Equipment fault risk prediction method and system based on rule base and fault information

Abstract

The invention is suitable for the field of equipment fault prediction, and particularly provides an equipment fault risk prediction method and system based on a rule base and fault information, and the prediction method comprises the steps: carrying out the feature extraction and screening of the fault information through a rule in the rule base, and obtaining a feature variable related to a fault through a Pearson correlation coefficient analysis method; building a fault prediction model based on an LSTM network, and optimizing parameters of the LSTM network by using an improved whale algorithm fusing a nonlinear strategy and chaotic mapping; taking the identified feature vector as the input of a fault prediction model after parameter optimization, and obtaining the feature vector of an equipment fault risk prediction result; and inputting the feature vector of the prediction result into an MLP classifier, the MLP classifier being used for mapping the feature vector of the prediction result to a corresponding equipment fault risk level to obtain a level result of equipment fault risk prediction. According to the method, the prediction performance and interpretability of the model are remarkably improved, and support is provided for equipment maintenance and management.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment fault prediction and analysis, and specifically to a method and system for predicting equipment fault risks based on a rule library and fault information. Background Art

[0002] Traditional equipment fault prediction mainly relies on manual experience and technical accumulation. This method is not only inefficient but also prone to misjudgment due to human factors. With the increasing complexity of equipment and the diversification of operating environments, this method can no longer meet the requirements of modern equipment fault prediction.

[0003] In recent years, with the development of big data and machine learning technologies, data-driven fault prediction methods have gradually emerged. These methods analyze and mine a large amount of equipment operation data and use machine learning algorithms to establish fault prediction models. However, these methods usually require a large amount of historical fault data for training, and the accuracy and robustness of the models are sensitive to the quality and quantity of the data. Some traditional fault diagnosis methods use a rule-based approach to summarize a series of fault diagnosis rules based on expert experience. These rules usually have a certain degree of interpretability and operability, but they often fall short when dealing with complex and non-linear fault relationships.

[0004] Therefore, the current fault prediction technologies have the following technical problems to be solved: First, the problem of insufficient data and poor quality: In practical applications, it is often difficult to obtain sufficient high-quality historical fault data. Problems such as data missing, noise, and inconsistency will affect the performance of data-driven fault prediction models.

[0005] Second, the adaptability and generalization ability of the models: Different equipment types, operating environments, and fault modes may lead to significant differences in fault characteristics. Existing fault prediction models often have difficulty adapting to these changes, resulting in insufficient generalization ability.

[0006] Third, the real-time and accuracy of fault prediction: For some key equipment, it is necessary to predict fault risks in a timely and accurate manner in order to take preventive measures. However, existing fault prediction methods often have difficulty balancing real-time and accuracy. Summary of the Invention

[0007] The purpose of the embodiments of the present invention is to provide a method and system for predicting equipment fault risks based on a rule library and fault information, aiming to solve the technical problems proposed in the above background art.

[0008] To achieve the above purpose, the present invention provides the following technical solutions.

[0009] An embodiment of the present invention provides a method for predicting equipment fault risks based on a rule base and fault information, including the following steps: S1. Extract and screen the features of the fault information using the rules in the rule base, and obtain the feature variables related to the fault by using the analysis method of Pearson correlation coefficient; S2. Build a fault prediction model based on the LSTM network, and use an improved whale algorithm that combines a non-linear strategy and chaotic mapping to optimize the parameters of the LSTM network; S3. Use the identified feature vector as the input of the fault prediction model with optimized parameters to obtain the feature vector of the equipment fault risk prediction result; S4. Input the feature vector of the prediction result into the MLP classifier, and the MLP classifier is used to map the feature vector of the prediction result to the corresponding equipment fault risk level to obtain the level result of the equipment fault risk prediction.

