Method for detecting abnormal operation of electrochemical energy storage power station based on deep learning

Abstract

The invention discloses an electrochemical energy storage power station operation anomaly detection method based on deep learning, and belongs to the technical field of electrochemical energy storage power station operation anomaly detection. Historical operation data of an electrochemical energy storage power station at a historical moment and a real power station operation state under the historical operation data are adopted to construct training data; the method comprises the following steps: acquiring real-time operation data of an electrochemical energy storage power station, training a deep learning model by adopting an improved parameter training algorithm on the basis of the training data, so that the deep learning model learns different operation states of the electrochemical energy storage power station to obtain an operation anomaly detection model, and finally inputting the real-time operation data into the operation anomaly detection model for identification. According to the method and the device, the operation anomaly detection result of the electrochemical energy storage power station is obtained, automatic anomaly detection of the operation of the electrochemical energy storage power station is realized, the anomaly detection efficiency and timeliness can be improved, and the problem of misrecognition caused by manual detection is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of abnormal operation detection of electrochemical energy storage power stations, and particularly relates to a method for detecting abnormal operation of electrochemical energy storage power stations based on deep learning. Background Art

[0002] With the wide application of renewable energy, electrochemical energy storage power stations, as key equipment for energy storage, are of great significance for ensuring the stability of the power grid and improving energy utilization efficiency. However, various abnormal operation conditions may occur during the operation of electrochemical energy storage power stations. The abnormal operation of electrochemical energy storage power stations refers to various abnormal states or faults that occur during the operation of the energy storage system. These abnormalities may originate from battery packs, battery management systems (BMS), power electronic devices, or auxiliary systems, etc. Common abnormalities include overcharging, over-discharging, overheating of batteries, and voltage or temperature inconsistencies. These abnormalities may lead to a decline in the performance of the energy storage system, a shortening of its lifespan, and even safety accidents. Therefore, real-time monitoring, early warning, and timely handling of abnormal operations are crucial. Operation and maintenance personnel need to analyze data to locate the fault points and take corresponding measures, such as adjusting operation parameters, replacing faulty components, etc., to ensure the safe, stable, and efficient operation of the power station. At the same time, regular maintenance and preventive detection are also important means to prevent abnormal operations. However, manual detection not only has low efficiency but also is prone to misidentification problems, resulting in the inability to detect the abnormal operation of electrochemical energy storage power stations in a timely manner. Summary of the Invention

[0003] The present invention provides a method for detecting abnormal operation of electrochemical energy storage power stations based on deep learning to solve the problems in the prior art that manual detection not only has low efficiency but also is prone to misidentification, resulting in the inability to detect the abnormal operation of electrochemical energy storage power stations in a timely manner.

[0004] A method for detecting abnormal operation of electrochemical energy storage power stations based on deep learning includes: Collecting historical operation data of the electrochemical energy storage power station at historical moments and the real operation state of the power station under the historical operation data; Constructing training data according to the historical operation data at the historical moments and the real operation state of the power station under the historical operation data; Based on the training data, training a deep learning model using an improved parameter training algorithm so that the deep learning model learns different operation states of the electrochemical energy storage power station to obtain an abnormal operation detection model; Collecting real-time operation data and inputting the real-time operation data into the abnormal operation detection model for identification to obtain the abnormal operation detection result of the electrochemical energy storage power station.

[0005] Further, collect the historical operation data of the electrochemical energy storage power station at historical moments and the real power station operation status under the historical operation data, including: Based on a preset data sampling frequency, collect the voltage, current, temperature, and SOC of the electrochemical energy storage power station at historical moments to obtain historical operation data; During the process of collecting historical operation data, synchronously collect the real power station operation status under the historical operation data.

[0006] Further, based on the historical operation data at the historical moments and the real power station operation status under the historical operation data, construct training data, including: Perform normalization processing on the historical operation data at the historical moments to obtain the historical operation data after normalization processing, and use the historical operation data after normalization processing as training samples; Construct the real power station operation status under the historical operation data as the training label corresponding to the training sample, and use the training sample and the training label together as training data.

