Energy storage power station battery soh estimation method and system

Abstract

The application provides an energy storage power station battery SOH estimation method and system in the field of electrochemical energy storage technology, which comprises the following steps: S1, obtaining and labeling cycle charging and discharging history data to construct a training data set; S2, constructing a test data set based on the training data set; S3, extracting first and second historical aggregate features based on the training data set and the test data set; S4, creating a first and second battery SOH estimation model, and training the first and second battery SOH estimation model through the first and second historical aggregate features and SOH label values; S5, verifying through a 5-fold cross-validation method during the training process, and generating five model files through reasoning of the first and second battery SOH estimation model respectively; S6, estimating SOH through each model file, and taking the average fusion of the reasoning results of the five model files as the final result. The application has the advantage that the accuracy of the energy storage power station battery SOH estimation is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of electrochemical energy storage technology, and in particular to a method and system for estimating the state of charge (SOH) of batteries in an energy storage power station. Background Technology

[0002] With the rapid development of renewable energy and the increasing demand for energy storage, energy storage power stations are playing an increasingly important role in the energy sector. The accuracy of State of Health (SOH) estimation for energy storage power station batteries is crucial for ensuring the reliability and performance of the battery system. Battery SOH represents the battery's health and lifespan, and is a key factor determining whether the battery can continue to operate reliably. However, in practical applications, battery SOH estimation is often limited and challenged by various factors, such as environmental factors like different temperatures, current loads, and charge / discharge cycle counts. The impact of these factors on battery SOH is often difficult to accurately account for, leading to instability in the estimated results. Traditionally, the following methods exist for estimating battery SOH:

[0003] Method 1: Model-based prediction method. This method not only requires a lot of expert knowledge, but also the model has high complexity. In practical applications, the model accuracy is insufficient and the generalization ability is poor due to the influence of actual outdoor temperature and outdoor environment.

[0004] Method 2: Data-driven prediction methods often use fragmented or fully charged / discharged battery data as model input, estimating the current SOH of the battery through feature extraction. However, this method ignores the continuity between fragments, meaning that the SOH state at the current moment is correlated with the SOH state at the previous N moments. This means that the method fails to fully utilize historical data, affecting model accuracy. If existing data-driven prediction methods are directly adopted, the model is often trained in the source domain and then predicted in the target domain. In this case, the differences in the charge / discharge curve feature distribution between the source and target domains due to different battery types and operating conditions may affect model performance, and thus affect the accuracy of SOH estimation.

[0005] Therefore, how to provide a method and system for estimating the state of energy (SOH) of batteries in energy storage power stations, and improve the accuracy of SOH estimation, has become an urgent technical problem to be solved. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for estimating the state of harm (SOH) of batteries in energy storage power stations, thereby improving the accuracy of SOH estimation for batteries in energy storage power stations.

[0007] In a first aspect, the present invention provides a method for estimating the state of harm (SOH) of a battery in an energy storage power station, comprising the following steps:

[0008] Step S1: Obtain a large amount of historical charge-discharge data of the energy storage power station battery, and construct a training dataset after labeling each of the historical charge-discharge data with SOH tag values;

[0009] Step S2: Construct a test dataset based on the training dataset;

[0010] Step S3: Based on the training dataset and the test dataset, extract the first historical aggregation feature and the second historical aggregation feature;

[0011] Step S4: Create a first battery SOH estimation model and a second battery SOH estimation model based on the lightGBM model. Train the first battery SOH estimation model using the first historical aggregation features and SOH label values, and train the second battery SOH estimation model using the second historical aggregation features and SOH label values.

[0012] Step S5: During the training process, the first battery SOH estimation model and the second battery SOH estimation model combine the early training stop strategy and the 5-fold cross-validation method. During the verification process using the 5-fold cross-validation method, the first battery SOH estimation model and the second battery SOH estimation model respectively generate 5 model files through inference.

[0013] Step S6: Estimate the SOH of the energy storage power station battery using the 10 model files mentioned above.

[0014] Furthermore, step S1 specifically includes:

[0015] Acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. The historical charge-discharge data includes at least the charge-discharge time, cycle period, battery status, first charging capacity, first discharging capacity, rated capacity, room temperature capacity, voltage, current, temperature, and SOC.

[0016] The cycle period refers to the number of times the battery is charged and discharged. A complete cycle period includes the entire process of charging, resting, and discharging. The battery state is charging, resting, or discharging. The first charging capacity is the accumulated charging capacity of the current cycle period. The first discharging capacity is the accumulated discharging capacity of the current cycle period. The room temperature capacity is the calibrated capacity of the battery at room temperature after thermal equilibrium is achieved.

[0017] The SOH tag value is labeled for each of the cycle charge and discharge history data. The calculation process of the SOH tag value is as follows: after performing bilinear difference on the room temperature capacity at intervals of N cycle cycles, divide by the rated capacity to obtain the SOH tag value with a value range of 0 to 1.

