A method and device for estimating the state of health of an energy storage device in coordination with a microgrid
Patent Information
- Application Number
- CN202611192866.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-07
- Publication Date
- 2026-10-30
Smart Images
Figure CN123019195A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of SOH estimation technology, specifically to a method and equipment for SOH estimation of energy storage devices under the coordination of distribution microgrids. Background Technology
[0002] With the increasing coordination between distribution networks and microgrids, energy storage systems are playing an increasingly important role in source-grid-load-storage coordination, renewable energy consumption, power quality maintenance, and emergency power support. During long-term operation, energy storage batteries are affected by factors such as cycle count, ambient temperature, charge / discharge rate, grid dispatch strategies, and load fluctuations, resulting in gradual capacity decay and internal resistance changes. Therefore, continuous estimation of their State of Health (SOH) is necessary. Traditional SOH estimation methods typically only collect local data from the battery itself (such as voltage, current, and temperature). However, energy storage devices operate in microgrids, and their aging process is significantly influenced by the global operating environment of the distribution network (such as load fluctuations, grid connection voltage deviations, and peak-valley dispatch strategies). Existing technologies neglect the coupled influence of global operating data from the distribution network on battery aging characteristics, leading to decreased SOH estimation accuracy under complex operating conditions. Summary of the Invention
[0003] The purpose of this invention is to provide a method and device for SOH estimation of energy storage devices under the coordination of distribution microgrids, which realizes high-precision and low-computing-power estimation of SOH of energy storage devices under the coordination of distribution microgrids.
[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: This invention provides a method for estimating the state of energy (SOH) of an energy storage device under the coordination of a distribution microgrid, comprising: S1. Collect local operating data of energy storage devices on the microgrid side and global operating data on the distribution network side; S2. Based on the local operating data, extract local lightweight features within the local voltage range of the energy storage device, and extract operating condition correction features based on the global operating data; S3. Using an attention mechanism, dynamically adjust the weights of the local lightweight features based on the working condition correction features, and then perform micro-cooperative feature fusion to generate a cooperative feature vector. S4. Call the pre-stored parameter template according to the battery type of the energy storage device, and dynamically adjust the noise parameters of the parameter template according to the operating condition correction characteristics to construct the corresponding state space model. S5. Based on the state-space model, perform Kalman filtering estimation on the cooperative feature vector to obtain the SOH estimation result.
[0005] Optionally, the step of extracting local lightweight features within the local voltage range of the energy storage device based on the local operating data, and extracting operating condition correction features based on the global operating data, includes: S21. Select the local voltage range of the microgrid-side energy storage device during the charging process based on the local operating data; S22. Extract local lightweight features within the selected local voltage range; the local lightweight features include incremental capacity peak features and constant voltage charging capacity features; the incremental capacity peak features characterize the capacity decay of the energy storage device; the constant voltage charging capacity features characterize the internal resistance change and capacity degradation of the energy storage device; S23. Extract operating condition correction features based on the global operating data. The operating condition correction features include load fluctuation coefficient, voltage deviation coefficient, charge-discharge coordination factor, and temperature influence coefficient.
[0006] Optionally, the step of using an attention mechanism to dynamically adjust the weights of the local lightweight features based on the working condition correction features, and then performing micro-cooperative feature fusion to generate a cooperative feature vector, includes: S31. Map the working condition correction features to query vectors, and map the local lightweight features to key vectors and value vectors respectively; S32. Calculate the relevance score between the query vector and the key vector, and generate dynamic attention weights based on the relevance score; S33. The value vector is weighted and fused based on the dynamic attention weights to generate the collaborative feature vector.
[0007] Optionally, the noise parameters include process noise parameters and observed noise parameters; the step of dynamically adjusting the noise parameters of the parameter template according to the operating condition correction characteristics includes: S41. Obtain the noise gain factor based on the load fluctuation coefficient and voltage deviation coefficient in the operating condition correction characteristics; S42. The process noise parameters and the observation noise parameters are dynamically adjusted and corrected using the noise gain factor.
[0008] Optionally, the corresponding state-space model can be constructed in the following ways: S43. Construct a dual observation channel based on the collaborative feature vector; the dual observation channel includes a first observation channel constructed based on incremental capacity peak features and a second observation channel constructed based on constant voltage charging capacity features; S44. Use the incremental capacity peak characteristics to drive the state update channel and recursively calculate the battery capacity state of the energy storage device. S45. The constant voltage charging capacity feature is used to drive the parameter adaptive channel to correct the parameters of the state space model in real time.