[0010] Further, in step S3, when inputting the identified feature vector into the fault prediction model for prediction, an online learning mechanism is introduced to update the model parameters in real time, so that the model can adapt to the dynamic changes of the equipment operating state. The process of real-time updating of the model parameters is expressed as:

[0011] In the formula, represents the newly monitored equipment operation data; represents the learning rate; represents the gradient of the loss function with respect to the parameters; represents the parameters of the prediction model at time t; represents the operating state of the equipment monitored at time t; represents the prediction output of the model for the input data under the parameters ; The Adagrad algorithm is used to adaptively update and adjust the learning rate , and the process of updating the prediction model parameters based on the adaptively updated and adjusted learning rate is expressed as:

[0012]

[0013] In the formula, represents the cumulative gradient square, represents a constant used to prevent the denominator from being zero; represents the gradient of the loss function, represents the prediction output of the model for the input data under the parameters , Represents the time step of the input data; Represents the time step of the target value, which is used to evaluate the difference between the predicted result of the model and the true value; t represents the current time step.

[0014] Furthermore, in step S4, during the process of mapping the feature vector of the predicted result to the device failure risk level through the MLP classifier, the predicted result of the MLP classifier is explained by combining the SHAP value and LIME, where the SHAP value is expressed as:

[0015] In the formula, represents the contribution of feature i to the output of the prediction model; M represents the total number of features; N represents the set of all features; represents the contribution of the feature subset S to the output of the prediction model; For a predicted state x of the device, LIME is used to explain the predicted result, which is expressed as:

[0016] In the formula, represents the approximation of the prediction model f at state x, G represents the space of the simple model, L represents the loss function, represents the weight function of state x; represents the penalty term for model complexity.

[0017] Furthermore, in step S1, the steps of obtaining the feature variables related to the fault using the analysis method of Pearson correlation coefficient include: Calculate the Pearson correlation coefficient between each feature variable and the device fault state, which is expressed as follows:

[0018] In the formula, r represents the correlation coefficient, r ∈ [-1, 1]; when the value of r is 1, it represents a perfect positive correlation; when the value of r is -1, it represents a perfect negative correlation; when the value of r is 0, it represents no correlation; and represent the observed values of two variables, and represent the means of the corresponding variables; Variables with the absolute value of the correlation coefficient not less than the threshold are used as the required feature vectors.

[0019] Furthermore, in step S2, the fault prediction model built based on the LSTM network includes an LSTM layer, an attention mechanism layer, and a fully connected layer, where: In the LSTM layer, there are multiple LSTM units, each LSTM unit contains 128 neurons, and a Dropout layer is added between each LSTM unit; In the attention mechanism layer, the hidden state output by the LSTM unit at each time step is used as the input of the attention mechanism, and the attention weight at each time step is calculated. The calculation formula of the attention weight is expressed as:

[0020]

[0021] In the formula, represents the attention score at time step t; represents the attention score at time step s; s represents the index of the time step; n_step represents the total number of steps of the time series; v represents the weight vector of the attention mechanism; W represents the weight matrix; represents the hidden state at time step t; represents concatenating the hidden states of the current time step and the previous time step together; In the fully connected layer, a low-dimensional feature vector is output, and this feature vector is used as the input of the MLP classifier.

[0022] Furthermore, the step of outputting a low-dimensional feature vector in the fully connected layer includes: adding two fully connected layers after the LSTM layer. One fully connected layer is used for feature integration and non-linear transformation, expressed as: , in the formula, represents the weight matrix, represents the bias term; represents the activation function, represents the feature vector output by the LSTM layer; the other fully connected layer is used to map the output of the feature integration to a low-dimensional feature vector, expressed as: , in the formula, represents the weight matrix, represents the bias term; represents the output of the feature integration.

[0023] Furthermore, in step S2, the steps of optimizing the parameters of the LSTM network by using an improved whale algorithm that fuses non-linear strategy and chaotic mapping include: Using a random number generator to randomly select the Logistic map or the Tent map to generate the initial population, mapping the chaotic sequence to the solution space, and obtaining the initial position of the whale population; the position of each whale individual represents a set of parameters of the LSTM network, including the number of neurons, the learning rate, etc.; Apply the LSTM parameters corresponding to the whale individual positions to the LSTM network, and calculate the fitness of the network on the training set. Among them, the fitness is measured by the mean square error (MSE) metric; Update the whale positions; During the iterative process of the whale optimization algorithm, dynamically adjust the convergence factor according to the current iteration number to control the search behavior of the algorithm. Among them, the convergence factor is calculated by the following formula:

[0024] In the formula, α(t) represents the convergence factor of the t-th iteration, m and n are both constants, and T max represents the maximum number of iterations; Introduce an adaptive inertia weight into the whale position update formula to adjust the way of updating the whale positions. Among them, the adaptive inertia weight is calculated by the following formula:

[0025] In the formula, represents the inertia weight; represents the position of the worst whale in the current whale population; represents the position of the best whale in the current whale population; and respectively represent the parameter variables upper and lower bounds; represents the iteration number of the current population; and represent two constants; Update the positions of the whales through each iteration until the maximum number of iterations is reached or the convergence condition is satisfied. When the maximum number of iterations is reached or the convergence condition is satisfied, output the optimal LSTM network parameters.

[0026] Furthermore, in step S4, the MLP classifier includes an input layer, a hidden layer, and an output layer; among them, in the hidden layer, the ReLU activation function is used, expressed as: , where represents the weight matrix of the hidden layer, represents the bias term; X represents the input feature vector; in the output layer, , where represents the weight matrix of the output layer, represents the bias term; represents the output of the hidden layer.

[0027] Another embodiment of the present invention provides a device failure risk prediction system based on a rule base and fault information, including the following modules: A feature extraction module, which is used to extract and screen the fault information by using the rules in the rule base, and obtain the feature variables related to the fault by using the analysis method of Pearson correlation coefficient; A model optimization module, which is used to build a fault prediction model based on the LSTM network, and use an improved whale algorithm that combines a nonlinear strategy and a chaotic map to optimize the parameters of the LSTM network; A prediction module, which is used to take the identified feature vector as the input of the fault prediction model with optimized parameters, and obtain the feature vector of the device fault risk prediction result; A classification module, which is used to input the feature vector of the prediction result into the MLP classifier. The MLP classifier is used to map the feature vector of the prediction result to the corresponding device fault risk level, and obtain the level result of the device fault risk prediction.

[0028] Compared with the prior art, the beneficial effects of the device fault risk prediction method and system based on the rule base and fault information of the present invention are: First, the present invention uses the rules in the rule base for feature extraction, which can more targeted screen out the features related to the fault; using the Pearson correlation coefficient analysis method to quantitatively analyze the linear correlation between the features and the fault from a statistical perspective, and further accurately screen out the most valuable feature variables for fault prediction, avoiding the deviation caused by subjective judgment and empiricism, and improving the scientificity and accuracy of feature selection; Second, the present invention builds a fault prediction model based on the LSTM network, and uses an improved whale algorithm that combines a nonlinear strategy and a chaotic map to optimize the parameters of the LSTM network. Compared with the traditional parameter initialization and optimization methods, it can more effectively search for the global optimal or approximate global optimal parameter combination, and improve the prediction accuracy and generalization ability of the LSTM network; the initial population generated by the chaotic map has good diversity and ergodicity, which can cover all regions of the solution space, making the algorithm not easy to fall into the local optimal solution during the optimization process; the introduction of the nonlinear strategy further improves the global search ability and local development ability of the algorithm, enabling the model to better adapt to the fault prediction requirements under different devices and different working conditions; the improved whale algorithm speeds up the convergence speed of the algorithm and improves the optimization efficiency by dynamically adjusting the convergence factor and introducing the adaptive inertia weight and other mechanisms. On the premise of ensuring the optimization effect, it reduces the calculation time and resource consumption, enabling the fault prediction model to respond to the actual needs more quickly; Third, taking the identified feature vector as the input of the fault prediction model with optimized parameters can make full use of the optimized LSTM network to perform deep learning and feature extraction on the features, obtaining a more representative and discriminative feature vector; by inputting the feature vector into the fault prediction model with optimized parameters, the continuity and stability of the model among different times and different data batches can be ensured; the fluctuations in the prediction results caused by the frequent changes of the model parameters are avoided, improving the reliability and consistency of the fault prediction. Fourth, the present invention inputs the feature vector of the prediction result into the MLP classifier, which can further explore the potential information in the feature vector and accurately map the equipment fault risk to different risk levels. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0030] Figure 1 is the implementation flowchart of the equipment fault risk prediction method based on the rule base and fault information of the present invention; Figure 2 is a sub - flowchart of the equipment fault risk prediction method based on the rule base and fault information of the present invention; Figure 3 is the structural block diagram of the equipment fault risk prediction system based on the rule base and fault information of the present invention; Figure 4 is the structural block diagram of a computer device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] In order to make the objectives, technical solutions and advantages of the present invention clearer, the following further elaborates on the present invention in conjunction with the drawings and embodiments. 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.

