A joint correction method for SOC / SOE of energy storage power station based on Kalman filter fusion and multi-parameter optimization

Abstract

The present invention discloses a joint correction method for the SOC / SOE of an energy storage power station by integrating Kalman filtering with multi-parameter optimization. The method comprises the steps of constructing a state model of an energy storage system, designing an observation model, initializing a Kalman filter, executing prediction and updating, introducing a multi-parameter optimization strategy, and outputting correction results. The method can realize accurate correction of the SOC / SOE in a dynamic environment through the Kalman filtering algorithm, improve the correction efficiency and reliability by combining multi-parameter optimization, overcome the shortcomings of the prior art in terms of multi-parameter coupling, signal delay, and adaptability to complex working conditions, provide a theoretical basis and practical support for the efficient operation of the energy storage power station, and significantly improve the correction accuracy and system performance.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage technology and battery management system technology, and in particular to a SOC / SOE joint correction method for an energy storage power station by integrating Kalman filtering with multi-parameter optimization. Background Art

[0002] In recent years, the rapid development of energy storage technology has driven the widespread application of energy storage power plants in power systems. As core parameters of energy storage plants, accurate calibration of SOC (State of Charge) and SOE (State of Energy) is crucial for improving their performance. However, existing SOC / SOE calibration methods still have limitations in terms of multi-parameter integration, adaptability to dynamic environments, and real-time optimization. These issues have, to a certain extent, impacted the reliability and efficiency of energy storage plant operations.

[0003] After searching, it was found that a patent with publication number CN113810227B, with a publication date of August 20, 2024, involves a master-slave switching method and a power station. This technology uses synchronization signals to achieve the switching and correction of the operating modes of the master and standby equipment, effectively improving equipment utilization and reducing manual operation costs. However, this solution is mainly aimed at the switching and fault recovery of equipment operating modes, and does not involve the specific method of SOC / SOE joint correction of energy storage power stations, and lacks the real-time perception and optimization capabilities of dynamic changes in multiple parameters. In addition, this solution relies on the judgment of synchronization signals. Under complex working conditions, the correction accuracy may decrease due to signal delays or interference, which makes it difficult to meet the needs of energy storage power stations for high-precision SOC / SOE correction.

[0004] Another patent with the publication number CN118350056B, published on September 13, 2024, proposes a method and system for automatic information acceptance of the substation virtual monitoring background. This technology calibrates the target monitoring information through full-phase spectrum correction technology, and combines it with automatic acceptance standards to complete the screening and review verification of the monitoring information, thereby realizing the automatic collection and acceptance of monitoring information. However, this solution is mainly used for the calibration and acceptance of monitoring information, and does not involve the core algorithm of the SOC / SOE joint correction of energy storage power stations. Its correction technology is limited to spectrum analysis, which makes it difficult to cope with the complexity of the multi-parameter coupling relationship in the energy storage system. At the same time, it lacks the integrated application of advanced algorithms such as Kalman filtering, resulting in certain limitations on adaptability and accuracy in dynamic environments.

[0005] The above issues indicate that existing technical solutions still have significant deficiencies in the joint SOC / SOE correction of energy storage power plants, particularly in terms of multi-parameter optimization fusion, adaptability to dynamic environments, and real-time correction accuracy. Therefore, the present invention proposes a joint SOC / SOE correction method for energy storage power plants based on Kalman filtering and multi-parameter optimization. This method aims to achieve precise correction in dynamic environments by introducing a Kalman filter algorithm, while simultaneously improving correction efficiency and reliability through the combination of a multi-parameter optimization strategy, thereby meeting the energy storage power plant's demand for efficient and accurate SOC / SOE management. Summary of the Invention

[0006] The purpose of this invention is to provide a method for joint SOC / SOE correction of energy storage power plants using a Kalman filter and multi-parameter optimization. This method uses a Kalman filter algorithm to achieve precise correction in dynamic environments, while also combining a multi-parameter optimization strategy to improve correction efficiency and reliability. This overcomes the shortcomings of existing technologies in terms of multi-parameter fusion, dynamic environment adaptability, and real-time optimization.