[0007] Further, based on the training data, use an improved parameter training algorithm to train the deep learning model, so that the deep learning model learns different operation states of the electrochemical energy storage power station to obtain an operation anomaly detection model, including: Use a chaotic initialization strategy to initialize the parameters of the deep learning model to obtain multiple different parameter individuals; For any parameter individual, based on the training data, obtain the fitness corresponding to each parameter individual, and determine the parameter individual with the maximum fitness as the optimal individual; Based on the current training times, obtain a balance factor for balancing global search and local search, and generate a decision factor between (0, 1); When the decision factor is greater than or equal to the balance factor, then according to the optimal individual, use a double local exploration strategy to perform local exploration on the parameter individual to obtain the parameter individual after one search; When the decision factor is less than the balance factor, then use the Levy global development strategy to perform global development on the parameter individual to obtain the parameter individual after one search; For any parameter individual after one search, use a surrounding optimization strategy to perform fine search on the parameter individual to obtain the parameter individual after two searches; For any parameter individual after two searches, use a position migration development strategy to perform adaptive position conversion on the parameter individual to obtain the parameter individual after three searches; Determine whether the current number of training times is greater than or equal to the preset maximum number of training times. If so, re-determine the optimal individual based on the parameter individuals after three searches; otherwise, return to the step of determining the optimal individual. Determine the final parameters of the deep learning model according to the re-determined optimal individual, and obtain the running anomaly detection model.

[0008] Furthermore, adopt a chaotic initialization strategy to initialize the parameters of the deep learning model, and obtain multiple different parameter individuals, including: For any parameter of the deep learning model, randomly initialize it between the upper limit and the lower limit corresponding to the parameter, and encode the initialized parameter as a vector to obtain a randomly initialized parameter individual. Based on the randomly initialized parameter individual, obtain multiple different parameter individuals as:

[0009] In the formula, represents the i th parameter individual, and when i = 1, represents the randomly initialized parameter individual, represents the i +1th parameter individual.

[0010] Furthermore, based on the current number of training times, obtain the balance factor for balancing global search and local search as:

[0011] In the formula, represents the balance factor, represents the initial value of the balance factor, represents the current number of training times, represents the preset maximum number of training times, represents the natural constant, represents the first constant term, and is set to 0.6.

[0012] Furthermore, according to the optimal individual, adopt a double local exploration strategy to perform local exploration on the parameter individuals to obtain the parameter individuals after one search, including: Based on the current number of training times, obtain the information exploration factor as:

[0013] In the formula, represents the information exploration factor, e represents the natural constant, represents the first random number between (0, 1), Represents a second random number between (0,1), and sin represents the sine function. Represents pi. According to the information exploration factor and the optimal individual, perform local exploration on the parameter individuals, and the parameter individuals after one search are:

[0014] In the formula, Represents the k th training process and the j th parameter individual, j i = 1, 2, …, NP, where NP represents the total number of parameter individuals, Represents the j th parameter individual after one search, Represents the first random parameter individual, Represents a third random number between (0,1), Represents a fourth random number between (0,1), Represents the optimal individual.

[0015] Furthermore, adopt the Levy global development strategy to conduct global development on the parameter individuals, and the parameter individuals after one search include: Adopt the Levy flight strategy to generate Levy flight parameters; Based on the current training times, generate the first learning factor as:

[0016] In the formula, Represents the first learning factor, Represents a fifth random number between (0,1), Represents the preset maximum number of training times; According to the learning factor and the Levy flight parameters, conduct global development on the parameter individuals, and the parameter individuals after one search are:

[0017] In the formula, Represents a sixth random number between (0,1), Represents a seventh random number between (0,1), Represents the Levy flight parameter, Represents the second random parameter individual.

[0018] Furthermore, for any parameter individual after one search, adopt the surrounding optimization strategy to conduct fine search on the parameter individual, and the parameter individuals after the second search include: For any parameter individual after the first search, obtain the fitness corresponding to the parameter individual, and arrange them in descending order of fitness to obtain the sorted parameter individuals; Based on the sorted parameter individuals, perform a fine search on the parameter individuals, and the parameter individuals after the second search are:

[0019] wherein, represents the k th sorted parameter individual in the m th training process, represents the k th training process and the m -1th sorted parameter individual, m = 1, 2, …, NP, represents the m th parameter individual after the second search, represents the eighth random number between [0, 2π], represents the ninth random number between [0, π], represents the third random parameter individual, represents the first fine search coefficient, and , represents the second fine search coefficient, and , represents the second constant term, and .