[0009] Optionally, it also includes: S6, uploading the collaborative feature vector, SOH estimation results, local lightweight features and operating condition correction features to the cloud platform; the cloud platform optimizes the parameter template of the state space model based on the uploaded data, and sends the optimized parameter template to the edge terminal for updating.
[0010] Optionally, the step of collecting local operating data of the energy storage device on the microgrid side and global operating data on the distribution network side further includes: S11, preprocessing the local operating data and the global operating data, wherein the preprocessing includes denoising, standardization, time-series alignment and missing data completion; and missing data completion is performed by combining interpolation and historical trends.
[0011] Optionally, the local operating data characterizes the operating status of the energy storage device, and the local operating data includes voltage data, current data, state of charge data, and temperature data; the global operating data characterizes the external operating environment status of the energy storage device, and the global operating data includes regional total load data, grid connection point voltage fluctuation data, grid frequency data, and peak-valley time period data.
[0012] Optionally, when the battery type of the energy storage device is a ternary lithium battery, the local voltage range is 3.7V to 4.1V.
[0013] The present invention also provides a terminal device, including a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described above.
[0014] The beneficial effects of this invention are as follows: 1. By collecting local operating data from the microgrid side and global operating data from the distribution network side, the limitation of relying solely on local data to estimate SOH is overcome. Utilizing an attention mechanism based on global operating condition correction features, dynamic weight adjustment and fusion of local lightweight features are achieved, realizing feature extraction through distribution-microgrid collaboration. This eliminates the interference of complex grid fluctuations on battery feature extraction, significantly improving feature robustness. Furthermore, by calling parameter templates according to battery type and dynamically adjusting the noise parameters of the state-space model in conjunction with operating condition features, the Kalman filter can adapt to different battery characteristics and changing operating environments, solving the problem of decreased estimation accuracy of fixed-parameter models under changing operating conditions. This method achieves high-precision, low-computing-power SOH estimation for energy storage devices under distribution-microgrid collaboration, balancing data privacy and real-time performance, and significantly improving the management efficiency and security of the energy storage system throughout its entire lifecycle.
[0015] 2. By extracting incremental capacity peak features and constant-voltage charging capacity features within a specific local voltage range during the charging process (3.7V to 4.1V), the core aging characteristics of the battery are accurately captured. Incremental capacity peak features reflect the loss of active lithium ions, characterizing capacity decay; constant-voltage charging capacity features reflect changes in battery internal resistance, thus comprehensively assessing battery capacity degradation and power decline. All extracted features are sensitive to aging. By introducing global operating data from the distribution network side, operating condition correction features such as load fluctuation coefficient, voltage deviation coefficient, charge-discharge synergy factor, and temperature influence coefficient are extracted. This comprehensively quantifies the impact of complex external environments on battery operation, enabling the model to perceive external grid conditions. Combining these two types of features allows SOH estimation to not only rely on the battery's own response but also to be dynamically corrected based on the actual operating environment (operating condition correction features). This suppresses the interference of complex operating conditions (such as severe load fluctuations and voltage deviations) on feature extraction, improving the robustness and generalization ability of the features, and laying a solid foundation for subsequent high-precision state estimation.
[0016] 3. A dynamic noise adjustment mechanism based on operating condition correction characteristics is introduced, solving the problem of traditional Kalman filtering's tendency to diverge under varying operating conditions. Specifically, the noise gain factor is calculated using the load fluctuation coefficient and voltage deviation coefficient, and this is used to correct the process noise and observation noise parameters of the state-space model in real time. This design allows the filter to adapt to changes in the power grid environment. When the load fluctuation is severe or the voltage deviation is large, the automatic gain factor increases the noise covariance to prevent the filter from diverging due to model mismatch; conversely, it reduces the noise weight to ensure the accuracy of steady-state estimation. The constructed dual observation channels achieve accurate decoupling and collaborative estimation of the battery aging physical process. The first observation channel, based on the incremental capacity peak characteristics, is dedicated to driving state updates and accurately capturing the battery's capacity decay trajectory; while the second observation channel, based on the constant voltage charging capacity characteristics, is responsible for driving the parameter adaptive channel, correcting model parameters in real time to track polarization effects such as internal resistance changes. This allows SOH estimation to not only distinguish different types of aging characteristics but also to achieve online model correction through the parameter adaptive channel, thereby maintaining high-accuracy estimation throughout the battery's entire lifespan. Attached Figure Description
[0017] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings.