[0018] A training dataset is constructed based on the labeled historical charge-discharge data of each cycle.

[0019] Furthermore, step S2 specifically includes:

[0020] From the training dataset, charge / discharge times, cycle periods, battery states, second charge capacity, second discharge capacity, and rated capacity are selected in days to construct a test dataset;

[0021] The second charging capacity is the cumulative charging capacity over one day; the second discharging capacity is the cumulative discharging capacity over one day.

[0022] Furthermore, step S3 specifically includes:

[0023] Based on the battery state statistics training dataset, the number of consecutive first charging segments, first discharging segments, and first stationary segments in each cycle are counted. The number of first charging cycles, first discharging cycles, and first stationary cycles in each cycle are calculated based on these counts. The first total number of segments and the first total number of charging cycles, discharging cycles, and stationary cycles within each cycle are calculated. Based on the ratio of the number of first charging segments, first discharging segments, and first stationary segments to the first total number of segments, and the ratio of the first charging cycles, first discharging cycles, and first stationary cycles to the first total number of segments, the first cycle aggregation feature for each cycle is extracted.

[0024] Based on the battery state statistics of the test dataset, the number of consecutive second charging segments, second discharging segments, and second resting segments for each day are counted. The number of second charging cycles, second discharging cycles, and second resting cycles for each day are calculated based on these counts. The total number of second charging segments, second discharging segments, and second resting segments for each day is calculated, as well as the total number of second charging cycles, second discharging cycles, and second resting cycles. Based on the ratio of the number of second charging segments, second discharging segments, and second resting segments to the total number of second segments, and the ratio of the number of second charging cycles, second discharging cycles, and second resting cycles to the total number of second segments, the second periodic aggregation features for each day are extracted.

[0025] Both the first and second cycle aggregation features include the proportion of charging times, the proportion of discharging times, the proportion of resting times, the proportion of charging capacity relative to rated capacity, the proportion of discharging capacity relative to rated capacity, the proportion of discharging capacity relative to charging capacity, and the difference between discharging capacity and charging capacity.

[0026] The first historical aggregated feature is obtained by accumulating the first period aggregated feature for each cycle, and the second historical aggregated feature is obtained by accumulating the second period aggregated feature for each day.

[0027] Furthermore, step S6 specifically includes:

[0028] Real-time charge-discharge cycle data for the energy storage power station to be estimated is obtained. The real-time charge-discharge cycle data is input into 10 model files for inference to obtain 10 inference results. The average value of the 10 inference results is taken to obtain the SOH estimation result, so as to estimate the SOH of the battery in the energy storage power station.

[0029] Secondly, the present invention provides a battery SOH estimation system for an energy storage power station, comprising the following modules:

[0030] The training dataset construction module is used to acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. After labeling each of the historical charge-discharge data with SOH tag values, the training dataset is constructed.

[0031] A test dataset construction module is used to construct a test dataset based on the training dataset.

[0032] The historical aggregation feature extraction module is used to extract a first historical aggregation feature and a second historical aggregation feature based on the training dataset and the test dataset.

[0033] The battery SOH estimation model training module is used to create a first battery SOH estimation model and a second battery SOH estimation model based on the lightGBM model. The first battery SOH estimation model is trained using the first historical aggregation features and SOH label values, and the second battery SOH estimation model is trained using the second historical aggregation features and SOH label values.

[0034] The model file inference module is used to generate 5 model files by combining the early stopping strategy and the 5-fold cross-validation method during the training process of the first battery SOH estimation model and the second battery SOH estimation model.

[0035] The SOH estimation module is used to estimate the SOH of batteries in an energy storage power station using the 10 model files mentioned above.

[0036] Furthermore, the training dataset construction module is specifically used for:

[0037] Acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. The historical charge-discharge data includes at least the charge-discharge time, cycle period, battery status, first charging capacity, first discharging capacity, rated capacity, room temperature capacity, voltage, current, temperature, and SOC.

[0038] The cycle period refers to the number of times the battery is charged and discharged. A complete cycle period includes the entire process of charging, resting, and discharging. The battery state is charging, resting, or discharging. The first charging capacity is the accumulated charging capacity of the current cycle period. The first discharging capacity is the accumulated discharging capacity of the current cycle period. The room temperature capacity is the calibrated capacity of the battery at room temperature after thermal equilibrium is achieved.

[0039] The SOH tag value is labeled for each of the cycle charge and discharge history data. The calculation process of the SOH tag value is as follows: after performing bilinear difference on the room temperature capacity at intervals of N cycle cycles, divide by the rated capacity to obtain the SOH tag value with a value range of 0 to 1.

[0040] A training dataset is constructed based on the labeled historical charge-discharge data of each cycle.