[0032] The following elaborates on the specific implementation of the present invention in detail with reference to specific embodiments.

[0033] Please refer to Figure 1 , in the embodiment of the present invention, there is provided an equipment fault risk prediction method based on a rule base and fault information. The prediction method includes the following steps: S1. Use the rules in the rule base to perform feature extraction and screening on the fault information, and use the analysis method of Pearson correlation coefficient to obtain the feature variables related to the fault; Specifically, in step S1 disclosed in the present invention, rules in the rule base can be used to extract features from the fault information to identify feature variables related to the fault. Exemplarily, the rule base may contain rules such as "If the temperature of the device exceeds a certain threshold, a fault may occur" and "If the pressure value of the device exceeds a certain threshold, a fault may occur". According to this rule, the device temperature can be extracted as a feature variable. Further, after extracting the feature variables, the Pearson correlation coefficient between each feature variable and the device fault state is calculated to measure the linear correlation between the feature variable and the device fault state, as shown below:

[0034] In the formula, r represents the correlation coefficient, ; when the value of r is 1, it represents a perfect positive correlation; when the value of r is -1, it represents a perfect negative correlation; when the value of r is 0, it represents no correlation; and represent the observed values of two variables, and represent the means of the corresponding variables; In one implementation manner of the present invention, a threshold of the correlation coefficient (for example, 0.3) is set, and variables whose absolute value of the correlation coefficient is not less than the threshold are used as the required feature vectors. In the embodiments of the present invention, the screened feature variables such as temperature and pressure can be used as the input of the LSTM initial model, and the fault features are used as the output to train the fault prediction model.

[0035] The present invention uses the rules in the rule base for feature extraction, which can more pertinently screen out the features related to the fault; uses the Pearson correlation coefficient analysis method to quantitatively analyze the linear correlation between the features and the fault from a statistical perspective, further accurately screen out the feature variables most valuable for fault prediction, avoid the deviation caused by subjective judgment and empiricism, and improve the scientificity and accuracy of feature selection.

[0036] Please continue to refer to Figure 1 , the device fault risk prediction method based on the rule base and fault information provided by the embodiments of the present invention further includes the following steps: S2. Build a fault prediction model based on the LSTM network, and use an improved whale algorithm that combines a nonlinear strategy and a chaotic map to optimize the parameters of the LSTM network; S3. Use the identified feature vectors as the input of the fault prediction model with optimized parameters to obtain the feature vectors of the device fault risk prediction results; S4. Input the feature vector of the prediction result into the MLP classifier, which is used to map the feature vector of the prediction result to the corresponding device failure risk level, and obtain the level result of the device failure risk prediction.

[0037] Specifically, in step S2 of the present invention, during the construction of the LSTM network, it includes the design of the model structure. In the model structure, the constructed LSTM model includes an input layer, an LSTM layer, a Dropout layer, and a fully connected layer. Among them, in the input layer, the preprocessed feature vector is input; in the LSTM layer, a multi-layer LSTM structure is used, with each layer containing 128 neurons, which is used to capture the long-term dependencies in the time series of the device operation data; in the Dropout layer, a Dropout layer is added between each LSTM layer to prevent overfitting; in the fully connected layer, one or more fully connected layers are added after the LSTM layer, which is used for feature integration and non-linear transformation. Specifically, in the embodiment of the present invention, the fault prediction model built based on the LSTM network includes an LSTM layer, an attention mechanism layer, and a fully connected layer, where: In the LSTM layer, it may include multiple LSTM units, and each LSTM unit contains 128 neurons. Further, a Dropout layer is added between each LSTM unit. In the attention mechanism layer, the hidden state output by the LSTM unit at each time step is used as the input of the attention mechanism to calculate the attention weight at each time step. The calculation formula of the attention weight is expressed as:

[0038]

[0039] In the formula, represents the attention score at time step t; represents the attention score at time step s; s represents the index of the time step; n_step represents the total number of steps in the time series; v represents the weight vector of the attention mechanism; W represents the weight matrix; h t represents the hidden state at time step t; represents splicing the hidden states of the current time step and the previous time step together; Further, in the fully connected layer, a low-dimensional feature vector is output, and this feature vector is used as the input of the MLP classifier.

[0040] Further, the step of outputting a low-dimensional feature vector in the fully connected layer includes: adding two fully connected layers after the LSTM layer. One fully connected layer is used for feature integration and non-linear transformation, which is expressed as: , in the formula, represents the weight matrix, represents the bias term; represents the activation function, represents the feature vector output by the LSTM layer; Another fully connected layer is used to map the output of the feature integration to a low-dimensional feature vector, denoted as: , where, represents the weight matrix, represents the bias term; represents the output of the feature integration.

[0041] Furthermore, as Figure 2 shown, in step S2, the steps of optimizing the parameters of the LSTM network by using an improved whale algorithm that combines a fusion non-linear strategy and a chaotic map include the following steps: S21. Use a random number generator to randomly select a Logistic map or a Tent map to generate an initial population, map the chaotic sequence into the solution space, and obtain the initial position of the whale population; The position of each whale individual represents a set of parameters of the LSTM network, including the number of neurons, the learning rate, etc.; In the embodiment of the present invention, the population is perturbed by using a chaotic map function, which can increase the diversity of the population and avoid falling into a local optimal solution; S22. Apply the LSTM parameters corresponding to the whale individual position to the LSTM network, and calculate the fitness of the network on the training set, where the fitness is measured by the mean square error MSE index; Among them, the expression of the fitness function is:

[0042] where, n represents the number of samples; represents the true value; represents the predicted value; S23. Update the whale position; S24. During the iterative process of the whale optimization algorithm, dynamically adjust the convergence factor according to the current iteration number to control the search behavior of the algorithm, where the convergence factor is calculated by the following formula:

[0043] where, α(t) represents the convergence factor of the t-th iteration, m and n are both constants, and T max represents the maximum number of iterations; S25. Introduce an adaptive inertia weight into the whale position update formula to adjust the way of updating the whale position, where the adaptive inertia weight is calculated by the following formula:

[0044] where, represents the inertia weight; represents the position of the worst whale in the current whale population; represents the position of the best whale in the current whale population; and respectively represent the upper and lower bounds of the parameter variable ; represents the number of iterations of the current population; and represent two constants; In the embodiments of the present invention, an adaptive inertia weight is introduced into the whale position update formula to accelerate the convergence speed of the algorithm and improve the solution accuracy; S26. Update the position of the whale by each iteration until the maximum number of iterations is reached or the convergence condition is satisfied. When the maximum number of iterations is reached or the convergence condition is satisfied, output the optimal LSTM network parameters.

[0045] Therefore, the present invention constructs a fault prediction model based on the LSTM network and optimizes the parameters of the LSTM network by using an improved whale algorithm that combines a nonlinear strategy and a chaotic map, which can more effectively search for the global optimal or approximate global optimal parameter combination, and improve the prediction accuracy and generalization ability of the LSTM network; The initial population generated based on the chaotic map in the embodiments of the present invention has good diversity and ergodicity, can cover all regions of the solution space, and makes the algorithm not easily fall into the local optimal solution during the optimization process; the introduction of the nonlinear strategy further improves the global search ability and local development ability of the algorithm, enabling the model to better adapt to the fault prediction requirements under different devices and different working conditions; the improved whale algorithm accelerates the convergence speed of the algorithm and improves the optimization efficiency by dynamically adjusting the convergence factor and introducing an adaptive inertia weight and other mechanisms. On the premise of ensuring the optimization effect, the calculation time and resource consumption are reduced, enabling the fault prediction model to respond to actual needs more quickly.