[0007] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0008] The present invention comprises the following steps:

[0009] S1: Constructing the energy storage system state model: Based on the physical characteristics of the energy storage system, a dynamic state equation including SOC and SOE is established;

[0010] S2: Design observation model: Based on the actual operating data of the energy storage system, establish observation equations to describe the relationship between the measured values ​​and the true values ​​of SOC and SOE;

[0011] S3: Initialize the Kalman filter: according to the initial state and measurement value of the energy storage system, set the initial state estimate and the initial error covariance matrix;

[0012] S4: Execute Kalman filter prediction: use the state equation to predict the state at the next moment and update the error covariance matrix;

[0013] S5: Execute Kalman filter update: modify the predicted state based on the observed value and update the error covariance matrix;

[0014] S6: Introducing a multi-parameter optimization strategy: Based on the Kalman filter, the parameters of the state transfer matrix and the observation matrix are adjusted through the optimization algorithm;

[0015] S7: Correction result output: Output the final corrected SOC and SOE values ​​to the energy storage management system;

[0016] S8: Dynamic environmental adaptability assessment: Evaluate the adaptability of the correction method in a complex environment by simulating the operation data of the energy storage system under different working conditions.

[0017] The beneficial effects of the present invention are:

[0018] This invention proposes a joint correction method for the SOC / SOE of an energy storage power station that combines Kalman filtering with multi-parameter optimization. Compared to existing technologies, this invention establishes a correction framework that combines Kalman filtering and multi-parameter optimization, which can be used to verify the applicability and accuracy of the correction method in dynamic environments. This framework achieves real-time correction of the SOC and SOE by modeling the state equation and observation equation of the energy storage system, further improving and developing existing SOC / SOE correction methods. Furthermore, a major innovation of this correction method lies in its simultaneous consideration of the dynamic characteristics of the energy storage system's state changes and the influence of multi-parameter coupling relationships, which has not been addressed in previous research. Correction methods that rely solely on spectral analysis have difficulty addressing the complex multi-parameter coupling relationships in energy storage systems and have poor adaptability to dynamic environments. Correction methods that rely solely on synchronization signals may reduce correction accuracy due to signal delays or interference. Therefore, to address the shortcomings of existing technologies, this invention proposes a joint correction method that combines Kalman filtering and multi-parameter optimization to explore the impact of the correction process on the operating performance of the energy storage system, thereby improving the reliability and efficiency of energy storage power stations.

[0019] In theory, the method of the present invention, on the one hand, expands the theoretical framework of SOC / SOE correction of energy storage power stations, broadens the application scope of the Kalman filter algorithm in energy storage systems, and makes it a complete correction system. On the other hand, it uses mathematical modeling methods to provide a basis for the efficient operation of energy storage power stations. In practical applications, this method enables the energy storage system to more accurately reflect its actual state during operation through accurate correction of SOC and SOE, which is conducive to improving the operating efficiency of energy storage power stations. Secondly, this mechanism can guide energy storage system designers to pay attention to the complexity of multi-parameter coupling relationships, thereby optimizing the overall performance of the energy storage system. Finally, this mechanism utilizes the adaptive characteristics of the Kalman filter algorithm, which is more flexible and effective than traditional spectrum analysis methods, and is conducive to achieving high-precision SOC / SOE correction under complex working conditions, meeting the needs of energy storage power stations for efficient and precise management. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic flow diagram of the present invention. DETAILED DESCRIPTION

[0021] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0022] The present invention comprises the following steps:

[0023] S1: Constructing the energy storage system state model: Based on the physical characteristics of the energy storage system, a dynamic state equation including SOC and SOE is established;

[0024] In practical applications, the state model construction step 1 is first performed. The core of this step is to establish a dynamic state equation including SOC and SOE based on the physical characteristics of the energy storage system:

[0025] Let the state vector at time k be defined as:

[0026]

[0027] in: Indicates the The state vector at time t, is the state of charge, is the energy state, which is the main state quantity. is the average temperature of the battery pack, which is used to consider the impact of temperature on battery capacity and internal resistance; It is the equivalent internal resistance correction value, which is used to compensate for the change of internal resistance due to aging and temperature;

[0028] Considering the nonlinear characteristics of the battery at different currents and temperatures, a first-order RC circuit model is used to describe the state transition:

[0029]

[0030]

[0031]

[0032]

[0033] in: The current measured at the last moment, a positive value indicates discharge, and a negative value indicates charge; It is a function of charge and discharge efficiency that changes with current and temperature, which can be approximated by looking up a table or using an empirical formula; is the effective capacity of the battery, which varies with both the temperature T and the state of health SOH (obtained by remaining capacity test or online estimation); It is a function of open circuit voltage and can be obtained through pre-offline calibration or empirical formula; is the equivalent internal resistance, considering the influence of temperature and SOC; is the total mass of the battery pack, is the specific heat capacity, is the thermal time constant, is the continuous change rate of internal resistance caused by aging, is the process noise;

[0034] Due to the nonlinearity of the above state transition, Unscented Transform is used for approximation. For the state with dimension n=4, 2n+1 Sigma points are generated:

[0035]

[0036]

[0037]

[0038] in:

[0039] is the posterior state estimate at the previous moment, is the covariance, is the scaling parameter of UKF, take Represents the matrix The lower triangular matrix obtained after Cholesky decomposition has its i-th column as the increase and decrease operators. These Sigma points are mapped by the nonlinear state transfer function and used to predict the state distribution at the next moment. In this step, the state of charge parameters need to be set according to the actual operating characteristics of the energy storage unit, such as the capacity decay rate of the energy storage unit and the impact of temperature changes on battery performance. The energy state is calculated based on parameters such as the charge and discharge efficiency and internal resistance of the energy storage system. The construction of this state model provides the basic framework for the subsequent implementation of the Kalman filter algorithm.

[0040] S2: Design observation model: Based on the actual operating data of the energy storage system, establish observation equations to describe the relationship between the measured values ​​and the true values ​​of SOC and SOE;

[0041] Next, we move on to observation model design 2. The goal of this step is to establish observation equations to describe the relationship between the measured values ​​and true values ​​of SOC and SOE. The observation equations are:

[0042] Let the observation vector be:

[0043]

[0044] in: is the SOC calculated by the ampere-hour integration method, is the SOC obtained by looking up the table of instantaneous open circuit voltage, is the SOE estimated by the energy conservation method (current × voltage integral);

[0045] The above three observations are combined into a comprehensive observation equation:

[0046]

[0047] in: is the observation fusion weight, and the sum of the SOC weights is constrained to be , and dynamically adjusts with sensor noise, temperature, and SOC range; To correct the offset, it is estimated by an online regression model;

[0048] Define the observation function :

[0049]

[0050] in: Indicates extracting the corresponding ampere-hour integral estimate from the state; Indicates that the SOC is checked by the voltage predicted by the state; Directly corresponds to the SOE item in the state; weight Adaptive update via recursive least squares (RLS) or online Kalman regression module;

[0051] The observation noise covariance is estimated online by:

[0052] First calculate the residual at each moment:

[0053]

[0054] in, are the observed forecasts generated by the UKF forecast step; an exponentially weighted sliding window is used:

[0055]

[0056] in, is the observation noise covariance matrix predicted at the kth moment, is the attenuation factor, which is used to filter out the drastic effect of the mutation point. In actual operation, the observation vector The parameters of need to be adjusted according to the characteristics of the energy storage system, such as the sensor accuracy and measurement error distribution. For example, in the energy storage system, SOC is usually measured by the ampere-hour integral method or the open circuit voltage method, while SOE may be estimated by the energy conservation equation. Therefore, the observation vector The design of the observation equation needs to comprehensively consider the error characteristics of different measurement methods to ensure that the observation equation can accurately reflect the relationship between the measured value and the true value.

