boiler parameters are in a controllable state.

[0060] In some embodiments, to improve the accuracy of safety detection, the initial state confidence threshold and the advanced state confidence threshold can be set to be relatively small. This allows boiler parameters with relatively high state confidence (i.e., state confidence exceeding the initial state confidence threshold or the advanced state confidence threshold) to be re-detected, ensuring the accuracy of boiler parameter detection and improving the safety of thermal power plant boilers.

[0061] Specifically, the controllable evaluation strategy for matching current boiler parameters can be a preset non-abnormal condition corresponding to the controllable evaluation strategy group for matching current boiler parameters. This preset non-abnormal condition can be set according to actual conditions, and this embodiment does not impose any restrictions on it. For example, the preset non-abnormal condition can be set as the number of controllable evaluation strategies in the controllable evaluation strategy group for matching current boiler parameters.

[0062] Optionally, in some embodiments, boiler parameters whose state confidence does not exceed the initial state confidence threshold and does not exceed the advanced state confidence threshold can be determined as the boiler controllable state. For example, the initial state confidence threshold can be denoted as Q1 and the advanced state confidence threshold can be denoted as Q2. Then, when the state confidence of the current boiler parameter does not exceed min(Q1, Q2), the current boiler parameter can be determined as the boiler controllable state.

[0063] The evaluation strategy and controllable evaluation strategy can be selected from the status evaluation strategies used in the boiler status evaluation model. Specifically, the status evaluation strategies that are called more frequently can be determined first. The frequency of calling indicates the importance of the status evaluation strategy in the safety detection of the boiler status evaluation model.

[0064] Optionally, the following examples are also provided in this embodiment:

[0065] The frequency of each state assessment strategy in the boiler condition assessment model is obtained by counting the strategies.

[0066] Based on the call frequency, the state evaluation strategy is extracted to obtain at least one pending evaluation strategy.

[0067] The evaluation strategy to be evaluated is determined by performing an evaluation strategy determination operation, resulting in an evaluation strategy and a controllable evaluation strategy.

[0068] Among them, the status evaluation strategy that is called more than a preset number of times can be determined as a pending evaluation strategy, and the preset number of times can be set according to the actual situation.

[0069] Specifically, each state assessment strategy in the boiler condition assessment model can be denoted as ri. The number of times each state assessment strategy ri appears in the boiler condition assessment model is counted as ti. When ti exceeds a preset number, the state assessment strategy ri corresponding to ti can be determined as a pending assessment strategy. Here, i is the label number of each state assessment strategy. If there are m state assessment strategies in the boiler condition assessment model, then i is a positive integer not exceeding m.

[0070] Optionally, the selection of the pending evaluation strategy may not be based on the frequency of invocation of the state evaluation strategy. This embodiment does not impose any restrictions on this, and other methods can also be used to screen the pending evaluation strategies. For example, a certain proportion of the state evaluation strategies appearing in the boiler state evaluation model can be randomly selected as the pending evaluation strategies, or the selection can be made by experimentally testing the actual prediction effect.

[0071] Optionally, in this embodiment, the step "perform an evaluation strategy determination operation on the given evaluation strategy to obtain the evaluation strategy and the controllable evaluation strategy" may include:

[0072] Obtain the training boiler parameters and the preset identifiers corresponding to the training boiler parameters. The preset identifiers indicate the probability that the training boiler parameters are in a boiler warning state. The training boiler parameters include boiler parameters in at least one category.

[0073] Feature vector extraction is performed on the boiler parameters of each category in the training boiler parameters to obtain the boiler state vector of the training boiler parameters, and the boiler state vector of the training boiler parameters under each pending evaluation strategy is determined from the boiler state vector of the training boiler parameters.

[0074] Based on the preset identifiers, statistical analysis is performed on the boiler state vectors corresponding to each pending evaluation strategy to obtain the contribution coefficients corresponding to each pending evaluation strategy.

[0075] Based on the contribution coefficient, an evaluation strategy determination operation is performed on the evaluation strategy to be determined, resulting in the evaluation strategy and the controllable evaluation strategy.

[0076] Specifically, the pre-defined identifier can be the predicted state confidence level of the trained boiler parameters. The boiler state vector corresponding to the trained boiler parameters under the pending evaluation strategy is also known as the boiler state vector in the trained boiler parameters that is related to the pending evaluation strategy. For example, if the boiler state vectors corresponding to the trained boiler parameters include A, B, and C, and for the pending evaluation strategy A > 5, the relevant boiler state vector is A.

[0077] One approach is to establish a regression model for the evaluation strategy to be determined, and use a regression algorithm to calculate the degree of influence of each evaluation strategy on the boiler parameter detection results. Specifically, the regression algorithm can be used to obtain the contribution coefficient corresponding to each evaluation strategy to be determined. The contribution coefficient represents the importance of the evaluation strategy to the detection results, and the evaluation strategy with high importance can be selected as the output.

[0078] Specifically, in this embodiment, the step "perform an evaluation strategy determination operation on the given evaluation strategy to obtain the evaluation strategy and the controllable evaluation strategy" may include:

[0079] Obtain the training boiler parameters and the preset identifier y corresponding to the training boiler parameters. The preset identifier y represents the probability that the training boiler parameters are in a boiler warning state. The training boiler parameters include boiler parameters in at least one category.

[0080] Feature vector extraction is performed on the boiler parameters of each category in the training boiler parameters to obtain the boiler state vector of the training boiler parameters, and the boiler state vector Rj corresponding to the training boiler parameters under each undetermined evaluation strategy j is determined from the boiler state vector of the training boiler parameters.

[0081] By using a preset regression algorithm, the boiler state vector Rj corresponding to each pending evaluation strategy is statistically analyzed to obtain the contribution coefficient wj corresponding to each pending evaluation strategy.

[0082] Based on the contribution coefficient wj, the evaluation strategy determination operation is performed on the evaluation strategy to be determined, resulting in the evaluation strategy and the controllable evaluation strategy.

[0083] Optionally, in this embodiment, the step "based on the contribution coefficient, perform an evaluation strategy determination operation on the evaluation strategy to be determined, to obtain the evaluation strategy and the controllable evaluation strategy" may include:

[0084] Based on the contribution coefficient, select a demand assessment strategy from the pending assessment strategies;

[0085] Demand assessment strategies with contribution coefficients exceeding preset coefficients are defined as assessment strategies, while demand assessment strategies with contribution coefficients not exceeding preset coefficients are defined as controllable assessment strategies.

