Power generation equipment fault prediction method and system based on mechanism model and AI dual verification

Abstract

The invention discloses a power generation equipment fault prediction method and system based on mechanism model and AI dual verification. The method comprises the following steps: carrying out double-path parallel feature processing on a temperature parameter, a pressure parameter and a vibration parameter of the power generation equipment to respectively obtain a mechanism feature vector and an AI feature vector; utilizing a mechanism model and an artificial intelligence model to predict the features to obtain two prediction state levels; and performing difference analysis on the prediction result, and triggering a secondary confirmation window mechanism to perform further verification when cross-level difference is detected. According to the method, the technical problem that existing single model prediction is insufficient in reliability is solved, and the accuracy and reliability of fault prediction are remarkably improved through double verification of the mechanism model and the AI model and a secondary confirmation mechanism; and meanwhile, a double-path parallel feature processing scheme is adopted, so that the response speed of the system is improved, and the engineering practical value is good.

Description

Technical Field

[0001] The present invention relates to the technical field of power generation equipment fault diagnosis, and in particular to a power generation equipment fault prediction method and system based on dual verification of mechanism model and artificial intelligence. Background Art

[0002] Power generation equipment is the core equipment of the power system, and its safe and stable operation is directly related to the reliability of power supply. As power generation equipment develops towards large-scale and complex directions, equipment failure prediction technology faces increasing challenges. At present, power generation equipment failure prediction mainly adopts two technical routes: traditional methods based on mechanism models and data-driven methods based on artificial intelligence.

[0003] Traditional mechanism model methods mainly perform fault analysis based on physical equations such as thermodynamics and vibration dynamics. Such methods have strong interpretability and can reflect the physical characteristics of the equipment, but they have limitations when dealing with complex working conditions and multi-parameter coupling problems. In recent years, with the development of artificial intelligence technology, fault prediction methods based on deep learning have been widely used. Such methods have powerful data processing capabilities and nonlinear feature extraction capabilities, but they are overly dependent on historical data, and the prediction results lack physical meaning.

[0004] The existing technology has the following main problems: first, the reliability of a single prediction model is insufficient, the mechanism model is difficult to adapt to complex working conditions, and the artificial intelligence model has a "black box" characteristic; second, there is a lack of an effective prediction result verification mechanism, which is prone to false positives or omissions; third, the traditional serial data processing method leads to a slow system response speed, which is difficult to meet the real-time monitoring needs. Therefore, it is urgent to develop a new method for power generation equipment fault prediction that is both reliable and real-time. Summary of the invention

[0005] In order to solve the problems of insufficient reliability of a single model, lack of a prediction result verification mechanism, and slow system response speed in the existing power generation equipment fault prediction technology, the present invention proposes a power generation equipment fault prediction method and system based on dual verification of a mechanism model and AI. This method significantly improves the accuracy and reliability of fault prediction through a dual verification mechanism of a mechanism model and an artificial intelligence model, combined with a secondary confirmation window strategy.

[0006] The technical solution adopted by the present invention is:

[0007] A method for predicting power generation equipment faults based on a mechanism model and AI dual verification, characterized by comprising the following steps:

[0008] Obtain temperature parameters, pressure parameters and vibration parameters of power generation equipment;

[0009] Performing dual-path parallel feature processing on the temperature parameter, the pressure parameter and the vibration parameter;

[0010] The first feature processing path performs data cleaning on the temperature parameter, the pressure parameter and the vibration parameter to obtain mechanism characteristics, and the data cleaning includes removing abnormal values, processing missing values ​​and eliminating noise signals;

[0011] The second feature processing path calculates the statistical features of the temperature parameter, the pressure parameter and the vibration parameter based on a preset time window, performs feature standardization processing to obtain a feature vector, wherein the feature standardization includes normalizing the statistical features and organizing the processed features into a vector form; the mechanism model and the artificial intelligence model are used to predict the mechanism features and the feature vector respectively; wherein the mechanism model prediction includes inputting the mechanism features into the mechanism model, performing calculations based on the mechanism equations of thermodynamics and vibration dynamics, and obtaining a first prediction state level;

