Fault Diagnosis Method and System for Energy Storage Power Station Based on Parameter Identification and Penalty Strategy

Abstract

The present invention discloses a fault diagnosis method and system for an energy storage power station based on parameter identification and penalty strategy, including Step 1: collecting training sample data for training a fault detection model and online real-time detection data; Step 2: according to the voltage and current collected at the current moment, using an online real-time parameter identification algorithm based on the alternating generalized least squares method with a forgetting factor to extract four identification features and construct a feature set; Step 3: based on the weights of different clustering algorithms, calculating the distances between the detection samples and different cluster centers of different clustering algorithms, and multiplying by the weights of the clustering algorithms to obtain the final distances to identify whether the samples are faulty; Step 4: updating the integrated weights of the clustering algorithms and the cluster centers depending on the penalty strategy of the integrated weights of the clustering algorithms and the update strategy of the cluster centers.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage power station fault diagnosis, and particularly to a fault diagnosis method, system, device and storage medium for an energy storage power station based on parameter identification and penalty strategy. Background Art

[0002] As a basic network module of an intelligent microgrid, the energy storage system supports the supply, storage and conversion of electric energy, provides guarantee for the safe and reliable operation of the power grid, and achieves the purpose of assisting the intelligent microgrid to realize smooth power transition, load balancing, frequency modulation and voltage regulation, peak shaving and valley filling. At the same time, it also plays a decisive role in the utilization efficiency and development prospect of the entire energy system, and is one of the research hotspots in the current intelligent microgrid.

[0003] Lithium-ion batteries are important carriers for storing and converting electric energy and are important guarantees for the normal operation of energy storage systems. With the rapid development and large-scale application of energy storage systems, ensuring the safe operation of lithium-ion batteries is of utmost importance for protecting personnel safety and reducing economic losses of energy storage systems. The faults of lithium-ion batteries mainly include external faults and internal faults. The internal faults cause higher harm and are difficult to detect and repair, becoming the current focus of attention. The internal faults are mainly divided into mechanical faults, thermal faults, overcharge faults and over-discharge faults. Among them, the main causes of overcharge and over-discharge are the inconsistency of single cells. Overcharge fault refers to the phenomenon of continuing to charge beyond the specified battery charging cut-off voltage, and over-discharge fault refers to the phenomenon of continuously discharging below the specified battery discharge cut-off voltage. Repeated overcharge and over-discharge of the battery will cause internal short circuit and thermal runaway. Therefore, accurate detection of overcharge and over-discharge faults of lithium-ion batteries is an important means for the safe operation and stable work of the energy storage power station system.

[0004] Fault diagnosis emphasizes timeliness. When a lithium-ion battery fails, early warning should be carried out as soon as possible to achieve rapid fault control and reduce losses. Low latency of diagnosis is necessary. At the same time, the main characteristic attributes of the energy storage power station monitored in real time are voltage and current. Using the parameter identification method, other parameters can be calculated to amplify the attribute characteristics and better infer the battery state.

[0005] Based on the parameters of the energy storage power station, relying on the parameter identification method and clustering algorithm, the present invention takes the overcharge fault and over-discharge fault of the battery cluster as the main research content to realize real-time monitoring and diagnosis of the faults of lithium-ion battery clusters in the energy storage system. Summary of the Invention

[0006] In order to overcome the deficiencies in the prior art, the present invention provides a fault diagnosis method, system, device and storage medium for an energy storage power station based on parameter identification and penalty strategy to realize real-time monitoring and diagnosis of the faults of energy storage cluster lithium batteries in the energy storage system.

[0007] To achieve the above-mentioned invention objectives and solve the technical problems, the following technical solutions are adopted:

[0008] The present invention proposes a fault diagnosis method for energy storage power stations based on parameter identification and penalty strategies, including the following steps:

[0009] Step 1: Collection of training sample data for training the fault detection model and online real-time detection data.

[0010] Step 2: According to the voltage and current collected at the current moment, use an online real-time parameter identification algorithm based on the alternating generalized least squares method with a forgetting factor to extract four identification features and construct a feature set.

[0011] Step 3: Based on the weights of different clustering algorithms, calculate the distances between the detection samples and different cluster centers of different clustering algorithms, and multiply by the weights of the clustering algorithms to obtain the final distances, and identify whether the samples are faulty.

[0012] Step 4: Update the integrated weights of the clustering algorithms and the cluster centers by relying on the penalty strategy of the integrated weights of the clustering algorithms and the update strategy of the cluster centers.