[0086] Specifically, any pending evaluation strategies whose contribution coefficient exceeds a preset contribution coefficient can be designated as demand evaluation strategies. The preset contribution coefficient can be set according to actual circumstances. The contribution coefficient of the pending evaluation strategy can be positive or negative.

[0087] Here, the preset coefficient can be 0, meaning that demand assessment strategies with positive contribution coefficients are defined as assessment strategies, and demand assessment strategies with negative contribution coefficients are defined as controllable assessment strategies. It can be understood that this preset coefficient can be set according to actual circumstances. Specifically, demand assessment strategies with positive contribution coefficients in the vector wj can be determined as effective assessment strategies, and this assessment strategy group can be denoted as K1; demand assessment strategies with negative contribution coefficients in the vector wj can be determined as controllable assessment strategies, and this controllable assessment strategy group can be denoted as K2.

[0088] This application can extract effective condition assessment strategies from machine learning models (i.e., boiler condition assessment models) using regression algorithms, and combine the boiler condition assessment model and effective condition assessment strategies to perform safety detection on boiler parameters.

[0089] Interpretability refers to the ability to explain which specific aspect of the boiler parameters presents a safety warning. This application can obtain anomalies in boiler parameters based on the extracted state assessment strategy. Specifically, after the current boiler parameters are predicted using the boiler state assessment model, further safety checks are required based on the assessment strategy group and the controllable assessment strategy group. The assessment strategies in the assessment strategy group and the controllable assessment strategies in the controllable assessment strategy group provide the solvability of the boiler parameters.

[0090] This application can also update and iterate the assessment strategy group and the controllable assessment strategy group at regular intervals to ensure the prevention and control effect of the boiler risk assessment device. The iteration cycle can be set according to the actual situation.

[0091] In the application scenario provided in this application, the boiler pressure to be detected can be obtained, and its boiler state vector can be extracted. A trained boiler state assessment model is then used to perform safety detection on the boiler pressure based on the boiler state vector, obtaining a predicted state confidence level. If the state confidence level does not exceed the initial state confidence threshold Q1 and the advanced state confidence threshold Q2, the boiler pressure can be determined as being within a controllable range (i.e., a controllable boiler state). If the boiler pressure exceeds the initial state confidence threshold Q1, it can be judged using the set of effective early warning rules (i.e., the assessment strategy group) K1. If the boiler pressure matches the assessment strategy in K1, it is determined as a state requiring early warning (i.e., a boiler early warning state); otherwise, it is a controllable state. If the boiler pressure exceeds the advanced state confidence threshold Q2, it can be judged using the set of controllable effective rules (i.e., the controllable assessment strategy group) K2. If the boiler pressure matches the controllable assessment strategy in K2, it is determined as a controllable state (i.e., a controllable boiler state); otherwise, it is a state requiring early warning.

[0092] Specifically, the process for safety testing of boiler parameters is as follows:

[0093] 1001. Obtain the current boiler parameters, and perform feature vector extraction on the boiler parameters of each category in the current boiler parameters to obtain the boiler state vector of the current boiler parameters;

[0094] 1002. Using the trained boiler condition assessment model, predict the state confidence of the current boiler parameters based on the boiler state vector;

[0095] 1003. When the state confidence exceeds the initial state confidence threshold Q1, obtain the evaluation strategy group, which includes at least one evaluation strategy.

[0096] 1004. Determine whether the current boiler parameters match the evaluation strategy. If yes, determine that the current boiler parameters are in a boiler early warning state; if not, determine that the current boiler parameters are in a boiler controllable state.

[0097] 1005. When the state confidence exceeds the advanced state confidence threshold Q2, obtain a controllable evaluation strategy group, which includes at least one controllable evaluation strategy.

[0098] 1006. Determine whether the current boiler parameters match the controllable assessment strategy. If yes, determine that the current boiler parameters are in a controllable state; otherwise, determine that the current boiler parameters are in a warning state.

[0099] 1007. When the state confidence level does not exceed or equal to the smaller value between the initial state confidence level threshold Q1 and the advanced state confidence level threshold Q2, the current boiler parameters are determined to be in a controllable state.

[0100] As can be seen from the above, the electronic device in this embodiment can acquire current boiler parameters, which include boiler parameters of at least one category; perform feature vector extraction on the boiler parameters of each category to obtain the boiler state vector of the current boiler parameters; through the boiler state assessment model, perform boiler state assessment on the current boiler parameters based on the boiler state vector to obtain the state confidence of the current boiler parameters. The state confidence represents the probability that the current boiler parameters are in a boiler warning state. The boiler state assessment model includes at least one state assessment sub-network, wherein each state assessment sub-network corresponds to at least one state assessment strategy; when the state confidence exceeds the initial state confidence threshold, execute the invocation command of the assessment strategy to obtain an assessment strategy group, which includes at least one assessment strategy selected from the state assessment strategies of the boiler state assessment model; when the current boiler parameters match the assessment strategy, determine that the current boiler parameters are in a boiler warning state. This application can combine the prediction results of the boiler state assessment model and the assessment strategy group to perform safety detection on boiler parameters, improving the detection efficiency and accuracy of boiler parameters.

[0101] Based on the method described in the preceding embodiments, the following will provide a more detailed explanation by taking the specific integration of the boiler risk assessment device into a server as an example.

[0102] This application provides a method for assessing the condition of a thermal power plant boiler. The specific process of this method is as follows:

[0103] 501. The server obtains the current boiler parameters and performs feature vector extraction on the boiler parameters of each category in the current boiler parameters to obtain the boiler state vector of the current boiler parameters.

[0104] This involves extracting feature vectors from boiler parameters across various categories within the current boiler parameters. Specifically, this can be done by constructing features from the original data; feature construction, also known as feature extraction, is a method of feature selection. Through feature construction, the boiler state vector corresponding to the current boiler parameters can be obtained. This boiler state vector may contain statistically relevant feature vectors, such as count features, density features, and deviation features.