[0020] Furthermore, for any parameter individual after the second search, adopt a position migration development strategy to perform adaptive position conversion on the parameter individual to obtain the parameter individual after the third search, including: Based on the current training times, generate the second learning factor as:

[0021] wherein, represents the second learning factor; According to the second learning factor, obtain the migration step size individual as:

[0022] wherein, represents the migration step size individual, represents the parameter lower limit individual, represents the parameter upper limit individual, and exp represents the exponential function with the natural constant e as the base; For any parameter individual after the second search, according to the migration step size individual, perform adaptive position conversion on the parameter individual to obtain the parameter individual after the third search as:

[0023] In the formula, represents the k parameter individual after the n th quadratic search during the th training process, n represents the parameter individual after the n th cubic search, where = 1, 2, …, NP, represents the tenth random number between (0, 1), represents the eleventh random number between (0, 1), represents the twelfth random number between (0, 1),

[0024] A method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning provided by the present invention constructs training data by using historical operation data of the electrochemical energy storage power station at historical moments and the true operation state of the power station under the historical operation data. Based on the training data, an improved parameter training algorithm is used to train a deep learning model, so that the deep learning model learns different operation states of the electrochemical energy storage power station to obtain an abnormal operation detection model. Finally, the real-time operation data is input into the abnormal operation detection model for identification to obtain the abnormal operation detection result of the electrochemical energy storage power station, realizing automatic abnormal detection of the operation of the electrochemical energy storage power station, improving the efficiency and timeliness of abnormal detection, and avoiding the problem of misidentification caused by manual detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings herein are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present invention and used together with the specification to explain the principles of the present invention.

[0026] Figure 1 FIG. is a schematic flow chart of a method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning provided by an embodiment of the present invention.

[0027] Through the above drawings, specific embodiments of the present invention have been shown, and there will be more detailed descriptions hereinafter. These drawings and text descriptions are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0029] Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0030] As Figure 1 shown, an embodiment of the present invention provides a method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning, including: S1. Collect historical operation data of the electrochemical energy storage power station at historical moments and the real power station operation state under the historical operation data; Optionally, multiple nodes can be set in the electrochemical energy storage power station to collect historical operation data to more comprehensively reflect the operation state of the electrochemical energy storage power station. The real power station operation state is a real normal operation state or a faulty state. There is an associated data relationship between the historical operation data and the real power station operation state. This associated data relationship can be learned through deep learning, so as to obtain an operation anomaly detection model with data recognition ability.

[0031] S2. Construct training data according to the historical operation data at the historical moments and the real power station operation state under the historical operation data; In an embodiment of the present invention, constructing training data according to the historical operation data at the historical moments and the real power station operation state under the historical operation data includes: Perform normalization processing on the historical operation data at the historical moments to obtain the historical operation data after normalization processing, and use the historical operation data after normalization processing as training samples; Construct the real power station operation state under the historical operation data as the training label corresponding to the training sample, and use the training sample and the training label together as training data.

[0032] It should be noted that the historical operation data is normalized here. In the subsequent recognition process, the data to be recognized also needs to be normalized to ensure accurate recognition of the data.

[0033] S3. Based on the training data, use an improved parameter training algorithm to train the deep learning model, so that the deep learning model learns different operation states of the electrochemical energy storage power station to obtain an operation anomaly detection model; The deep learning model can identify the associated data relationships. After being trained with training data, an operation anomaly detection model can be obtained, and then the operation anomaly detection model can be used for automatic detection to achieve the operation anomaly detection of the electrochemical energy storage power station.

[0034] In the prior art, the parameter training algorithm of the deep learning model generally adopts the gradient descent algorithm. Although this algorithm has good convergence accuracy and training speed, it is extremely easy to fall into the local optimum, resulting in the deep learning model finally obtained by training being unable to accurately complete the data recognition task. Therefore, the embodiment of the present invention adopts an improved parameter training algorithm to train the deep learning model to improve the training effect and enhance the recognition accuracy of the operation anomaly detection model for data.