[0012] Wherein, the artificial intelligence model prediction includes inputting the feature vector into the artificial intelligence model, processing it through a feedforward neural network, and obtaining a second prediction state level;

[0013] Performing a difference analysis on the first prediction state level and the second prediction state level, and executing a corresponding processing strategy according to a preset difference level determination rule;

[0014] The first prediction state level and the second prediction state level are as follows: normal state is recorded as level 1, warning state is recorded as level 2, and dangerous state is recorded as level 3 from low to high;

[0015] The difference level determination rule includes: when the level difference between the first prediction state level and the second prediction state level is 1, it is determined as an adjacent level difference; when the level difference between the first prediction state level and the second prediction state level is 2, it is determined as a cross-level difference;

[0016] When the cross-level difference is detected, a secondary confirmation window mechanism is triggered for further verification, and a final fault prediction result is obtained based on the result of the difference analysis and the verification result of the secondary confirmation window mechanism.

[0017] A further technical solution is that the mechanism model includes:

[0018] Temperature mechanism model, which calculates the temperature field deviation value based on the energy balance equation;

[0019] Pressure mechanism model, calculates pressure deviation value based on oil film load equation;

[0020] Vibration mechanism model, which calculates the vibration deviation value based on the vibration response equation;

[0021] Wherein, the mechanism model includes a temperature preset threshold, a pressure preset threshold and a vibration preset threshold, and the temperature preset threshold, the pressure preset threshold and the vibration preset threshold all include a first gear value and a second gear value;

[0022] Wherein, the mechanism model makes a state judgment by comparing the temperature field deviation value with the temperature preset threshold, the pressure deviation value with the pressure preset threshold, and the vibration deviation value with the vibration preset threshold. When the deviation value is less than the first gear value corresponding to the preset threshold, it is judged as the level 1; when the deviation value is greater than or equal to the first gear value corresponding to the preset threshold and less than the second gear value corresponding to the preset threshold, it is judged as the level 2; when the deviation value is greater than or equal to the second gear value corresponding to the preset threshold, it is judged as the level 3;

[0023] After the mechanism model completes independent determination of the temperature state, the pressure state and the vibration state, the highest level is selected as the first predicted state level.

[0024] A further technical solution is that the statistical features include:

[0025] The mean parameter, standard deviation parameter and peak parameter of the temperature parameter;

[0026] The mean parameter, standard deviation parameter and peak parameter of the pressure parameter;

[0027] Mean parameter, standard deviation parameter and peak parameter of the vibration parameter;

[0028] A further technical solution is that the artificial intelligence model is a feedforward neural network, including:

[0029] an input layer having a plurality of input nodes, the number of the input nodes matching the dimension of the feature vector, for receiving and processing the feature vector;

[0030] A hidden layer having a plurality of hidden nodes, wherein the hidden nodes perform a nonlinear transformation on the information from the input layer through the first activation function, wherein the first activation function adopts a ReLU function for extracting a nonlinear representation of a feature;

[0031] An output layer having a plurality of output nodes, wherein the output nodes calculate and output the probability distribution of level 1, level 2 and level 3 through the second activation function, wherein the second activation function adopts a Softmax function for converting the output into probability values ​​of three levels, each level corresponds to a probability value between 0 and 1, and the sum of the three probability values ​​is 1;

[0032] The specific method for determining the prediction state based on the maximum probability value is to select the level corresponding to the maximum value from the three output probability values ​​as the prediction result. If the maximum probability value corresponds to level 1, level 1 is output; if the maximum probability value corresponds to level 2, level 2 is output; if the maximum probability value corresponds to level 3, level 3 is output.

[0033] A further technical solution is that the fusion processing includes:

[0034] When there is an adjacent level difference between the first prediction state level and the second prediction state level, taking the higher level as the final prediction state level;

[0035] When the first predicted state level or the second predicted state level is the dangerous state, taking the dangerous state as the final predicted state level;

[0036] When there is the cross-level difference between the first predicted state level and the second predicted state level and both are not the dangerous state, the secondary confirmation window mechanism is activated.