[0041] Furthermore, the test dataset construction module is specifically used for:

[0042] From the training dataset, charge / discharge times, cycle periods, battery states, second charge capacity, second discharge capacity, and rated capacity are selected in days to construct a test dataset;

[0043] The second charging capacity is the cumulative charging capacity over one day; the second discharging capacity is the cumulative discharging capacity over one day.

[0044] Furthermore, the historical aggregation feature extraction module is specifically used for:

[0045] Based on the battery state statistics training dataset, the number of consecutive first charging segments, first discharging segments, and first stationary segments in each cycle are counted. The number of first charging cycles, first discharging cycles, and first stationary cycles in each cycle are calculated based on these counts. The first total number of segments and the first total number of charging cycles, discharging cycles, and stationary cycles within each cycle are calculated. Based on the ratio of the number of first charging segments, first discharging segments, and first stationary segments to the first total number of segments, and the ratio of the first charging cycles, first discharging cycles, and first stationary cycles to the first total number of segments, the first cycle aggregation feature for each cycle is extracted.

[0046] Based on the battery state statistics of the test dataset, the number of consecutive second charging segments, second discharging segments, and second resting segments for each day are counted. The number of second charging cycles, second discharging cycles, and second resting cycles for each day are calculated based on these counts. The total number of second charging segments, second discharging segments, and second resting segments for each day is calculated, as well as the total number of second charging cycles, second discharging cycles, and second resting cycles. Based on the ratio of the number of second charging segments, second discharging segments, and second resting segments to the total number of second segments, and the ratio of the number of second charging cycles, second discharging cycles, and second resting cycles to the total number of second segments, the second periodic aggregation features for each day are extracted.

[0047] Both the first and second cycle aggregation features include the proportion of charging times, the proportion of discharging times, the proportion of resting times, the proportion of charging capacity relative to rated capacity, the proportion of discharging capacity relative to rated capacity, the proportion of discharging capacity relative to charging capacity, and the difference between discharging capacity and charging capacity.

[0048] The first historical aggregated feature is obtained by accumulating the first period aggregated feature for each cycle, and the second historical aggregated feature is obtained by accumulating the second period aggregated feature for each day.

[0049] Furthermore, the SOH estimation module is specifically used for:

[0050] Real-time charge-discharge cycle data for the energy storage power station to be estimated is obtained. The real-time charge-discharge cycle data is input into 10 model files for inference to obtain 10 inference results. The average value of the 10 inference results is taken to obtain the SOH estimation result, so as to estimate the SOH of the battery in the energy storage power station.

[0051] The advantages of this invention are:

[0052] A training dataset was constructed by acquiring a large amount of historical charge-discharge data of batteries in an energy storage power station and labeling it with State of Health (SOH) values. A test dataset was then built based on the training dataset. Next, based on the training and test datasets, first and second historical aggregated features were extracted. Then, a first battery SOH estimation model and a second battery SOH estimation model were created based on the LightGBM model. The first battery SOH estimation model was trained using the first historical aggregated features and SOH label values, and the second battery SOH estimation model was trained using the second historical aggregated features and SOH label values. During training, an early stop strategy and a 5-fold cross-validation method were combined. During the validation process using the 5-fold cross-validation method, the first and second battery SOH estimation models each generated 5 model files. Finally, the SOH of the energy storage power station batteries was estimated using 10 model files. Since the test dataset was constructed based on the training dataset, only charge-discharge times, cycle periods, battery states, second charging capacity, second discharging capacity, and rated capacity were extracted from the training dataset. The method avoids selecting room temperature capacity, voltage, current, temperature, and SOC, thus avoiding the impact of differences in input data distribution between the training dataset (source domain) and the test dataset (target domain) under different charge and discharge conditions, which could lead to a decrease in model performance and improve the generalization of battery SOH estimation. Based on the first and second cycle aggregation features, the first and second historical aggregation features are extracted respectively. That is, the historical charge and discharge data (cycle aggregation features) are accumulated and analyzed to extract features, which can make full use of historical data and leverage the advantages of continuous monitoring of energy storage battery data. By accumulating and analyzing features from the data of the entire battery life cycle (charging, resting, discharging), the comprehensiveness of feature extraction is ensured, which ultimately greatly improves the accuracy of SOH estimation of energy storage power station batteries. Attached Figure Description

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0054] Figure 1 This is a flowchart of a method for estimating the state of energy (SOH) of a battery in an energy storage power station according to the present invention.

