A comprehensive energy fault accurate and rapid diagnosis method and system
By using a CNN-LSTM neural network model and local mean decomposition technology in integrated energy systems, combined with the dung beetle optimization algorithm to optimize hyperparameters and feature selection, the problems of data redundancy and insufficient hyperparameter optimization in fault diagnosis of integrated energy systems are solved, and rapid and accurate fault diagnosis is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 国网山东综合能源服务有限公司
- Filing Date
- 2023-06-29
- Publication Date
- 2026-07-24
AI Technical Summary
Fault diagnosis of integrated energy systems faces challenges such as difficulty in data acquisition, excessive redundancy, and insufficient hyperparameter optimization, resulting in low diagnostic efficiency and low accuracy.
We employ a CNN-LSTM neural network model, combined with local mean decomposition and dung beetle optimization algorithm, to optimize hyperparameters and feature selection dimension. By simulating fault dataset expansion and noise reduction reconstruction, we construct a dual objective function to improve diagnostic accuracy and reduce training time.
It enables rapid and accurate diagnosis of faults in integrated energy systems, improves the accuracy of diagnostic models, reduces training time, and enhances the applicability and efficiency of the system.
Smart Images
Figure CN116776205B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of integrated energy system fault diagnosis technology, and in particular to a method and system for accurate and rapid diagnosis of integrated energy system faults. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] An integrated energy system connects the production, transmission, consumption, and storage of various energy sources, including electricity, cooling, heating, gas, and hydrogen, achieving organic coordination and optimization of these energy sources. Compared to single-energy systems, integrated energy systems operate under more complex conditions; a failure in one unit can even affect the normal operation of the entire system, causing systemic damage. Therefore, fault diagnosis of equipment in integrated energy systems is crucial for ensuring their safe and reliable operation.
[0004] Integrated energy systems are characterized by limited fault data and a wide variety of fault types, making data acquisition more difficult compared to single-energy systems. The efficiency of subsequent diagnosis hinges on obtaining sufficient integrated energy system fault data. Furthermore, the acquired integrated energy system fault data contains significant redundancy. Extracting useful information and selecting appropriate feature dimensions to ensure both diagnostic model accuracy and reduced training time is crucial. In recent years, with the rise of artificial intelligence and big data, deep learning has experienced rapid development in fault diagnosis. However, research on hyperparameter optimization for fault diagnosis models still has many shortcomings. Manually adjusting hyperparameters significantly impacts both diagnostic accuracy and training time. Summary of the Invention
[0005] To address the aforementioned issues, this invention proposes a comprehensive method and system for accurate and rapid diagnosis of energy faults. A CNN-LSTM neural network model is established as the fault diagnosis model, and the model's hyperparameters are initialized. The dual objectives are maximizing model accuracy and minimizing training time. The model's hyperparameters and feature selection dimension are optimized using the dung beetle optimization algorithm to obtain the Pareto solution set for the dual objectives. Appropriate hyperparameters are selected by limiting accuracy and training time.
[0006] In some implementations, the following technical solutions are adopted:
[0007] A comprehensive method for accurate and rapid diagnosis of energy faults includes:
[0008] Obtain operational data of the integrated energy system and use the trained fault diagnosis model to obtain fault classification results;
[0009] The fault diagnosis model is a CNN-LSTM model, and the training process for the fault diagnosis model is as follows:
[0010] By simulating faults in the integrated energy system, a fault dataset of each device in the integrated energy system is obtained; the fault dataset is then expanded to obtain a new fault data sample set.
[0011] The local mean decomposition method is used to denoise and reconstruct the new fault data sample set. The reconstructed fault data sample set is then input into the CNN-LSTM model to initialize the hyperparameters of the CNN-LSTM model.
[0012] We construct a dual objective function that maximizes model accuracy and minimizes training time. We then optimize the hyperparameters of the CNN-LSTM model using the dung beetle optimization algorithm to obtain a well-trained CNN-LSTM model.