[0037] A further technical solution is that the specific steps of the secondary confirmation window mechanism are:

[0038] A confirmation window for setting a preset duration, wherein the preset duration is T minutes;

[0039] During the confirmation window, continuously collecting the temperature parameter, the pressure parameter and the vibration parameter at a preset sampling frequency of f times / minute;

[0040] Performing feature processing on the collected parameters, respectively predicting through the mechanism model and the artificial intelligence model, and obtaining a prediction sequence within the confirmation window;

[0041] Performing statistical analysis based on the prediction sequence, when the prediction state level is consistent for n consecutive times in the prediction sequence, using the consistent prediction result as the final prediction state level;

[0042] If n consecutive consistent prediction status levels still do not appear at the end of the confirmation window, the first prediction status level is adopted as the final prediction status level.

[0043] The present invention also discloses a power generation equipment fault prediction system based on a mechanism model and AI dual verification, which is characterized by comprising:

[0044] Parameter collection unit, used to collect temperature parameters, pressure parameters and vibration parameters of power generation equipment in real time;

[0045] A mechanism characteristic processing unit, used for performing data cleaning on the parameters to obtain characteristic data for mechanism analysis;

[0046] An AI feature processing unit, used to calculate the statistical features of the parameters and perform standardization to obtain a feature vector for AI analysis;

[0047] A mechanism prediction unit, including a temperature mechanism model, a pressure mechanism model and a vibration mechanism model, is used to analyze the mechanism characteristic data based on a physical equation and output a first prediction state level;

[0048] An AI prediction unit, comprising a feedforward neural network having an input layer, a hidden layer, and an output layer, for analyzing the feature vector and outputting a second prediction state level;

[0049] The prediction fusion unit is used to perform difference analysis on the first prediction state level and the second prediction state level, and trigger a secondary confirmation mechanism when a cross-level difference is detected, and finally output a fault prediction result.

[0050] The present invention has the following beneficial effects:

[0051] (1) Through the dual verification mechanism of the mechanism model and the AI ​​model, the advantages of the prediction model are complementary. The mechanism model is based on physical equations for prediction, has strong interpretability and stability, and can accurately reflect the basic physical characteristics of the equipment; the AI ​​model has powerful nonlinear feature extraction and learning capabilities, and can adapt to complex working conditions. The two models verify and complement each other, which not only significantly improves the accuracy and reliability of fault prediction, but also effectively overcomes the problem of insufficient prediction ability of a single model under complex working conditions.

[0052] (2) The system innovatively designed the difference level judgment rules and the secondary confirmation window mechanism, and built a complete prediction result verification system. When differences between adjacent levels are detected, the system will use the higher level as the prediction result; when differences across levels occur, the system will automatically trigger the secondary confirmation window mechanism, continuously collect data and verify within the preset time, and ensure the reliability of the final prediction result through consistency analysis of multiple consecutive predictions, effectively reducing the risk of false positives and false negatives.

[0053] (3) A dual-path parallel feature processing solution is adopted to break through the efficiency bottleneck of the traditional serial processing method. The mechanism feature processing path obtains feature data that meets the requirements of the physical model through data cleaning; the AI ​​feature processing path calculates statistical features based on the preset time window and performs standardized processing. The parallel execution of the two processing paths not only optimizes the computing efficiency of the system, but also realizes the simultaneous extraction of mechanism features and AI features, significantly improving the system response speed, so that the fault prediction system can meet the strict timeliness requirements of real-time monitoring of power generation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of the power generation equipment fault prediction method based on mechanism model and AI dual verification.

[0055] Figure 2 Two-way parallel flow chart for feature processing.

[0056] Figure 3 This is a diagram of the dual prediction architecture of the mechanism model and the AI ​​model.