[0057] S3: Initialize the Kalman filter: According to the initial state and measurement value of the energy storage system, set the initial state estimate and initial error covariance matrix; specifically:

[0058] Initial state estimate:

[0059] : The open circuit voltage is measured after uniform charge and discharge, and combined with the OCV curve provided by the manufacturer, after stabilization at room temperature for 10 minutes; : After the initial SOC is measured, a simple mapping is performed based on the nominal energy curve and the current SOC:

[0060]

[0061] in: It is the rated capacity multiplied by the rated voltage;

[0062] Initial covariance matrix P0:

[0063]

[0064] Each component is set according to offline calibration data or empirical values; The detection value of the temperature sensor; For a priori estimation, the factory nominal value can be taken or the value calculated through short-time constant current discharge experiment;

[0065] UKF weight initialization:

[0066] It is the scale parameter, proportion parameter and distribution parameter used to generate Sigma points; observation fusion weight Initialize to a uniform distribution:

[0067]

[0068] Initial noise covariance estimate:

[0069] Using experience value:

[0070]

[0071] Then, adaptive updates are performed based on the online mechanisms in S4 and S5.

[0072] S4: Execute Kalman filter prediction: Use the state equation to predict the state at the next moment and update the error covariance matrix:

[0073] Sigma point propagation: The predicted Sigma point is obtained through nonlinear state equation mapping:

[0074]

[0075] in: For control input, including current , ambient temperature ;

[0076] Predicted states and covariances:

[0077] Predicted state mean:

[0078]

[0079] in It represents the mean value of the predicted state at the kth moment;

[0080] Forecast covariance matrix:

[0081]

[0082] in: is the UKF weight coefficient, is the process noise covariance after online learning / correction in the previous step;

[0083] Adaptive process noise update:

[0084] If the current working condition is detected to exceed the preset threshold, including temperature mutation and load mutation, the small sample online learning module is started. The small sample online learning module uses a lightweight LSTM to convert the actual residual sequence of the last few moments into As input, predict the noise covariance increment at the next moment ;

[0085] renew:

[0086]

[0087] If there is no sudden change, the update is smoothed using exponential weighting:

[0088]

[0089] The matrix on the right can be regarded as the one-time estimated process noise sample covariance and is updated smoothly.

[0090] S5: Execute Kalman filter update: modify the predicted state based on the observed value and update the error covariance matrix;

[0091] After the state model and observation model are constructed, the Kalman filter prediction is initialized and the two key steps of Kalman filter prediction 3 and Kalman filter update 4 are entered. In Kalman filter prediction 3, the state equation is used to predict the state at the next moment and the error covariance matrix is ​​updated. Specifically:

[0092] Predicted observed Sigma points:

[0093] Will Observed function Mapping to obtain the observed Sigma point:

[0094]

[0095] Forecast observation mean and covariance:

[0096] Observed predicted mean:

[0097]

[0098] Observation covariance:

[0099]

[0100] State-observation covariance:

[0101]

[0102] Observation noise covariance matrix renew:

[0103] Compute the instantaneous observation residuals:

[0104]

[0105] Update with sliding index:

[0106]

[0107] If an abnormal residual increase is detected, including N consecutive residuals exceeding the threshold, the online regression model is triggered and the initial value of R_k is re-estimated using short-term LSTM or weighted regression;

[0108] Calculate the Kalman gain and state update:

[0109] Kalman gain:

[0110]

[0111] is the inverse matrix of the observation covariance at the kth moment;

[0112] Posterior state estimate:

[0113]

[0114] Posterior covariance:

[0115]

[0116] Observation fusion weight renew:

[0117] According to the latest residual sequence, the fusion weights are updated through the online least squares or Kalman regression algorithm so that the fused observations have the minimum variance. The current fusion model is:

[0118]

[0119] Then, the weighted sum of squares of the recent M-step residuals is minimized:

[0120]

[0121] in Approximate output from short-term high-precision calibration or reference equipment. When this is not available, EKF / SOF technology can be used to roughly estimate it as a label.

[0122] Similar update To fusion SOE estimation, constraints: .

[0123] In practical applications, this prediction process needs to be combined with the real-time operating data of the energy storage system for calculation. For example, when the energy storage system is in a fast charging and discharging state, the observed prediction mean Will be affected by large process noise, so the observation covariance needs to be adjusted dynamically To improve the prediction accuracy, the Kalman filter update step 4 is then entered, which corrects the predicted state based on the observation value and updates the error covariance matrix.