[0013] In other embodiments, the following technical solutions are adopted:
[0014] A comprehensive energy fault accurate and rapid diagnosis system includes:
[0015] The data acquisition module is used to acquire operational data of the integrated energy system;
[0016] The fault classification module is used to obtain fault classification results using the trained fault diagnosis model;
[0017] The fault diagnosis model is a CNN-LSTM model, and the training process for the fault diagnosis model is as follows:
[0018] By simulating faults in the integrated energy system, a fault dataset of each device in the integrated energy system is obtained; the fault dataset is then expanded to obtain a new fault data sample set.
[0019] The local mean decomposition method is used to denoise and reconstruct the new fault data sample set. The reconstructed fault data sample set is then input into the CNN-LSTM model to initialize the hyperparameters of the CNN-LSTM model.
[0020] We construct a dual objective function that maximizes model accuracy and minimizes training time. We then optimize the hyperparameters of the CNN-LSTM model using the dung beetle optimization algorithm to obtain a well-trained CNN-LSTM model.
[0021] In other embodiments, the following technical solutions are adopted:
[0022] A terminal device includes a processor and a memory, the processor being used to implement instructions; the memory being used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide the aforementioned comprehensive energy fault accurate and rapid diagnosis method.
[0023] Compared with the prior art, the beneficial effects of the present invention are:
[0024] (1) This invention considers using local mean decomposition technology to optimize the initial dataset of integrated energy system faults, reduce dataset noise, and uses the whale algorithm to automatically optimize the main parameters of the local mean decomposition mechanism to enhance system applicability.
[0025] (2) This invention constructs a fault diagnosis model based on a CNN-LSTM model, using model diagnosis accuracy and training time as dual objective construction functions to determine hyperparameters and the dimension of fault feature vectors. An improved dung beetle optimization algorithm is used to optimize the hyperparameters, improving diagnosis accuracy, reducing training time, and enabling faster and more accurate fault diagnosis of integrated energy systems.
[0026] Other features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0027] Figure 1 This is a schematic diagram of the fault diagnosis model training process in an embodiment of the present invention;
[0028] Figure 2 This is a schematic diagram illustrating the noise reduction and reconstruction process using the local mean decomposition method in an embodiment of the present invention.
[0029] Figure 3 This is a schematic diagram illustrating the process of optimizing the hyperparameters of a CNN-LSTM model using the dung beetle optimization algorithm in an embodiment of the present invention. Detailed Implementation
[0030] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0031] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0032] Example 1
[0033] In one or more embodiments, a method for accurate and rapid diagnosis of comprehensive energy faults is disclosed, specifically including the following process:
[0034] Obtain operational data of the integrated energy system and use the trained fault diagnosis model to obtain fault classification results;
[0035] The fault diagnosis model is a CNN-LSTM model, combined with... Figure 1 The specific training process for the fault diagnosis model is as follows:
[0036] (1) By simulating the faults of the integrated energy system, the fault dataset of each device in the integrated energy system is obtained.
[0037] In this embodiment, a model of the integrated energy system of the park is built in MATLAB / SIMULINK to simulate common integrated energy system failures. The failure dataset of each device in the integrated energy system is collected through real-time simulation using RT-LAB.
[0038] Specifically, the integrated energy system model of the park includes a thermodynamic system module and an electrical system module.
[0039] The equipment in the thermodynamic system module mainly includes: electric refrigeration units, lithium bromide absorption chiller units, heat pumps, waste heat boilers, steam-gas combined cycle power plants (CHP units), and waste heat boilers.
[0040] The main equipment in the power system module includes: wind turbines, photovoltaic units, battery banks, hydrogen fuel cells, electrolyzers, diesel engines, and power distribution networks.
[0041] The collected data for each device in the integrated energy system includes: cooling power, power consumption, COP, cooling capacity, chilled water supply temperature, and cooling water return temperature of the electric chiller unit; power generation, heating power, thermal efficiency, heating capacity, flue gas temperature, hot water supply temperature, output three-phase voltage, and three-phase current of the CHP unit; output three-phase voltage and three-phase current of the photovoltaic unit; output three-phase voltage and three-phase current of the wind turbine unit; output three-phase voltage and three-phase current of the battery pack; and gas temperature, gas inlet pressure, and current of the hydrogen fuel cell.