[0105] 502. The server uses the boiler status assessment model to perform boiler status assessment on the current boiler parameters based on the boiler status vector, and obtains the status confidence of the current boiler parameters. The status confidence represents the probability that the current boiler parameters are in a boiler warning state. The boiler status assessment model includes at least one status assessment sub-network, wherein each status assessment sub-network corresponds to at least one status assessment strategy.

[0106] The boiler early warning status refers to the state where safety warning information exists, and the status confidence level can be regarded as the initial prediction result of the current boiler parameters. Optionally, in this embodiment, the current boiler parameters can be further tested for safety based on the status confidence level, combined with the evaluation strategy group and the controllable evaluation strategy group.

[0107] Optionally, in this embodiment, the step "using the boiler condition assessment model to perform boiler condition assessment on the current boiler parameters based on the boiler condition vector, and obtaining the condition confidence of the current boiler parameters" may include:

[0108] For each state assessment subnetwork in the boiler condition assessment model, the boiler condition assessment operation is performed on the current boiler parameters based on the boiler condition vector through the state assessment subnetwork to obtain the undetermined state confidence of the current boiler parameters.

[0109] The confidence scores of the undetermined states corresponding to each state assessment subnetwork are integrated to obtain the state confidence scores of the current boiler parameters.

[0110] 503. When the state confidence exceeds the initial state confidence threshold, the server executes the command to call the evaluation strategy to obtain an evaluation strategy group. The evaluation strategy group includes at least one evaluation strategy, which is selected from the state evaluation strategies of the boiler state evaluation model. When the current boiler parameters match the evaluation strategy, the current boiler parameters are determined to be in a boiler warning state.

[0111] Specifically, the current boiler parameter matching and evaluation strategy can be a preset abnormal condition corresponding to the current boiler parameter matching and evaluation strategy group. This preset abnormal condition can be set according to actual conditions, and this embodiment does not impose any restrictions on it. For example, the preset abnormal condition can be the number of evaluation strategies in the current boiler parameter matching and evaluation strategy group.

[0112] 504. When the state confidence exceeds the advanced state confidence threshold, the server executes the command to call the controllable evaluation strategy to obtain a controllable evaluation strategy group. The controllable evaluation strategy group includes at least one controllable evaluation strategy, which is selected from the state evaluation strategies of the boiler state evaluation model. When the current boiler parameters match the controllable evaluation strategy, the current boiler parameters are determined to be in a controllable state.

[0113] The confidence threshold for advanced states can be set according to actual conditions, and this embodiment does not impose any restrictions on it. Specifically, the confidence threshold for advanced states can generally be set larger than the confidence threshold for the initial state. The initial state confidence threshold and the confidence threshold for advanced states are mainly determined by the actual performance of the model during testing, and it is necessary to observe the model's accuracy, recall, and other evaluation metrics under different threshold settings.

[0114] Specifically, the controllable evaluation strategy for matching current boiler parameters can be a preset non-abnormal condition corresponding to the controllable evaluation strategy group for matching current boiler parameters. This preset non-abnormal condition can be set according to actual conditions, and this embodiment does not impose any restrictions on it. For example, the preset non-abnormal condition can be set as the number of controllable evaluation strategies in the controllable evaluation strategy group for matching current boiler parameters.

[0115] The evaluation strategy and controllable evaluation strategy can be selected from the state evaluation strategies used by the trained boiler state evaluation model through the model parsing module. The model parsing module can include a rule extractor, a model parser, and a key feature extractor. It can parse the trained boiler state evaluation model and extract the evaluation strategy and controllable evaluation strategy from it. Specifically, the rule extractor can select some important state evaluation strategies as pending evaluation strategies from the state evaluation strategies used by the boiler state evaluation model. For example, it can select pending evaluation strategies based on the call frequency. The obtained pending evaluation strategies are used as input to the model parser. The model parser can perform statistical analysis on the pending evaluation strategies to obtain the contribution coefficients corresponding to each pending evaluation strategy. Then, the key feature extractor determines the evaluation strategy and controllable evaluation strategy from the pending evaluation strategies based on the contribution coefficients. The evaluation strategy and controllable evaluation strategy can be used in the application stage of the boiler evaluation method for this thermal power plant.

[0116] The training process of the boiler condition assessment model is as follows: First, raw data, i.e., training boiler parameters (specifically, historical boiler pressures) and their corresponding labels (used to indicate whether the boiler pressure is risky), is obtained. The feature extraction module constructs features from the sample boiler states, resulting in a boiler state vector for the training boiler parameters. The boiler condition assessment model then performs safety checks on the training boiler parameters based on this boiler state vector, obtaining the true state confidence score. The boiler condition assessment model is then trained based on the true state confidence score and labels. This training process can also be combined with an offline model evaluation module. The offline evaluation module assesses the training effect; if the effect is unsatisfactory, iterative retraining is required. If the effect meets expectations, the trained boiler condition assessment model is output. For the trained boiler condition assessment model, effective rules (i.e., assessment strategies and controllable assessment strategies) can be extracted using the model parsing module. In practical applications, the boiler condition assessment model and a strategy pool (containing assessment strategies and controllable assessment strategies) are combined to perform safety checks on boiler pressure.

[0117] The overall framework diagram provided in this application is shown below. The specific process of the model training phase can be referred to the description in the above embodiments, and will not be repeated here. In the model application phase, the boiler pressure information to be detected can be input, and features are constructed from it through the feature extraction module. The derived boiler state vector is input into the trained boiler state assessment model. The boiler state assessment model predicts the input feature vector and outputs the state confidence score. If the state confidence score does not exceed the initial state confidence threshold Q1 and the advanced state confidence threshold Q2, the boiler pressure can be determined as a controllable state (i.e., a controllable boiler state). If the boiler pressure exceeds the initial state confidence threshold Q1, it can be judged by the set of effective early warning rules (i.e., the assessment strategy group) K1. If the boiler pressure matches the assessment strategy in K1, it is determined as a state requiring early warning (i.e., a boiler early warning state); otherwise, it is a controllable state. If the boiler pressure exceeds the advanced state confidence threshold Q2, it can be judged by the set of effective controllable rules (i.e., the controllable assessment strategy group) K2, indicating a state requiring early warning.