[0018] Figure 1 This is a flowchart of a method for estimating the State of Harmony (SOH) of an energy storage device under the coordination of a microgrid, as described in this invention. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0020] As one implementation method, such as Figure 1 As shown, this invention provides a method for estimating the State of Health (SOH) of an energy storage device under the coordination of a distribution microgrid. By fusing local charging interval data from the microgrid side and global operation data from the distribution network side, it achieves online estimation of the energy storage battery's health state at the edge terminal with low computational complexity, including: S1. Collect local operating data of the energy storage device on the microgrid side and global operating data on the distribution network side. Local data on the microgrid side includes at least voltage, current, state of charge (SOC), and temperature data of the energy storage battery during charging. Global data on the distribution network side includes at least the regional total load, voltage fluctuation at the grid connection point, grid frequency, and peak / valley time information. Microgrid-side data characterizes the battery's own operating status, while distribution network-side data characterizes the external operating environment and coordinated dispatch conditions of the energy storage. Preprocessing is performed on both local and global operating data, including noise reduction, standardization, time-series alignment, and missing data completion. Missing data is completed using a combination of interpolation and historical trend analysis.
[0021] For continuous data, this could be real-time charging voltage data, charging / discharging current data, or ambient temperature data of the energy storage battery. For example, the voltage data of the energy storage battery can be processed using the z-score normalization method. ; in, This is the raw data, namely the energy storage battery voltage value. This is the historical average, i.e., the average voltage of this battery model under the standard charging curve. The historical standard deviation is used to characterize the dispersion of voltage data. This is the standardized data.
[0022] S2. Based on the local operating data, extract local lightweight features within the local voltage range of the energy storage device, and extract operating condition correction features based on the global operating data.
[0023] S3. Using an attention mechanism, dynamically adjust the weights of the local lightweight features based on the working condition correction features, and then perform micro-cooperative feature fusion to generate a cooperative feature vector.
[0024] S4. Based on the battery type of the energy storage device, call the pre-stored parameter template, and dynamically adjust the noise parameters of the parameter template according to the operating condition correction characteristics to construct the corresponding state space model.
[0025] S5. Based on the state-space model, perform Kalman filtering estimation on the cooperative feature vector to obtain the SOH estimation result.
[0026] It should be noted that the battery type of an energy storage device refers to the specific electrochemical system or battery chemical composition used in the energy storage device. In practical applications, energy storage devices can include, but are not limited to, lithium iron phosphate batteries, ternary lithium batteries, lithium titanate batteries, lead-acid batteries, or flow batteries. Since different types of batteries exhibit significant differences in electrochemical characteristics, open-circuit voltage curves, and aging mechanisms, this invention, by identifying the battery type, can specifically call upon preset parameter templates that match the electrochemical system (e.g., preset a specific initial noise covariance for the flat voltage region characteristics of lithium iron phosphate), thereby laying the foundation for subsequently constructing a high-precision state-space model.
[0027] Local lightweight features refer to feature data that can be extracted within a specific local voltage range (e.g., 3.7V-4.1V) during the operation of energy storage devices, without requiring a complete full charge-discharge cycle. Lightweight features are characterized by low computational complexity and small data volume, making them suitable for real-time processing at edge terminals with limited computing power. They also effectively reflect the battery's capacity decay and internal resistance trends. Operating condition correction features are quantitative indicators calculated based on global operating data from the distribution network side. Their function is to characterize the degree of interference caused by the external grid environment (such as severe load fluctuations and voltage deviations) on the operating state of the energy storage battery. This feature is mainly used for dynamically adjusting the weights of local features in subsequent steps and adaptively correcting the noise parameters of the state-space model to offset the negative impact of external environmental fluctuations on the accuracy of SOH estimation.
[0028] Distribution-microgrid collaborative feature fusion refers to the process of organically combining local lightweight features reflecting the internal aging state of a battery (microgrid-side data) with operating condition correction features reflecting external operating environment disturbances (distribution network-side data) using an attention mechanism. Through this fusion, the generated collaborative feature vector contains both the battery's own health information and contextual information from external operating conditions, achieving cross-level collaborative perception of distribution network and microgrid data. The pre-stored parameter template is a set of basic model parameters pre-calibrated and stored for batteries with different electrochemical systems. This template includes at least: initial state transition coefficients matching this type of battery, initial values of basic process noise covariance, initial values of basic observation noise covariance, and basic curve data of the open-circuit voltage-state-of-charge mapping relationship.
