Energy storage system insulation detection method based on artificial intelligence

Abstract

The invention discloses an energy storage system insulation detection method based on artificial intelligence, and the method comprises the steps: collecting the electric digital data and text data of an energy storage system, and carrying out the vectorization processing of the text data based on a BERT model, so as to facilitate the subsequent modeling analysis, then, key features related to the insulation state are extracted from the electrical digital data and the vectorized text data through a GBDT algorithm; based on the selected features, constructing a GBDT model for predicting the insulation state of the energy storage system; preprocessing the collected data, and dividing the data into a training set and a test set; adopting a K-fold cross validation and grid search method to train the GBDT model; and evaluating the trained model by using test set data, and selecting MAE and Ras evaluation indexes. According to the invention, the accuracy and prediction capability of insulation detection are obviously improved, preventive maintenance is realized, the service life of equipment is prolonged, and the overall safety management level of the energy storage system is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage insulation detection, and in particular to an energy storage system insulation detection method based on artificial intelligence. Background Art

[0002] With the continuous development of energy technology, energy storage systems, as a key link in energy conversion and storage, play an increasingly important role in power systems. However, the safety and stability of energy storage systems are directly related to the safety and reliability of power grid operation. Therefore, it is particularly important to conduct efficient insulation testing on energy storage systems.

[0003] The traditional insulation detection method of energy storage system mainly relies on comparing the data collected by the sensor with the preset threshold to determine the insulation status. Although this method is simple and easy to implement, it has significant limitations: on the one hand, it cannot fully consider the comprehensive impact of multiple factors such as environmental factors, equipment aging, and changes in operating conditions on insulation performance, resulting in low accuracy of the detection results and prone to missed detection or misjudgment; on the other hand, the traditional method lacks intelligence and adaptability, and it is difficult to adapt to the complex and changeable system operating environment, which limits its role in improving the safety and stability of energy storage systems.

[0004] In order to overcome the shortcomings of traditional methods, in recent years, the rise of artificial intelligence technology, especially deep learning and ensemble learning algorithms, has provided new ideas and technical means for insulation detection of energy storage systems. Artificial intelligence technology can discover hidden laws and features in data through deep mining and learning of large amounts of data, thereby achieving a more accurate and comprehensive judgment of the insulation status.

[0005] Specifically, deep learning technology can automatically extract high-level features from insulation detection data and achieve high-precision identification of insulation status by building complex neural network models. This technology not only improves the accuracy of detection, but also enhances the self-learning and adaptive capabilities of the system, enabling it to better cope with complex and changing operating environments. At the same time, integrated learning algorithms such as gradient boosted decision trees (GBDT) can effectively handle nonlinear relationships between features, improve the generalization and robustness of the model, and further enhance the reliability of insulation detection.

[0006] Although the application prospects of artificial intelligence technology in insulation detection of energy storage systems are broad, research in this field still faces many challenges. For example, most existing studies focus on the application of a single technology and lack a comprehensive solution that effectively integrates multiple artificial intelligence technologies; at the same time, in-depth analysis and modeling of the complex relationship between characteristic parameters and insulation status of energy storage systems also need to be strengthened. Moreover, most studies are based on modeling and analysis of digital data, and these studies have ignored some important text information. Therefore, proposing a reasonable and reliable method for predicting the insulation status of energy storage systems based on artificial intelligence technology not only has important academic value, but also has far-reaching practical significance for improving the safety and stability of energy storage systems and promoting the further development of energy storage technology. Summary of the invention

[0007] In view of this, the present invention proposes an energy storage system insulation detection method based on artificial intelligence to solve the problem that traditional methods in the prior art are difficult to fully and accurately reflect the insulation status of the energy storage system and lack intelligent improvement solutions.