[0046] Further, in step S3, when inputting the identified feature vector into the fault prediction model for prediction, an online learning mechanism is introduced to update the model parameters in real time, so that the model can adapt to the dynamic changes of the device operation state. The process of real-time updating of the model parameters is expressed as:

[0047] In the formula, represents the newly monitored device operation data; represents the learning rate; represents the gradient of the loss function with respect to the parameters; represents the parameters of the prediction model at time t; represents the operation state of the monitored device at time t; denotes the predicted output of the model for the input data under the parameter ; Furthermore, the present invention uses the Adagrad algorithm to adaptively update and adjust the learning rate , and the prediction model parameter update process based on the adaptive update and adjustment of the learning rate is expressed as:

[0048]

[0049] In the formula, denotes the cumulative gradient square, denotes a constant used to prevent the denominator from being zero; denotes the gradient of the loss function, denotes the predicted output of the model for the input data under the parameter ; denotes the input data at time step ; denotes the target value at time step , which is used to evaluate the difference between the predicted result of the model and the true value; t represents the current time step.

[0050] Furthermore, in step S4, in the process of mapping the feature vector of the prediction result to the device failure risk level through the MLP classifier, the prediction result of the MLP classifier is explained by combining the SHAP value and LIME, where the SHAP value is expressed as:

[0051] In the formula, denotes the contribution of feature i to the output of the prediction model; M denotes the total number of features; N denotes the set of all features; denotes the contribution of the feature subset S to the output of the prediction model; For a predicted state x of the device, LIME is used to explain the prediction result, which is expressed as:

[0052] In the formula, denotes the approximation of the prediction model f at state x, G denotes the space of simple models, L denotes the loss function, denotes the weight function of state x; denotes the penalty term for model complexity.

[0053] In the embodiments of the present invention, an interpretable method using SHAP values can explain the model from a global perspective. By calculating the Shapley values of features, the contribution of each feature to the model prediction is measured; while LIME used can explain the model from a local perspective. By approximating the complex model with a simple model in a local area, a single prediction result is explained. The present invention can more comprehensively explain the prediction model and improve the credibility of the prediction results of the prediction model.

[0054] Further, in step S4, the MLP classifier includes an input layer, a hidden layer, and an output layer; wherein, in the hidden layer, the ReLU activation function is used, expressed as: , where represents the weight matrix of the hidden layer, represents the bias term; X represents the input feature vector; in the output layer, , where represents the weight matrix of the output layer, represents the bias term; represents the output of the hidden layer.

[0055] The present invention takes the identified feature vector as the input of the fault prediction model after parameter optimization, can make full use of the optimized LSTM network to perform deep learning and feature extraction on the features, and obtain a more representative and discriminative feature vector; by inputting the feature vector into the fault prediction model after parameter optimization, the continuity and stability of the model between different times and different data batches can be ensured; the fluctuation of the prediction results caused by the frequent change of the model parameters is avoided, and the reliability and consistency of the fault prediction are improved; The present invention inputs the feature vector of the prediction result into the MLP classifier, can further mine the potential information in the feature vector, and accurately map the equipment fault risk to different risk levels.

[0056] In summary, the present invention realizes efficient, accurate, and interpretable equipment fault risk prediction by combining a rule base, Pearson correlation coefficient analysis, an optimized LSTM network, and an MLP classifier. Compared with the prior art, it significantly improves the accuracy of feature selection, the prediction performance, and the interpretability of the model, can better adapt to the complex and changeable equipment operating conditions, and provides strong support for equipment maintenance and management.

[0057] As Figure 3 shown, another embodiment of the present invention provides an equipment fault risk prediction system based on a rule base and fault information, including the following modules: A feature extraction module 101, configured to extract and screen fault information using the rules in the rule base, and obtain feature variables related to faults using the analysis method of Pearson correlation coefficient; The model optimization module 102 is used to build a fault prediction model based on the LSTM network, and use an improved whale algorithm that combines a non-linear strategy and chaotic mapping to optimize the parameters of the LSTM network; The prediction module 103 is used to take the identified feature vector as the input of the fault prediction model with optimized parameters, and obtain the feature vector of the device fault risk prediction result; The classification module 104 is used to input the feature vector of the prediction result into the MLP classifier. The MLP classifier is used to map the feature vector of the prediction result to the corresponding device fault risk level to obtain the level result of the device fault risk prediction.