[0013] Further, in Step 1, in the laboratory environment, different charging or discharging parameters are set for the lithium iron phosphate battery to simulate overcharging and over-discharging faults of the battery in actual working conditions, and the corresponding current and voltage are recorded. The data samples in actual working conditions and the samples in the laboratory environment are combined as training samples to expand the data volume; the voltage and current of online real-time detection are collected in real time through the BMS acquisition device.

[0014] Further, in Step 2, the current and voltage of the data samples collected in real time by the BMS acquisition device are input into an online real-time parameter identification algorithm based on the alternating generalized least squares method with a forgetting factor. This algorithm identifies the ohmic internal resistance , polarization resistance , polarization capacitance and the open-circuit voltage of the battery of the four features, and combine them with the collected voltage and current to form a feature set for battery cluster fault diagnosis. Formulas (1) and (2) represent the calculation methods of the various parameters in the equivalent circuit model when the battery working behavior is characterized by the equivalent circuit model;

[0015] The voltage at the battery terminal can be obtained from the equivalent circuit model as:

[0016] (1) where represents the open-circuit voltage of the battery, represents the voltage across the polarization resistance;

[0017] The voltage across the capacitor The calculation method is as follows:

[0018] (2) Among them, 、 is positive in the charging state and negative in the discharging state; after identifying 、 、 、 four parameters by the alternating generalized least squares method with forgetting factor, the ohmic internal resistance , polarization resistance , polarization capacitance and the open-circuit voltage of the battery can be identified and used for the fault diagnosis of the battery cluster. Some calculation formulas are as shown in Equation (3):

[0019] (3).

[0020] Furthermore, in step 3, the extracted features are input into the clustering algorithm that has been trained with training samples, and the distances between the features and the clustering centers of different clustering algorithms are calculated respectively. The final distances between the features and the clustering centers of overcharge fault, over-discharge fault, and normal state are calculated through an integration strategy , and the category to which the battery cluster belongs is determined according to the distance magnitude, completing the fault diagnosis of the battery cluster, and classifying and storing the battery cluster data into the database The calculation method of

[0021] (4) is as shown in the formula (4):

[0022] Among them, represents the weight of the K-means algorithm, represents the weight of the spectral clustering algorithm, represents the distance between the battery cluster and the overcharge fault clustering center in the K-means algorithm, represents the distance between the battery cluster and the over-discharge fault clustering center in the K-means algorithm, represents the distance between the battery cluster and the normal state clustering center in the K-means algorithm; represents the distance between the battery cluster and the overcharge fault clustering center in the spectral clustering algorithm, represents the distance between the battery cluster and the over-discharge fault clustering center in the spectral clustering algorithm, represents the distance between the battery cluster and the normal state clustering center in the spectral clustering algorithm. The distance between the battery cluster and the clustering center uses the Euclidean distance.

[0023] Further, in step 4, a penalty strategy for the integrated weights of the clustering algorithms and an update strategy for the cluster centers are set according to the final diagnosis result of the sample and the storage size of the sample: when the final category of the sample is the same as the categories calculated by the K-means algorithm and the spectral clustering algorithm or when both algorithms are different, the weight of the K-means clustering algorithm and the weight of the spectral clustering algorithm remain unchanged at their current values;

[0024] In the initial state, and are both 0.5;

[0025] When the K-means algorithm is the same and the spectral clustering algorithm is different, the weight of the K-means algorithm is reduced by 0.1 based on the current value, and the weight of the spectral clustering algorithm is increased by 0.1 based on the current value;

[0026] When the spectral clustering algorithm is the same and the K-means algorithm is different, the weight of the spectral clustering algorithm is reduced by 0.1 based on the current value, and the weight of the K-means algorithm is increased by 0.1 based on the current value;

[0027] The increase and decrease amplitudes of the weights are both 0.1, and the sum of the two weights remains 1 unchanged. Based on the final detection result of the data of the sample to be detected, the weights of the two clustering algorithms are modified according to the above weight update strategy for the state discrimination of the subsequent battery clusters to be detected.

[0028] Further, when the integrated weight of any one clustering algorithm drops to 0 or the number of detected data samples reaches a multiple of 50, the cluster centers of the K-means algorithm and the spectral clustering algorithm are recalculated using the currently stored data samples respectively, and the integrated weight values of the two algorithms also become 0.5. Thus, the update of the clustering algorithm parameters is completed.

