Energy storage power station safety assessment method and system

Abstract

The invention relates to the technical field of energy storage power station safety monitoring, and discloses an energy storage power station safety assessment method and system, and the method comprises the steps: collecting historical operation data and regular sampling inspection sample data of an energy storage power station, and carrying out the data preprocessing and feature extraction; and respectively constructing a security evaluation model A based on a decision tree algorithm and a security prediction model B based on a long short-term memory (LSTM) network algorithm. A model A learns a safety mode and a fault mode in historical data, and a model B predicts a future safety state change trend. Sampling inspection data are input into the two models to obtain evaluation and prediction results, and a preliminary fusion evaluation result is obtained through comprehensive analysis of a fusion algorithm. And according to a set safety threshold, determining a power station safety level and a potential risk point. And finally generating a safety assessment report including the overall safety condition, the potential risk points and the suggested measures. According to the invention, the accuracy and timeliness of safety assessment of the energy storage power station are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring of energy storage power stations, and specifically provides a safety assessment method and system for energy storage power stations. Background Art

[0002] With the transformation of the global energy structure and the rapid development of renewable energy, energy storage power stations, as a key link in energy storage and conversion, are crucial for the stable operation of the power system. However, energy storage power stations, especially battery energy storage-based systems, face many safety challenges due to their complex electrochemical processes, variable operating environments, and interactions between devices. These challenges include, but are not limited to, thermal runaway, short circuit, overcharge / overdischarge of batteries, and the impact of environmental factors (such as temperature and humidity) on device performance.

[0003] In the prior art, there are already some safety monitoring methods based on data analysis. However, most of these methods focus on data analysis in a single dimension, such as only paying attention to electrical parameters such as the voltage and current of the battery, or only considering the impact of environmental factors on device performance, without comprehensively considering historical operation data, time series changes, and the interaction relationship between devices, resulting in limited accuracy and reliability of the evaluation results. Summary of the Invention

[0004] The purpose of the present invention is to provide a safety assessment method and system for energy storage power stations to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A safety assessment method for an energy storage power station, the method includes;

[0006] Step 1: Collect historical operation data of the energy storage power station; at the same time, regularly obtain spot-check sample data of the energy storage power station, including battery samples, environmental samples, and equipment samples;

[0007] Step 2: Preprocess the data collected in Step 1, including data cleaning, missing value processing, outlier detection and elimination, and data standardization, and then perform feature extraction to form a historical feature data set and a spot-check feature data set;

[0008] Step 3: Construct a safety assessment model A based on the decision tree algorithm, and this model is trained using the historical feature data set to learn the safety patterns and fault patterns in the historical data;

[0009] Step 4: Construct a safety prediction model B based on the long short-term memory network (LSTM) algorithm, and this model uses the time series data in the historical feature data set to predict the change trend of the safety state of the energy storage power station in the future for a period of time;

[0010] Step 5: Input the sampled feature dataset into Model A and Model B respectively to obtain the evaluation results based on the decision tree and the prediction results based on LSTM;

[0011] Step 6: Define a fusion algorithm to conduct a fusion analysis on the evaluation results of Model A and the prediction results of Model B, and integrate historical experience and future trends to obtain preliminary fusion evaluation results;

[0012] Step 7: Set a safety threshold, and determine the safety level and potential risk points of the energy storage power station according to the comparison between the fusion evaluation results and the safety threshold;

[0013] Step 8: Generate a safety evaluation report according to the safety evaluation results, including the overall safety condition of the energy storage power station, potential risk points, and corresponding recommended measures.

[0014] Preferably, the historical operation data includes battery state data, environmental monitoring data, equipment operation parameters, fault records, charge and discharge records, temperature records, humidity records, and voltage and current records.