[0057] Figure 4 This is a flow chart for determining the difference level. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0059] Reference Figure 1 The main steps of the method provided by the present invention are as follows:

[0060] (1) Obtaining temperature parameters, pressure parameters and vibration parameters of power generation equipment;

[0061] (2) Performing dual-path parallel feature processing on temperature parameters, pressure parameters, and vibration parameters;

[0062] (3) using a mechanism model and an artificial intelligence model to predict the mechanism characteristics and the characteristic vector respectively;

[0063] (4) performing a difference analysis on the first prediction status level and the second prediction status level, and executing a corresponding processing strategy according to a preset difference level determination rule;

[0064] As a preferred embodiment of the present invention, in step (1), the process of parameter collection is as follows:

[0065] (1-1) Sensor deployment. The system implements real-time monitoring of the equipment’s operating status by deploying high-precision sensor arrays at key locations of the power generation equipment. Specifically, it includes three types of sensors: temperature sensors, pressure sensors, and vibration sensors. Among them, temperature sensors are mainly installed at hot spots such as bearings, stators, and rotors; pressure sensors are deployed at key nodes of the oil system and sealing system; and vibration sensors are fixed at vibration-sensitive locations such as bearing seats and frames.

[0066] (1-2) Data acquisition. A distributed data acquisition system is used to synchronously collect various sensor signals. Preferably, the acquisition frequency of temperature parameters is set to 2 times per minute, the acquisition frequency of pressure parameters is set to 4 times per minute, and the acquisition frequency of vibration parameters is set to 10 times per minute. The system transmits the collected parameters to the central processing unit in real time through a high-speed industrial bus to ensure the synchronization and integrity of the acquisition of various parameters. These parameters will serve as the basic data for subsequent feature processing and fault prediction.

[0067] As a preferred embodiment of the present invention, refer to Figure 2 , in step (2), the feature processing process is as follows:

[0068] The present invention adopts an innovative dual-path parallel feature processing solution to process the collected parameters from two dimensions: mechanism analysis and statistical analysis. This parallel processing architecture not only improves the computational efficiency of the system, but also provides support for different feature dimensions for subsequent dual prediction.

[0069] (2-1) First-pass processing: mechanism feature extraction

[0070] The first feature processing mainly obtains feature data suitable for mechanism analysis through data cleaning. The data cleaning process includes three key steps:

[0071] First, the system uses an outlier detection method based on statistical principles to remove outliers. In the specific implementation, a sliding window is used to calculate the mean and standard deviation of the parameters, and data points that deviate from the mean by more than three times the standard deviation are marked as outliers and removed. This method can effectively identify and remove abnormal data caused by factors such as sensor failure and external interference.

[0072] Secondly, for the missing values ​​detected, the system uses linear interpolation to process them. Preferably, in practical applications, different interpolation strategies are selected according to the time span of the missing data: when the missing time is short (less than 3 sampling periods), linear interpolation is used; when the missing time is long, the statistical mean of historical data is used for filling.

[0073] Finally, the system uses the wavelet transform method to reduce the noise of the signal. Preferably, the db4 wavelet basis function is used to decompose the temperature, pressure and vibration signals into 4 layers respectively, and the high-frequency noise components are filtered by the threshold method, and then the cleaned signals are reconstructed. This method not only retains the effective components of the signal, but also effectively eliminates the influence of random noise.

[0074] (2-2) Second processing: AI feature extraction

[0075] The second feature processing is mainly for artificial intelligence models, which obtain feature vectors by calculating statistical features and performing normalization. The specific implementation process is as follows:

[0076] First, preferably, the system sets a 10-minute sliding time window, and calculates the statistical characteristics of the three types of parameters within this window. For each type of parameter (temperature, pressure, vibration), calculate separately:

[0077] Mean parameter, reflecting the average level of the parameter within the time window;

[0078] The standard deviation parameter characterizes the degree of fluctuation of the parameter;

[0079] Peak parameter, record the maximum deviation value of the parameter;

[0080] Secondly, the calculated statistical features are normalized. Preferably, the maximum and minimum normalization method is used to normalize each statistical feature to the interval [0,1] to ensure that features of different dimensions can be processed uniformly. The normalization formula is:

[0081]

[0082] Among them, x is the original eigenvalue that needs to be normalized, x min is the minimum value of the feature in the data set, x max is the maximum value of the feature in the data set, x normalized is the normalized eigenvalue, ranging from [0,1].