[0029] The method of this invention achieves SOH estimation that is deployable at the edge, has available local data, adapts to changes in operating conditions, and is tolerant of noise and missing data by extracting features of local charging intervals, introducing global operating conditions features of the distribution network, fusing lightweight collaborative features, using a Kalman filter plus a dual observation channel estimation, and employing multi-type battery parameter templates and fine-tuning mechanisms.
[0030] In one implementation, the method of the present invention can be deployed on an edge terminal. The SOH estimation method further includes: S6, uploading the collaborative feature vector, SOH estimation result, local lightweight features, and operating condition correction features to a cloud platform. The cloud platform optimizes the parameter template of the state-space model based on the uploaded data and sends the optimized parameter template to the edge terminal for updating. After receiving the parameters, the edge terminal does not need to retrain the entire model; it only needs to update the local model parameters to continue performing SOH estimation.
[0031] As one implementation, the step of extracting local lightweight features within the local voltage range of the energy storage device based on the local operating data, and extracting operating condition correction features based on the global operating data, includes: S21. Based on the local operating data, select the local voltage range of the microgrid-side energy storage device during the charging process. When the battery type of the energy storage device is a ternary lithium battery, the local voltage range can be 3.7V to 4.1V.
[0032] S22. Extract local lightweight features within the selected local voltage range. These local lightweight features include peak incremental capacity characteristic (PICC) and constant voltage charging capacity characteristic (CVCC). The peak incremental capacity characteristic characterizes the capacity decay of the energy storage device, while the constant voltage charging capacity characteristic characterizes the internal resistance change and capacity degradation of the energy storage device.
[0033] S23. Based on the global operating data, extract operating condition correction features, including load fluctuation coefficient, voltage deviation coefficient, charge-discharge coordination factor, and temperature influence coefficient. The load fluctuation coefficient reflects the severity of load changes on the distribution network side and can be obtained by collecting the total load power sequence of the distribution network within a time window and calculating the standard deviation of its rate of change. The voltage deviation coefficient reflects the impact of grid voltage stability on the operating stress of the energy storage converter and can be obtained by calculating the normalized deviation between the actual operating voltage and the rated voltage. The charge-discharge coordination factor characterizes the matching degree between the distribution network dispatch command and the current state of the energy storage device, i.e., the degree of distribution-micro-coordination, and can be obtained by calculating the correlation between the grid dispatch power command and the actual response power of the energy storage device based on the Pearson correlation coefficient. The temperature influence coefficient can be a correction parameter for the effect of ambient temperature on the battery chemical reaction rate and can be obtained by mapping the ambient temperature into a digitized influence coefficient.
[0034] As one implementation method, the step of dynamically adjusting the weights of the local lightweight features based on the working condition correction features using an attention mechanism, and then performing micro-cooperative feature fusion to generate a cooperative feature vector, includes: S31. Map the working condition correction features to the query vector, and map the local lightweight features to key vectors and value vectors respectively. Key vector generation: Multiply the local lightweight feature vector by a pre-trained key weight matrix to obtain the key vector. This component is mainly used for similarity matching with the query vector to determine the degree of attention of the feature under the current working condition. Value vector generation: Multiply the local lightweight feature vector by a pre-trained value weight matrix to obtain the value vector. This component is mainly used to provide specific feature numerical information during the weighted fusion stage. The key weight matrix and value weight matrix are pre-trained using historical local lightweight feature samples and historical working condition correction feature samples. The training input includes historical local lightweight feature samples and historical working condition correction feature samples. The training objective is to minimize the working condition relevance matching error between the query vector and the key vector.
[0035] S32. Calculate the relevance score between the query vector and the key vector, and generate dynamic attention weights based on the relevance score. The query vector could be a large load fluctuation coefficient of the power grid. If the relevance score between the query key vector "battery temperature" and "load fluctuation coefficient" is high, then the specific numerical information of "battery temperature" is weighted and fused to obtain a collaborative feature vector.