[0015] Preferably, the decision tree algorithm adopts the CART algorithm. By recursively dividing the dataset, a decision tree model is constructed. Each node is split according to a certain feature until the stopping condition is met; the following formula is used for node splitting in the decision tree algorithm:

[0016]

[0017] where G(D,A) represents the information gain of dataset D on feature A, H(D) represents the entropy of dataset D, Values(A) represents all possible values of feature A, D v represents the subset of dataset D where the value of feature A is v, and |D| and |D v | respectively represent the number of samples in dataset D and subset D v .

[0018] Preferably, the network structure of the LSTM algorithm includes an input layer, a hidden layer, and an output layer. The hidden layer contains at least one LSTM unit, and each LSTM unit is internally provided with a forget gate, an input gate, and an output gate to capture long-term dependencies in time series data.

[0019] Preferably, the fusion algorithm adopts a method based on Bayesian fusion to conduct a fusion analysis on the evaluation results of the decision tree Model A and the prediction results of the LSTM Model B.

[0020] Preferably, the implementation steps of the fusion algorithm include:

[0021] F1: Define the evaluation result of decision tree model A as D_A, whose value is a discrete security level or a continuous security score; define the prediction result of LSTM model B as D_B, whose value is the predicted value of the energy storage power station's security state or the predicted security level / score within a future period of time.

[0022] F2: For each sampled inspection sample, calculate the credibility P(A|D) of the evaluation result D_A of decision tree model A, estimated based on the accuracy rate of model A in historical data; assume the accuracy rate of model A in historical data is Acc_A, then P(A|D) = Acc_A.

[0023] F3: Calculate the credibility P(B|D) of the prediction result D_B of LSTM model B, estimated based on the prediction accuracy of model B in historical data; assume the prediction accuracy of model B in historical data is Acc_B, then P(B|D) = Acc_B;

[0024] F4: Adopt the Bayesian fusion method, combine the credibilities of the two models, and calculate the fused security evaluation result D_Fusion; for the discrete security level, use Bayes' theorem to calculate the posterior probability of each level and select the level with the highest probability as the final evaluation result; for the continuous security score, use the weighted average method, with the weights being the credibilities of each model.

[0025] F5: According to the fused security evaluation result D_Fusion, determine the security level or potential risk points of the energy storage power station; if the discrete level is adopted, directly determine according to the level with the highest probability of D_Fusion; if the continuous score is adopted, divide the security level according to the preset score threshold.

[0026] Preferably, the steps for training decision tree model A include:

[0027] S1: Use the historical feature dataset as the training set, and set the parameters of the decision tree depth and the minimum sample splitting number;

[0028] S2: Adopt the CART algorithm to recursively divide the dataset and construct the decision tree;

[0029] S3: Use the validation set to verify the decision tree model and adjust the parameters to optimize the model performance;

[0030] S4: When the model performance reaches the preset standard, stop training to obtain decision tree model A.

[0031] Preferably, the steps for training LSTM model B include:

[0032] P1: Divide the time series data in the historical feature dataset into a training set, a validation set, and a test set;

[0033] P2: Set the number of layers of the LSTM network, the number of LSTM units in each layer, the learning rate, and the optimizer;

[0034] P3: Use the training set to iteratively train the LSTM model, and optimize the model parameters through the backpropagation algorithm and the gradient descent method;

[0035] P4: In each iteration, calculate the loss between the model prediction value and the true value, and update the model parameters according to the loss function;

[0036] P5: When the loss converges or reaches the preset number of iterations, stop the training to obtain the LSTM model B.

[0037] Preferably, the safety threshold is set according to the safety standards and historical experience of the energy storage power station, including the battery health threshold, the equipment failure rate threshold, and the environmental temperature threshold, and is used to judge whether the energy storage power station is in a safe state.