[0008] The specific technical solutions of the present invention are as follows: An energy storage system insulation detection method based on artificial intelligence, comprising: Step 1: Collect the electrical digital data of the energy storage system and the text data of the energy storage system at the same time, vectorize the text data based on the BERT model, and then use the GBDT algorithm to extract key features related to the insulation state from the fusion vector of the electrical digital data and text data; Step 2, based on the features selected in step 1, a GBDT model is constructed to predict the insulation state of the energy storage system; Step 3, preprocessing the collected digital data and vectorized text data to divide them into training sets and test sets; Step 4: Use K-fold cross validation and grid search methods to train the GBDT model to optimize model parameters and improve prediction performance; Step 5: Use the test set data to evaluate the trained model, and select MAE and R² as evaluation indicators to measure the accuracy of model prediction.

[0009] Furthermore, the method also includes: step 6, using the feature importance evaluation function of the GBDT model to analyze the influence of each feature on the model prediction results, and displaying the interaction between features and the model decision path through visualization tools; based on the analysis results, proposing model optimization suggestions to improve the interpretability and credibility of the model results.

[0010] Furthermore, in step 1, the electrical digital data of the energy storage system includes: voltage data, covering the voltage of each part of the energy storage system to reflect the voltage stability and change; current data, recording the current of each part of the system to understand the current load and change trend; temperature data, collecting the temperature of each component inside the system to reflect the heat distribution and temperature change; frequency data, monitoring the frequency stability and fluctuation of the system; humidity data, recording the humidity of the system environment, because humidity changes may affect the insulation state; battery status data, including charge state and discharge state, for systems containing battery components; working mode data, recording the working state of the system, including charging and discharging modes; fault history data, including past fault records and maintenance history; text data including the model, basic configuration, and text information of historical faults of the energy storage device.

[0011] Furthermore, in the feature extraction process of step 1, iterative learning is performed by constructing multiple decision trees, and each decision tree is trained based on the residual of the previous tree to minimize the overall prediction error, thereby extracting the most critical feature set for predicting the insulation state of the energy storage system.

[0012] Furthermore, the preprocessing in step 3 includes: flexibly adjusting the ratio of the training set and the test set, dynamically setting the ratio of the training set and the test set according to the specific requirements of the energy storage system insulation detection task and the characteristics of the data, so as to balance the model training effect and the evaluation accuracy.

[0013] Furthermore, the preprocessing in step 3 includes: identifying and analyzing the pattern of missing values, and selecting the most appropriate filling strategy according to the specific distribution of the missing values, including filling the predicted value of the mean, median or mode.

[0014] Furthermore, step 4 also introduces random search or Bayesian optimization methods, combined with grid search, to more efficiently explore the optimal combination of GBDT model parameters.

[0015] Furthermore, the K-fold cross-validation in step 4 is to divide the training set data into K subsets on average, select K-1 subsets as training data each time, and the remaining subset as validation data, repeat K times, use the average validation error as the evaluation criterion of model performance, and combine with the grid search method to automatically adjust the parameters of the GBDT model to achieve the best prediction performance.

[0016] Furthermore, the parameters of the GBDT model include learning rate, number of trees, and depth of trees.

[0017] Furthermore, step 6 also includes: calculating and outputting the global importance ranking of each feature in the GBDT model to quantify the influence of each factor on the insulation state prediction result of the energy storage system.

[0018] The beneficial effects of the present invention are: (1) Improve detection accuracy and prediction ability: The GBDT model constructed by the present invention can establish a mapping relationship between the electrical digital data of the energy storage system and its insulation status with high precision, significantly improving the accuracy and prediction ability of insulation detection. This feature enables engineers to monitor and predict the insulation status of the energy storage system in real time, discover and identify potential safety hazards in a timely manner, buy valuable time for taking preventive measures, and effectively reduce the probability of accidents.

[0019] (2) Realize preventive maintenance and extend equipment life: Through a data-driven approach, the present invention realizes preventive monitoring of the insulation status of the energy storage system. Discovering potential problems in advance and taking corresponding maintenance measures not only avoids downtime losses caused by sudden failures, but also significantly extends the service life of the energy storage equipment and reduces the overall operation and maintenance costs.