[0124] S6: Introducing a multi-parameter optimization strategy: Based on the Kalman filter, the parameters of the state transfer matrix and the observation matrix are adjusted through the optimization algorithm;

[0125] To further improve the correction accuracy, a multi-parameter optimization strategy 5 is introduced. Based on the Kalman filter, this step adjusts the parameters of the state transfer matrix and the observation matrix through the optimization algorithm to further improve the correction effect. The optimization objective function of the optimization algorithm is defined as a dual objective function:

[0126]

[0127] in: Represents the algorithm parameter set The joint estimation vector of SOC / SOE at the next k-th moment; The nominal values ​​given for high-precision reference systems, including laboratory calibration values; (1) measures the accuracy of the estimation, (2) ensures smooth and continuous estimation results, and (3) measures the stability of the algorithm under different working conditions; As weight, through cross-validation or experience setting, set

[0128] Algorithm parameter set include:

[0129]

[0130] Among them: UKF weight parameter Affects the distribution of Sigma points; and Model coefficients affect observation mapping;

[0131] Particle Swarm Optimization PSO:

[0132] Initialize the population: randomly generate P particles within a reasonable range, each particle corresponds to a set of θ parameter vectors, and is randomly assigned an initial velocity;

[0133] Fitness calculation: For each particle, run the UKF+ online update algorithm for a period of time to calculate the objective function ;

[0134] Update individual and global optimality: Update the individual optimality of particles according to fitness and global optimal ;

[0135] Velocity and position updates:

[0136]

[0137] in: is the inertia factor, which is used to decrease generation by generation to improve convergence. is the learning factor, .

[0138] Iterate until convergence: When the global optimal fitness Stop when the change is less than a threshold or the maximum number of iterations is reached.

[0139] In practical applications, the optimization algorithm can use either gradient descent or genetic algorithms. The specific choice should be balanced based on the operating characteristics of the energy storage system and the computational resources. For example, if a large deviation is found between the measured and predicted values ​​of SOC and SOE during energy storage system operation, the error can be reduced by adjusting the parameters of the state transition matrix. At the same time, the setting of the regularization parameter should comprehensively consider the balance between correction accuracy and the magnitude of parameter variation to avoid overfitting and reduced correction effect.

[0140] S7: Correction result output: The final corrected SOC and SOE values ​​are output to the energy storage management system; the output frequency of the final corrected SOC and SOE values ​​is adaptive: when the energy storage battery is in a rapid charge and discharge state (|I| ≥ 1C), the output frequency is increased to 1Hz; when the battery is in a low-speed operating condition (|I| ≤ 0.1C), it is reduced to below 0.1Hz to reduce computing overhead; when the temperature suddenly changes (ΔT > 3°C / min), the output frequency is temporarily increased to 5Hz for more precise tracking.

[0141] Real-time residual monitoring and closed-loop self-checking:

[0142] Residual error continues to exceed the limit detection: If M consecutive frames (The threshold value can be set based on experience or calibration), it is considered that the estimate has an instability risk.

[0143] “Hysteresis Window Revaluation”: When the above-mentioned instability risk is detected:

[0144] Take the current time k as the starting point of the window, look back L frames (such as L=50), run UKF (or change to EKF) again to batch re-estimate these L frames, and re-initialize with new Q, R or updated weights;

[0145] If the residuals recover to more than 80% of the normal value (≤ threshold) after batch revaluation, the system will continue to operate online; otherwise, a "manual maintenance reminder" or "switch to the backup estimation model" will be triggered (for example, switching to the SOE fast method based on the electrochemical equivalent circuit within seconds).

[0146] Output format and interface:

[0147] Local CAN bus: broadcasts current SOC, SOE, temperature, internal resistance correction and other status variables using CAN protocol;

[0148] Upper-level EMS / SCADA system: The estimated value with time stamp is regularly pushed via Modbus / TCP or uFTP, along with "health indicator" and "Uncertainty Estimate" (uncertainty output).

[0149] Alarms and logs: When the algorithm is obviously inaccurate or a closed-loop self-check is triggered, a log is automatically generated and sent to the remote backend via Ethernet.

[0150] S8: Dynamic environmental adaptability assessment: Evaluate the adaptability of the correction method in a complex environment by simulating the operation data of the energy storage system under different working conditions.