[0055] Figure 2 This is a schematic diagram of the structure of a battery SOH estimation system for an energy storage power station according to the present invention. Detailed Implementation

[0056] The overall approach of the technical solution in this application is as follows: The test dataset only takes the charging and discharging time, cycle period, battery state, second charging capacity, second discharging capacity, and rated capacity from the training dataset. Room temperature capacity, voltage, current, temperature, and SOC are not selected, thus avoiding the impact of differences in input data distribution between the training dataset (source domain) and the test dataset (target domain) under different charging and discharging conditions, which could lead to a decrease in model performance. First historical aggregated features and second historical aggregated features are extracted based on the aggregated features of each first cycle and the aggregated features of each second cycle, i.e., feature accumulation analysis and extraction are performed on historical charging and discharging data, which can fully utilize historical data. By performing feature accumulation analysis and extraction on data throughout the entire battery lifecycle, the comprehensiveness of feature extraction is ensured, thereby improving the accuracy of SOH estimation for energy storage power stations.

[0057] Please refer to Figures 1 to 2 As shown, a preferred embodiment of the SOH estimation method for batteries in an energy storage power station according to the present invention includes the following steps:

[0058] Step S1: Obtain a large amount of historical charge-discharge data of the energy storage power station battery, and construct a training dataset after labeling each of the historical charge-discharge data with SOH tag values;

[0059] Step S2: Construct a test dataset based on the training dataset;

[0060] Step S3: Based on the training dataset and the test dataset, extract the first historical aggregation feature and the second historical aggregation feature;

[0061] Step S4: Create a first battery SOH estimation model and a second battery SOH estimation model based on the lightGBM model. Train the first battery SOH estimation model using the first historical aggregated features and SOH label values, and train the second battery SOH estimation model using the second historical aggregated features and SOH label values. The lightGBM model reduces the depth of the decision tree, which helps to reduce the complexity of the model and reduce the risk of overfitting.

[0062] Step S5: During the training process of the first battery SOH estimation model and the second battery SOH estimation model, an early stopping strategy and the 5-fold cross-validation method of the lightGBM model are combined. During the validation process using the 5-fold cross-validation method, the first battery SOH estimation model and the second battery SOH estimation model respectively generate 5 model files. The early stopping strategy can stop the training process in time to prevent the model from overfitting. The 5-fold cross-validation method helps to evaluate the performance of the model on different data subsets and ensures that the model can generalize to new datasets.

[0063] Step S6: Estimate the SOH of the energy storage power station battery using the 10 model files mentioned above.

[0064] Step S1 specifically involves:

[0065] Acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. The historical charge-discharge data includes at least the charge-discharge time (time stamp), cycle period, battery status, first charge capacity (unit Ah), first discharge capacity (unit Ah), rated capacity (unit Ah), room temperature capacity (unit Ah), voltage (unit V), current (unit A), temperature (unit °C), and SOC (%).

[0066] The cycle period refers to the number of times the battery is charged and discharged. A complete cycle period includes the entire process of charging, resting, and discharging. The battery state is charging, resting, or discharging. The first charging capacity is the accumulated charging capacity of the current cycle period. The first discharging capacity is the accumulated discharging capacity of the current cycle period. The room temperature capacity is the calibrated capacity of the battery at room temperature after thermal equilibrium is achieved.

[0067] The SOH tag value is labeled for each of the aforementioned historical charge-discharge cycles. The calculation process of the SOH tag value is as follows: after performing bilinear interpolation on the room temperature capacity at intervals of N cycles, the value is divided by the rated capacity to obtain the SOH tag value with a range of 0 to 1; that is, the SOH uses the capacity decay under standard room temperature environment as the tag value to be estimated.

[0068] A training dataset is constructed based on the labeled historical charge-discharge data of each cycle.

[0069] Step S2 specifically involves:

[0070] From the training dataset, charge / discharge times, cycle periods, battery states, second charge capacity, second discharge capacity, and rated capacity are selected in days to construct a test dataset;

[0071] The second charging capacity is the cumulative charging capacity over one day; the second discharging capacity is the cumulative discharging capacity over one day.

[0072] Step S3 specifically involves:

[0073] Based on the battery state statistics training dataset, the number of consecutive first charging segments, first discharging segments, and first stationary segments in each cycle are counted. The number of first charging cycles, first discharging cycles, and first stationary cycles in each cycle are calculated based on these counts. The first total number of segments and the first total number of charging cycles, discharging cycles, and stationary cycles within each cycle are calculated. Based on the ratio of the number of first charging segments, first discharging segments, and first stationary segments to the first total number of segments, and the ratio of the first charging cycles, first discharging cycles, and first stationary cycles to the first total number of segments, the first cycle aggregation feature for each cycle is extracted.

[0074] Based on the battery state statistics of the test dataset, the number of consecutive second charging segments, second discharging segments, and second resting segments for each day are counted. The number of second charging cycles, second discharging cycles, and second resting cycles for each day are calculated based on these counts. The total number of second charging segments, second discharging segments, and second resting segments for each day is calculated, as well as the total number of second charging cycles, second discharging cycles, and second resting cycles. Based on the ratio of the number of second charging segments, second discharging segments, and second resting segments to the total number of second segments, and the ratio of the number of second charging cycles, second discharging cycles, and second resting cycles to the total number of second segments, the second periodic aggregation features for each day are extracted.