[0042] The simulated integrated energy system fault types mainly include: chiller evaporator overheating fault, chiller evaporator low pressure fault, chiller expansion valve fouling fault, chiller condenser overheating fault, chiller condenser fouling fault, CHP compressor fouling fault, CHP compressor pressure ratio reduction fault, CHP circulating pump low pressure fault, CHP turbine generator single-phase short circuit fault, CHP turbine generator multi-phase short circuit fault, fan multi-phase short circuit fault, photovoltaic inverter multi-phase short circuit fault, battery multi-phase short circuit fault, hydrogen fuel cell membrane electrode fault, and temperature control valve fault.
[0043] (2) Expand the fault dataset to obtain a new fault data sample set.
[0044] In this embodiment, based on the dataset obtained from simulation, a variational autoencoder is used to perform data augmentation to obtain an expanded sample set. The expanded sample set is then merged with the sample set obtained from simulation to form a new fault data sample set.
[0045] (3) The local mean decomposition method is used to denoise and reconstruct the new fault data sample set.
[0046] In this embodiment, the Local Mean Decomposition (LTD) method is selected to reduce the noise in the fault data sample set and to reconstruct the fault dataset. The Whale Algorithm is used to automatically optimize the main parameters of the LTD method, including the number of PF components k, to remove the influence of noise on feature extraction.
[0047] Specifically, in combination Figure 2 The local mean decomposition method mainly includes the following steps:
[0048] Step 301: Initialize the parameters of the whale algorithm, including population size, number of iterations, number of PF components k, and correlation coefficient.
[0049] Step 302: Use the local mean decomposition method to process the signal using the k PF components determined as needed.
[0050] Step 303: Calculate the envelope signal value based on the corresponding parameters of each individual. With minimizing the energy of the envelope signal as the objective function of the algorithm, iteratively optimize the local mean decomposition, select the optimal number k for decomposition, and obtain the decomposed result.
[0051] Step 304: Decompose k PF components and a margin from the original signal one by one, and then reconstruct the signal through the autocorrelation principle to obtain the denoised signal.
[0052] In this embodiment, the individual in the whale algorithm is the number k of the Power Factor (PF) components in the Local Means Decomposition (LTD) method, and the population is the set of values for each individual. The number k of PF components is an important parameter of the LTD method, which can decompose the collected comprehensive energy fault signal into several k PF components. The number of PF components affects the final noise reduction effect. Therefore, this embodiment uses the whale algorithm to optimize the number k of PF components in the LTD method, selecting the number k of PF components with the best noise reduction effect.
[0053] (4) Input the reconstructed fault data sample set into the CNN-LSTM model and initialize the hyperparameters of the CNN-LSTM model.
[0054] In this embodiment, the CNN-LSTM model is a neural network model for fault diagnosis of integrated energy systems. CNN stands for Convolutional Neural Network, which mainly includes convolutional layers, pooling layers, and fully connected layers; LSTM stands for Long Short-Term Memory Network, which mainly includes input gates, memory units, and output gates.
[0055] The CNN-LSTM network embeds LSTM between the last pooling layer and the fully connected layer of the CNN. The final fully connected layer connects to the output layer to output the fault diagnosis results.
[0056] Accuracy is determined by the hyperparameters of the neural network, and the kernel size determines the feature dimension at the time of final diagnosis, which in turn determines the fault diagnosis time. In this embodiment, the solution set for accuracy should be selected from 95% and above, and solutions with an accuracy of less than 95% should be discarded.
[0057] The hyperparameters of the optimized CNN-LSTM model mainly include the learning rate η, the number of hidden layers l, the number of neurons in each layer j, the regularization parameter λ, the optimizer type, the activation function type, and the kernel size of each layer. These hyperparameters constitute the hyperparameter set φ.
[0058] To optimize the fault feature dimension, the first step is to construct a fault sample set from the denoised input data and then input it into the network for feature extraction. The construction of fault feature samples for integrated energy system equipment includes the following steps:
[0059] Step 401: Perform information fusion on the denoised data to obtain a multi-source fault data sample set S, including chiller units, CHP units, photovoltaic power generation systems, wind turbine power generation systems, and battery energy storage systems. denoise .