[0118] Optionally, in one specific embodiment, the boiler risk assessment device may further include a model parsing module, which includes a rule extractor, a model parser, and a key feature extractor; embodiments of this application may also include:

[0119] The rule extractor is invoked to count the state evaluation strategies in the trained boiler state evaluation model, thereby obtaining the invocation frequency of each state evaluation strategy in the trained boiler state evaluation model. Based on the invocation frequency, the state evaluation strategy is extracted to obtain at least one undetermined evaluation strategy.

[0120] Obtain the training boiler parameters and the preset identifiers corresponding to the training boiler parameters. The preset identifiers indicate the probability that the training boiler parameters are in a boiler warning state. The training boiler parameters include boiler parameters in at least one category.

[0121] Feature vector extraction is performed on the boiler parameters of each category in the training boiler parameters to obtain the boiler state vector of the training boiler parameters, and the boiler state vector of the training boiler parameters under each pending evaluation strategy is determined from the boiler state vector of the training boiler parameters.

[0122] The model parser is invoked to perform statistical analysis on the boiler state vectors corresponding to each pending evaluation strategy based on the preset identifier, and the contribution coefficients corresponding to each pending evaluation strategy are obtained.

[0123] The key feature extractor is invoked, and the evaluation strategy determination operation is performed based on the contribution coefficient to obtain the evaluation strategy and the controllable evaluation strategy.

[0124] As can be seen from the above, this embodiment can obtain the current boiler parameters through the server and perform feature vector extraction operations on the boiler parameters of each category in the current boiler parameters to obtain the boiler state vector of the current boiler parameters; the server uses the boiler state assessment model to perform boiler state assessment operations on the current boiler parameters according to the boiler state vector to obtain the state confidence of the current boiler parameters. The state confidence represents the probability that the current boiler parameters are in a boiler warning state. The boiler state assessment model includes at least one state assessment sub-network, wherein each state assessment sub-network corresponds to at least one state assessment strategy; when the state confidence exceeds the initial state confidence threshold, the command to call the assessment strategy is executed to obtain an assessment strategy group. The assessment strategy group includes at least one assessment strategy, which is selected from the state assessment strategies of the boiler state assessment model; when the current boiler parameters match the assessment strategy, the current boiler parameters are determined to be in a boiler warning state. When the state confidence level exceeds the advanced state confidence level threshold, the command to invoke the controllable assessment strategy is executed, resulting in a controllable assessment strategy group. This group includes at least one controllable assessment strategy selected from the state assessment strategies of the boiler state assessment model. When the current boiler parameters match a controllable assessment strategy, the current boiler parameters are determined to be in a controllable state. This application can combine the prediction results of the boiler state assessment model and the assessment strategy group to perform safety detection on boiler parameters, improving the detection efficiency and accuracy.

[0125] To more clearly describe the solution provided in the embodiments of this application, the aforementioned step 204 can be implemented through the following detailed steps.

[0126] 301. Obtain the current boiler parameters of the current boiler equipment and the target early warning and solution strategy for the current boiler parameters.

[0127] Among them, the current boiler parameters of the current boiler equipment are also the relevant information on the current fault risk of the current boiler equipment, and the target early warning and solution strategy is the relevant information on the solution strategy used to solve the current fault risk of the current boiler equipment.

[0128] 302. Based on the historical maintenance logs of the current boiler equipment, determine the equipment status information vector of the current boiler equipment. The historical maintenance logs include the historical boiler parameters of the current boiler equipment and the historical early warning and resolution strategies for the historical boiler parameters.

[0129] The historical boiler parameters of the current boiler equipment represent information about past boiler warning faults, while the historical warning resolution strategies represent information about the strategies adopted by the current boiler equipment in the past. The equipment status information vector of the current boiler equipment can reflect the historical warning risks and historical warning handling strategies of the current boiler equipment.

[0130] 303. Based on the current boiler equipment status information vector, current boiler parameters, and target early warning solution strategy, determine the correlation coefficient. The correlation coefficient is used to represent the compatibility between the target early warning solution strategy and the current boiler equipment.

[0131] The higher the compatibility score, the more applicable the solution strategy corresponding to the warning strategy is to the current boiler equipment; the lower the compatibility score, the less applicable the solution strategy corresponding to the warning strategy is to the current boiler equipment.

[0132] 304. If the correlation coefficient is greater than the preset correlation coefficient threshold, execute the target early warning solution strategy.

[0133] In this embodiment, the equipment status information vector of the current boiler equipment is determined through the historical maintenance logs of the current boiler equipment. This equipment status information vector reflects the historical early warning risks and handling strategies of the current boiler equipment. Based on the equipment status information vector, current boiler parameters, and the target early warning resolution strategy, predictions can be made to obtain the correlation coefficient between the target early warning resolution strategy and the current boiler equipment. The prediction process does not rely on rules, thus reducing auxiliary decision-making costs.

[0134] The above steps 301-304 are a simplified description of the embodiments of this application. The method provided by the embodiments of this application will be described in more detail below with some examples. The method includes the following steps.

[0135] 401. Obtain the current boiler parameters of the current boiler equipment and the target early warning and solution strategy for the current boiler parameters.

[0136] Among them, the current boiler parameters of the current boiler equipment are also the relevant information about the current fault risk of the current boiler equipment, such as the name of the current fault risk of the current boiler equipment. The number of current fault risks of the current boiler equipment can be one or multiple, and this application embodiment does not limit this; the target early warning solution strategy is the relevant information about the solution strategy used to solve the current fault risk of the current boiler equipment.

[0137] 402. Obtain the historical maintenance log of the current boiler equipment. The historical maintenance log includes the historical boiler parameters of the current boiler equipment and the historical early warning and resolution strategies for the historical boiler parameters.

[0138] 403. Based on the historical maintenance logs of the current boiler equipment, determine the equipment status information vector of the current boiler equipment.