[0038] Preferably, an energy storage power station safety assessment system, the system includes:

[0039] A data collection module, used to collect the historical operation data of the energy storage power station, and regularly obtain the spot-check sample data of the energy storage power station, and the spot-check sample data includes battery samples, environmental samples, and equipment samples;

[0040] A data preprocessing module, connected to the data collection module, used to preprocess the collected data, including data cleaning, missing value processing, outlier detection and elimination, and data standardization, and further perform feature extraction to form a historical feature data set and a spot-check feature data set;

[0041] A decision tree safety assessment model construction module, connected to the data preprocessing module, used to construct a safety assessment model A based on the decision tree algorithm, and this model is trained using the historical feature data set to learn the safety patterns and failure patterns in the historical data;

[0042] An LSTM safety prediction model construction module, connected to the data preprocessing module, used to construct a safety prediction model B based on the long short-term memory network LSTM algorithm, and this model uses the time series data in the historical feature data set to predict the safety state change trend of the energy storage power station in the future for a period of time;

[0043] An evaluation and prediction result acquisition module, connected to the decision tree safety assessment model construction module and the LSTM safety prediction model construction module, used to input the spot-check feature data set into model A and model B respectively to obtain the evaluation result based on the decision tree and the prediction result based on the LSTM;

[0044] The fusion analysis module, connected to the evaluation and prediction result acquisition module, is used to define and execute a fusion algorithm to perform a fusion analysis on the evaluation results of Model A and the prediction results of Model B, synthesize historical experience and future trends, and generate preliminary fusion evaluation results;

[0045] The safety level and risk determination module, connected to the fusion analysis module, is used to set safety thresholds and determine the safety level and potential risk points of the energy storage power station based on the comparison between the fusion evaluation results and the safety thresholds;

[0046] The report generation module, connected to the safety level and risk determination module, is used to generate a safety assessment report containing the overall safety status of the energy storage power station, potential risk points, and corresponding recommended measures based on the safety assessment results.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] By collecting the historical operation data of the energy storage power station and regularly sampling inspection sample data, the present invention can comprehensively consider various factors such as batteries, environment, and equipment, avoiding the limitations of traditional methods that only focus on single-dimensional data. The data preprocessing and feature extraction steps ensure the accuracy and reliability of the data, providing a solid foundation for subsequent model training and prediction. The application of the decision tree algorithm can learn the safety patterns and fault patterns in historical data, while the LSTM algorithm can predict the change trend of the safety state in the future for a period of time. The combination of the two makes the evaluation results more comprehensive and accurate.

[0049] By constructing a safety prediction model based on LSTM, the present invention can real-time predict the change trend of the safety state of the energy storage power station, timely discover potential risks, and provide a scientific basis for operation and maintenance management. The regularly obtained sampling inspection sample data can reflect the latest state of the energy storage power station, ensuring the timeliness of the evaluation results.

[0050] By setting safety thresholds, the present invention can accurately judge the safety level of the energy storage power station and precisely locate potential risk points. The safety assessment report details the overall safety status of the energy storage power station, potential risk points, and corresponding recommended measures, providing clear guidance for operation and maintenance personnel. The method and system of the present invention can be applied to energy storage power stations of different scales and complexities, having good adaptability. With the continuous accumulation of the operation data of the energy storage power station, the model can continuously learn and optimize, improving the accuracy and reliability of the evaluation, showing good scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is the working principle diagram of a safety assessment method for an energy storage power station according to the present invention;

[0052] Figure 2 It is the step flow chart for training Decision Tree Model A;

[0053] Figure 3 It is a flow chart for implementing the comprehensive analysis of the output results of two models based on the Bayesian fusion method. Specific implementation manners

[0054] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] Please refer to Figures 1 - 3 , the present invention provides a technical solution: a safety assessment method for an energy storage power station, and the system includes:

[0056] Step 1, data collection: Collect historical operation data of the energy storage power station. These data include but are not limited to electrical parameters such as the voltage, current, temperature, and internal resistance of the battery, as well as the operation log and maintenance record of the energy storage power station. At the same time, regularly obtain the spot-check sample data of the energy storage power station, including battery samples, environmental samples (such as temperature, humidity, etc.) and equipment samples (such as equipment vibration, noise, etc.) to ensure the comprehensiveness and representativeness of the data.