[0075] Both the first and second cycle aggregation features include the proportion of charging times, the proportion of discharging times, the proportion of resting times, the proportion of charging capacity relative to rated capacity, the proportion of discharging capacity relative to rated capacity, the proportion of discharging capacity relative to charging capacity, and the difference between discharging capacity and charging capacity.

[0076] Charging cycle ratio = Number of charging cycles / (Number of charging cycles + Number of discharging cycles + Number of resting cycles);

[0077] Discharge cycle ratio = Number of discharges / (Number of charges + Number of discharges + Number of rest periods);

[0078] Percentage of resting times = Number of resting times / (Number of charging times + Number of discharging times + Number of resting times);

[0079] The ratio of charging capacity to rated capacity = cumulative charging capacity / rated capacity;

[0080] The ratio of discharge capacity to rated capacity = cumulative discharge capacity / rated capacity;

[0081] The ratio of discharge capacity to charge capacity = cumulative discharge capacity / cumulative charge capacity;

[0082] The difference between discharge capacity and charging capacity = cumulative discharge capacity - cumulative charging capacity;

[0083] The first historical aggregated feature is obtained by accumulating the first period aggregated feature for each cycle, and the second historical aggregated feature is obtained by accumulating the second period aggregated feature for each day.

[0084] By introducing the correlation between historical data (periodic aggregation features) through historical aggregation features, the differences in charge and discharge curve features caused by different charge and discharge conditions are avoided.

[0085] Step S6 specifically involves:

[0086] Real-time charge-discharge cycle data for the energy storage power station to be estimated is obtained. The real-time charge-discharge cycle data is input into 10 model files for inference to obtain 10 inference results. The average value of the 10 inference results is taken to obtain the SOH estimation result, so as to estimate the SOH of the battery in the energy storage power station.

[0087] A preferred embodiment of the SOH estimation system for a battery in an energy storage power station according to the present invention includes the following modules:

[0088] The training dataset construction module is used to acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. After labeling each of the historical charge-discharge data with SOH tag values, the training dataset is constructed.

[0089] A test dataset construction module is used to construct a test dataset based on the training dataset.

[0090] The historical aggregation feature extraction module is used to extract a first historical aggregation feature and a second historical aggregation feature based on the training dataset and the test dataset.

[0091] The battery SOH estimation model training module is used to create a first battery SOH estimation model and a second battery SOH estimation model based on the lightGBM model. The first battery SOH estimation model is trained using the first historical aggregated features and SOH label values, and the second battery SOH estimation model is trained using the second historical aggregated features and SOH label values. The lightGBM model reduces the depth of the decision tree, which helps to reduce the complexity of the model and reduce the risk of overfitting.

[0092] The model file inference module is used during the training of the first and second battery SOH estimation models. It combines an early stopping strategy with the 5-fold cross-validation method of the lightGBM model. During the validation process using the 5-fold cross-validation method, the first and second battery SOH estimation models each generate five model files. The early stopping strategy can stop the training process in a timely manner to prevent overfitting. The 5-fold cross-validation method helps evaluate the model's performance on different data subsets, ensuring that the model can generalize to new datasets.

[0093] The SOH estimation module is used to estimate the SOH of batteries in an energy storage power station using the 10 model files mentioned above.

[0094] The training dataset construction module is specifically used for:

[0095] Acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. The historical charge-discharge data includes at least the charge-discharge time (time stamp), cycle period, battery status, first charge capacity (unit Ah), first discharge capacity (unit Ah), rated capacity (unit Ah), room temperature capacity (unit Ah), voltage (unit V), current (unit A), temperature (unit °C), and SOC (%).

[0096] The cycle period refers to the number of times the battery is charged and discharged. A complete cycle period includes the entire process of charging, resting, and discharging. The battery state is charging, resting, or discharging. The first charging capacity is the accumulated charging capacity of the current cycle period. The first discharging capacity is the accumulated discharging capacity of the current cycle period. The room temperature capacity is the calibrated capacity of the battery at room temperature after thermal equilibrium is achieved.

[0097] The SOH tag value is labeled for each of the aforementioned historical charge-discharge cycles. The calculation process of the SOH tag value is as follows: after performing bilinear interpolation on the room temperature capacity at intervals of N cycles, the value is divided by the rated capacity to obtain the SOH tag value with a range of 0 to 1; that is, the SOH uses the capacity decay under standard room temperature environment as the tag value to be estimated.

[0098] A training dataset is constructed based on the labeled historical charge-discharge data of each cycle.