[0139] In one possible implementation, the terminal determines multiple target boiler parameter nodes corresponding to the current boiler equipment in a boiler state graph structure based on the historical boiler parameters of the current boiler equipment. The boiler state graph structure describes the interaction relationships between the multiple boiler parameter nodes. The terminal also determines multiple target early warning strategy nodes corresponding to the current boiler equipment in an early warning strategy graph structure based on the historical early warning resolution strategies of the current boiler equipment. The early warning strategy graph structure describes the interaction relationships between the multiple early warning strategy nodes. Finally, the terminal determines the equipment state information vector of the current boiler equipment based on the multiple target boiler parameter nodes and the multiple target early warning strategy nodes. Both the boiler state graph structure and the early warning strategy graph structure are vertical domain knowledge graphs. The early warning strategy graph structure includes the interaction relationships between early warning strategy nodes, such as strategy number, strategy content, and applicable scenarios.

[0140] In this implementation, the terminal can associate the historical boiler parameters of the current boiler equipment with the boiler status diagram structure to determine multiple target boiler parameter nodes corresponding to the current boiler equipment. The terminal also associates the historical early warning resolution strategies of the current boiler equipment with the early warning strategy diagram structure to determine multiple target early warning strategy nodes corresponding to the current boiler equipment. Based on the determined target boiler parameter nodes and target early warning strategy nodes, the terminal determines the equipment status information vector of the current boiler equipment, which can reflect the current equipment condition to a certain extent. Subsequently, the terminal can determine the compatibility between the target early warning resolution strategies and the current boiler equipment based on the equipment status information vector.

[0141] To provide a clearer explanation of the above embodiments, the following description will be divided into three parts.

[0142] Part 1 describes the method by which the terminal determines multiple target boiler parameter nodes corresponding to the current boiler equipment in the boiler state diagram structure based on the historical boiler parameters of the current boiler equipment.

[0143] In one possible implementation, the terminal obtains at least one boiler parameter indication from the historical boiler parameters of the current boiler equipment. The boiler parameter indication includes at least one of the following: the name of the required boiler parameter node, the relationship between the required boiler parameter node and other nodes in the historical boiler parameters, the semantic data of the required boiler parameter node, and the graph structure parameters corresponding to the required boiler parameter node. The required boiler parameter node is also the node corresponding to a risk present in the current boiler equipment. Based on the boiler parameter indication, the terminal determines multiple target boiler parameter nodes corresponding to the current boiler equipment in the boiler state graph structure. For example, the terminal performs entity recognition on the historical boiler parameters of the current boiler equipment to determine at least one required boiler parameter node. In some embodiments, the required boiler parameter node is the name of a historical early warning fault of the current boiler equipment. The terminal obtains the boiler parameter indication of this required boiler parameter node. Based on the boiler parameter indication, the terminal determines at least one of the following: the name vector, the association vector, the semantic vector, and the graph vector of the required boiler parameter node. Based on the name vector, the association vector, the semantic vector, and the graph vector, the terminal obtains the pending boiler equipment vector of the current boiler equipment. The terminal performs a comparison operation between the vector of the boiler equipment to be determined and the vector of the boiler equipment of multiple boiler parameter nodes in the boiler state diagram structure. Multiple target boiler parameter nodes are determined from the multiple boiler parameter nodes. The vector distance between the boiler equipment vector of the target boiler parameter node and the vector of the boiler equipment to be determined is less than a preset vector distance threshold.

[0144] For example, the terminal uses a boiler parameter node template to match the historical boiler parameters of the current boiler equipment, and identifies at least one required boiler parameter node from the historical boiler parameters. The terminal determines the information associated with the required boiler parameter node in the historical boiler parameters as the boiler parameter indication. The terminal performs feature extraction on the boiler parameter indication to obtain at least one of the following: name vector, association vector, semantic vector, and graph vector of the required boiler parameter node. Specifically, the name embedding feature of the required boiler parameter node is the name vector obtained by the terminal embedding and encoding the name of the required boiler parameter node. The association vector of the required boiler parameter node is the association vector obtained by the terminal embedding and encoding the relationship between the required boiler parameter node and other nodes in the historical boiler parameters. In some embodiments, the terminal embeds and encodes the relationship based on a knowledge graph embedding model to obtain the association vector. The semantic vector of the required boiler parameter node is obtained by the terminal performing convolution processing on the semantic data of the required boiler parameter node. In some embodiments, the terminal can use a CNN to perform convolution processing on the semantic data of the required boiler parameter node to obtain the text embedding feature of the required boiler parameter node. For the graph vector of the required boiler parameter node, it is obtained by the terminal after extracting features from the graph structure parameters corresponding to the required boiler parameter node. In some embodiments, the terminal can use D-res-Net (Deep Residual Network) to extract features from the graph structure parameters to obtain the graph vector of the required boiler parameter node. The terminal performs an integration operation based on at least one of the name vector, association vector, semantic vector, and graph vector to obtain the pending boiler device vector of the current boiler equipment. If there is only one embedded feature, the terminal can directly use the embedded feature as the pending boiler device vector of the current boiler equipment; if there are at least two embedded features, the terminal can perform a weighted sum of the at least two embedded features to obtain the pending boiler device vector of the current boiler equipment. The weights are set by the technicians according to the actual situation, and this application embodiment does not limit this. For multiple boiler parameter nodes in the boiler state graph structure, the terminal can perform the aforementioned embedding feature extraction process on multiple boiler parameter nodes in advance, that is, obtain the boiler parameter node information of each boiler parameter node. The boiler parameter node information includes at least one of the following: the name of each boiler parameter node, the relationship between each boiler parameter node and other nodes in the historical boiler parameters, the semantic data of each boiler parameter node, and the graph structure parameters corresponding to each boiler parameter node. The terminal performs feature extraction on the boiler parameter nodes of each boiler parameter node to obtain at least one of the following: name vector, association vector, semantic vector, and graph vector for each boiler parameter node.The terminal obtains the boiler equipment vector for each boiler parameter node based on at least one of the following: name vector, association vector, semantic vector, and graph vector. The terminal determines the cosine similarity between the proposed boiler equipment vector and the boiler equipment vectors of multiple boiler parameter nodes, and identifies boiler parameter nodes with a cosine similarity greater than or equal to a similarity threshold as target boiler parameter nodes.

[0145] Part Two describes the method by which the terminal determines the multiple target early warning strategy nodes corresponding to the current boiler equipment in the early warning strategy graph structure based on the historical early warning resolution strategies of the current boiler equipment.