[0057] Step 2, data preprocessing and feature extraction: Preprocess the data collected in Step 1, including data cleaning to remove invalid or incorrect data, missing value processing to fill or process missing data, outlier detection and removal to exclude the influence of abnormal data on model training, and data standardization to ensure the consistency and comparability of the data. Then, perform feature extraction. According to the data characteristics and safety assessment requirements, extract key features to form a historical feature data set and a spot-check feature data set.

[0058] Step 3, construct a safety assessment model A: Construct a safety assessment model A based on the decision tree algorithm. Use the historical feature data set to train model A so that the model can learn the safety patterns and fault patterns in the historical data. By adjusting the branch conditions of the decision tree, optimize the classification performance of the model to ensure that the model can accurately identify the safety state of the energy storage power station.

[0059] Step 4, construct a safety prediction model B: Construct a safety prediction model B based on the long short-term memory network (LSTM) algorithm. Use the time series data in the historical feature data set to train model B so that the model can predict the change trend of the safety state of the energy storage power station in the future period. By adjusting parameters such as the number of layers and neurons of the LSTM network, optimize the prediction performance of the model to ensure the accuracy and reliability of the prediction results.

[0060] Step 5, Model Evaluation and Prediction: Input the randomly selected feature dataset into Model A and Model B respectively to obtain the evaluation result based on the decision tree and the prediction result based on LSTM. The evaluation result reflects the current safety status of the energy storage power station, and the prediction result reflects the changing trend of the safety status of the energy storage power station in a future period of time.

[0061] Step 6, Fusion Analysis: Define a fusion algorithm to conduct a fusion analysis on the evaluation result of Model A and the prediction result of Model B. The fusion algorithm can comprehensively consider historical experience and future trends, and through methods such as weighted average and logistic regression, fuse the results of the two models to obtain a preliminary fusion evaluation result.

[0062] Step 7, Safety Level Determination and Risk Point Identification: Set a safety threshold, and determine the safety level of the energy storage power station according to the comparison between the fusion evaluation result and the safety threshold. At the same time, identify potential risk points of the energy storage power station based on abnormalities or trend changes in the fusion evaluation result, providing a basis for subsequent operation and maintenance management.

[0063] Step 8, Generate a Safety Evaluation Report: Generate a safety evaluation report according to the safety evaluation result. The content of the report includes the overall safety status of the energy storage power station, including the safety level, potential risk points, and corresponding recommended measures. The recommended measures can include repairs, replacements, and adjustments of operating parameters for potential risk points to ensure the safe and stable operation of the energy storage power station.

[0064] The present invention will be further described below in conjunction with Embodiments 1 to 3:

[0065] Embodiment 1:

[0066] When constructing the safety evaluation model A based on the decision tree algorithm, the CART (Classification and Regression Trees) algorithm is adopted, and the decision tree model is constructed by recursively dividing the dataset. Each node is split according to a certain feature until the stopping condition is met. The basis for node splitting is the information gain, and its calculation formula is as follows:

[0067]

[0068] Among them, G(D,A) represents the information gain of dataset D on feature A, H(D) represents the entropy of dataset D, Values(A) represents all possible values of feature A, D v represents the subset of dataset D where the value of feature A is v, |D| and |D v | respectively represent the number of samples in dataset D and subset D v .

[0069] The specific steps for training the decision tree model A are as follows:

[0070] S1: Use the historical feature dataset as the training set, which contains various features and their corresponding safety states during the past operation of the energy storage power station. At the same time, set the parameters of the decision tree depth and the minimum sample split number, which are used to control the complexity of the decision tree and avoid overfitting.