[0083] Finally, the standardized 9 features (3 types of parameters × 3 statistics) are organized into a feature vector in a predefined order. Preferably, the specific order is: temperature mean, temperature standard deviation, temperature peak, pressure mean, pressure standard deviation, pressure peak, vibration mean, vibration standard deviation, vibration peak. This unified feature vector format provides a standardized input form for subsequent AI model predictions.

[0084] Through the above-mentioned two-way parallel feature processing, the system not only obtains cleaned data suitable for mechanism analysis, but also obtains standardized feature vectors that are convenient for AI model processing, laying a good data foundation for subsequent dual predictions.

[0085] As a preferred embodiment of the present invention, refer to Figure 3 , in step (3), the state prediction process is as follows:

[0086] The present invention adopts a dual prediction mechanism of mechanism model and artificial intelligence model to comprehensively evaluate the equipment status through two models with different principles. This dual prediction mechanism gives full play to the interpretability of mechanism analysis and the nonlinear feature extraction ability of artificial intelligence.

[0087] (3-1) Mechanism model prediction

[0088] The mechanism model predicts the equipment status based on physical equations and includes three sub-models: temperature mechanism model, pressure mechanism model and vibration mechanism model.

[0089] The temperature mechanism model is analyzed using the energy balance equation. The system establishes a thermodynamic model of the power generation equipment and calculates the temperature field distribution of key components. The temperature field deviation value is obtained by comparing the actual measured temperature with the theoretical calculated temperature. Preferably, the preset temperature threshold includes two gears: the first gear value is set to 1.15 times the theoretical temperature, and the second gear value is set to 1.3 times the theoretical temperature.

[0090] The pressure mechanism model is analyzed based on the oil film load equation. The system evaluates the working status of the bearing system by calculating the oil film pressure distribution. The difference between the measured pressure value and the theoretical calculated pressure value is the pressure deviation value. Preferably, the first gear value of the pressure preset threshold is ±10% of the standard working pressure, and the second gear value is ±20% of the standard working pressure.

[0091] The vibration mechanism model is analyzed using the vibration response equation. The system calculates the theoretical vibration response of the equipment under different working conditions and compares it with the measured vibration data to obtain the vibration deviation value. Preferably, the first gear value of the vibration preset threshold is set to 1.5 times the reference amplitude, and the second gear value is set to 2.5 times the reference amplitude.

[0092] For each sub-model, the system determines the status level based on the comparison result between the deviation value and the preset threshold:

[0093] When the deviation value is less than the first gear value, it is judged as level 1 (normal state);

[0094] When the deviation value is between the first gear value and the second gear value, it is determined to be level 2 (warning state);

[0095] When the deviation value is greater than or equal to the second gear value, it is judged as level 3 (dangerous state)

[0096] Ultimately, the system selects the highest level from the three judgment results of temperature state, pressure state and vibration state as the first predicted state level, reflecting the system's safety priority principle.

[0097] (3-2) Artificial Intelligence Model Prediction

[0098] The artificial intelligence model adopts a three-layer feedforward neural network structure and realizes state prediction through deep learning of standardized feature vectors.

[0099] Preferably, the input layer is designed with 9 input nodes, which completely correspond to the dimensions of the feature vector, and receive the mean, standard deviation and peak characteristics of three types of parameters: temperature, pressure and vibration. Each input node is equipped with a data normalization processing unit to ensure the uniform scale of the input data.

[0100] Preferably, the hidden layer uses 64 neuron nodes, and the nonlinear transformation of features is achieved through the ReLU activation function. The expression of the ReLU function is:

[0101] f(u)=max(0,u)

[0102] Among them, u is the input value of the hidden layer neuron in the neural network, f(u) is the output value of the activation function, when u≥0, the output is equal to the input value u, when u<0, the output is 0.