[0146] In one possible implementation, the terminal obtains at least one resolution strategy indication from the historical early warning resolution strategies of the current boiler equipment. The resolution strategy indication includes at least one of the following: the name of the demand early warning strategy node, the relationship between the demand early warning strategy node and other nodes in the historical early warning resolution strategies, the semantic data of the demand early warning strategy node, and the graph structure parameters corresponding to the demand early warning strategy node. Based on the resolution strategy indication, the terminal determines multiple target early warning strategy nodes corresponding to the current boiler equipment in the early warning strategy graph structure, wherein the demand early warning strategy node is the entity corresponding to the resolution strategy object of the current boiler equipment.

[0147] For example, the terminal performs entity recognition on the historical early warning resolution strategies of the current boiler equipment, and determines at least one demand early warning strategy node from the historical early warning resolution strategies. In some embodiments, the demand early warning strategy node is the name of the strategy historically adopted by the current boiler equipment. The terminal obtains the resolution strategy indication of the demand early warning strategy node. Based on the resolution strategy indication, the terminal determines at least one of the name vector, association vector, semantic vector, and graph vector of the demand early warning strategy node. Based on at least one of the name vector, association vector, semantic vector, and graph vector, the terminal obtains the pending early warning strategy vector of the current boiler equipment. The pending early warning strategy vector is compared with the boiler equipment vectors of multiple early warning strategy nodes in the early warning strategy graph structure, and multiple target early warning strategy nodes are determined from the multiple early warning strategy nodes. The vector distance between the boiler equipment vector of the target early warning strategy node and the pending early warning strategy vector is less than a preset vector distance threshold. In some embodiments, the terminal can represent the embedded features in the form of vectors, and the above-mentioned embedded features can be denoted as embedded vectors.

[0148] For example, the terminal uses an early warning strategy node template to match the historical early warning resolution strategies of the current boiler equipment, and identifies at least one demand early warning strategy node from the historical early warning resolution strategies. The terminal determines the information associated with the demand early warning strategy node in the historical early warning resolution strategies as the resolution strategy indication. The terminal performs feature extraction on the resolution strategy indication to obtain at least one of the following: name vector, association vector, semantic vector, and graph vector of the demand early warning strategy node. Specifically, the name embedding feature of the demand early warning strategy node is the name vector obtained by embedding and encoding the name of the demand early warning strategy node. The association vector of the demand early warning strategy node is the association vector obtained by embedding and encoding the relationship between the demand early warning strategy node and other nodes in the historical early warning resolution strategies. In some embodiments, the terminal embeds and encodes the relationship based on the TransE model to obtain the association vector. The semantic vector of the demand early warning strategy node is obtained by convolutional processing of the semantic data of the demand early warning strategy node. In some embodiments, the terminal can use a CNN to convolutionally process the semantic data of the demand early warning strategy node to obtain the text embedding feature of the demand early warning strategy node. For the graph vector of the demand warning strategy node, it is obtained by the terminal after extracting features from the graph structure parameters corresponding to the demand warning strategy node. In some embodiments, the terminal can extract features from the graph structure parameters using D-res-Net to obtain the graph vector of the demand warning strategy node. The terminal performs an integration operation based on at least one of the name vector, association vector, semantic vector, and graph vector to obtain the pending warning strategy vector of the current boiler equipment. If there is only one embedded feature, the terminal can directly use the embedded feature as the pending warning strategy vector of the current boiler equipment; if there are at least two embedded features, the terminal can perform a weighted sum of the at least two embedded features to obtain the pending warning strategy vector of the current boiler equipment. The weights are set by the technicians according to the actual situation, and this application embodiment does not limit this. For multiple early warning strategy nodes in the early warning strategy graph structure, the terminal can perform the aforementioned embedded feature extraction process on multiple early warning strategy nodes in advance, that is, obtain the early warning strategy node information of each early warning strategy node. The early warning strategy node information includes at least one of the following: the name of each early warning strategy node, the relationship between each early warning strategy node and other nodes in the historical early warning resolution strategy, the semantic data of each early warning strategy node, and the graph structure parameters corresponding to each early warning strategy node. The terminal performs feature extraction on the early warning strategy node of each early warning strategy node, obtaining at least one of the following: name vector, association vector, semantic vector, and graph vector. Based on at least one of the following: name vector, association vector, semantic vector, and graph vector, the terminal obtains the boiler equipment vector of each early warning strategy node.The terminal determines the cosine similarity between the vector of the pending early warning strategy and the boiler equipment vectors of multiple early warning strategy nodes, and identifies the early warning strategy nodes whose cosine similarity is greater than or equal to the similarity threshold as the target early warning strategy nodes.

[0149] Part Three explains the method by which the terminal determines the current boiler equipment status information vector based on multiple target boiler parameter nodes and multiple target early warning strategy nodes.

[0150] In one possible implementation, the terminal performs an integration operation on multiple boiler parameter vectors and multiple early warning strategy vectors to obtain the current boiler equipment status information vector, where the boiler parameter vector is the feature vector corresponding to the target boiler parameter node, and the early warning strategy vector is the feature vector corresponding to the target early warning strategy node.

[0151] In the third part described above, the terminal uses boiler parameter vectors and early warning strategy vectors to obtain the current boiler equipment status information vector. The boiler parameter vector is the feature vector corresponding to the target boiler parameter node, which belongs to the boiler state graph structure. The early warning strategy vector is the feature vector corresponding to the target early warning strategy node, which belongs to the early warning strategy graph structure. In some embodiments, the terminal can perform feature extraction on multiple boiler parameter nodes in the boiler state graph structure in advance to obtain the feature vector of each boiler parameter node. Similarly, it can perform feature extraction on multiple early warning strategy nodes in the early warning strategy graph structure to obtain the feature vector of each early warning strategy node. By performing the above steps, the terminal can directly obtain the boiler parameter vector and early warning strategy vector, which is highly efficient. The following describes the methods for the terminal to extract features from multiple boiler parameter nodes in the boiler state graph structure and from multiple early warning strategy nodes in the early warning strategy graph structure.

[0152] In one possible implementation, the terminal performs graph convolutional processing on the boiler state graph structure to obtain feature vectors for multiple boiler parameter nodes in the boiler state graph structure. The terminal then uses a GCN network to extract feature vectors for multiple early warning strategy nodes in the early warning strategy graph structure.