[0099] The test dataset construction module is specifically used for:

[0100] From the training dataset, charge / discharge times, cycle periods, battery states, second charge capacity, second discharge capacity, and rated capacity are selected in days to construct a test dataset;

[0101] The second charging capacity is the cumulative charging capacity over one day; the second discharging capacity is the cumulative discharging capacity over one day.

[0102] The historical aggregation feature extraction module is specifically used for:

[0103] Based on the battery state statistics training dataset, the number of consecutive first charging segments, first discharging segments, and first stationary segments in each cycle are counted. The number of first charging cycles, first discharging cycles, and first stationary cycles in each cycle are calculated based on these counts. The first total number of segments and the first total number of charging cycles, discharging cycles, and stationary cycles within each cycle are calculated. Based on the ratio of the number of first charging segments, first discharging segments, and first stationary segments to the first total number of segments, and the ratio of the first charging cycles, first discharging cycles, and first stationary cycles to the first total number of segments, the first cycle aggregation feature for each cycle is extracted.

[0104] Based on the battery state statistics of the test dataset, the number of consecutive second charging segments, second discharging segments, and second resting segments for each day are counted. The number of second charging cycles, second discharging cycles, and second resting cycles for each day are calculated based on these counts. The total number of second charging segments, second discharging segments, and second resting segments for each day is calculated, as well as the total number of second charging cycles, second discharging cycles, and second resting cycles. Based on the ratio of the number of second charging segments, second discharging segments, and second resting segments to the total number of second segments, and the ratio of the number of second charging cycles, second discharging cycles, and second resting cycles to the total number of second segments, the second periodic aggregation features for each day are extracted.

[0105] Both the first and second cycle aggregation features include the proportion of charging times, the proportion of discharging times, the proportion of resting times, the proportion of charging capacity relative to rated capacity, the proportion of discharging capacity relative to rated capacity, the proportion of discharging capacity relative to charging capacity, and the difference between discharging capacity and charging capacity.

[0106] Charging cycle ratio = Number of charging cycles / (Number of charging cycles + Number of discharging cycles + Number of resting cycles);

[0107] Discharge cycle ratio = Number of discharges / (Number of charges + Number of discharges + Number of rest periods);

[0108] Percentage of resting times = Number of resting times / (Number of charging times + Number of discharging times + Number of resting times);

[0109] The ratio of charging capacity to rated capacity = cumulative charging capacity / rated capacity;

[0110] The ratio of discharge capacity to rated capacity = cumulative discharge capacity / rated capacity;

[0111] The ratio of discharge capacity to charge capacity = cumulative discharge capacity / cumulative charge capacity;

[0112] The difference between discharge capacity and charging capacity = cumulative discharge capacity - cumulative charging capacity;

[0113] The first historical aggregated feature is obtained by accumulating the first period aggregated feature for each cycle, and the second historical aggregated feature is obtained by accumulating the second period aggregated feature for each day.

[0114] By introducing the correlation between historical data (periodic aggregation features) through historical aggregation features, the differences in charge and discharge curve features caused by different charge and discharge conditions are avoided.

[0115] The SOH estimation module is specifically used for:

[0116] Real-time charge-discharge cycle data for the energy storage power station to be estimated is obtained. The real-time charge-discharge cycle data is input into 10 model files for inference to obtain 10 inference results. The average value of the 10 inference results is taken to obtain the SOH estimation result, so as to estimate the SOH of the battery in the energy storage power station.

[0117] In summary, the advantages of this invention are:

[0118] A training dataset was constructed by acquiring a large amount of historical charge-discharge data of batteries in an energy storage power station and labeling it with State of Health (SOH) values. A test dataset was then built based on the training dataset. Next, based on the training and test datasets, first and second historical aggregated features were extracted. Then, a first battery SOH estimation model and a second battery SOH estimation model were created based on the LightGBM model. The first battery SOH estimation model was trained using the first historical aggregated features and SOH label values, and the second battery SOH estimation model was trained using the second historical aggregated features and SOH label values. During training, an early stop strategy and a 5-fold cross-validation method were combined. During the validation process using the 5-fold cross-validation method, the first and second battery SOH estimation models each generated 5 model files. Finally, the SOH of the energy storage power station batteries was estimated using 10 model files. Since the test dataset was constructed based on the training dataset, only charge-discharge times, cycle periods, battery states, second charging capacity, second discharging capacity, and rated capacity were extracted from the training dataset. The method avoids selecting room temperature capacity, voltage, current, temperature, and SOC, thus avoiding the impact of differences in input data distribution between the training dataset (source domain) and the test dataset (target domain) under different charge and discharge conditions, which could lead to a decrease in model performance and improve the generalization of battery SOH estimation. Based on the first and second cycle aggregation features, the first and second historical aggregation features are extracted respectively. That is, the historical charge and discharge data (cycle aggregation features) are accumulated and analyzed to extract features, which can make full use of historical data and leverage the advantages of continuous monitoring of energy storage battery data. By accumulating and analyzing features from the data of the entire battery life cycle (charging, resting, discharging), the comprehensiveness of feature extraction is ensured, which ultimately greatly improves the accuracy of SOH estimation of energy storage power station batteries.