[0035] S4. Collect real-time operation data, and input the real-time operation data into the operation anomaly detection model for recognition to obtain the operation anomaly detection result of the electrochemical energy storage power station.

[0036] It should be noted that in order to ensure that the data can be accurately recognized, the data structures of the real-time operation data and the historical operation data should be the same.

[0037] The embodiment of the present invention provides a method for detecting operation anomalies of an electrochemical energy storage power station based on deep learning. By using the historical operation data of the electrochemical energy storage power station at historical moments and the real power station operation state under the historical operation data to construct training data, and based on the training data, an improved parameter training algorithm is used to train the deep learning model, enabling the deep learning model to learn different operation states of the electrochemical energy storage power station to obtain an operation anomaly detection model. Finally, the real-time operation data is input into the operation anomaly detection model for recognition to obtain the operation anomaly detection result of the electrochemical energy storage power station, realizing the automatic anomaly detection of the operation of the electrochemical energy storage power station, which can improve the anomaly detection efficiency and timeliness and avoid the misrecognition problem caused by manual detection.

[0038] In the embodiment of the present invention, collecting the historical operation data of the electrochemical energy storage power station at historical moments and the real power station operation state under the historical operation data includes: Based on a preset data sampling frequency, collect the voltage, current, temperature, and SOC (state of charge) of the electrochemical energy storage power station at historical moments to obtain historical operation data; During the process of collecting historical operation data, synchronously collect the real power station operation state under the historical operation data.

[0039] It should be noted that the above historical operation data is only a preferred example in the embodiment of the present invention, and other data can also be used to construct historical operation data.

[0040] Optionally, the historical operation data may include data on multiple data acquisition nodes, which may be distributed at different locations in the electrochemical energy storage power station. Then, according to a preset data sampling frequency, historical operation data at multiple consecutive time points may be collected, and a multi-dimensional data matrix can be obtained. Each row of the multi-dimensional data matrix may be the historical operation data on a data acquisition node.

[0041] In an embodiment of the present invention, based on the training data, an improved parameter training algorithm is used to train a deep learning model, enabling the deep learning model to learn different operation states of the electrochemical energy storage power station, and obtaining an operation anomaly detection model, including: Using a chaotic initialization strategy to initialize the parameters of the deep learning model to obtain multiple different parameter individuals; Optionally, the deep learning model may be set as a convolutional neural network model and its corresponding improved model to achieve the recognition of the data matrix. Therefore, the parameters of the deep learning model may be its weight parameters.

[0042] For any one parameter individual, based on the training data, obtain the fitness corresponding to each parameter individual, and determine the parameter individual with the maximum fitness as the optimal individual; Based on the current training times, obtain a balance factor for balancing global search and local search, and generate a decision factor between (0, 1); When the decision factor is greater than or equal to the balance factor, then according to the optimal individual, use a double local exploration strategy to perform local exploration on the parameter individuals to obtain the parameter individuals after one search; When the decision factor is less than the balance factor, then use the Levy global development strategy to perform global development on the parameter individuals to obtain the parameter individuals after one search; For any one parameter individual after one search, use a surrounding optimization strategy to perform fine search on the parameter individual to obtain the parameter individuals after two searches; For any one parameter individual after two searches, use a position migration development strategy to perform adaptive position conversion on the parameter individual to obtain the parameter individuals after three searches; Determine whether the current training times are greater than or equal to the preset maximum training times. If so, based on the parameter individuals after three searches, re-determine the optimal individual. Otherwise, return to the step of determining the optimal individual; Based on the re-determined optimal individual, determine the final parameters of the deep learning model to obtain an operation anomaly detection model.

[0043] Optionally, an out-of-bounds handling strategy may be introduced to control the training process, thereby ensuring that the parameters are always within the valid range.

[0044] In an embodiment of the present invention, a chaotic initialization strategy is adopted to initialize the parameters of a deep learning model, and multiple different parameter individuals are obtained, including: For any parameter of the deep learning model, it is randomly initialized between the upper limit and the lower limit corresponding to the parameter, and the parameter after initialization is encoded as a vector to obtain a randomly initialized parameter individual; Based on the randomly initialized parameter individual, multiple different parameter individuals are obtained as:

[0045] where represents the i th parameter individual, and when i = 1, represents the randomly initialized parameter individual, represents the i + 1th parameter individual.