[0153] For example, the terminal trains a first GCN to extract features from the boiler state graph structure using the GCN network, obtaining feature vectors for multiple boiler parameter nodes in the boiler state graph structure. The terminal then trains a second GCN to extract features from the early warning strategy graph structure using the GCN network, obtaining feature vectors for multiple early warning strategy nodes in the early warning strategy graph structure.

[0154] 404. Based on the current boiler equipment status information vector, current boiler parameters, and target early warning solution strategy, determine the correlation coefficient. The correlation coefficient is used to represent the compatibility between the target early warning solution strategy and the current boiler equipment.

[0155] The higher the compatibility score, the more applicable the solution strategy corresponding to the warning strategy is to the current boiler equipment; the lower the compatibility score, the less applicable the solution strategy corresponding to the warning strategy is to the current boiler equipment.

[0156] In one possible implementation, the terminal obtains the equipment reference coefficient between the current boiler equipment's equipment status information vector and the current boiler parameters. The terminal also obtains the strategy reference coefficient between the current boiler equipment's equipment status information vector and the target early warning resolution strategy. Based on the equipment reference coefficient and the strategy reference coefficient, the terminal performs an integration operation on the current boiler parameters and the target early warning resolution strategy to obtain the matching vector for the current boiler equipment. Finally, the terminal determines the correlation coefficient based on the current boiler equipment's equipment status information vector and its matching vector.

[0157] In this embodiment, the equipment status information vector of the current boiler equipment is determined through the historical maintenance logs of the current boiler equipment. This vector reflects the historical early warning risks and handling strategies of the current boiler equipment. Based on the equipment status information vector, current boiler parameters, and the target early warning resolution strategy, a correlation coefficient between the target early warning resolution strategy and the current boiler equipment can be obtained. This prediction process does not rely on rules, reducing auxiliary decision-making costs. The terminal displays the correlation coefficient on a page, allowing users to determine the compatibility between the target early warning resolution strategy and the current boiler equipment. This facilitates timely adjustments to the target early warning resolution strategy when the compatibility is low, resulting in high efficiency in human-computer interaction.

[0158] This invention provides a server 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the server 100 executes the aforementioned artificial intelligence-based safety monitoring method for thermal power plants. Figure 2 As shown, Figure 2 This is a structural block diagram of a server 100 provided in an embodiment of the present invention. The server 100 includes a memory 111, a processor 112, and a communication unit 113.

[0159] To enable data transmission or interaction, the memory 111, processor 112, and communication unit 113 are electrically connected to each other, either directly or indirectly. For example, these components can be electrically connected to each other via one or more communication buses or signal lines.

[0160] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the present disclosure and to employ various embodiments with different modifications to suit a particular intended application. For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the present disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A safety monitoring method for thermal power plants based on artificial intelligence, characterized in that, include: In response to the start command of the safety monitoring system of the thermal power plant, generate periodic safety assessment instructions; According to the periodic safety assessment instruction, the current boiler parameters are obtained at preset time intervals; Call the pre-trained boiler condition assessment model, input the current boiler parameters into the boiler condition assessment model, and determine whether the current boiler parameters are in a boiler early warning state; If so, then obtain and execute the target early warning solution strategy for the current boiler parameters; If not, generate a boiler operation safety report for the current period and send the boiler operation safety report to the preset safety terminal; The step of calling a pre-trained boiler condition assessment model, inputting the current boiler parameters into the boiler condition assessment model, and determining whether the current boiler parameters are in a boiler early warning state includes: The frequency of each state assessment strategy in the boiler state assessment model is obtained by counting the strategies. Based on the call frequency, the state evaluation strategy is subjected to an evaluation strategy extraction operation to obtain at least one undetermined evaluation strategy. Obtain training boiler parameters and preset identifiers corresponding to the training boiler parameters. The preset identifiers indicate the probability that the training boiler parameters are in a boiler warning state. The training boiler parameters include boiler parameters in at least one category. Feature vector extraction is performed on the boiler parameters of each category in the training boiler parameters to obtain the boiler state vector of the training boiler parameters, and the boiler state vector corresponding to the training boiler parameters under each pending evaluation strategy is determined from the boiler state vector of the training boiler parameters. Based on the preset identifier, statistical analysis is performed on the boiler state vector corresponding to each pending evaluation strategy to obtain the contribution coefficient corresponding to each pending evaluation strategy. Based on the contribution coefficient, a demand assessment strategy is selected from the pending assessment strategies; The demand assessment strategy whose contribution coefficient exceeds the preset coefficient is determined as the assessment strategy, and the demand assessment strategy whose contribution coefficient does not exceed the preset coefficient is determined as the controllable assessment strategy.

2. The method according to claim 1, characterized in that, The step of calling a pre-trained boiler condition assessment model, inputting the current boiler parameters into the boiler condition assessment model, and determining whether the current boiler parameters are in a boiler early warning state further includes: Obtain the current boiler parameters, which include boiler parameters in at least one category; Feature vector extraction is performed on the boiler parameters of each category to obtain the boiler state vector of the current boiler parameters; The boiler status assessment model performs a boiler status assessment operation on the current boiler parameters based on the boiler status vector to obtain the status confidence of the current boiler parameters. The status confidence represents the probability that the current boiler parameters are in a boiler warning state. The boiler status assessment model includes at least one status assessment sub-network, wherein each status assessment sub-network corresponds to at least one status assessment strategy. When the state confidence exceeds the initial state confidence threshold, the command to call the evaluation strategy is executed to obtain an evaluation strategy group. The evaluation strategy group includes at least one evaluation strategy, which is selected from the state evaluation strategies of the boiler state evaluation model. When the current boiler parameters match the evaluation strategy, the current boiler parameters are determined to be in a boiler early warning state.

3. The method according to claim 2, characterized in that, The step of calling a pre-trained boiler condition assessment model, inputting the current boiler parameters into the boiler condition assessment model, and determining whether the current boiler parameters are in a boiler early warning state further includes: When the state confidence exceeds the advanced state confidence threshold, the command to call the controllable evaluation strategy is executed to obtain a controllable evaluation strategy group. The controllable evaluation strategy group includes at least one controllable evaluation strategy, which is selected from the state evaluation strategies of the boiler state evaluation model. When the current boiler parameters match the controllable evaluation strategy, the current boiler parameters are determined to be in a controllable state.