[0060] Step 402: Transfer the multi-source fault data sample set S denoiseInputting into a CNN-LSTM, the signal is processed sequentially through convolutional layers and pooling layers for convolutional computation and signal compression. Each time, the corresponding fault feature vector is output according to the required optimized convolutional kernel size.
[0061] Step 403: Construct a feature vector sample set S based on the feature vectors extracted by CNN-LSTM. feature This serves as the input for the final fault diagnosis and classification.
[0062] (5) Construct a dual objective function with the highest model accuracy and shortest training time, and optimize the hyperparameters of the CNN-LSTM model using the dung beetle optimization algorithm to finally obtain the trained CNN-LSTM model.
[0063] In this embodiment, a dual objective function is constructed with the goal of maximizing model accuracy and minimizing training time, specifically:
[0064]
[0065] In the formula, n correct The number of correctly diagnosed samples is n, the total number of samples is n, and the model training time is t.
[0066] This embodiment optimizes the hyperparameters and feature selection dimension of the CNN-LSTM model using an improved dung beetle optimization algorithm; to comprehensively enhance population diversity, a Logistic chaotic mapping is used to optimize the initial population of the algorithm, and its mathematical model is as follows:
[0067] y n+1 =μy n (1-y n )
[0068] In the formula, y n ∈(0,1), μ∈[0,4], when μ=4, the chaotic mapping system is in a completely chaotic state, and the mapping distribution is most uniform at this moment.
[0069] To improve the convergence speed of the algorithm, the dung beetle optimization algorithm in this embodiment has better global exploitation capabilities in the early stages of iteration and good local search capabilities in the later stages of iteration. An adaptive t-distribution is used to optimize the dung beetle population update position, and its expression is:
[0070] X i =X i +X i ·t(iter)
[0071] In the formula, X i ' is the position updated after adaptive t-distribution perturbation, X i This is the position of normal update, and t(iter) is a t-distribution with the number of iterations as the degree of freedom parameter.
[0072] Applying the adaptive t-distribution mutation operator to all individuals in each iteration of the algorithm increases computation time and hinders the full utilization of the algorithm's inherent advantages. Therefore, the algorithm uses a dynamic selection probability p to adjust the application of the adaptive t-distribution mutation operator, expressed as:
[0073] p = w1 - w2 × (T) max -iter) / T max
[0074] In the formula, T max w1 represents the maximum number of iterations, w2 represents the current number of iterations, w1 represents the upper limit of the dynamic selection probability, and w2 represents the lower limit of the dynamic selection probability.
[0075] The dynamic selection probability p is a probability, or rather, the frequency of perturbing the optimal solution using the adaptive t-distribution. This allows the algorithm to use the adaptive t-distribution mutation operator to perturb the position more frequently in the early stages of iteration, thus improving the tendency of the original algorithm to converge to the optimal solution in the early stages of iteration. At the same time, in the later stages of iteration, the algorithm's good local exploitation ability is fully utilized, and the t-distribution mutation with a smaller probability is used as a supplement to improve the convergence speed of the algorithm.
[0076] In this embodiment, the dung beetle optimization algorithm is used to optimize the hyperparameters of the CNN-LSTM model. Each individual represents the hyperparameter value of the CNN-LSTM model to be optimized, and the population is a set of solutions for each individual. The surrogate refers to an individual, the dung beetle position represents the value of each individual, the value of each individual is the hyperparameter to be optimized, the fitness of the individual (surrogate) is the model's accuracy and training time, and the optimal solution is the individual when the objective function is optimal.
[0077] Combination Figure 3 The process of the improved dung beetle optimization algorithm is as follows:
[0078] Step 501: Initialize population parameters such as the dung beetle colony location X and the maximum number of iterations T. max Population size N, optimization upper and lower bounds U b and L b Logistic chaotic mapping is used to optimize the population distribution of the algorithm.
[0079] Step 502: Calculate the fitness value of each agent based on the objective function.