[0058] As Figure 4 shown, in one embodiment, a computer device is provided. The computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the device fault risk prediction method based on the rule base and fault information provided in the above embodiment.

[0059] In one embodiment, a computer-readable storage medium is provided. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the processor is caused to execute the device fault risk prediction method based on the rule base and fault information provided in the above embodiment.

[0060] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages does not necessarily have to be sequential, but can be executed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.

[0061] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0062] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0063] The above embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention should be subject to the appended claims.

[0064] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the device failure risk based on a rule base and fault information, characterized in that, It includes the following steps: S1. Extract and screen the fault information using the rules in the rule library, and use the analysis method of Pearson correlation coefficient to obtain the characteristic variables related to the fault; S2. Build a fault prediction model based on the LSTM network, and use an improved whale algorithm that combines a nonlinear strategy and chaotic mapping to optimize the parameters of the LSTM network; S3. Take the identified feature vector as the input of the fault prediction model with optimized parameters to obtain the feature vector of the equipment fault risk prediction result; S4. Input the feature vector of the prediction result into the MLP classifier, and the MLP classifier is used to map the feature vector of the prediction result to the corresponding equipment fault risk level to obtain the level result of the equipment fault risk prediction.

2. The device fault risk prediction method based on a rule base and fault information according to claim 1, wherein In step S3, when inputting the identified feature vector into the fault prediction model for prediction, an online learning mechanism is introduced to update the model parameters in real time, so that the model can adapt to the dynamic changes of the equipment operation state. The process of real-time updating of the model parameters is expressed as: ， In the formula, represents the newly monitored device operation data; represents the learning rate; represents the gradient of the loss function with respect to the parameters; represents the parameters of the prediction model at time t; represents the operating state of the monitored device at time t; represents the model with parameters for the input data predicted output; Adagrad algorithm is used to adaptively update and adjust the learning rate , and the update process of the prediction model parameters based on the adaptive update and adjustment of the learning rate is expressed as: ， ， In the formula, represents the cumulative squared gradient, represents a constant used to prevent the denominator from being zero; represents the gradient of the loss function, represents the prediction output of the model for the input data under the parameter ; represents the time step of the input data; represents the time step of the target value, which is used to evaluate the difference between the prediction result of the model and the true value; t represents the current time step.

3. The device fault risk prediction method based on a rule base and fault information according to claim 2, wherein In step S4, in the process of mapping the feature vector of the prediction result to the equipment fault risk level through the MLP classifier, the SHAP value and LIME are combined to explain the prediction result of the MLP classifier, where the SHAP value is expressed as: ， In the formula, represents the contribution of feature i to the output of the prediction model; M represents the total number of features; N represents the set of all features; represents the contribution of the feature subset S to the output of the prediction model; For a predicted state x of the equipment, LIME is used to explain the prediction result, which is expressed as: ， In the formula, represents the approximation of the prediction model f at state x, G represents the space of the simple model, and L represents the loss function. represents the weight function of state x; represents the penalty term for model complexity.

4. The method for predicting equipment failure risk based on a rule base and fault information according to claim 3, characterized in that In step S1, the steps of using the analysis method of Pearson correlation coefficient to obtain the characteristic variables related to the fault include: Calculate the Pearson correlation coefficient between each characteristic variable and the equipment fault state, which is expressed as follows: ， Wherein, r represents the correlation coefficient, r ∈ [-1, 1]; when the value of r is 1, it indicates a perfect positive correlation; when the value of r is -1, it indicates a perfect negative correlation; when the value of r is 0, it indicates no correlation; and represent the observed values of two variables, and represent the mean values of the corresponding variables; Take the variables whose absolute value of the correlation coefficient is not less than the threshold as the required feature vectors.