[0071] S2: Adopt the CART algorithm to recursively divide the dataset. For the current node, traverse all features and all possible values of the features, calculate the information gain corresponding to each feature and value. Select the feature and value with the largest information gain as the splitting condition of the current node, divide the dataset into two subsets, and recursively divide the subsets respectively until the stopping condition is met (such as reaching the maximum depth, the number of samples in the subset is less than the minimum sample split number, etc.).

[0072] S3: Use the validation set to validate the decision tree model and evaluate the performance of the model (such as accuracy, recall, etc.). According to the validation results, adjust the parameters such as the decision tree depth and the minimum sample split number to optimize the model performance. Through multiple iterations of training and validation, find the parameter combination with the best performance.

[0073] S4: When the model performance reaches the preset standard (such as the accuracy exceeds a certain threshold), stop training to obtain the final decision tree model A. This model can classify and evaluate the safety state based on the feature data of the energy storage power station.

[0074] Example 2:

[0075] This example mainly introduces the network structure of the safety prediction model B based on the long short-term memory (LSTM) algorithm, which mainly includes an input layer, a hidden layer, and an output layer. Among them, the hidden layer contains at least one layer of LSTM units, and each LSTM unit is carefully designed with a forget gate, an input gate, and an output gate inside. These gating mechanisms cooperate together to effectively capture the long-term dependencies in the time series data. The following are the detailed steps for training the safety prediction model B:

[0076] P1. Data preparation and division: Carefully divide the time series data in the historical feature dataset into a training set, a validation set, and a test set. The training set is used for the iterative training of the model, the validation set is used for the evaluation and parameter adjustment of the model performance, and the test set is used for the final evaluation of the generalization ability of the model. This division ensures the independence of model training and evaluation, thereby improving the objectivity of model evaluation.

[0077] P2. Network Structure and Parameter Settings: Clearly define the number of layers of the LSTM network and the specific number of LSTM units in each layer. These settings need to be flexibly adjusted according to the complexity of the task and the characteristics of the data. At the same time, select an appropriate learning rate and optimizer. The learning rate controls the step size of model parameter updates, while the optimizer determines the specific way of parameter updates. Both jointly affect the training efficiency and effect of the model.

[0078] P3. Model Training and Parameter Optimization: Use the training set to iteratively train the LSTM model. In each iteration, calculate the gradient of the loss function with respect to the model parameters through the backpropagation algorithm, and then use the gradient descent method to update the model parameters to gradually reduce the gap between the predicted value and the true value. This process needs to be repeated until the model performance reaches stability or meets the preset stopping conditions.

[0079] P4. Loss Calculation and Parameter Update: In each iteration process, it is necessary to calculate the loss between the model predicted value and the true value, which is usually achieved by defining an appropriate loss function (such as mean square error, cross entropy, etc.). According to the gradient information calculated by the loss function, use the optimizer to update the model parameters to continuously approach the optimal solution.

[0080] P5. Training Stop and Model Output: When the loss function value converges to a certain stable level or reaches the preset number of iterations, stop the training process. Obtain the trained LSTM model B. This model can accurately predict the future safety state change trend of the energy storage power station based on historical time series data, thus providing strong decision-making support for the operation and maintenance management of the power station.

[0081] Example 3:

[0082] This example describes the method of comprehensively analyzing the output results of two models by using the Bayesian fusion method in the construction of a safety assessment system integrating the decision tree model A and the LSTM model B. The following are the specific implementation steps of this fusion algorithm:

[0083] F1. Define Model Outputs: Clearly define that the evaluation result of the decision tree model A is denoted as D_A, which can be a discrete safety level (such as "safe", "average", "dangerous", etc.) or a continuous safety score (such as a value between 0 and 100). Similarly, the prediction result of the LSTM model B is denoted as D_B, which represents the predicted value of the safety state of the energy storage power station in the future for a period of time, and can also be the predicted safety level or score.