[0036] S33. Weighted fusion of value vectors based on dynamic attention weights to generate collaborative feature vectors, as follows: ; in, Represents the collaborative feature vector. Indicates the first Each input feature Indicates the first The corresponding weights of each input feature satisfy the following conditions: , This represents the total number of input features involved in the fusion. Index variables representing features. Among them, distribution network features are mainly used to correct the deviation of local features under different operating conditions, that is, by adjusting the weights to meet the needs of local feature deviation correction under different operating conditions.
[0037] As one implementation method, the noise parameters include process noise parameters and observation noise parameters. S4, based on the battery type of the energy storage device, a pre-stored parameter template is called, and the noise parameters of the parameter template are dynamically adjusted according to the operating condition correction characteristics to construct a corresponding state-space model. This also includes: S41. The noise gain factor is obtained by weighted summation of the load fluctuation coefficient and voltage deviation coefficient in the operating condition correction characteristics.
[0038] S42. Dynamically adjust and correct process noise parameters and observation noise parameters using noise gain factors.
[0039] S43. Construct a dual observation channel based on collaborative feature vectors. The dual observation channel includes a first observation channel constructed based on incremental capacity peak features and a second observation channel constructed based on constant voltage charging capacity features.
[0040] S44. Use the incremental capacity peak characteristics to drive the state update channel and recursively calculate the battery capacity state of the energy storage device.
[0041] S45. The constant voltage charging capacity characteristic is used to drive the parameter adaptive channel to correct the parameters of the state space model in real time.
[0042] Specifically, an adaptive state-space model is established based on different types of energy storage batteries. The battery type is identified based on the battery nameplate information or a preset type identifier, and the corresponding parameter template is selected. The state-space equation can be simplified as follows: ; ; ; in, Indicates the first Capacity state at an estimated time point. Indicates the first Capacity state at an estimated time point. Indicates the first The collaborative feature vector at each estimated time point, and These represent the dual observation channels consisting of PICC and CVCC, respectively. , Let A, B, C, and D represent the incremental capacity peak characteristic and constant voltage charging capacity characteristic at the i-th estimation time, respectively. A, B, C, and D are model parameters related to the battery type. , , They represent the first The process noise, first observation channel noise, and second observation channel noise are estimated at each time point. The process noise, first observation channel noise, and second observation channel noise can be zero-mean Gaussian white noise, and their covariance matrix is dynamically adjusted according to the operating condition correction characteristics of the distribution microgrid to adapt to the filtering accuracy requirements under different load fluctuations and voltage deviations.
[0043] Parameter A is the recursive coefficient of capacity state, reflecting the natural degradation trend of battery capacity; parameter B is the influence coefficient of the cooperative feature vector on capacity state, preset or corrected online according to the aging characteristics of battery type (e.g., lithium iron phosphate, ternary lithium). Parameter C is the incremental capacity peak characteristic. To the observed value The mapping coefficients are used to convert capacity decay characteristics into a relationship with capacity state. The observed values are of the same dimension, and their values are determined based on the incremental capacity characteristics of different battery types. Parameter D represents the constant voltage charging capacity characteristic. To the observed value The mapping coefficients are used to convert the internal resistance and capacity degradation characteristics into the capacity state. The observed values of the same dimension are determined based on the constant voltage charging characteristics of different battery types.
[0044] Subsequently, in the improved Kalman filter plus dual-observation-channel estimation step, state prediction is performed first: ; In the formula, Indicates at time The prior prediction of the current battery capacity state is based on the state estimate of the previous moment. Indicates time The posterior capacity state estimate at the previous time step The optimal estimate of battery capacity, obtained after correcting for changes in the observed data at the time, is used as the initial baseline value for this state prediction. (The first value in the subscript...) Indicates the estimated target time; the second in the subscript. (After the vertical line) indicates the cutoff time of the observation data used for estimation. This parameter reflects the recursive nature of Kalman filtering, meaning that the optimal result at the previous time step serves as the starting point for prediction at the next time step.
[0045] Then, based on the dual observation channels, Kalman gain is calculated and updated posteriorly. The Kalman gain can be expressed as: ; in, Indicates time The Kalman gain matrix, Indicates time The predicted covariance matrix, Represents the observation matrix. This represents the transpose of the observation matrix. The observation noise covariance matrix represents the posterior state update, which can be expressed as: ; in, Indicates time The posterior capacity state estimate is the optimal estimate after fusing observation information. The fused observations formed by the dual observation channels are composed of both observation channels (incremental capacity peak and constant voltage charging capacity). To enhance anti-interference capability, the process noise covariance and observation noise covariance are dynamically adjusted based on the current data quality, so that the filter appropriately reduces its dependence on unreliable observations in scenarios with high noise or missing data.