[0046] By initializing the parameters of the deep learning model using the above chaotic initialization strategy, the parameter individuals can be evenly distributed in the solution space at the initial moment, improving the training effect of the algorithm.

[0047] In an embodiment of the present invention, based on the current number of training times, a balance factor for balancing global search and local search is obtained as:

[0048] where represents the balance factor, represents the initial value of the balance factor, represents the current number of training times, represents the preset maximum number of training times, represents the natural constant, represents the first constant term, and is set to 0.6.

[0049] The balance factor for balancing global search and local search provided by the embodiment of the present invention can slowly decrease in the middle and early stages of the algorithm, and achieve a rapid decrease in the later stage of the algorithm, so as to achieve a certain global search in the middle and early stages of the algorithm, and convert to a more refined local search in the later stage of the algorithm, comprehensively improving the performance of the algorithm.

[0050] In an embodiment of the present invention, according to the optimal individual, a double local exploration strategy is adopted to perform local exploration on the parameter individuals, and the parameter individuals after one search are obtained, including: Based on the current number of training times, an information exploration factor is obtained as:

[0051] In the formula, represents the information exploration factor, e represents the natural constant, represents the first random number between (0, 1), represents the second random number between (0, 1), sin represents the sine function, represents the pi; According to the information exploration factor and the optimal individual, perform local exploration on the parameter individuals, and the parameter individuals after one search are:

[0052] In the formula, represents the k th j parameter individual in the j th training process, = 1, 2, …, NP, NP represents the total number of parameter individuals, j represents the th parameter individual after one search, represents the first random parameter individual, represents the third random number between (0, 1), represents the optimal individual.

[0053] The above double local exploration strategy can enable the parameter individuals to search in two different ways while learning the optimal information. The search processes intersect with each other and are independent, so as to improve the ability to search for better solutions in the neighborhood. In the early stage of the algorithm, learning more optimal information can improve the training speed of the algorithm. In the later stage of the algorithm, it tends to the information exchange between different parameter individuals, ensuring the training diversity and making the algorithm not easily fall into the local optimum.

[0054] In the embodiment of the present invention, the Levy global exploration strategy is adopted to perform global exploration on the parameter individuals, and the parameter individuals after one search are obtained, including: Generate Levy flight parameters by using the Levy flight strategy; Based on the current training times, generate the first learning factor as:

[0055] In the formula, represents the first learning factor, represents the fifth random number between (0, 1), represents the preset maximum training times; According to the learning factor and the Levy flight parameters, perform global exploration on the parameter individuals, and the parameter individuals after one search are:

[0056] In the formula, represents the sixth random number between (0, 1), represents the seventh random number between (0, 1), represents the Levy flight parameter, represents the second random parameter individual.

[0057] The above Levy global development strategy provides a certain global search ability and ensures that the algorithm will not easily fall into local optima.

[0058] In the embodiment of the present invention, for any parameter individual after a primary search, a fine search is performed on the parameter individual by adopting a surrounding optimization strategy to obtain the parameter individual after a secondary search, including: For any parameter individual after a primary search, obtain the fitness corresponding to the parameter individual and arrange them in descending order of fitness to obtain the sorted parameter individuals; Based on the sorted parameter individuals, perform a fine search on the parameter individuals, and the parameter individuals after a secondary search are:

[0059] In the formula, represents the k th m sorted parameter individual in the th k training process, m represents the m -1 th m sorted parameter individual in the th training process, = 1, 2,..., NP, represents the th sorted parameter individual after a secondary search, represents the eighth random number between [0, 2π], represents the ninth random number between [0, π], represents the third random parameter individual,

[0060] The above surrounding optimization strategy can enable the parameter individuals to search in a way of overall cooperative search around their own positions, which is more conducive to finding better solutions and improving the search accuracy of the algorithm.