[0103] This activation function not only speeds up the training of the network, but also effectively alleviates the gradient vanishing problem and improves the performance of the model.

[0104] Preferably, the output layer includes three output nodes, corresponding to the three states of level 1, level 2 and level 3. The output of the neural network is converted into a probability distribution through the Softmax activation function to ensure that the sum of the probability values ​​output by the three nodes is 1, where the expression of the Softmax function is:

[0105]

[0106] Among them, z i is the original value of the ith output node (i.e., the output value of the last layer of the neural network), m is the total number of output nodes (preferably, in this system, m = 3, corresponding to three state levels), P(i) is the probability of the state corresponding to the ith output node, and its value range is [0,1]. The sum of the probabilities of all output nodes is equal to 1, i.e., ΣP(i) = 1.

[0107] The system determines the second predicted state level based on the probability values ​​of the three nodes in the output layer. The specific method is to select the state with the largest probability value as the prediction result. This probability-based prediction method not only gives the judgment result of the device state, but also provides prediction confidence information.

[0108] Through the above-mentioned dual prediction mechanism, the system not only ensures the physical rationality of the prediction results, but also realizes the effective identification of complex nonlinear characteristics, laying a solid foundation for subsequent prediction fusion.

[0109] As a preferred embodiment of the present invention, refer to Figure 4 , in step (4), the prediction fusion process is as follows:

[0110] The present invention designs a complete prediction fusion mechanism, which realizes the reliable fusion of the prediction results of the mechanism model and the artificial intelligence model through difference analysis and secondary confirmation window mechanism.

[0111] (4-1) Difference Analysis

[0112] The system first analyzes the difference between the prediction results of the mechanism model and the artificial intelligence model. The prediction status level is divided into three levels: level 1 represents normal status, level 2 represents warning status, and level 3 represents dangerous status. The system calculates the level difference between the two prediction results and adopts corresponding processing strategies according to the degree of difference.

[0113] When the level difference between the two prediction results is 1, the system determines it as an adjacent level difference. For example, when the mechanism model predicts level 1 (normal state) and the artificial intelligence model predicts level 2 (warning state), it is an adjacent level difference. In this case, the system follows the principle of safety first and selects the higher prediction level as the final result.

[0114] When the level difference between the two prediction results is 2, the system determines it as a cross-level difference. For example, when the mechanism model predicts level 1 (normal state) and the artificial intelligence model predicts level 3 (dangerous state), it is a cross-level difference. This significant prediction difference indicates that there may be special circumstances that require further verification.

[0115] (4-2) Processing strategy execution

[0116] The system executes corresponding processing strategies according to the results of difference analysis. The specific implementation plan is as follows:

[0117] When differences are found between adjacent levels, the system directly adopts the higher predicted level as the final result. This strategy reflects the preventive principle of the system, which would rather warn of the actual failure than miss the potential failure risk.

[0118] When any model predicts a dangerous state (level 3), the system will immediately use the dangerous state as the final prediction result. This rapid response mechanism ensures that the system can issue an alarm in time when there is a risk of serious failure.

[0119] When there is a cross-level difference and both forecast results are not in a dangerous state, the system will start the secondary confirmation window mechanism for in-depth verification. This cautious approach ensures that judgments will not be made lightly when there are major differences in forecast results.

[0120] (4-3) Second confirmation window mechanism

[0121] When the system triggers the secondary confirmation window mechanism, the following verification process will be executed:

[0122] Preferably, the system first sets a 10-minute confirmation time window (T=10). Within this time window, preferably, the system continuously collects temperature, pressure and vibration parameter data at a sampling frequency of 2 times per minute (f=2). This moderate sampling frequency ensures the temporal continuity of the data and reserves sufficient computing time for data processing and model prediction.

[0123] The collected parameter data is processed with features, and continuous prediction is performed through the mechanism model and artificial intelligence model to form a prediction sequence. The system monitors the state level changes in the prediction sequence in real time and looks for continuous and consistent prediction results.