[0080] Step 503: Update the positions of all types of dung beetles. Use an adaptive t-distribution mutation operator to perturb the dung beetle positions, and use the perturbated and updated positions as the final newly generated positions for this iteration.
[0081] Step 504: Determine whether each agent has exceeded the boundary.
[0082] Step 505: Determine if the newly generated position is better than the original position. If so, update the position; otherwise, keep the original position.
[0083] Step 506: Compare the latest positions of each type of dung beetle and select the current optimal solution and fitness value.
[0084] Step 507: Determine if the number of iterations has reached the upper limit. If yes, proceed to step 509; otherwise, proceed to step 408.
[0085] Step 508: Iteration count iter = iter + 1, return to step 502.
[0086] Step 509: Output the global optimal solution and its fitness value.
[0087] The above hyperparameters were optimized using an improved dung beetle optimization algorithm, resulting in an optimized fault diagnosis model. This improved the diagnostic accuracy, reduced training time, and enabled faster and more accurate fault diagnosis of integrated energy systems.
[0088] Example 2
[0089] In one or more embodiments, a comprehensive energy fault accurate and rapid diagnosis system is disclosed, comprising:
[0090] The data acquisition module is used to acquire operational data of the integrated energy system;
[0091] The fault classification module is used to obtain fault classification results using the trained fault diagnosis model;
[0092] The fault diagnosis model is a CNN-LSTM model, and the training process for the fault diagnosis model is as follows:
[0093] By simulating faults in the integrated energy system, a fault dataset of each device in the integrated energy system is obtained; the fault dataset is then expanded to obtain a new fault data sample set.
[0094] The local mean decomposition method is used to denoise and reconstruct the new fault data sample set. The reconstructed fault data sample set is then input into the CNN-LSTM model to initialize the hyperparameters of the CNN-LSTM model.
[0095] We construct a dual objective function that maximizes model accuracy and minimizes training time. We then use the dung beetle optimization algorithm to optimize the hyperparameters and feature selection dimension of the CNN-LSTM model, ultimately obtaining a well-trained CNN-LSTM model.
[0096] The specific implementation methods of the above modules have been described in detail in Example 1, and will not be repeated here.
[0097] Example 3
[0098] In one or more embodiments, a terminal device is disclosed, including a server. The server includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the comprehensive energy fault accurate and rapid diagnosis method of Embodiment 1. For the sake of brevity, further details are omitted here.
[0099] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.
[0100] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.
[0101] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.
[0102] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.
Claims
1. A comprehensive energy fault accurate and rapid diagnosis method, characterized in that, include: Obtain operational data of the integrated energy system and use the trained fault diagnosis model to obtain fault classification results; The fault diagnosis model is a CNN-LSTM model, and the training process for the fault diagnosis model is as follows: By simulating faults in the integrated energy system, a fault dataset of each device in the integrated energy system is obtained; the fault dataset is then expanded to obtain a new fault data sample set. The local mean decomposition method is used to denoise and reconstruct the new fault data sample set. The reconstructed fault data sample set is then input into the CNN-LSTM model to initialize the hyperparameters of the CNN-LSTM model. We construct a dual objective function with the highest model accuracy and shortest training time, and optimize the hyperparameters of the CNN-LSTM model using the dung beetle optimization algorithm to finally obtain a well-trained CNN-LSTM model. The optimization of the hyperparameters and feature selection dimension of the CNN-LSTM model using the dung beetle optimization algorithm is as follows: Initialize population parameters; Calculate the fitness value of each agent based on the objective function; Update the positions of all types of dung beetles by perturbing the dung beetle positions using an adaptive t-distribution mutation operator, and use the perturbed and updated positions as the final newly generated positions for this iteration; Determine if each agent exceeds the boundary, and determine if the newly generated position is better than the original position. If so, update the position; otherwise, keep the original position. Compare the latest locations of each type of dung beetle, and select the current optimal solution and fitness value; Continue until the required number of iterations is met, then output the global optimal solution and its fitness value. The use of a dynamically selected probability p to adjust the application of the adaptive t-distribution mutation operator is as follows: in, The maximum number of iterations, This represents the current iteration number. To dynamically select the upper limit of the probability. This is the lower bound for dynamically selecting the probability.