[0020] (3) Enhance the safety management level of energy storage systems: This paper deeply analyzes the influence of key factors such as voltage, current, and temperature on the insulation state of energy storage systems, providing an important basis for improving the safety management level of energy storage systems. By comprehensively evaluating the multi-dimensional characteristic indicators of energy storage systems, it is possible to more comprehensively grasp the changing trend of insulation status, providing strong support for the formulation of targeted safety management strategies.

[0021] (4) Optimizing equipment management and maintenance services: The present invention comprehensively considers the complex relationship between the characteristic parameters of different dimensions of the energy storage system and the insulation status, and provides users with a more comprehensive and accurate equipment management and maintenance service solution. This comprehensive consideration method can more accurately capture the multi-faceted characteristics of system operation, thereby improving the accuracy and reliability of insulation status prediction, and providing a solid guarantee for the stable operation of the energy storage system.

[0022] (5) Promote the improvement of intelligence level: By applying advanced machine learning algorithms, the present invention continuously optimizes the insulation detection model and gradually improves the intelligence level of the detection method. This intelligent trend not only improves system performance and stability, but also provides users with a more convenient and efficient operation and maintenance experience, and promotes the continuous progress and development of insulation detection technology for energy storage systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0024] Figure 1It is a flow chart of the insulation detection method of energy storage system based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0025] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0026] The present invention proposes an energy storage system insulation detection method based on artificial intelligence, such as Figure 1 As shown. This method cleverly combines the GBDT (gradient boosted decision tree) algorithm with expertise in the field of power systems, aiming to significantly improve the safety and reliability of energy storage systems. High-precision sensors are used to collect key electrical digital data such as voltage, current, and temperature of the energy storage system in real time and store them in a database. Then, the powerful feature extraction capability of the GBDT algorithm is used to accurately screen out feature parameters closely related to the insulation status from massive data. After data preprocessing, multiple decision trees are iteratively constructed, and each tree is fitted on the residual of the previous tree. The model structure is continuously optimized, the prediction accuracy is gradually improved, and the model performance is optimized by adjusting hyperparameters (the number of trees, depth, and learning rate, etc.). Finally, the generalization ability of the model is verified using the test set to ensure its accuracy on unknown data. After the model training is completed, the model prediction results are analyzed and interpreted to help engineers understand the insulation status of the system, take maintenance measures in a timely manner, and ensure the safe operation of the system.

[0027] The specific steps of this method are as follows: Step 1: Comprehensive data collection and efficient feature extraction to build an insulation prediction model for energy storage systems.

[0028] First, it is necessary to comprehensively collect the electrical digital data of the energy storage system, which includes information in multiple dimensions such as voltage, current, temperature, frequency, and humidity. Voltage data reflects the voltage stability and changes of each part of the energy storage system, and is one of the important indicators for evaluating the insulation status of the system; current data records the current load and change trend of each part of the system, which helps to understand the operating status of the system; temperature data reveals the heat distribution and temperature changes of the internal components of the system, which is an important basis for judging the insulation status; frequency data helps monitor the frequency stability and fluctuations of the system; humidity data takes into account the impact of ambient humidity on the insulation status, which is a characteristic parameter that cannot be ignored. At the same time, collect text data such as the model, basic configuration, and fault history description of the energy storage system, and vectorize the text data based on the BERT model.

[0029] In addition, if the energy storage system includes battery components, the battery status data such as charge state, discharge state, etc. are also important independent variables, which are directly related to the health status and insulation performance of the battery. At the same time, it is also crucial to record the system's operating mode such as charging mode, discharging mode, etc., because the characteristics and insulation state of the system may be different under different operating modes. In addition, past fault records and maintenance history data are also valuable resources. They can help the model learn the failure modes and laws of the system and improve the accuracy and reliability of predictions.

[0030] In order to ensure the accuracy and completeness of the data, these data need to be collected in real time by high-precision and high-sensitivity sensors, and undergo preliminary data cleaning to remove outliers and noise. Subsequently, these data are stored in the database for subsequent processing and analysis.