[0029] Based on the same concept, the present invention also provides an energy storage power station fault diagnosis system based on parameter identification and penalty strategy, including:

[0030] An energy storage power station battery cluster voltage and current acquisition module for online real-time detection of the acquisition of battery cluster state data and inputting the data collected by the BMS acquisition device in real time into the feature extraction module;

[0031] A feature set construction module for fault detection, which is used to construct a feature set for detecting the fault state of the battery cluster, input the voltage and current of the battery cluster into the online real-time parameter identification algorithm of the alternating generalized least squares method with a forgetting factor, and obtain the ohmic internal resistance and polarization resistance , polarization capacitance and the open-circuit voltage of the battery Four identification features, together with the directly collected current and voltage, form a feature set for fault detection;

[0032] The battery cluster fault diagnosis module is used to output the fault detection status of the battery cluster, including overcharge status, overdischarge status, and normal status; specifically, it includes: inputting the feature set data of the constructed battery cluster into the K-means algorithm and spectral clustering algorithm that have been trained with training samples respectively, and integrating the output results of the two algorithms according to the set integration rules to output the final detection status of the battery cluster;

[0033] The clustering center and integration weight update module is used to update the clustering center and integration weight. Among them, when the number of battery cluster samples detected by the fault detection model in real time reaches a multiple of 50 or when the integration weight of any one clustering algorithm drops to 0, the clustering center is updated; the integration weights of the two clustering algorithms are updated respectively according to the integration weight penalty strategy.

[0034] Based on the same concept, the present invention also provides a computer device, including:

[0035] A memory, where the memory is used to store a processing program;

[0036] A processor, when the processor executes the processing program, it implements the energy storage power station fault diagnosis method based on parameter identification and penalty strategy described in any one of the above.

[0037] Based on the same concept, the present invention also provides a readable storage medium, on which a processing program is stored, and when the processing program is executed by a processor, it implements the energy storage power station fault diagnosis method based on parameter identification and penalty strategy described in any one of the above.

[0038] Due to the above technical solutions, the present invention has the following advantages and positive effects compared with the prior art:

[0039] Fault diagnosis method for energy storage power station based on parameter identification and penalty strategy. By combining overcharge fault, over-discharge fault, and normal state data samples collected respectively under actual working conditions and laboratory environment, the amount of training sample data is expanded to prevent overfitting or underfitting of the fault detection model. The voltage and current collected in real time are input into the online real-time parameter identification algorithm of the recursive least squares method with forgetting factor to identify four characteristics reflecting the characteristics of the battery cluster, and combined to form a feature set for battery cluster fault diagnosis. This parameter identification algorithm has the accuracy and stability of identification under variable temperature and different working conditions, high identification accuracy and can be calculated in real time, expands the feature set of battery cluster fault diagnosis, and at the same time reduces the delay of detecting faults. Based on the idea of ensemble learning, the final result of the fault detection model is obtained by synthesizing the discriminant results of different fault diagnosis clustering algorithms to improve the accuracy of diagnosis. At the same time, the integrated weight penalty strategy and clustering center update strategy of the clustering algorithm are set to maximize the robustness and reliability of the fault detection model. Finally, the battery cluster data samples after diagnosis are stored by fault and the laboratory data corresponding to the faults are deleted to expand the amount of training sample data and improve the adaptability of the fault detection model to actual working condition data. Brief Description of the Drawings

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative efforts. In the drawings:

[0041] Figure 1 is the Thevenin equivalent circuit model diagram of the present invention;

[0042] Figure 2 is the flow chart of the online parameter identification algorithm based on the recursive least squares method with forgetting factor of the present invention;

[0043] Figure 3 is the flow chart of the penalty strategy for the weight of the clustering algorithm of the present invention;

[0044] Figure 4 is the framework schematic diagram of the fault diagnosis method for energy storage power station based on parameter identification and penalty strategy of the present invention. Detailed Embodiments

[0045] The technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0046] The first embodiment

[0047] Figures 1-4 As shown, this embodiment discloses a fault diagnosis method for an energy storage power station based on parameter identification and a penalty strategy, including the following steps:

[0048] Step 1: Acquisition of training sample data and online real-time detection data for training a fault detection model;

[0049] Specifically, in Step 1, the data samples are divided into training samples and test samples. The voltage and current of the battery cluster collected need to be input into the parameter identification algorithm. Among them, the training samples are used to determine the clustering center and related parameters in the clustering algorithm. The test samples can be regarded as the samples input when the fault detection model is officially running, and are the voltage and current of the battery cluster collected in real time by the BMS acquisition device. Therefore, the training samples are labeled samples, and the test samples are unlabeled.