2. The comprehensive energy fault accurate and rapid diagnosis method as described in claim 1, characterized in that, Obtain operational data from integrated energy systems, specifically including: Refrigeration unit: cooling power, electrical power, COP, cooling capacity, chilled water supply temperature and cooling water return temperature; CHP unit: power generation, heating power, thermal efficiency, heating capacity, flue gas temperature and hot water supply temperature, output three-phase voltage and three-phase current; photovoltaic unit: output three-phase voltage and three-phase current; wind turbine unit: output three-phase voltage and three-phase current; battery bank: output three-phase voltage and three-phase current; hydrogen fuel cell: gas temperature, gas inlet pressure and current.
3. The comprehensive energy fault accurate and rapid diagnosis method as described in claim 1, characterized in that, By simulating faults in the integrated energy system, a fault dataset of each device in the integrated energy system is obtained, specifically: A model of the integrated energy system of the park was built in MATLAB / SIMULINK to simulate common integrated energy system failures. Fault datasets of various devices in the integrated energy system were collected through real-time simulation using RT-LAB. The integrated energy system model of the park includes a thermodynamic system module and an electrical system module. The equipment in the thermodynamic system module includes electric chillers, lithium bromide absorption chillers, heat pumps, waste heat boilers, steam-gas combined cycle power plants, and waste heat boilers. The equipment in the electrical system module includes wind turbines, photovoltaic units, battery banks, hydrogen fuel cells, electrolyzers, diesel engines, and power distribution networks.
4. The comprehensive energy fault accurate and rapid diagnosis method as described in claim 1, characterized in that, The new fault data sample set is denoised and reconstructed using the local mean decomposition method, specifically as follows: The collected comprehensive energy fault signal is decomposed into k PF components and a margin by the local mean decomposition method, and then the signal is reconstructed by the autocorrelation principle to obtain the denoised signal. Among them, the whale algorithm is used to optimize the number k of PF components to obtain the optimal k value.
5. The method for accurate and rapid diagnosis of integrated energy faults as described in claim 1, characterized in that, The CNN-LSTM model is specifically as follows: The LSTM is embedded between the last pooling layer and the fully connected layer of the CNN, and the final fully connected layer is connected to the output layer to output the fault diagnosis results. For the input denoised dataset, the CNN-LSTM model sequentially passes through convolutional layers and pooling layers to perform convolution calculations and signal compression, outputting the corresponding fault feature vector each time according to the required optimized convolutional kernel size; a feature vector sample set is constructed based on the fault feature vectors extracted each time, which serves as the input for the final fault diagnosis and classification.
6. The comprehensive energy fault accurate and rapid diagnosis method as described in claim 1, characterized in that, Construct a dual objective function that maximizes model accuracy and minimizes training time, specifically: in, The number of samples that were correctly diagnosed. The total number of samples, This refers to the model training time.
7. A comprehensive energy fault accurate and rapid diagnosis system, implementing the comprehensive energy fault accurate and rapid diagnosis method as described in any one of claims 1-6, characterized in that, include: The data acquisition module is used to acquire operational data of the integrated energy system; The fault classification module is used to obtain fault classification results using the trained fault diagnosis model; The fault diagnosis model is a CNN-LSTM model, and the training process for the fault diagnosis model is as follows: By simulating faults in the integrated energy system, a fault dataset of each device in the integrated energy system is obtained; the fault dataset is then expanded to obtain a new fault data sample set. The local mean decomposition method is used to denoise and reconstruct the new fault data sample set. The reconstructed fault data sample set is then input into the CNN-LSTM model to initialize the hyperparameters of the CNN-LSTM model. We construct a dual objective function that maximizes model accuracy and minimizes training time. We then optimize the hyperparameters of the CNN-LSTM model using the dung beetle optimization algorithm to obtain a well-trained CNN-LSTM model.
8. A terminal device comprising a processor and a memory, the processor for implementing instructions; the memory for storing multiple instructions, characterized in that, The instructions are adapted to be loaded by a processor and executed as described in any one of claims 1-6, representing a comprehensive energy fault accurate and rapid diagnosis method.