5. The method for predicting equipment failure risk based on a rule library and fault information according to claim 4, characterized in that In step S2, the fault prediction model built based on the LSTM network includes an LSTM layer, an attention mechanism layer, and a fully connected layer, where: In the LSTM layer, there are multiple LSTM cells, each LSTM cell contains 128 neurons, and a Dropout layer is added between each LSTM cell; In the attention mechanism layer, take the hidden state output by the LSTM cell at each time step as the input of the attention mechanism, and calculate the attention weight at each time step. The calculation formula of the attention weight is expressed as: ， ， In the formula, represents the attention score at time step t; represents the attention score at time step s; s represents the index of the time step; n_step represents the total number of steps in the time series; v represents the weight vector of the attention mechanism; W represents the weight matrix; represents the hidden state at time step t; represents concatenating the hidden states of the current time step and the previous time step; In the fully connected layer, output a low-dimensional feature vector, and the low-dimensional feature vector is used as the input of the MLP classifier.

6. The method for predicting equipment failure risk based on a rule base and failure information according to claim 5, wherein In the fully connected layer, the steps to output a low-dimensional feature vector include: adding two fully connected layers after the LSTM layer. One fully connected layer is used for feature integration and non-linear transformation, expressed as: , where represents the weight matrix, represents the bias term; represents the activation function, represents the feature vector output by the LSTM layer; The other fully connected layer is used to map the output of feature integration to a low-dimensional feature vector, expressed as: , where represents the weight matrix, represents the bias term; represents the output of feature integration.

7. The method for predicting equipment failure risk based on a rule base and fault information according to claim 6, characterized in that In step S2, the steps of using an improved whale algorithm that combines a nonlinear strategy and chaotic mapping to optimize the parameters of the LSTM network include: Use a random number generator to randomly select a Logistic map or a Tent map to generate an initial population, map the chaotic sequence to the solution space to obtain the initial position of the whale population; the position of each whale individual represents a set of parameters of the LSTM network; Apply the LSTM parameters corresponding to the whale individual position to the LSTM network and calculate the fitness of the network on the training set; Update the whale position; During the iteration process of the whale optimization algorithm, the convergence factor is dynamically adjusted according to the current iteration number to control the search behavior of the algorithm. The convergence factor is calculated by the following formula: ， Where, α(t) represents the convergence factor of the t-th iteration, both m and n represent constants, and T max represents the maximum number of iterations; An adaptive inertia weight is introduced into the whale position update formula to adjust the way of updating the whale position. The adaptive inertia weight is calculated by the following formula: ， Where ω represents the inertia weight; represents the position of the worst whale in the current whale population; represents the position of the best whale in the current whale population; and respectively represent the upper and lower bounds of the parameter variable ; represents the number of iterations of the current population; and represent two constants; The position of the whale is updated in each iteration until the maximum iteration number is reached or the convergence condition is satisfied. When the maximum iteration number is reached or the convergence condition is satisfied, the optimal LSTM network parameters are output.

8. The method for predicting equipment failure risk based on a rule base and fault information according to claim 7, wherein In step S4, the MLP classifier includes an input layer, a hidden layer, and an output layer; among them, in the hidden layer, the ReLU activation function is used, expressed as: , where represents the weight matrix of the hidden layer, represents the bias term; X represents the input feature vector; in the output layer, , where represents the weight matrix of the output layer, represents the bias term; represents the output of the hidden layer.

9. A prediction system for implementing the device fault risk prediction method based on a rule base and fault information according to any one of claims 1 to 8, characterized in that The prediction system includes the following modules: A feature extraction module, which is used to extract and screen fault information features using the rules in the rule base, and uses the Pearson correlation coefficient analysis method to obtain feature variables related to faults; A model optimization module, which is used to build a fault prediction model based on the LSTM network and optimize the parameters of the LSTM network using an improved whale algorithm that combines a nonlinear strategy and chaotic mapping; A prediction module, which is used to input the identified feature vector as the input of the fault prediction model with optimized parameters to obtain the feature vector of the device fault risk prediction result; A classification module, which is used to input the feature vector of the prediction result into the MLP classifier. The MLP classifier is used to map the feature vector of the prediction result to the corresponding device fault risk level to obtain the level result of the device fault risk prediction.

Images