[0046] Finally, SOH is calculated based on the posterior capacity estimate: ; in, Indicates time Battery health status estimate, This indicates the initial rated capacity.
[0047] As one implementation method, in a microgrid collaborative operation scenario, a lithium iron phosphate energy storage system is configured within a microgrid, while the distribution network side manages multiple microgrid energy storage units simultaneously. Edge terminals collect real-time data on the voltage, current, and state of charge of the energy storage batteries within the 3.7V to 4.1V range during charging. Simultaneously, they obtain information from the distribution network side regarding the total regional load, grid connection point voltage deviation, and peak / valley time periods. After synchronizing and preprocessing the local and global data, the edge terminals extract local charging characteristics (PICC and CVCC), as well as load fluctuation coefficients, voltage deviation coefficients, charge / discharge coordination factors, and temperature influence coefficients.
[0048] Suppose that in a certain estimation process, the extracted six-dimensional features are as follows: PICC = 0.82, CVCC = 0.76, load fluctuation coefficient = 0.35, voltage deviation coefficient = 0.08, charge / discharge synergy factor = 0.42, and temperature influence coefficient = 0.10. If the corresponding weights are set to PICC = 0.30, CVCC = 0.25, load fluctuation coefficient = 0.15, voltage deviation coefficient = 0.15, charge / discharge synergy factor = 0.10, and temperature influence coefficient = 0.05, then the synergy feature fusion result can be calculated using a weighted method as follows: ; The calculation yielded: ; If the energy storage system is identified as a lithium iron phosphate battery, then the lithium iron phosphate parameter template is invoked. Assume the posterior capacity estimate from the previous time step is... In the parameter template , ,but The time-predicted capacity can be expressed as: ; We can obtain: ; Subsequently, the system constructs a dual observable based on PICC and CVCC, and performs posterior correction on the predicted values. If the posterior update yields... The initial rated capacity Then the estimated value of SOH is: ; In noisy environments, if the acquired signal experiences additional measurement noise or missing data, the edge terminal improves signal quality through preprocessing. It also adaptively increases the observation noise covariance based on the current data quality by using an improved Kalman filter and a dual-observation channel, thereby reducing the impact of abnormal observations on the estimation results. If the edge terminal accumulates new cooperative features and SOH estimation results after running continuously for a day, this data is uploaded to the cloud platform. The cloud platform then re-estimates the parameter template based on historical data from multiple microgrids and issues an update. After receiving the update, the edge terminal does not need to be retrained entirely; it only needs to update its local parameters to continue estimation. This embodiment demonstrates that the present invention can fully cover the entire process of multi-source data acquisition for distribution microgrids, local area feature extraction, cooperative feature fusion, multi-type battery adaptation, improved Kalman filter and dual-observation channel estimation, and cloud-edge iterative updates.
[0049] Without departing from the core concept of this invention, other alternative methods can be used to achieve the same inventive objective. For example, in the lightweight filtering estimation part, an extended Kalman filter or an unscented Kalman filter suitable for nonlinear state spaces can be used instead of a dual Kalman filter. In the local feature extraction part, other local interval features that can reflect capacity decay and internal resistance changes can be used. In the collaborative feature fusion part, dimensionality compression or principal component extraction can be used instead of simple weighted fusion. In the iterative optimization part, event-triggered parameter updates can be used instead of fixed-period updates. In the multi-type battery adaptation part, classification recognition plus parameter regression can be used instead of template calling and fine-tuning. The above alternative solutions are all reasonable variations of the technical concept of this invention.
[0050] The present invention also provides a terminal device, including a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described above.
[0051] The above-described specific embodiments are preferred embodiments of the SOH estimation method and equipment for energy storage devices under microgrid coordination according to this application, and are not intended to limit the specific implementation scope of this application. The scope of this application includes but is not limited to the specific embodiments described herein. All equivalent changes made in accordance with the shape and structure of this application are within the protection scope of this application.