[0151] Real-time online testing of digital twins:

[0152] Build a digital twin model: Utilize various operating condition data (charge and discharge rates, temperature, SOC range, etc.) collected offline from a real energy storage power station (or battery pack), combined with a high-precision electrochemical-thermal coupling model (for example, based on the SPMeT model), to build a "digital twin" that corresponds one-to-one with the actual object.

[0153] The digital twin runs on the cloud / edge server and synchronizes input current, power, temperature, environmental status, etc. with the actual on-site system in real time.

[0154] Online parallel simulation:

[0155] While the actual system performs the UKF estimation at each moment, the digital twin runs a "shadow UKF + optimization algorithm" in parallel in the background. The digital twin's input comes from the same set of sensor data, resulting in a different set of estimation results.

[0156] The digital twin estimate is subtracted from the on-site UKF estimate. If the difference continues to exceed the allowable threshold, it is considered that the strategy's "generalization ability" for the current working conditions has deviated.

[0157] Adaptive Feedback:

[0158] If the digital twin error continues to be too large, the “secondary parameter renormalization” mechanism is automatically triggered (see online adaptive noise and weight updates in S4 and S5) until the agreement between the digital twin and the on-site UKF estimate returns to a normal range.

[0159] At the same time, the deviation information is uploaded to the background for incremental training of offline or online deep learning models to ensure that the system has stronger adaptability to similar working conditions in the future.

[0160] Adversarial scenario generation and robustness testing

[0161] Scene parameter space definition

[0162] The key environmental factors that the energy storage system may encounter (charge and discharge rate, temperature, aging degree, ambient temperature mutation rate, etc.) are taken as a multidimensional parameter space.

[0163] Charge and discharge rate: 0.1C–3C;

[0164] Battery temperature: –20°C–60°C;

[0165] Temperature mutation rate: 0–5°C / min;

[0166] State of aging (SOH): 0.6–1.0;

[0167] Environmental interference: such as sensor error amplitude, communication delay, etc.

[0168] Adversarial Example Generation

[0169] Genetic Programming or Monte Carlo Tree Search (MCTS) is used to automatically generate adversarial scenarios that can best penetrate the estimation algorithm. For example:

[0170] A sudden switch from 2°C to 0.1°C input is accompanied by a temperature change from 20°C to –10°C.

[0171] The sensor output drifts from zero error briefly for 5–10 seconds.

[0172] Network communication delay 200ms + Gaussian jitter perturbation;

[0173] When the battery load bypass is triggered, the analog output is powered off for 3 seconds before recovering.

[0174] Each adversarial scenario is simulated on the digital twin to test the performance of the algorithm under this extreme condition.

[0175] Quantitative robustness indicators:

[0176] For each scenario, a set of indicators is defined:

[0177] Maximum instantaneous residual ;

[0178] Estimated response time : The time required for the algorithm to converge to the normal residual range again after the deviation occurs;

[0179] RMSE distribution under uniform distribution: In multiple start-stop tests in this scenario, the RMSE generated exhibits distribution characteristics, extracting the 80%-tile, 95%-tile, etc.

[0180] Variance increase in adversarial scenarios: Estimate the variance increase percentage compared to normal operating conditions.

[0181] Parallel evaluation of multiple scenarios:

[0182] All adversarial scenarios are divided into several parallel simulation threads, batch tested in the cloud / edge, and various robustness indicators are counted and reports are automatically generated.

[0183] If the performance indicators do not meet the standards in certain scenarios, they will be automatically marked as "high-risk conditions" and fed back to algorithm developers or system operation and maintenance personnel for optimization.

[0184] Online self-learning and indicator regression analysis:

[0185] Self-learning modules:

[0186] The difference data between the “true UKF estimate” and the “digital twin reference estimate” is continuously collected online to accumulate and form a training set.

[0187] The lightweight one-dimensional convolutional neural network (1D-CNN) or long short-term memory network (LSTM) is regularly triggered to learn the difference, generate the "noise covariance correction strategy" and "observation weight mapping function", etc., and automatically send them to the on-site UKF for parameter update.

[0188] Indicator regression analysis (offline / near-line):

[0189] The adversarial scenario indicators collected online are sent to the back-end big data platform together with the normal scenario indicators on a daily / weekly basis. Based on multivariate linear regression or decision tree regression analysis, the operating condition factor with the greatest impact on the algorithm performance is determined.