[0031] In the feature selection stage, the advantages of the GBDT (gradient boosted decision tree) algorithm are used to mine key features related to insulation status from a large amount of electrical digital data and text data. As an integrated learning method, GBDT trains multiple decision tree models in series and combines their prediction results to improve overall performance. In the feature extraction process, GBDT can iteratively optimize the model, and each training will pay special attention to samples that were predicted incorrectly by the previous model, so as to extract more discriminative features. This method not only improves the accuracy and stability of the model, but also provides a powerful feature extraction capability for insulation detection of energy storage systems, ensuring that the model can better understand the data and mine the most representative features, thereby achieving accurate insulation status prediction.

[0032] Step 2: Construction of insulation status prediction model of energy storage system based on GBDT.

[0033] In this step, carefully collected and preprocessed data (including voltage, current, temperature, humidity, battery status, working mode and fault history data of the energy storage system) will be used as input features of the GBDT (gradient boosting decision tree) model, aiming to train a model that can accurately predict the insulation status of the energy storage system. The construction of this model will significantly improve the safety and reliability of the energy storage system, and provide strong guarantee for the stable operation of the system by timely warning of potential insulation failures.

[0034] The specific steps of the GBDT algorithm to build an insulation prediction model for energy storage systems are as follows: Input: training set , x i ∈R n ,y i ∈R, where R nrepresents n-dimensional real number space, R represents one-dimensional real number space, i represents the number of data, N represents the total amount of data, and n represents the dimension; the independent variable x i It represents a vector, and the value of each dimension in the vector represents the value of the voltage, current, temperature, battery status, working mode and fault history of the energy storage system; for The insulation state of the energy storage system corresponds to the following.

[0035] Output: Energy storage system insulation status prediction model .

[0036] After training, the model is able to receive new electrical digital data input and output the corresponding insulation status prediction results.

[0037] Step 21, Initialization ,in is the loss function; c represents the residual value to be fitted (similar to linear regression). The meaning of this formula is to select the residual value that minimizes the total loss, that is, to traverse all samples and estimate a better c so that the total loss calculated on c for all samples is the smallest, that is, the total residual fit is the best.

[0038] Step 22, iterative training (for ,in M is the number of iterations).

[0039] (1) For each sample in the training set , calculate the residual (The residual here is approximated by the negative gradient of the loss function): .

[0040] (2) Based on the previous step Fit a new regression tree, whose leaf node area is recorded as ,in J Representative m The number of leaf nodes in a tree.

[0041] (3) Yes ,calculate: .

[0042] (4) Update the current learner and obtain: .

[0043] Step 23: Obtain the final energy storage system insulation state prediction model This model is an integrated model composed of multiple decision trees, and each tree contributes to the final prediction result. When predicting the insulation status of a new sample, the model traverses all trees and performs a weighted summation of the prediction values ​​of each tree to finally get the prediction result.

[0044] Through the above steps, the GBDT prediction model of the insulation state of the energy storage system was successfully constructed. This model can make full use of the information in the data, capture complex nonlinear relationships, and achieve accurate prediction of the insulation state.

[0045] Step 3: Fine data preprocessing and division to adapt to the characteristics of the GBDT model.

[0046] In the process of using the gradient boosted decision tree (GBDT) method to build an energy storage system insulation state prediction model, data preprocessing is an indispensable link, which is directly related to the subsequent stability and prediction performance of the model. For a given data set, the following key data preprocessing steps need to be performed: Step 31, handling missing values: First, you must carefully check whether there are missing values ​​in the data set. The presence of missing values ​​will interfere with the training process of the model and may lead to inaccurate prediction results. In order to ensure the integrity of the data, missing values ​​need to be handled. A common and effective method is to use the mean value of the feature (or other statistics, such as median, mode, etc., depending on the distribution characteristics of the data) to fill in missing values. This method is based on the assumption that missing values ​​can be approximately represented by the average level of the feature, thereby preserving the overall distribution characteristics of the data to a certain extent.