[0050] In order to make the fault detection model more robust and stable, a large number of training samples with different battery cluster state labels are required for training to iteratively update the clustering parameters. In the present invention, in addition to collecting the voltage and current data of battery clusters with different faults in actual working conditions, the corresponding battery cluster voltage and current can also be obtained by simulating battery faults in the laboratory. The steps for designing battery over-discharge and over-charge faults in the laboratory environment are described as follows:

[0051] In the over-discharge experiment, first charge the battery to full (SOC = 100%) by the constant current and constant voltage method, and then discharge the battery with a fixed-rate current without controlling the discharge cut-off voltage, and control the discharge time to achieve over-discharge of the battery with different discharge depths. Different values can be set for the discharge depth, such as 100%, 105%, 110%, 115%, etc. The corresponding discharge capacity can be obtained through the discharge depth. Based on the above settings, the battery is discharged cyclically and the over-discharge current and voltage of the battery are recorded.

[0052] In the over-charge experiment, first charge the battery to full (SOC = 100%) by the constant current and constant voltage method, and then charge the battery with a fixed-rate current at a constant current. Set the SOC of the over-charge amount and the charge cut-off voltage respectively. For example, the SOC can be set to 105%, 110%, 115%, and the charge cut-off voltage can be set to 4.71V, 5.11V, 5.12V, etc. Based on the above settings, the battery is charged cyclically and the over-charge voltage and current of the battery are recorded.

[0053] The data samples in the actual working conditions and the data samples in the laboratory environment are centralized and used as training samples together to expand the data volume, so that the fault detection model can be better trained and prevent overfitting and underfitting of the fault detection model.

[0054] Step 2: According to the voltage and current collected at the current moment, use an online real-time parameter identification model based on the alternating generalized least squares method with a forgetting factor to extract four identification features and construct a feature set;

[0055] Furthermore, in Step 2, the accuracy of lithium battery dynamic modeling is a prerequisite for the stable and reliable operation of the energy storage system. Through parameter identification methods, the state information of the energy storage power station can be obtained, including parameters such as the internal resistance, capacity, and health status of the battery cluster. The currently commonly used lithium-ion battery model is the Thevenin equivalent circuit model, which converts the lithium-ion battery into a circuit model composed of electrical components such as a voltage source, resistor, and capacitor. By measuring parameters such as the voltage and current of the battery under different conditions, algorithms such as the least squares method and Kalman filter are used to perform online identification of the parameters in the equivalent circuit model. The equivalent circuit model can relatively accurately reflect the dynamic behavior of the battery and is suitable for real-time monitoring and management of the battery. The Thevenin equivalent circuit model has high identification accuracy, low complexity, and good results in lithium battery modeling. Therefore, in the present invention, the Thevenin equivalent circuit model is used to model the energy storage power station. The Thevenin equivalent circuit model is as Figure 1 shown.

[0056] The voltage and current data of the battery cluster are collected through the BMS acquisition device and input into the online parameter identification algorithm of the alternating generalized least squares method with a forgetting factor to obtain the ohmic internal resistance , polarization resistance , polarization capacitance , and open-circuit voltage of the battery cluster These four identification parameters. Finally, the features used for battery fault diagnosis consist of the voltage, current, , , , These six features of the battery cluster. Formulas (1) and (2) represent the calculation methods of the various parameters in the equivalent circuit model when using the equivalent circuit model to characterize the battery operating behavior:

[0057] From the equivalent circuit model, the voltage at the battery terminal is:

[0058] (1)

[0059] where represents the battery open-circuit voltage, and represents the voltage across the polarization resistance;

[0060] The voltage across the capacitor is calculated as follows:

[0061] (2)

[0062] Among them, and are positive in the charging state and negative in the discharging state;

[0063] The least squares method, the recursive least squares method, or the recursive least squares method with a forgetting factor are often used in the online parameter identification of energy storage power stations based on the equivalent circuit model. However, the above algorithms are greatly affected by external noise and have low robustness, which easily leads to a reduction in the parameter identification accuracy in the equivalent circuit model. Based on previous research, the present invention uses the alternating generalized least squares method with a forgetting factor to complete the parameter identification of the energy storage power station. This algorithm has high robustness and can achieve relatively accurate online parameter identification.

[0064] The alternating generalized least squares method with a forgetting factor solves the characteristic of time-varying parameters by introducing a forgetting factor The forgetting factor assigns weights to data at different times, weakens the role of past data depending on the weight value, strengthens the role of new data, and avoids the covariance matrix and the gain matrix from decaying and causing the algorithm to lose its correction function.