[0084] F2. Calculate the credibility of decision tree model A: For each randomly inspected sample, it is necessary to evaluate the credibility of the evaluation result \(D_A\) of decision tree model A. This credibility is estimated based on the accuracy rate of model A in historical data. Assume that through verification of historical data, the accuracy rate of model A is \(Acc_A\), then for the current sample, the credibility \(P(A|D)\) of the evaluation result of decision tree model A is set as \(Acc_A\).

[0085] F3. Calculate the credibility of LSTM model B: Similarly, calculate the credibility of the prediction result \(D_B\) of LSTM model B. This credibility is estimated based on the prediction accuracy of model B in historical data. Assume that the prediction accuracy of model B in historical data is \(Acc_B\), then for the current sample, the credibility \(P(B|D)\) of the prediction result of LSTM model B is set as \(Acc_B\).

[0086] F4. Bayesian fusion calculation:

[0087] Adopt the Bayesian fusion method, combine the credibilities of the two models, and calculate the fused safety evaluation result \(D_{Fusion}\). Specifically as follows:

[0088] If both \(D_A\) and \(D_B\) are discrete safety levels, then apply Bayes' theorem to calculate the posterior probability of each safety level. For each possible level, calculate its joint probability, and select the level with the maximum probability as the final safety evaluation result \(D_{Fusion}\).

[0089] If \(D_A\) and \(D_B\) are continuous safety scores, then use the weighted average method to calculate the fused score. The weights of the weighted average are the credibilities \(P(A|D)\) and \(P(B|D)\) of decision tree model A and LSTM model B respectively. That is, \(D_{Fusion}=P(A|D)*D_A + P(B|D)*D_B\).

[0090] F5. Determine the safety level or potential risk points:

[0091] According to the fused safety evaluation result \(D_{Fusion}\), determine the safety level of the energy storage power station or identify potential risk points.

[0092] If discrete safety levels are adopted, directly determine the safety state of the energy storage power station according to the maximum probability level indicated by \(D_{Fusion}\).

[0093] If continuous safety scores are adopted, it is necessary to divide different safety levels according to the preset score thresholds. For example, it can be set that above 90 points is "safe", 70 - 89 points is "average", below 70 points is "dangerous", etc. According to the score of \(D_{Fusion}\), classify it into the corresponding safety level, and evaluate the safety status and potential risks of the energy storage power station accordingly.

[0094] The present invention also includes a safety assessment system for an energy storage power station, and the system includes:

[0095] A data collection module, configured to collect historical operation data of the energy storage power station and regularly obtain spot-check sample data of the energy storage power station, where the spot-check sample data includes battery samples, environmental samples, and equipment samples;

[0096] A data preprocessing module, connected to the data collection module, for preprocessing the collected data, including data cleaning, missing value processing, outlier detection and elimination, and data standardization, and further performing feature extraction to form a historical feature data set and a spot-check feature data set;

[0097] A decision tree safety assessment model construction module, connected to the data preprocessing module, for constructing a safety assessment model A based on the decision tree algorithm, and this model is trained using the historical feature data set to learn the safety patterns and fault patterns in the historical data;

[0098] An LSTM safety prediction model construction module, connected to the data preprocessing module, for constructing a safety prediction model B based on the long short-term memory network (LSTM) algorithm, and this model uses the time series data in the historical feature data set to predict the change trend of the safety state of the energy storage power station in a future period of time;

[0099] An evaluation and prediction result acquisition module, connected to the decision tree safety assessment model construction module and the LSTM safety prediction model construction module, for respectively inputting the spot-check feature data set into model A and model B to obtain the evaluation result based on the decision tree and the prediction result based on the LSTM;

[0100] A fusion analysis module, connected to the evaluation and prediction result acquisition module, for defining and executing a fusion algorithm to perform fusion analysis on the evaluation result of model A and the prediction result of model B, and comprehensively considering historical experience and future trends to generate a preliminary fusion evaluation result;

[0101] A safety level and risk determination module, connected to the fusion analysis module, for setting a safety threshold and determining the safety level and potential risk points of the energy storage power station according to the comparison between the fusion evaluation result and the safety threshold;

[0102] A report generation module, connected to the safety level and risk determination module, for generating a safety assessment report including the overall safety condition of the energy storage power station, potential risk points, and corresponding recommended measures according to the safety assessment result.