[0047] Step 32, divide the training set and test set: Next, the processed data set needs to be divided into a training set and a test set. The purpose of this step is to evaluate the performance of the model on unseen data, that is, the generalization ability of the model. Usually, it is divided according to the ratio of 80% training set and 20% test set, but this ratio is not fixed and can be adjusted according to the specific data set size and project requirements. When dividing, the independence between the training set and the test set must be ensured to avoid any potential data leakage problems.

[0048] It is worth noting that although data standardization is a common step before model training in the field of machine learning, aiming to eliminate dimensional differences and scale effects between different features, the gradient boosted decision tree (GBDT) has a unique advantage: it does not require data standardization. This is because GBDT is a nonlinear model based on a tree structure, and its construction process mainly depends on the relative order of features rather than the specific numerical size. In the process of node division of the decision tree, the focus is on which feature can bring the greatest information gain or purity improvement when dividing, rather than the absolute value of the feature. Therefore, data standardization is not necessary for the construction and prediction of the GBDT model. This feature enables GBDT to show good robustness when dealing with features of different scales and ranges, without the need for complex linear transformations of features, so that it can more flexibly and effectively capture complex nonlinear relationships in the data.

[0049] Step 4: Training and parameter optimization of the insulation state prediction model of the energy storage system based on the gradient boosting decision tree.

[0050] When building an insulation state prediction model for energy storage systems, this application selected the gradient boosted decision tree (GBDT) as the core algorithm. During the training process of the GBDT model, its performance is affected by multiple key parameters. The reasonable setting of these parameters is crucial to improving the prediction accuracy and generalization ability of the model. The following are the main parameters involved in the GBDT model training process and their detailed descriptions: (1) n_estimators: This parameter represents the number of decision trees, that is, the total number of trees that the model will build during the training process, which can also be regarded as the number of model iterations. Generally speaking, the larger the value of n_estimators, the more details in the data the model can capture, thereby enhancing the model's fitting ability. However, an excessively large n_estimators value may cause the model to perform too well on the training set, while performance may degrade on unseen data, which is a phenomenon called overfitting. Therefore, when selecting n_estimators, it is necessary to comprehensively consider the model's fit and training efficiency to achieve the best balance. By default, the value of n_estimators is 100.

[0051] (2) learning_rate: The learning rate represents the contribution weight of each base learner (here each tree) in the overall iteration process of the model. It is used to control the step size of each tree model update, also known as the step size factor. The size of the learning rate directly affects the convergence speed and stability of the model. A smaller learning rate means that the model updates the weights at a smaller pace in each iteration, requiring more iterations to achieve the ideal fitting effect, but helps reduce the risk of overfitting. On the contrary, a larger learning rate can speed up training, but may increase the risk of overfitting. Therefore, when adjusting the learning rate, it is necessary to comprehensively consider the number of iterations and the size of the learning rate to find the optimal parameter combination. By default, the value of learning_rate is 0.1.

[0052] (3) subsample: This parameter is used to control the proportion of subsamples used by each base learner during training. By randomly sampling samples, subsample helps to increase the generalization ability of the model and reduce the occurrence of overfitting. The value range of subsample is between (0,1], and it is generally recommended to set it in the range of [0.5,0.8]. When the subsample value is 1, it means that in each iteration, the model will use the complete training set for training, that is, no samples will be sampled.

[0053] (4) max_depth: max_depth limits the depth of each decision tree and is an important parameter for controlling model complexity. When the depth is 1, the generated trees are all binary trees with only two leaf nodes, so there is no interaction between the variables. When the depth is 2, the generated tree model has interaction between two variables. By analogy, deeper base trees can explore very complex variable interactions, but they are also more likely to cause overfitting. Shallower trees (smaller max_depth) are computationally efficient, but may not fully capture the complex relationships in the data, resulting in underfitting of the model. On the contrary, overly deep trees (excessively large max_depth) can capture more details, but are prone to over-performing on the training set and degrading performance on the test set, i.e. overfitting. Therefore, choosing an appropriate max_depth value is crucial to balancing the model's fitting ability and complexity. By default, the value of max_depth is 3.