[0119] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for estimating the state of harm (SOH) of a battery in an energy storage power station, characterized in that: Includes the following steps: Step S1: Obtain a large amount of historical charge-discharge data of the energy storage power station battery, and construct a training dataset after labeling each of the historical charge-discharge data with SOH tag values; Step S2: Construct a test dataset based on the training dataset; Step S3: Based on the training dataset and the test dataset, extract the first historical aggregation feature and the second historical aggregation feature; Step S4: Create a first battery SOH estimation model and a second battery SOH estimation model based on the lightGBM model. Train the first battery SOH estimation model using the first historical aggregation features and SOH label values, and train the second battery SOH estimation model using the second historical aggregation features and SOH label values. Step S5: During the training process, the first battery SOH estimation model and the second battery SOH estimation model combine the early training stop strategy and the 5-fold cross-validation method. During the verification process using the 5-fold cross-validation method, the first battery SOH estimation model and the second battery SOH estimation model respectively generate 5 model files through inference. Step S6: Estimate the SOH of the battery in the energy storage power station using the 10 model files mentioned above; Step S3 specifically involves: Based on the battery state statistics in the training dataset, the number of consecutive first charging segments, first discharging segments, and first static segments in each cycle are counted. The number of first charging cycles, first discharging cycles, and first static cycles in each cycle are calculated based on these counts. The total number of first charging segments, first discharging segments, and first static segments, as well as the total number of first charging cycles, first discharging cycles, and first static cycles, are calculated within each cycle. Based on the ratio of the number of first charging segments, first discharging segments, and first static segments to the total number of segments, and the ratio of the number of first charging cycles, first discharging cycles, and first static cycles to the total number of cycles, the first cycle aggregation features of each cycle are extracted. Based on the battery state statistics of the test dataset, the number of consecutive second charging segments, second discharging segments, and second static segments for each day are counted. The number of second charging cycles, second discharging cycles, and second static cycles for each day are calculated based on these counts. The total number of second charging segments, second discharging segments, and second static segments, as well as the total number of second charging cycles, second discharging cycles, and second static cycles for each day are calculated. Based on the ratio of the number of second charging segments, second discharging segments, and second static segments to the total number of second segments, and the ratio of the number of second charging cycles, second discharging cycles, and second static cycles to the total number of second segments, the second cycle aggregation features for each day are extracted. Both the first and second cycle aggregation features include the proportion of charging times, the proportion of discharging times, the proportion of resting times, the proportion of charging capacity relative to rated capacity, the proportion of discharging capacity relative to rated capacity, the proportion of discharging capacity relative to charging capacity, and the difference between discharging capacity and charging capacity. The first historical aggregated feature is obtained by accumulating the first period aggregated feature for each cycle, and the second historical aggregated feature is obtained by accumulating the second period aggregated feature for each day.

2. The method for estimating the state of harm (SOH) of a battery in an energy storage power station as described in claim 1, characterized in that: Step S1 specifically involves: Acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. The historical charge-discharge data includes at least the charge-discharge time, cycle period, battery status, first charging capacity, first discharging capacity, rated capacity, room temperature capacity, voltage, current, temperature, and SOC. The cycle period refers to the number of times the battery is charged and discharged. A complete cycle period includes the entire process of charging, resting, and discharging. The battery state is charging, resting, or discharging. The first charging capacity is the accumulated charging capacity of the current cycle period. The first discharging capacity is the accumulated discharging capacity of the current cycle period. The room temperature capacity is the calibrated capacity of the battery at room temperature after thermal equilibrium is achieved. The SOH tag value is labeled for each of the cycle charge and discharge history data. The calculation process of the SOH tag value is as follows: after performing bilinear difference on the room temperature capacity at intervals of N cycle cycles, divide by the rated capacity to obtain the SOH tag value with a value range of 0~1. A training dataset is constructed based on the labeled historical charge-discharge data of each cycle.

3. The method for estimating the state of harm (SOH) of a battery in an energy storage power station as described in claim 1, characterized in that: Step S2 specifically involves: From the training dataset, charge / discharge times, cycle periods, battery states, second charge capacity, second discharge capacity, and rated capacity are selected in days to construct a test dataset; The second charging capacity is the cumulative charging capacity for one day; The second discharge capacity is the cumulative discharge capacity over one day.

4. The method for estimating the state of harm (SOH) of a battery in an energy storage power station as described in claim 1, characterized in that: Step S6 specifically involves: Real-time charge-discharge cycle data for the energy storage power station to be estimated is obtained. The real-time charge-discharge cycle data is input into 10 model files for inference to obtain 10 inference results. The average value of the 10 inference results is taken to obtain the SOH estimation result, so as to estimate the SOH of the battery in the energy storage power station.