[0065] is an empirically selected parameter, and its value range is (0, 1]. The smaller it is, the faster the old data is forgotten. The recursive formula of the alternating generalized least squares method with a forgetting factor is as follows:

[0066] (3)

[0067] Among them:

[0068] (4)

[0069] (5)

[0070] and are the parameter estimation values of the non-stationary system, represents the difference between the actual value and the predicted value, also known as the residual, and represent the measured values of the system, and are the intermediate quantities at the (N + 1)th moment;

[0071] From the Thevenin equivalent circuit model, the relationship between the voltage U and the current in the complex frequency domain is as follows:

[0072] (6)

[0073] The discrete calculation of the above formula using the bilinear formula gives:

[0074] (7)

[0075] Where:

[0076] (8)

[0077] (9)

[0078] (10)

[0079] Based on the above formula, an unsteady system equation for the Thevenin equivalent circuit model is obtained. is the time parameter, which can be set according to the actual situation. At the same time, the open-circuit voltage within a period of time is set as a constant value, and the formula is as shown in Equation (11):

[0080] (11)

[0081] Let

[0082] , and is also a parameter to be identified;

[0083] (12)

[0084] is the measured value of the system output, that is, the terminal voltage of the battery cluster,

[0085] is the parameter vector to be identified,

[0086] is the input and output vector of the battery system. In Equation (12), , , , are the identified parameters. The online parameter identification algorithm process based on the alternating generalized least squares method with a forgetting factor is as shown in Figure 2 , where represents the unit delay operator, represents the difference between the actually measured parameter value and the identified parameter, also known as the residual.

[0087] In the present invention, the two basic characteristics of the current and voltage of the battery cluster of the energy storage power station measured directly, as well as the ohmic internal resistance identified in real time online using the Thevenin equivalent circuit model and the alternating generalized least squares method with a forgetting factor, the polarization resistance , polarization capacitance , open circuit voltage of the battery The four features are integrated together as a feature set for anomaly detection of the battery cluster in the energy storage power station. The total number of features in this feature set is 6. To prevent low detection accuracy caused by too few features, the battery cluster is detected from multiple dimensions to improve the detection accuracy and stability.

[0088] Step 3: Based on the weights of different clustering algorithms, calculate the distances between the detection samples and different cluster centers of different clustering algorithms, and multiply by the weights of the clustering algorithms to obtain the final distances, and identify whether the samples are faulty;

[0089] Furthermore, in step 3, in order to reduce the misjudgment of a single clustering algorithm and improve the reliability of the detection results, the idea of ensemble learning is applied in the present invention, and the K-means algorithm and the spectral clustering algorithm are respectively used to complete the anomaly detection of the energy storage power station. The initial values of the influence weights of the K-means algorithm and the spectral clustering algorithm on the final result are both 0.5, and the sum of the weights of the two algorithms is 1. In the present invention, the states of the energy storage power station are divided into three categories, namely overcharge fault, over-discharge fault, and normal state. Therefore, in the clustering algorithm, the number of cluster centers is selected as 3. The following is a detailed description of the method flow of this part:

[0090] The clustering steps of the K-means algorithm are as follows:

[0091] (1) Initialization of cluster centers: Set the number of initial cluster centers to 3, and randomly select 3 data samples as the initial cluster centers;

[0092] (2) Assignment of data samples: Calculate the distances between each data point and the cluster centers, and assign it to the cluster center with the smallest distance. The distance is calculated using the Euclidean distance, and the calculation method is shown in formula (13):

[0093] (13)

[0094] where n is the total number of features, represents the Euclidean distance between point x and point y, represents the i-th attribute of sample x, represents the i-th attribute of sample y;

[0095] (3) Update of cluster centers: Recalculate the cluster centers of each cluster, and take the average value of all data points within the cluster as the new cluster center;

[0096] (4) Iteration: Repeat steps (2) and (3) until the cluster centers no longer change significantly or reach the preset number of iterations.

[0097] In the spectral clustering algorithm, in addition to determining the clustering centers, the method for generating the similarity matrix and the graph cut method need to be determined. In the present invention, a fully connected method based on Gaussian kernel distance is selected to generate the similarity matrix, and the Ncut method is used to complete the graph cut. The steps of this clustering algorithm are as follows:

[0098] (1) Construct the similarity matrix S of the samples based on the fully connected method of Gaussian kernel distance;

[0099] (2) Construct the adjacency matrix according to the similarity matrix, and generate the corresponding degree matrix D;

[0100] (3) Calculate the Laplacian matrix L;

[0101] (4) Standardize the Laplacian matrix to generate the standardized Laplacian matrix , and the calculation method is shown in formula (14):

[0102] (14)

[0103] (5) Perform eigen-decomposition on the matrix

[0104] to obtain the corresponding eigenvectors. Usually, take the first k non-zero eigenvectors of the matrix

[0105] , where K is the number of clustering centers;

[0106] (6) Use the eigenvectors as the new data representation, and use the fuzzy C-means clustering algorithm to cluster these eigenvectors, and finally divide the data points into k different clustering clusters;