[0054] In the process of model parameter adjustment, in order to find the optimal parameter combination, this application adopts K-fold cross-validation and grid search methods. K-fold cross-validation divides the data set into K non-overlapping equal subsets, and uses one subset as the test set in each iteration, and the remaining K-1 subsets as the training set to evaluate the performance of the model. The final result is the average of all cross-validation results. This application chooses 10-fold cross-validation, that is, K=10, to ensure the stability and reliability of the evaluation results. Combined with grid search technology, this application first determines a wide range of parameter combinations, and then conducts a detailed search within this range to find the optimal parameter combination through multiple training and verification. In the end, a GBDT model that is neither overfitting nor too simple is obtained, which can accurately predict the insulation state of the energy storage system.

[0055] Step 5: Accurately evaluate the performance of the energy storage system insulation state prediction model.

[0056] Model evaluation plays a vital role in the task of predicting the insulation status of energy storage systems. As a key link in the machine learning process, a reasonable evaluation process can not only help us judge the pros and cons of the model, but also provide a clear direction for subsequent model optimization. In order to comprehensively evaluate the performance of the model, it is first necessary to clarify the task type and select appropriate evaluation indicators accordingly.

[0057] Given that the insulation state prediction of energy storage systems is essentially a regression task, that is, predicting a continuous insulation state value, it is natural to use regression indicators to measure the performance of the model. Among the many regression indicators, MAE (mean absolute error) and R² (coefficient of determination) are widely used because of their intuitiveness and practicality. MAE can directly reflect the overall error level of the model prediction by calculating the average absolute value of the difference between the predicted value and the true value; while R² measures the goodness of fit of the model to the data by comparing the degree of variation between the model predicted value and the true value.

[0058] In the energy storage system insulation state prediction model of the present invention, MAE and R² are specially selected as evaluation indicators. MAE can help understand the absolute size of the model prediction error, so as to judge the performance of the model in terms of prediction accuracy; while R² can evaluate the degree of fit of the model to the data as a whole, and further verify the prediction ability of the model. By calculating the values ​​of these two indicators, the performance of the model can be quantitatively evaluated, and the problems existing in the model, such as overfitting, underfitting or improper feature selection, can be found out accordingly, so as to determine the subsequent optimization direction.

[0059] Step 6: Deeply analyze the prediction results of the GBDT model to accurately locate the key influencing factors of the insulation status of the energy storage system.

[0060] In the energy storage system insulation status prediction project, the interpretability of the model results is crucial for engineers to understand the system operation mechanism and formulate effective maintenance strategies. The present invention selects the GBDT (gradient boosted decision tree) model as a prediction tool, not only because of its excellent performance in prediction accuracy, but also because of its unique interpretability advantage. The GBDT model is based on the ensemble learning principle of the decision tree, so that the decision-making process of each tree can be clearly interpreted. At the same time, the GBDT algorithm has a rich visualization tool, which can be used to show the importance of each feature to the model and the degree of influence of a single feature or two features on the model's dependent variable.

[0061] The following analyzes the importance of voltage, current, temperature, etc. in the energy storage system to the insulation status of the energy storage system: GBDT is an iterative ensemble learning algorithm. The global importance of a single feature can be obtained by calculating the average value of the local importance of the corresponding feature in each subtree. The global importance calculation formula of a single feature is as follows: in, and Respectively represent the global importance of feature j and the importance of feature j in the subtree The local importance of , and N represents the number of subtrees.

[0062] Feature j in subtree The calculation formula of local importance in is as follows: Among them, M is a subtree The number of leaf nodes, because the GBDT model is a binary tree, so the subtree The number of leaf nodes is represented by M-1. is the reduction in the residual sum of squares of non-leaf nodes after t splits. is the discriminant function. When the feature that a non-leaf node depends on at split t is feature j, the function value is 1; otherwise, the function value is 0.