[0061] In the embodiment of the present invention, for any parameter individual after the second search, an adaptive position transformation is performed on the parameter individual by using a position migration development strategy to obtain a parameter individual after the third search, including: Based on the current number of training times, a second learning factor is generated as:

[0062] In the formula, represents the second learning factor; According to the second learning factor, a migration step size individual is obtained as:

[0063] In the formula, represents the migration step size individual, represents the parameter lower limit individual, represents the parameter upper limit individual, and exp represents the exponential function with the natural constant e as the base; For any parameter individual after the second search, according to the migration step size individual, an adaptive position transformation is performed on the parameter individual to obtain a parameter individual after the third search as:

[0064] In the formula, represents the k th parameter individual after the second search in the n th training process, represents the n th parameter individual after the third search, n = 1, 2, …, NP, represents the tenth random number between (0, 1), represents the eleventh random number between (0, 1), represents the twelfth random number between (0, 1), represents the fourth random parameter individual.

[0065] The above position migration development strategy can make the position of the parameter individual undergo adaptive migration, with more migration in the early stage of the algorithm, providing a strong global search ability.

[0066] Optionally, an annealing simulation algorithm can also be used to control the position migration development strategy to ensure the training speed of the algorithm.

[0067] Through the mutual cooperation of the above various strategies, the training effect of the algorithm is effectively improved, ensuring that the algorithm can find the global optimal solution in the high-dimensional solution space, thereby ensuring that the trained running anomaly detection model can accurately implement running anomaly detection.

[0068] Other embodiments of the present invention will be readily apparent to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the invention following the general principles of the invention and including known common general knowledge or conventional technical means in the technical field not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. An abnormal operation detection method for an electrochemical energy storage power station based on deep learning, characterized in that, Including: Collect the historical operation data of the electrochemical energy storage power station at historical moments and the real power station operation state under the historical operation data; Construct training data according to the historical operation data at the historical moments and the real power station operation state under the historical operation data; Based on the training data, use an improved parameter training algorithm to train the deep learning model, enabling the deep learning model to learn different operation states of the electrochemical energy storage power station, and obtain an operation anomaly detection model; Collect real-time operation data and input the real-time operation data into the operation anomaly detection model for identification to obtain the operation anomaly detection result of the electrochemical energy storage power station.

2. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 1, wherein Collect the historical operation data of the electrochemical energy storage power station at historical moments and the real power station operation state under the historical operation data, including: Based on a preset data sampling frequency, collect the voltage, current, temperature, and SOC of the electrochemical energy storage power station at historical moments to obtain historical operation data; During the process of collecting historical operation data, synchronously collect the real power station operation state under the historical operation data.

3. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 1, characterized in that Construct training data according to the historical operation data at the historical moments and the real power station operation state under the historical operation data, including: Perform normalization processing on the historical operation data at the historical moments to obtain the historical operation data after normalization processing, and use the historical operation data after normalization processing as training samples; Construct the real power station operation state under the historical operation data into training labels corresponding to the training samples, and use the training samples and the training labels together as training data.

4. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 1, wherein Based on the training data, use an improved parameter training algorithm to train the deep learning model, enabling the deep learning model to learn different operation states of the electrochemical energy storage power station, and obtain an operation anomaly detection model, including: Use a chaotic initialization strategy to initialize the parameters of the deep learning model to obtain multiple different parameter individuals; For any parameter individual, based on the training data, obtain the fitness corresponding to each parameter individual, and determine the parameter individual with the maximum fitness as the optimal individual; Based on the current training times, obtain a balance factor for balancing global search and local search, and generate a decision factor between (0, 1); When the decision factor is greater than or equal to the balance factor, perform local exploration on the parameter individuals according to the optimal individual using a double local exploration strategy to obtain the parameter individuals after one search; When the decision factor is less than the balance factor, perform global development on the parameter individuals using the Levy global development strategy to obtain the parameter individuals after one search; For any parameter individual after one search, perform fine search on the parameter individual using a surrounding optimization strategy to obtain the parameter individuals after two searches; For any parameter individual after two searches, perform adaptive position conversion on the parameter individual using a position migration development strategy to obtain the parameter individuals after three searches; Determine whether the current number of training times is greater than or equal to the preset maximum number of training times. If so, re-determine the optimal individual according to the parameter individuals after three searches; otherwise, return to the step of determining the optimal individual. Determine the final parameters of the deep learning model according to the re-determined optimal individual, and obtain the running anomaly detection model.

5. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 4, wherein, Initialize the parameters of the deep learning model using the chaotic initialization strategy to obtain multiple different parameter individuals, including: For any parameter of the deep learning model, randomly initialize it between the upper and lower limits corresponding to the parameter, and encode the initialized parameter as a vector to obtain a randomly initialized parameter individual. Based on the randomly initialized parameter individual, obtain multiple different parameter individuals as: In the formula, represents the i th parameter individual, and i when = 1, represents the i ( i + 1)th parameter individual.

6. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 5, wherein Based on the current number of training times, obtain the balance factor for balancing global search and local search as: In the formula, represents the balance factor, represents the initial value of the balance factor, represents the current number of training times, represents the preset maximum number of training times, represents the natural constant, represents the first constant term, and is set to 0.

6.

7. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 5, characterized in that, According to the optimal individual, perform local exploration on the parameter individuals using the double local exploration strategy to obtain the parameter individuals after one search, including: Based on the current number of training times, obtain the information exploration factor as: In the formula, represents the information exploration factor, e represents the natural constant, represents the first random number between (0, 1), represents the second random number between (0, 1), sin represents the sine function, represents the pi; According to the information exploration factor and the optimal individual, perform local exploration on the parameter individuals to obtain the parameter individuals after one search as: In the formula, represents the k th parameter individual in the j th training process, j = 1, 2, …, NP, where NP represents the total number of parameter individuals, represents the parameter individual after the j th first search, represents the first random parameter individual, represents the third random number between (0, 1), represents the fourth random number between (0, 1), represents the optimal individual.

8. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 7, wherein Perform global exploitation on the parameter individuals using the Levy global exploitation strategy to obtain the parameter individuals after one search, including: Generate Levy flight parameters using the Levy flight strategy. Based on the current number of training times, generate the first learning factor as: In the formula, represents the first learning factor, represents the fifth random number between (0, 1), represents the preset maximum number of training times; According to the learning factor and the Levy flight parameters, perform global exploitation on the parameter individuals to obtain the parameter individuals after one search as: In the formula, represents the sixth random number between (0, 1), represents the seventh random number between (0, 1), represents the Levy flight parameter, represents the second random parameter individual.

9. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 8, wherein For any parameter individual after one search, perform fine search on the parameter individual using the surrounding optimization strategy to obtain the parameter individuals after two searches, including: For any parameter individual after one search, obtain the fitness corresponding to the parameter individual, and arrange them in descending order of fitness to obtain the sorted parameter individuals. Based on the sorted parameter individuals, perform fine search on the parameter individuals to obtain the parameter individuals after two searches as: Wherein, represents the k th sorted parameter individual in the m th training process, represents the k th training process and the m -1th sorted parameter individual, m = 1, 2, …, NP, represents the m th parameter individual after the second search, represents the eighth random number between [0, 2π], represents the ninth random number between [0, π], represents the third random parameter individual, represents the first fine search coefficient, and , represents the second fine search coefficient, and , represents the second constant term, and .

10. The method for detecting abnormal operation of an electrochemical energy storage power station based on deep learning according to claim 9, characterized in that, For any parameter individual after two searches, perform adaptive position transformation on the parameter individual using the position migration exploitation strategy to obtain the parameter individuals after three searches, including: Based on the current number of training times, generate the second learning factor as: In the formula, represents the second learning factor; According to the second learning factor, obtain the migration step size individuals as: wherein, represents the individual migration step size, represents the individual parameter lower limit, represents the individual parameter upper limit, and exp represents the exponential function with the natural constant e as the base; For any parameter individual after two searches, perform adaptive position transformation on the parameter individual according to the migration step size individuals to obtain the parameter individuals after three searches as: Wherein, represents the parameter individual after the k th secondary search in the n th training process, represents the parameter individual after the n th tertiary search, n = 1, 2, …, NP, represents the tenth random number between (0, 1), represents the eleventh random number between (0, 1), represents the twelfth random number between (0, 1), represents the fourth random parameter individual.

Images