[0107] Cluster the existing data samples using the K-means algorithm and the spectral clustering algorithm respectively. Each clustering algorithm can obtain 3 different clustering centers, namely the clustering center of overcharge fault, the clustering center of over-discharge fault, and the clustering center of normal state;

[0108] Complete parameter identification for the current and voltage data of the battery cluster collected in real time to obtain 4 identification features. Based on the above 6 features, calculate the distances between this sample and the 3 clustering centers under the K-means algorithm, and the distances between this sample and the 3 clustering centers under the spectral clustering algorithm. The distances under different algorithms are multiplied by the corresponding clustering algorithm weights and then added to obtain the final distance between this sample and each clustering center. Assign the category of this sample to the clustering center with the closest distance to obtain the status label of this sample. Thus, the status diagnosis of this battery cluster at the current moment is completed, and the feature set data and status label of this sample are stored in the database. One data sample obtained based on the laboratory environment is deleted from the database. The calculation method of the final distance from the sample to the clustering center is shown in formula (15):

[0109] (15)

[0110] Wherein, represents the weight of the K-means algorithm, represents the weight of the spectral clustering algorithm, represents the distance from the battery cluster to the overcharge fault clustering center in the K-means algorithm, represents the distance from the battery cluster to the over-discharge fault clustering center in the K-means algorithm, represents the distance from the battery cluster to the normal state clustering center in the K-means algorithm; represents the distance from the battery cluster to the overcharge fault clustering center in the spectral clustering algorithm, represents the distance from the battery cluster to the over-discharge fault clustering center in the spectral clustering algorithm, represents the distance from the battery cluster to the normal state clustering center in the spectral clustering algorithm. The distance from the battery cluster to the clustering center uses the Euclidean distance.

[0111] Design a strategy for calculating the final battery cluster fault diagnosis result based on the integrated weight of the clustering algorithm according to the idea of ensemble learning, and comprehensively consider the discriminant advantages of different clustering algorithms to effectively improve the discriminant accuracy of the fault detection model.

[0112] Step 4: Update the integrated weight of the clustering algorithm and the clustering center by relying on the penalty strategy of the integrated weight of the clustering algorithm and the update strategy of the clustering center.

[0113] Furthermore, in Step 4, according to the final diagnosis result of the sample and the size of the sample storage, a penalty strategy for the integrated weight of the clustering algorithm and an update strategy for the clustering center are respectively set, and the clustering parameters are updated according to these two strategies to improve the robustness and accuracy of the fault detection model, as Figure 3 shown:

[0114] When the final category of the sample is the same as the categories calculated by the K-means algorithm and the spectral clustering algorithm or different from both algorithms, the weight of the K-means algorithm and the weight of the spectral clustering algorithm remain unchanged at the current value. In the initial state, and are both 0.5;

[0115] When the K-means algorithm is the same and the spectral clustering algorithm is different, the weight of the K-means algorithm is reduced by 0.1 based on the current value, and the weight of the spectral clustering algorithm is increased by 0.1 based on the current value;

[0116] When the spectral clustering algorithm is the same and the K-means algorithm is different, the weight of the spectral clustering algorithm is reduced by 0.1 based on the current value, and the weight of the K-means algorithm is increased by 0.1 based on the current value;

[0117] The increase and decrease amplitude of the weights are both 0.1, and the sum of the two weights remains 1 unchanged. The final detection result based on the data of the sample to be detected modifies the weights of the two clustering algorithms according to the above weight update strategy for the subsequent state discrimination of the battery cluster to be detected.

[0118] When the weight of any algorithm drops to 0 or the number of detected data samples reaches a multiple of 50 (for example, the number of detected samples is 50, 100, 150,... etc.), the clustering centers of the K-means algorithm and the spectral clustering algorithm are recalculated respectively using the currently stored data samples, and the weight values of the two algorithms also become 0.5. The framework of the energy storage power station fault diagnosis method based on parameter identification and penalty strategy is as Figure 4 shown.