[0124] Preferably, when there are three consecutive (n=3) prediction state levels that are consistent in the prediction sequence, the system will use the consistent prediction result as the final fault prediction level. Preferably, within a 10-minute time window, the system can complete a maximum of 20 predictions (10 minutes × 2 times / minute), which provides a sufficient sample size for observing three consecutive consistent predictions.

[0125] Preferably, if no consistent prediction status levels are observed for three consecutive times at the end of the 10-minute confirmation window, the system will use the prediction result of the mechanism model (the first prediction status level) as the final prediction result. This conservative approach reflects the system's emphasis on the interpretability of the mechanism model.

[0126] Through the above prediction fusion mechanism, the present invention realizes the intelligent fusion of different prediction results, which not only ensures the timeliness of the prediction, but also ensures the reliability of the prediction results. This mechanism can effectively handle prediction differences and provide a verification method for prediction results through a secondary confirmation window mechanism, which significantly improves the overall performance of the fault prediction system.

[0127] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be within the protection scope of the present invention.

Claims

1. A method for predicting power generation equipment failure based on a mechanism model and AI dual verification, characterized in that: The following steps are involved: Obtain temperature parameters, pressure parameters and vibration parameters of power generation equipment; Performing dual-path parallel feature processing on the temperature parameter, the pressure parameter and the vibration parameter; The first feature processing path performs data cleaning on the temperature parameter, the pressure parameter and the vibration parameter to obtain mechanism characteristics, and the data cleaning includes removing abnormal values, processing missing values ​​and eliminating noise signals; The second feature processing path calculates the statistical features of the temperature parameter, the pressure parameter and the vibration parameter based on a preset time window, performs feature standardization processing to obtain a feature vector, wherein the feature standardization includes normalizing the statistical features and organizing the processed features into a vector form; the mechanism model and the artificial intelligence model are used to predict the mechanism features and the feature vector respectively; wherein the mechanism model prediction includes inputting the mechanism features into the mechanism model, performing calculations based on the mechanism equations of thermodynamics and vibration dynamics, and obtaining a first prediction state level; Wherein, the artificial intelligence model prediction includes inputting the feature vector into the artificial intelligence model, processing it through a feedforward neural network, and obtaining a second prediction state level; Performing a difference analysis on the first prediction state level and the second prediction state level, and executing a corresponding processing strategy according to a preset difference level determination rule; The first prediction state level and the second prediction state level are as follows: normal state is recorded as level 1, warning state is recorded as level 2, and dangerous state is recorded as level 3 from low to high; The difference level determination rule includes: when the level difference between the first prediction state level and the second prediction state level is 1, it is determined as an adjacent level difference; when the level difference between the first prediction state level and the second prediction state level is 2, it is determined as a cross-level difference; When the cross-level difference is detected, a secondary confirmation window mechanism is triggered for further verification, and a final fault prediction result is obtained based on the result of the difference analysis and the verification result of the secondary confirmation window mechanism.

2. According to claim 1, a method for predicting power generation equipment failure based on mechanism model and AI dual verification is characterized in that: The mechanism model includes: Temperature mechanism model, which calculates the temperature field deviation value based on the energy balance equation; Pressure mechanism model, calculates pressure deviation value based on oil film load equation; Vibration mechanism model, which calculates the vibration deviation value based on the vibration response equation; Wherein, the mechanism model includes a temperature preset threshold, a pressure preset threshold and a vibration preset threshold, and the temperature preset threshold, the pressure preset threshold and the vibration preset threshold all include a first gear value and a second gear value; Wherein, the mechanism model makes a state judgment by comparing the temperature field deviation value with the temperature preset threshold, the pressure deviation value with the pressure preset threshold, and the vibration deviation value with the vibration preset threshold. When the deviation value is less than the first gear value corresponding to the preset threshold, it is judged as the level 1; when the deviation value is greater than or equal to the first gear value corresponding to the preset threshold and less than the second gear value corresponding to the preset threshold, it is judged as the level 2; when the deviation value is greater than or equal to the second gear value corresponding to the preset threshold, it is judged as the level 3; After the mechanism model completes independent determination of the temperature state, the pressure state and the vibration state, the highest level is selected as the first predicted state level.