[0190] Based on the regression analysis results, the UKF fusion model or adaptive update rules are upgraded (e.g., adjusting weight update window length M, etc.).

[0191] Assessment Report and Visualization:

[0192] Output an interactive "Dynamic Environmental Adaptability Assessment Report": including performance curves, residual trend graphs, parameter sensitivity analysis graphs, etc. under various scenarios;

[0193] Generate an "algorithm risk map": Use a two-dimensional plane to map the operating space into "high-risk-low-risk" areas, helping operation and maintenance engineers intuitively understand which operating conditions may pose the greatest challenges to SOC / SOE estimation.

[0194] Finally, enter the correction result output 6, and output the final corrected SOC and SOE values ​​to the energy storage management system for subsequent decision-making. The output frequency of the correction results can be set to once per second or once per minute according to actual needs. In actual application scenarios, the output of the correction results needs to be dynamically adjusted in combination with the operating conditions of the energy storage system. For example, when the energy storage system is in a high-frequency charging and discharging state, the output frequency of the correction results should be appropriately increased to ensure that the energy storage management system can obtain accurate status information in a timely manner. In addition, the output of the correction results needs to be evaluated in combination with the operating environment of the energy storage system. For example, under complex working conditions, the adaptability of the correction method in a dynamic environment can be evaluated by simulating the operating data of the energy storage system under different working conditions. The evaluation indicators include correction accuracy, computational efficiency, and robustness. The analysis results of these indicators can be used to guide the optimal design and operation management of the energy storage system.

[0195] Throughout the implementation process, the connections and coordination between the various steps are crucial. State model construction 1 and observation model design 2 provide the basic framework for Kalman filter prediction 3 and Kalman filter update 4, while the Kalman filter prediction and update process provides data support for the multi-parameter optimization strategy 5. The multi-parameter optimization strategy 5 further improves the correction accuracy by adjusting the parameters of the state transfer matrix and the observation matrix, thereby providing more accurate state information for the correction result output 6. This interlocking implementation method ensures the efficiency and reliability of the present method in practical applications.

[0196] In the specific implementation process, the method of the present invention can be applied to a variety of energy storage system scenarios. For example, in large-scale energy storage power stations, the number and scale of energy storage units are large, and accurate correction of SOC and SOE is crucial to improving the operating efficiency of the energy storage system. Through the method of the present invention, real-time status monitoring and correction of energy storage units can be achieved, thereby optimizing the overall performance of the energy storage system. In addition, in distributed energy storage systems, the method of the present invention can also be used to solve the complexity problem of multi-parameter coupling relationships and improve the operating reliability and efficiency of energy storage systems. By combining the adaptive characteristics of the Kalman filter algorithm and the multi-parameter optimization strategy, the method of the present invention can achieve high-precision SOC and SOE correction under complex working conditions, meeting the needs of energy storage power stations for efficient and accurate management.

[0197] In order to better enable relevant personnel in this technical field to fully understand and implement the present invention, the specific implementation principle of the present invention is further explained below in conjunction with a specific application scenario. Figure 1 This is a flow chart of the method of the present invention, showing the overall implementation process of the SOC / SOE joint correction method of the energy storage power station based on Kalman filter fusion multi-parameter optimization, including key steps such as state model construction 1, observation model design 2, Kalman filter prediction 3, Kalman filter update 4, multi-parameter optimization strategy 5 and correction result output 6.

[0198] In a large-scale energy storage power station, the energy storage system operates in a complex and ever-changing environment, requiring real-time monitoring and correction of the state of charge (SOC) and state of energy efficiency (SOE) to ensure stable and efficient system operation. First, in state model construction 1, a dynamic state equation is established based on the actual operating characteristics of the energy storage unit. For example, the capacity decay rate and temperature changes of the energy storage unit that affect battery performance are incorporated into the state of charge parameter settings. Simultaneously, the energy state is calculated in conjunction with parameters such as the energy storage system's charge and discharge efficiency and internal resistance, forming a complete state model. This model provides the basic framework for the subsequent implementation of the Kalman filter algorithm. This model enables the energy storage system to accurately describe the temporal variations of the SOC and SOE, while also accounting for the influence of external input variables such as charge and discharge power and current.