5. A battery SOH estimation system for an energy storage power station, characterized in that: Includes the following modules: The training dataset construction module is used to acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. After labeling each of the historical charge-discharge data with SOH tag values, the training dataset is constructed. A test dataset construction module is used to construct a test dataset based on the training dataset. The historical aggregation feature extraction module is used to extract a first historical aggregation feature and a second historical aggregation feature based on the training dataset and the test dataset. The battery SOH estimation model training module is used to create a first battery SOH estimation model and a second battery SOH estimation model based on the lightGBM model. The first battery SOH estimation model is trained using the first historical aggregated features and SOH label values, and the second battery SOH estimation model is trained using the second historical aggregated features and SOH label values. The model file inference module is used to generate 5 model files by combining the early stopping strategy and the 5-fold cross-validation method during the training process of the first battery SOH estimation model and the second battery SOH estimation model. The SOH estimation module is used to estimate the SOH of batteries in energy storage power stations using the 10 model files mentioned above. The historical aggregation feature extraction module is specifically used for: Based on the battery state statistics in the training dataset, the number of consecutive first charging segments, first discharging segments, and first static segments in each cycle are counted. The number of first charging cycles, first discharging cycles, and first static cycles in each cycle are calculated based on these counts. The total number of first charging segments, first discharging segments, and first static segments, as well as the total number of first charging cycles, first discharging cycles, and first static cycles, are calculated within each cycle. Based on the ratio of the number of first charging segments, first discharging segments, and first static segments to the total number of segments, and the ratio of the number of first charging cycles, first discharging cycles, and first static cycles to the total number of cycles, the first cycle aggregation features of each cycle are extracted. Based on the battery state statistics of the test dataset, the number of consecutive second charging segments, second discharging segments, and second static segments for each day are counted. The number of second charging cycles, second discharging cycles, and second static cycles for each day are calculated based on these counts. The total number of second charging segments, second discharging segments, and second static segments, as well as the total number of second charging cycles, second discharging cycles, and second static cycles for each day are calculated. Based on the ratio of the number of second charging segments, second discharging segments, and second static segments to the total number of second segments, and the ratio of the number of second charging cycles, second discharging cycles, and second static cycles to the total number of second segments, the second cycle aggregation features for each day are extracted. Both the first and second cycle aggregation features include the proportion of charging times, the proportion of discharging times, the proportion of resting times, the proportion of charging capacity relative to rated capacity, the proportion of discharging capacity relative to rated capacity, the proportion of discharging capacity relative to charging capacity, and the difference between discharging capacity and charging capacity. The first historical aggregated feature is obtained by accumulating the first period aggregated feature for each cycle, and the second historical aggregated feature is obtained by accumulating the second period aggregated feature for each day.

6. The SOH estimation system for a battery in an energy storage power station as described in claim 5, characterized in that: The training dataset construction module is specifically used for: Acquire a large amount of historical charge-discharge data of batteries in energy storage power stations. The historical charge-discharge data includes at least the charge-discharge time, cycle period, battery status, first charging capacity, first discharging capacity, rated capacity, room temperature capacity, voltage, current, temperature, and SOC. The cycle period refers to the number of times the battery is charged and discharged. A complete cycle period includes the entire process of charging, resting, and discharging. The battery state is charging, resting, or discharging. The first charging capacity is the accumulated charging capacity of the current cycle period. The first discharging capacity is the accumulated discharging capacity of the current cycle period. The room temperature capacity is the calibrated capacity of the battery at room temperature after thermal equilibrium is achieved. The SOH tag value is labeled for each of the cycle charge and discharge history data. The calculation process of the SOH tag value is as follows: after performing bilinear difference on the room temperature capacity at intervals of N cycle cycles, divide by the rated capacity to obtain the SOH tag value with a value range of 0~1. A training dataset is constructed based on the labeled historical charge-discharge data of each cycle.

7. The SOH estimation system for a battery in an energy storage power station as described in claim 5, characterized in that: The test dataset construction module is specifically used for: From the training dataset, charge / discharge times, cycle periods, battery states, second charge capacity, second discharge capacity, and rated capacity are selected in days to construct a test dataset; The second charging capacity is the cumulative charging capacity for one day; The second discharge capacity is the cumulative discharge capacity over one day.

8. The SOH estimation system for a battery in an energy storage power station as described in claim 5, characterized in that: The SOH estimation module is specifically used for: Real-time charge-discharge cycle data for the energy storage power station to be estimated is obtained. The real-time charge-discharge cycle data is input into 10 model files for inference to obtain 10 inference results. The average value of the 10 inference results is taken to obtain the SOH estimation result, so as to estimate the SOH of the battery in the energy storage power station.

Images