[0063] Analyzing and interpreting the model prediction results can help engineers gain a deeper understanding of the system's operation. By analyzing the basis and key features of the model predictions, engineers can understand which factors have the greatest impact on the system's insulation status, thereby performing targeted system maintenance and improvements.

[0064] In addition, the interpretation of the model prediction results can also help engineers develop maintenance plans and preventive measures. By analyzing the insulation status of the model output, engineers can develop corresponding maintenance plans based on the current situation of the system, repair potential insulation problems in a timely manner, and ensure the safe operation of the system.

[0065] The beneficial effects of the present invention are: (1) In-depth analysis of improving detection accuracy and prediction capabilities: The GBDT model constructed by the present invention, with its powerful nonlinear fitting ability and self-learning ability, can deeply explore the hidden patterns and laws in the electrical digital data of the energy storage system, thereby constructing a complex mapping relationship between the electrical digital data and its insulation state with high precision. This high precision is not only reflected in the accurate judgment of the current insulation state, but also in the accurate prediction of future insulation state changes. With the help of this model, engineers can grasp the insulation health status of the energy storage system in real time, and provide early warning for possible insulation degradation or failure risks, ensuring that effective measures are taken before the problem worsens, greatly reducing the risk of accidents caused by insulation failures.

[0066] (2) Specific measures to achieve preventive maintenance and extend equipment life: The preventive maintenance mechanism of the present invention relies on a data-driven real-time monitoring system. By continuously collecting and analyzing the operating data of the energy storage system, the system can automatically identify slight changes in the insulation state and predict the time and impact of potential faults. This forward-looking maintenance strategy enables engineers to intervene in advance and take targeted maintenance measures, such as adjusting operating parameters and replacing aging components, thereby effectively preventing the occurrence and expansion of faults. This preventive maintenance not only reduces downtime caused by sudden failures, but also avoids the high cost of fault repair, significantly extending the service life of energy storage equipment and improving overall economic benefits.

[0067] (3) Comprehensive considerations to enhance the safety management level of energy storage systems: Based on an in-depth analysis of the impact of key factors such as voltage, current, and temperature on the insulation status of energy storage systems, the present invention further expands the field of safety management. By constructing a multi-dimensional feature evaluation system, the system can comprehensively examine the operating status of the energy storage system and identify various factors that may lead to a decrease in insulation performance. This comprehensive evaluation is not limited to the changes in a single parameter, but also focuses on the interactions and influences between parameters, providing solid data support for the formulation of more scientific and reasonable safety management strategies. At the same time, the system can also automatically adjust the monitoring frequency and warning threshold according to the evaluation results to ensure that a keen insight into the insulation status is always maintained in a complex and changing operating environment.

[0068] (4) Innovative practice in optimizing equipment management and maintenance services: The innovation of the present invention in equipment management and maintenance services is mainly reflected in the comprehensive consideration of characteristic parameters of different dimensions of the energy storage system. By integrating data from multiple sensors, the system can construct a multi-dimensional feature space that comprehensively reflects the operating status of the energy storage system. On this basis, the system can use advanced machine learning algorithms to automatically identify and extract the most critical feature combinations for insulation status prediction, thereby achieving a more accurate prediction of the insulation status. This method of comprehensively considering characteristic parameters of different dimensions not only improves the accuracy of the prediction, but also provides users with a more personalized and customized equipment management and maintenance service solution, further improving user experience and satisfaction.

[0069] (5) Continuous driving force for improving the level of intelligence: The improvement of the level of intelligence of the present invention is due to the in-depth application and continuous optimization of machine learning algorithms. As the system operating time increases and data continues to accumulate, the system can automatically learn and adapt to new operating environments and conditions, and continuously improve the accuracy and generalization ability of the model. At the same time, the system also has the ability to self-optimize and iteratively upgrade, and can continuously optimize algorithm parameters and model structure based on user feedback and actual needs to ensure that it always remains at the forefront of industry technology. This continuous intelligent upgrade not only improves system performance and stability, but also lays a solid foundation for the future development of insulation detection technology for energy storage systems.