[0103] The implementation manner of this system refers to the above-mentioned embodiment and will not be elaborated in the description.

[0104] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variation thereof are intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0105] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for safety assessment of an energy storage power station, characterized in that: The method comprises: Step 1: Collect historical operation data of the energy storage power station; at the same time, regularly obtain random inspection sample data of the energy storage power station, including battery samples, environmental samples and equipment samples; Step 2: Preprocess the data collected in step 1, including data cleaning, missing value processing, outlier detection and elimination, and data standardization, and then perform feature extraction to form historical feature data sets and sampling feature data sets; Step 3: Build a safety assessment model A based on the decision tree algorithm. The model is trained using the historical feature data set to learn the safety mode and failure mode in the historical data. Step 4: Construct a safety prediction model B based on the long short-term memory network LSTM algorithm. This model uses the time series data in the historical feature data set to predict the safety status change trend of the energy storage power station in the future. Step 5: Input the sampling feature data set into model A and model B respectively to obtain the evaluation results based on the decision tree and the prediction results based on the LSTM; Step 6: Define the fusion algorithm, perform fusion analysis on the evaluation results of model A and the prediction results of model B, and obtain preliminary fusion evaluation results by integrating historical experience and future trends; Step 7: Set the safety threshold, and determine the safety level and potential risk points of the energy storage power station based on the comparison between the fusion assessment results and the safety threshold; Step 8: Generate a safety assessment report based on the safety assessment results, including the overall safety status of the energy storage power station, potential risk points and corresponding recommended measures.

2. A method for safety assessment of an energy storage power station according to claim 1, characterized in that: The historical operation data includes battery status data, environmental monitoring data, equipment operation parameters, fault records, charge and discharge records, temperature records, humidity records and voltage and current records.

3. A method for safety assessment of an energy storage power station according to claim 1, characterized in that: The decision tree algorithm adopts the CART algorithm, and constructs a decision tree model by recursively dividing the data set. Each node is split according to a certain feature until the stopping condition is met. The decision tree algorithm uses the following formula to split the node: Among them, G(D,A) represents the information gain of data set D on feature A, H(D) represents the entropy of data set D, Values(A) represents all possible values ​​of feature A, and D v represents the subset of feature A in dataset D whose value is v, |D| and |D v | respectively represent the dataset D and subset D v The number of samples.

4. A method for safety assessment of an energy storage power station according to claim 1, characterized in that: The network structure of the LSTM algorithm includes an input layer, a hidden layer and an output layer, wherein the hidden layer contains at least one layer of LSTM units, and each LSTM unit is provided with a forget gate, an input gate and an output gate to capture long-term dependencies in time series data.

5. A method for safety assessment of an energy storage power station according to claim 1, characterized in that: The fusion algorithm adopts a Bayesian fusion method to perform fusion analysis on the evaluation results of the decision tree model A and the prediction results of the LSTM model B.