3. According to claim 1, a method for predicting power generation equipment failure based on mechanism model and AI dual verification is characterized in that: The statistical features include: The mean parameter, standard deviation parameter and peak parameter of the temperature parameter; The mean parameter, standard deviation parameter and peak parameter of the pressure parameter; The mean parameter, standard deviation parameter and peak parameter of the vibration parameter.

4. According to claim 1, a method for predicting power generation equipment failure based on mechanism model and AI dual verification is characterized in that: The artificial intelligence model is a feedforward neural network, comprising: an input layer having a plurality of input nodes, the number of the input nodes matching the dimension of the feature vector, for receiving and processing the feature vector; A hidden layer having a plurality of hidden nodes, wherein the hidden nodes perform a nonlinear transformation on the information from the input layer through the first activation function, wherein the first activation function adopts a ReLU function for extracting a nonlinear representation of a feature; An output layer having a plurality of output nodes, wherein the output nodes calculate and output the probability distribution of level 1, level 2 and level 3 through the second activation function, wherein the second activation function adopts a Softmax function for converting the output into probability values ​​of three levels, each level corresponds to a probability value between 0 and 1, and the sum of the three probability values ​​is 1; The specific method for determining the prediction state based on the maximum probability value is to select the level corresponding to the maximum value from the three output probability values ​​as the prediction result. If the maximum probability value corresponds to level 1, level 1 is output; if the maximum probability value corresponds to level 2, level 2 is output; if the maximum probability value corresponds to level 3, level 3 is output.

5. According to claim 1, a method for predicting power generation equipment failure based on mechanism model and AI dual verification is characterized in that: The fusion process includes: When there is an adjacent level difference between the first prediction state level and the second prediction state level, taking the higher level as the final prediction state level; When the first predicted state level or the second predicted state level is the dangerous state, taking the dangerous state as the final predicted state level; When there is the cross-level difference between the first predicted state level and the second predicted state level and both are not the dangerous state, the secondary confirmation window mechanism is activated.

6. A method for predicting power generation equipment failure based on mechanism model and AI dual verification according to claim 5, characterized in that: The specific steps of the secondary confirmation window mechanism are: A confirmation window for setting a preset duration, wherein the preset duration is T minutes; During the confirmation window, continuously collecting the temperature parameter, the pressure parameter and the vibration parameter at a preset sampling frequency of f times / minute; Performing feature processing on the collected parameters, respectively predicting through the mechanism model and the artificial intelligence model, and obtaining a prediction sequence within the confirmation window; Performing statistical analysis based on the prediction sequence, when the prediction state level is consistent for n consecutive times in the prediction sequence, using the consistent prediction result as the final prediction state level; If n consecutive consistent prediction status levels still do not appear at the end of the confirmation window, the first prediction status level is adopted as the final prediction status level.

7. A power generation equipment fault prediction system based on mechanism model and AI dual verification, characterized in that: include: Parameter collection unit, used to collect temperature parameters, pressure parameters and vibration parameters of power generation equipment in real time; A mechanism characteristic processing unit, used for performing data cleaning on the parameters to obtain characteristic data for mechanism analysis; An AI feature processing unit, used to calculate the statistical features of the parameters and perform standardization to obtain a feature vector for AI analysis; A mechanism prediction unit, including a temperature mechanism model, a pressure mechanism model and a vibration mechanism model, is used to analyze the mechanism characteristic data based on a physical equation and output a first prediction state level; An AI prediction unit, comprising a feedforward neural network having an input layer, a hidden layer, and an output layer, for analyzing the feature vector and outputting a second prediction state level; The prediction fusion unit is used to perform difference analysis on the first prediction state level and the second prediction state level, and trigger a secondary confirmation mechanism when a cross-level difference is detected, and finally output a fault prediction result.

Images