Claims
1. A method for estimating the State of Harmony (SOH) of an energy storage device under microgrid coordination, characterized in that, include: S1. Collect local operating data of energy storage devices on the microgrid side and global operating data on the distribution network side; S2. Based on the local operating data, extract local lightweight features within the local voltage range of the energy storage device, and extract operating condition correction features based on the global operating data; S3. Using an attention mechanism, dynamically adjust the weights of the local lightweight features based on the working condition correction features, and then perform micro-cooperative feature fusion to generate a cooperative feature vector. S4. Call the pre-stored parameter template according to the battery type of the energy storage device, and dynamically adjust the noise parameters of the parameter template according to the operating condition correction characteristics to construct the corresponding state space model. S5. Based on the state-space model, perform Kalman filtering estimation on the cooperative feature vector to obtain the SOH estimation result.
2. The SOH estimation method for an energy storage device under microgrid coordination according to claim 1, characterized in that, The steps of extracting local lightweight features within the local voltage range of the energy storage device based on the local operating data, and extracting operating condition correction features based on the global operating data, include: S21. Select the local voltage range of the microgrid-side energy storage device during the charging process based on the local operating data; S22. Extract local lightweight features within the selected local voltage range; the local lightweight features include incremental capacity peak features and constant voltage charging capacity features; the incremental capacity peak features characterize the capacity decay of the energy storage device; the constant voltage charging capacity features characterize the internal resistance change and capacity degradation of the energy storage device; S23. Extract operating condition correction features based on the global operating data. The operating condition correction features include load fluctuation coefficient, voltage deviation coefficient, charge-discharge coordination factor, and temperature influence coefficient.
3. The SOH estimation method for an energy storage device under microgrid coordination according to claim 1, characterized in that, The step of using an attention mechanism to dynamically adjust the weights of the local lightweight features based on the working condition correction features, and then performing micro-cooperative feature fusion to generate a cooperative feature vector, includes: S31. Map the working condition correction features to query vectors, and map the local lightweight features to key vectors and value vectors respectively; S32. Calculate the relevance score between the query vector and the key vector, and generate dynamic attention weights based on the relevance score; S33. The value vector is weighted and fused based on the dynamic attention weights to generate the collaborative feature vector.
4. The SOH estimation method for an energy storage device under microgrid coordination according to claim 2, characterized in that, The noise parameters include process noise parameters and observation noise parameters; The step of dynamically adjusting the noise parameters of the parameter template based on the operating condition correction characteristics includes: S41. Obtain the noise gain factor based on the load fluctuation coefficient and voltage deviation coefficient in the operating condition correction characteristics; S42. The process noise parameters and the observation noise parameters are dynamically adjusted and corrected using the noise gain factor.
5. The SOH estimation method for an energy storage device under microgrid coordination according to claim 4, characterized in that, The corresponding state-space model is constructed in the following way: S43. Construct a dual observation channel based on the collaborative feature vector; the dual observation channel includes a first observation channel constructed based on incremental capacity peak features and a second observation channel constructed based on constant voltage charging capacity features; S44. Use the incremental capacity peak characteristics to drive the state update channel and recursively calculate the battery capacity state of the energy storage device. S45. The constant voltage charging capacity feature is used to drive the parameter adaptive channel to correct the parameters of the state space model in real time.
6. The SOH estimation method for an energy storage device under microgrid coordination according to claim 1, characterized in that, Also includes: S6. Upload the collaborative feature vector, SOH estimation results, local lightweight features, and operating condition correction features to the cloud platform; the cloud platform optimizes the parameter template of the state space model based on the uploaded data, and sends the optimized parameter template to the edge terminal for updating.
7. The SOH estimation method for an energy storage device under microgrid coordination according to claim 1, characterized in that, The steps of collecting local operating data of the microgrid-side energy storage device and global operating data of the distribution network side further include: S11, preprocessing the local operating data and the global operating data, wherein the preprocessing includes noise reduction, standardization, time-series alignment and missing data completion; and missing data completion is performed by combining interpolation and historical trends.
8. The SOH estimation method for an energy storage device under microgrid coordination according to claim 1, characterized in that, The local operating data characterizes the operating status of the energy storage device, and includes voltage data, current data, state of charge data, and temperature data; the global operating data characterizes the external operating environment of the energy storage device, and includes regional total load data, grid connection point voltage fluctuation data, grid frequency data, and peak-valley time period data.
9. The SOH estimation method for an energy storage device under microgrid coordination according to claim 2, characterized in that, When the energy storage device uses a ternary lithium battery, the local voltage range is 3.7V to 4.1V.
10. A terminal device, characterized in that, The method includes a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the method as described in any one of claims 1 to 9.