[0119] The second embodiment

[0120] This embodiment provides an energy storage power station fault diagnosis system based on parameter identification and penalty strategy, including:

[0121] An energy storage power station battery cluster voltage and current acquisition module, which is used for online real-time detection of the acquisition of battery cluster state data, and inputs the data collected by the BMS acquisition device in real time into the feature extraction module, where the model is configured in the EMS system;

[0122] A feature set construction module for fault detection, which is used to construct a feature set for detecting the fault state of the battery cluster, inputs the voltage and current of the battery cluster into the online real-time parameter identification algorithm of the alternating generalized least squares method with a forgetting factor, and obtains the ohmic internal resistance , polarization resistance , polarization capacitance and the open circuit voltage of the battery 4 identification features, and jointly form a feature set for fault detection with the directly collected current and voltage;

[0123] A battery cluster fault diagnosis module, which is used to output the fault detection state of the battery cluster, including overcharge state, over-discharge state, and normal state; specifically including: respectively inputting the constructed feature set data of the battery cluster into the K-means algorithm and the spectral clustering algorithm that have been trained by training samples, and integrating the output results of the two algorithms according to the set integration rule to output the final detection state of the battery cluster;

[0124] The clustering center and integrated weight update module is used to update the clustering center and integrated weight. Among them, when the number of battery cluster samples detected by the fault detection model in real time reaches a multiple of 50 or when the integrated weight of any one clustering algorithm drops to 0, the clustering center is updated; the integrated weights of the two clustering algorithms are updated according to the integrated weight penalty strategy.

[0125] The energy storage power station fault diagnosis system based on parameter identification and penalty strategy in this embodiment uses the online parameter identification method of the alternating generalized least squares method with a forgetting factor, which has the accuracy and stability of identification under variable temperature and different working conditions, and can identify the ohmic internal resistance online in real time. , polarization resistance , polarization capacitance , battery open-circuit voltage These four features expand the feature set for battery cluster fault diagnosis and also reduce the delay in detecting faults. By integrating two clustering algorithms and setting a penalty strategy for the integrated weights of the two algorithms and an update strategy for the clustering center, the false detection rate of battery cluster faults is reduced. By storing the detected data samples in the database and deleting the data in the previous laboratory environment, the data samples are getting closer to the actual working conditions, increasing the robustness of the model to the actual working condition data. This system is deployed on the EMS side.

[0126] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated here.

[0127] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs, etc., which can store program codes.

[0128] Those of ordinary skill in the art will realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A professional technician can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0129] In the embodiments provided in this application, it should be understood that the disclosed device / network device and method can be implemented in other ways. For example, the device / network device embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical or other form.

[0130] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0131] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of this application, and should all be included in the protection scope of this application.

Claims

1. A fault diagnosis method for energy storage power station based on parameter identification and penalty strategy, characterized in that: The following steps are involved: Step 1: Collect training sample data and online real-time detection data for training fault detection models; Step 2: According to the voltage and current collected at the current moment, an online real-time parameter identification algorithm based on the alternating generalized least squares method with a forgetting factor is used to extract four identification features and construct a feature set; Step 3: Based on the weights of different clustering algorithms, calculate the distance between the detection sample and different clustering centers of different clustering algorithms, and multiply it by the weight of the clustering algorithm to get the final distance to identify whether the sample is faulty; Step 4: Update the clustering algorithm ensemble weights and cluster centers based on the penalty strategy of the clustering algorithm ensemble weights and the update strategy of the cluster centers.

2. The energy storage power station fault diagnosis method based on parameter identification and penalty strategy according to claim 1 is characterized in that: In step 1, different charging or discharging parameters are set for the lithium iron phosphate battery in a laboratory environment to simulate the overcharging and over-discharging faults of the battery in actual working conditions and record the corresponding current and voltage. The data samples in the actual working conditions and the samples in the laboratory environment are combined as training samples to expand the data volume; the voltage and current detected online in real time are collected in real time through the BMS acquisition device.

3. The energy storage power station fault diagnosis method based on parameter identification and penalty strategy according to claim 1 is characterized in that: In step 2, the current and voltage of the data samples collected in real time by the BMS acquisition device are input into an online real-time parameter identification algorithm based on the alternating generalized least squares method with a forgetting factor, and the algorithm identifies the ohmic internal resistance , Polarization resistance , polarized capacitor and battery open circuit voltage The four features, combined with the collected voltage and current, form a feature set for battery cluster fault diagnosis. Formulas (1) and (2) represent the calculation method of each parameter in the equivalent circuit model when the equivalent circuit model is used to characterize the battery working behavior; The voltage at the battery terminal can be obtained from the equivalent circuit model: for: （1） in, Indicates the battery open circuit voltage, Represents the voltage across the polarization resistor; The voltage across the capacitor The calculation method is as follows: （2） in, , It is positive in the charging state and negative in the discharging state; Based on the alternating generalized least squares method with forgetting factor, , , , After four parameters, the ohmic internal resistance can be identified , polarization resistance , polarized capacitance and battery open circuit voltage For battery cluster fault diagnosis, some calculation formulas are as follows: （3）。 4. The energy storage power station fault diagnosis method based on parameter identification and penalty strategy according to claim 3 is characterized in that: In step 3, the extracted features are input into the clustering algorithm that has been trained with the training samples, and the distances between the clustering centers of different clustering algorithms are calculated respectively. The final distances between the three clustering centers of overcharge fault, over-discharge fault, and normal state are calculated through the integrated strategy. , determine the category of the battery cluster according to the distance, complete the fault diagnosis of the battery cluster, and store the battery cluster data in the database by category, The calculation method is shown in formula (4): （4） in, represents the weight of the K-means algorithm, represents the weight of the spectral clustering algorithm, represents the distance from the battery cluster to the overcharge fault cluster center in the K-means algorithm, Represents the distance from the battery cluster to the over-discharge fault cluster center in the K-means algorithm, Indicates the distance from the battery cluster to the normal state cluster center in the K-means algorithm; represents the distance from the battery cluster to the overcharge fault cluster center in the spectral clustering algorithm, represents the distance from the battery cluster to the over-discharge fault cluster center in the spectral clustering algorithm, Represents the distance from the battery cluster to the normal state cluster center in the spectral clustering algorithm. The distance from the battery cluster to the cluster center uses the Euclidean distance.