4. The method according to claim 2, characterized in that, Before obtaining the state confidence level of the current boiler parameters by performing a boiler state assessment operation on the current boiler parameters based on the boiler state vector using the boiler state assessment model, the method further includes: Obtain a sample training set, which includes training boiler parameters and prediction state confidence corresponding to the training boiler parameters. The training boiler parameters include boiler parameters in at least one category. Feature vector extraction is performed on the boiler parameters of each category in the training boiler parameters to obtain the boiler state vector of the training boiler parameters; Using the initial boiler condition assessment model, the boiler condition assessment operation is performed on the training boiler parameters according to the boiler condition vector to obtain the true state confidence of the training boiler parameters as the boiler early warning state. Based on the confidence levels of the actual and predicted states of the trained boiler parameters, the parameters of the initial boiler condition assessment model are adjusted to obtain the trained boiler condition assessment model.

5. The method according to claim 2, characterized in that, The step of using a boiler condition assessment model to perform a boiler condition assessment operation on the current boiler parameters based on the boiler condition vector, and obtaining the condition confidence level of the current boiler parameters, includes: For each state assessment subnetwork in the boiler condition assessment model, the boiler condition assessment operation is performed on the current boiler parameters based on the boiler condition vector through the state assessment subnetwork to obtain the undetermined state confidence of the current boiler parameters. The confidence scores of the undetermined states corresponding to each state evaluation subnetwork are integrated to obtain the state confidence scores of the current boiler parameters.

6. The method according to claim 5, characterized in that, The step of integrating the confidence scores of the undetermined states corresponding to each state evaluation sub-network to obtain the state confidence scores of the current boiler parameters includes: Determine the contribution coefficient of the confidence of the undetermined state corresponding to each state evaluation subnetwork; Based on the contribution coefficient, a weighted average operation is performed on the confidence scores of the undetermined states corresponding to each state evaluation subnetwork to obtain the state confidence scores of the current boiler parameters.

7. The method according to claim 1, characterized in that, The step of acquiring and executing the target early warning solution strategy for the current boiler parameters includes: Obtain the current boiler parameters of the current boiler equipment and the target early warning and resolution strategy for the current boiler parameters; Based on the historical maintenance log of the current boiler equipment, the equipment status information vector of the current boiler equipment is determined. The historical maintenance log includes the historical boiler parameters of the current boiler equipment and the historical early warning and resolution strategies for the historical boiler parameters. Based on the current boiler parameters, at least one reference boiler parameter node corresponding to the current boiler equipment in the boiler state diagram structure is determined. Based on the dot product between the current boiler equipment status information vector and the equipment reference vector, the equipment reference coefficient is obtained, and the equipment reference vector is the feature vector corresponding to the reference boiler parameter node. Obtain the strategy reference coefficient between the current boiler equipment status information vector and the target early warning resolution strategy; Based on the equipment reference coefficient and the strategy reference coefficient, an integration operation is performed on the current boiler parameters and the target early warning solution strategy to obtain the matching vector of the current boiler equipment; Based on the device status information vector of the current boiler equipment and the matching vector of the current boiler equipment, a correlation coefficient is determined. The correlation coefficient is used to represent the compatibility between the target early warning solution strategy and the current boiler equipment. If the correlation coefficient is greater than a preset correlation coefficient threshold, the target early warning solution strategy is executed.

8. The method according to claim 7, characterized in that, The step of determining the equipment status information vector of the current boiler equipment based on the historical maintenance log of the current boiler equipment includes: From the historical boiler parameters of the current boiler equipment, at least one boiler parameter indication is obtained. The boiler parameter indication includes at least one of the following: the name of the demand boiler parameter node, the relationship between the demand boiler parameter node and other nodes in the historical boiler parameters, the semantic data of the demand boiler parameter node, and the graph structure parameters corresponding to the demand boiler parameter node. Based on the boiler parameter indication, determine at least one of the name vector, association vector, semantic vector, and graph vector of the required boiler parameter node; Based on at least one of the name vector, the association vector, the semantic vector, and the graph vector, obtain the pending boiler equipment vector of the current boiler equipment; The vector of the undetermined boiler equipment is compared with the boiler equipment vectors of multiple boiler parameter nodes in the boiler state diagram structure. Multiple target boiler parameter nodes are determined from the multiple boiler parameter nodes. The vector distance between the boiler equipment vector of the target boiler parameter node and the vector of the undetermined boiler equipment is less than a preset vector distance threshold. The boiler state diagram structure is used to describe the interaction relationship between multiple boiler parameter nodes. From the historical early warning resolution strategies of the current boiler equipment, at least one resolution strategy indication is obtained. The resolution strategy indication includes at least one of the following: the name of the demand early warning strategy node, the relationship between the demand early warning strategy node and other nodes in the historical early warning resolution strategies, the semantic data of the demand early warning strategy node, and the graph structure parameters corresponding to the demand early warning strategy node. Based on the solution strategy indication, at least one of the name vector, association vector, semantic vector, and graph vector of the demand early warning strategy node is determined; Based on at least one of the name vector, the association vector, the semantic vector, and the graph vector, obtain the pending early warning strategy vector for the current boiler equipment; The undetermined early warning strategy vector is compared with the boiler equipment vectors of multiple early warning strategy nodes in the early warning strategy graph structure. Multiple target early warning strategy nodes are determined from the multiple early warning strategy nodes. The vector distance between the boiler equipment vector of the target early warning strategy node and the undetermined early warning strategy vector is less than a preset vector distance threshold. The early warning strategy graph structure is used to describe the interaction relationship between multiple early warning strategy nodes. The GCN network is called to extract the feature vectors of multiple boiler parameter nodes in the boiler state diagram structure. The GCN network is called to extract the feature vectors of multiple early warning strategy nodes in the early warning strategy graph structure. Multiple boiler parameter vectors and multiple early warning strategy vectors are integrated to obtain the current boiler equipment status information vector. The boiler parameter vector is the feature vector corresponding to the target boiler parameter node, and the early warning strategy vector is the feature vector corresponding to the target early warning strategy node.

9. A server system, characterized in that, Includes a server, said server being used to perform the method of any one of claims 1-8.

Images