[0070] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An energy storage system insulation detection method based on artificial intelligence, characterized in that: include: Step 1: Collect the electrical digital data and related text data of the energy storage system, vectorize the text data based on the BERT model, and then use the GBDT algorithm to extract key features related to the insulation status from the electrical digital data and the vectorized text data; Step 2, based on the features selected in step 1, a GBDT model is constructed to predict the insulation state of the energy storage system; Step 3, preprocessing the collected digital data and vectorized text data to divide them into training sets and test sets; Step 4: Use K-fold cross validation and grid search methods to train the GBDT model to optimize model parameters and improve prediction performance; Step 5: Use the test set data to evaluate the trained model, and select MAE and R² as evaluation indicators to measure the accuracy of model prediction.

2. The energy storage system insulation detection method based on artificial intelligence according to claim 1, characterized in that: The method also includes: step 6, using the feature importance evaluation function of the GBDT model to analyze the influence of each feature on the model prediction result, and displaying the interaction between features and the model decision path through a visualization tool; based on the analysis results, proposing model optimization suggestions to improve the interpretability and credibility of the model results.

3. The energy storage system insulation detection method based on artificial intelligence according to claim 1, characterized in that: In step 1, the electrical digital data of the energy storage system includes: voltage data, covering the voltage of each part of the energy storage system to reflect the voltage stability and change; current data, recording the current of each part of the system to understand the current load and change trend; temperature data, collecting the temperature of each component inside the system to reflect the heat distribution and temperature change; frequency data, monitoring the frequency stability and fluctuation of the system; humidity data, recording the humidity of the system environment, because humidity changes may affect the insulation state; battery status data, including charge state and discharge state, for systems containing battery components; working mode data, recording the working state of the system, including charging and discharging modes; fault history data, including past fault records and maintenance history; text data including the model of the energy storage device, basic configuration, and text information content of historical fault information.

4. The energy storage system insulation detection method based on artificial intelligence according to claim 1, characterized in that: In the feature extraction process of step 1, iterative learning is performed by constructing multiple decision trees, and each decision tree is trained based on the residual of the previous tree to minimize the overall prediction error, thereby extracting the most critical feature set for predicting the insulation state of the energy storage system.

5. The energy storage system insulation detection method based on artificial intelligence according to claim 1, characterized in that: The preprocessing in step 3 includes: flexibly adjusting the ratio of the training set to the test set, and dynamically setting the ratio of the training set to the test set according to the specific requirements of the energy storage system insulation detection task and the characteristics of the data, so as to balance the model training effect and the evaluation accuracy.

6. The energy storage system insulation detection method based on artificial intelligence according to claim 1, characterized in that: The preprocessing in step 3 includes: identifying and analyzing the pattern of missing values, and selecting the most appropriate filling strategy according to the specific distribution of the missing values, including filling the predicted value of the mean, median or mode.

7. The energy storage system insulation detection method based on artificial intelligence according to claim 1, characterized in that: The step 4 also introduces random search or Bayesian optimization method, combined with grid search, to more efficiently explore the optimal combination of GBDT model parameters.

8. The energy storage system insulation detection method based on artificial intelligence according to claim 1, characterized in that: The K-fold cross-validation in step 4 is to divide the training set data into K subsets on average, select K-1 subsets as training data each time, and use the remaining subset as validation data, repeat K times, use the average validation error as the evaluation criterion of model performance, and automatically adjust the parameters of the GBDT model in combination with the grid search method to achieve the best prediction performance.

9. The energy storage system insulation detection method based on artificial intelligence according to claim 8, characterized in that: The parameters of the GBDT model include learning rate, number of trees, and depth of trees.

10. The energy storage system insulation detection method based on artificial intelligence according to claim 2, characterized in that: The step 6 also includes: calculating and outputting the global importance ranking of each feature in the GBDT model to quantify the influence of each factor on the insulation state prediction result of the energy storage system.

Images