5. The energy storage power station fault diagnosis method based on parameter identification and penalty strategy according to claim 1 is characterized in that: In step 4, the penalty strategy for clustering algorithm integration weight and the update strategy for cluster center are set according to the final diagnosis result of the sample and the storage size of the sample: When the final category of the sample is the same as the category calculated by the K-means algorithm and the spectral clustering algorithm, or the two algorithms are different, the K-means clustering algorithm weight and spectral clustering algorithm weights Keep the current value unchanged; in the initial state, and All are 0.5; When the K-means algorithms are the same and the spectral clustering algorithms are different, the K-means algorithm weight Reduce the weight of the spectral clustering algorithm by 0.1 based on the current value Add 0.1 to the current value; When the spectral clustering algorithms are the same and the K-means algorithms are different, the weight of the spectral clustering algorithm is Reduce the K-means algorithm weight by 0.1 based on the current value Add 0.1 to the current value; The increase or decrease of the weight is 0.1, and the sum of the two weights remains unchanged at 1. The weights of the two clustering algorithms are modified according to the above weight update strategy based on the final detection results of the sample data to be detected, which are used for the subsequent state judgment of the battery cluster to be detected.

6. The energy storage power station fault diagnosis method based on parameter identification and penalty strategy according to claim 5 is characterized in that: When the integration weight of any clustering algorithm drops to 0 or the number of detected data samples reaches a multiple of 50, the cluster centers of the K-means algorithm and the spectral clustering algorithm are recalculated using the currently stored data samples, and the integration weight values ​​of the two algorithms are also changed to 0.

5. At this point, the clustering algorithm parameter update is completed.

7. The energy storage power station fault diagnosis system based on parameter identification and penalty strategy is characterized by: include: The battery cluster voltage and current acquisition module of the energy storage power station is used to collect the battery cluster status data online in real time, and input the data collected by the BMS acquisition device in real time into the feature extraction module; The feature set construction module for fault detection is used to construct a feature set for detecting the fault status of the battery cluster. The voltage and current of the battery cluster are input into the online real-time parameter identification algorithm of the alternating generalized least squares method with forgetting factor to obtain the ohmic internal resistance. , Polarization resistance , polarized capacitor and battery open circuit voltage The four identification features, together with the directly collected current and voltage, form a feature set for fault detection; The battery cluster fault diagnosis module is used to output the fault detection status of the battery cluster, including overcharge status, over-discharge status, and normal status. Specifically, the feature set data of the battery cluster that has been constructed is input into the K-means algorithm and the spectral clustering algorithm that have been trained with training samples, and the output results of the two algorithms are integrated according to the set integration rules to output the final detection status of the battery cluster. The cluster center and integrated weight update module is used to update the cluster center and integrated weight. When the number of battery cluster samples detected by the fault detection model in real time reaches a multiple of 50 or when the integrated weight of any clustering algorithm drops to 0, the cluster center is updated; the integrated weights of the two clustering algorithms are updated separately according to the integrated weight penalty strategy.

8. A computer device, characterized in that: include: A memory, the memory being used to store a processing program; A processor, wherein when executing the processing program, the processor implements the energy storage power station fault diagnosis method based on parameter identification and penalty strategy as described in any one of claims 1 to 6.

9. A readable storage medium, characterized in that: The readable storage medium stores a processing program, and when the processing program is executed by the processor, the energy storage power station fault diagnosis method based on parameter identification and penalty strategy as described in any one of claims 1 to 6 is implemented.

Images