6. A method for safety assessment of an energy storage power station according to claim 5, characterized in that: The implementation steps of the fusion algorithm include: F1: Define the evaluation result of decision tree model A as D_A, whose value is a discrete safety level or a continuous safety score; define the prediction result of LSTM model B as D_B, whose value is the predicted value of the safety status of the energy storage power station in the future or the predicted safety level / score; F2: For each sample, calculate the credibility P(A|D) of the evaluation result D_A of the decision tree model A, and estimate it based on the accuracy of model A in historical data; assuming that the accuracy of model A in historical data is Acc_A, then P(A|D)=Ac_A; F3: Calculate the credibility P(B|D) of the prediction result D_B of LSTM model B, and estimate it based on the prediction accuracy of model B in historical data; assuming that the prediction accuracy of model B in historical data is Acc_B, then P(B|D)=Ac_B; F4: The Bayesian fusion method is used to combine the credibility of the two models to calculate the fused safety assessment result D_Fusion; for discrete safety levels, the Bayesian theorem is used to calculate the posterior probability of each level, and the level with the highest probability is selected as the final assessment result; for continuous safety scores, the weighted average method is used, and the weight is the credibility of each model; F5: Determine the safety level or potential risk points of the energy storage power station based on the fused safety assessment result D_Fusion; if a discrete level is used, it is directly determined based on the maximum probability level of D_{Fusion}; if a continuous score is used, the safety level is divided according to the preset score threshold.

7. A method for safety assessment of an energy storage power station according to claim 3, characterized in that: The steps for training decision tree model A include: S1: Use the historical feature dataset as the training set and set the depth and minimum sample split number parameters of the decision tree; S2: Use the CART algorithm to recursively divide the data set and build a decision tree; S3: Use the validation set to validate the decision tree model and adjust parameters to optimize model performance; S4: When the model performance reaches the preset standard, stop training and obtain decision tree model A.

8. A method for safety assessment of an energy storage power station according to claim 4, characterized in that: The steps for training LSTM model B include: P1: Divide the time series data in the historical feature dataset into training set, validation set and test set; P2: Set the number of LSTM network layers, the number of LSTM units per layer, the learning rate, and the optimizer; P3: Iteratively train the LSTM model using the training set and optimize the model parameters through the back propagation algorithm and gradient descent method; P4: In each iteration, the loss between the model prediction value and the true value is calculated, and the model parameters are updated according to the loss function; P5: When the loss converges or reaches the preset number of iterations, stop training and obtain LSTM model B.

9. A method for safety assessment of an energy storage power station according to claim 1, characterized in that: The safety threshold is set according to the safety standards and historical experience of the energy storage power station, including the battery health threshold, the equipment failure rate threshold and the ambient temperature threshold, which are used to determine whether the energy storage power station is in a safe state.

10. A safety assessment system for an energy storage power station, characterized in that: The system comprises: A data collection module is used to collect historical operation data of the energy storage power station and regularly obtain random inspection sample data of the energy storage power station, wherein the random inspection sample data includes battery samples, environmental samples and equipment samples; The data preprocessing module is connected to the data collection module and is used to preprocess the collected data, including data cleaning, missing value processing, outlier detection and elimination, and data standardization, and further perform feature extraction to form a historical feature data set and a sampling feature data set; A decision tree safety assessment model building module, connected to the data preprocessing module, is used to build a safety assessment model A based on a decision tree algorithm. The model is trained using a historical feature data set to learn safety modes and failure modes in historical data. The LSTM safety prediction model building module is connected to the data preprocessing module and is used to build a safety prediction model B based on the long short-term memory network LSTM algorithm. The model uses the time series data in the historical feature data set to predict the safety status change trend of the energy storage power station in the future. The evaluation and prediction result acquisition module is connected to the decision tree security evaluation model construction module and the LSTM security prediction model construction module, and is used to input the sampling feature data set into model A and model B respectively to obtain the evaluation results based on the decision tree and the prediction results based on the LSTM; The fusion analysis module is connected to the evaluation and prediction result acquisition module, and is used to define and execute the fusion algorithm, perform fusion analysis on the evaluation results of model A and the prediction results of model B, and generate preliminary fusion evaluation results by integrating historical experience and future trends; The safety level and risk determination module is connected to the fusion analysis module and is used to set the safety threshold and determine the safety level and potential risk points of the energy storage power station based on the comparison between the fusion assessment results and the safety threshold; The report generation module is connected to the safety level and risk determination module and is used to generate a safety assessment report including the overall safety status of the energy storage power station, potential risk points and corresponding recommended measures based on the safety assessment results.

Images