[0199] Next, we proceed to observation model design 2. The goal of this step is to establish observation equations to describe the relationship between the measured values ​​and the true values ​​of SOC and SOE. In actual operation, the SOC of the energy storage system is usually measured using the ampere-hour integration method or the open-circuit voltage method, while the SOE is estimated using the energy conservation equation. The error characteristics of these measurement methods are comprehensively considered and used to adjust the parameters of the observation noise covariance. For example, when the sensor accuracy is low or the measurement noise is large, the parameters of the observation noise covariance will be appropriately adjusted to reduce the impact of measurement error on the correction results. Through this observation model, the energy storage system can accurately reflect the true state of SOC and SOE, providing reliable data support for subsequent Kalman filter prediction and updates.

[0200] After the state model and observation model are constructed, the two key steps of Kalman filter prediction 3 and Kalman filter update 4 are entered. In Kalman filter prediction 3, the state equation is used to predict the state at the next moment and update the error covariance matrix. For example, when the energy storage system is in a rapid charging and discharging state, the predicted state and covariance will be affected by large process noise. Therefore, the predicted state and covariance are dynamically adjusted to improve prediction accuracy. Then, Kalman filter update 4 is entered. This step corrects the predicted state based on the observation value and updates the error covariance matrix. For example, when the observation noise is large, the calculation of the Kalman gain will be adjusted according to the characteristics of the observation noise covariance matrix and the predicted observation mean and covariance to ensure that the observation fusion weight can accurately reflect the actual state of the energy storage system. Through these two steps, the energy storage system can achieve real-time correction of SOC and SOE in a dynamic environment.

[0201] To further improve correction accuracy, a multi-parameter optimization strategy 5 is introduced. This step uses an optimization algorithm based on Kalman filtering to adjust the parameters of the state of charge and state of energy. For example, during the operation of the energy storage system, if a large deviation is found between the measured and predicted values ​​of SOC and SOE, the error can be reduced by adjusting the state of charge parameters. At the same time, the setting of the average battery pack temperature requires a comprehensive consideration of the balance between correction accuracy and the amplitude of parameter changes to avoid a decrease in correction effect due to overfitting. Through this optimization strategy, the energy storage system can further improve correction accuracy and meet the high-precision requirements under complex operating conditions.

[0202] Finally, the correction result output 6 is entered, and the final corrected SOC and SOE values ​​are output to the energy storage management system for subsequent decision-making. During the correction result output process, the output frequency is dynamically adjusted according to the operating conditions of the energy storage system. For example, under high-frequency charging and discharging conditions, the output frequency of the correction results is appropriately increased to ensure that the energy storage management system can obtain accurate status information in a timely manner. In addition, under complex operating conditions, the adaptability of the correction method in a dynamic environment can be evaluated by simulating the operating data of the energy storage system under different operating conditions. Evaluation indicators include correction accuracy, computational efficiency, and robustness. The analysis results of these indicators can be used to guide the optimized design and operation management of the energy storage system.

[0203] Throughout the implementation process, the connections and coordination between the various steps are crucial. State model construction 1 and observation model design 2 provide the basic framework for Kalman filter prediction 3 and Kalman filter update 4, while the Kalman filter prediction and update process provides data support for the multi-parameter optimization strategy 5. The multi-parameter optimization strategy 5 further improves the correction accuracy by adjusting the parameters of the state of charge and energy state, thereby providing more accurate state information for the correction result output 6. This interlocking implementation method ensures the efficiency and reliability of the present method in practical applications.

[0204] Simulation experiments were conducted on the operating data of the energy storage system under different working conditions to evaluate the dynamic environmental adaptability of the correction method. Specific working conditions include:

[0205] Condition 1: Low Charge / Discharge Rate (0.2C): The energy storage system was charged and discharged at a rate of 0.2C for 10 hours at a 25°C ambient temperature, with a sampling frequency of 1Hz. Under this condition, the root mean square error (RMSE) of the corrected SOC estimation was 0.012, and the RMSE of the SOE estimation was 0.018, demonstrating the high accuracy of the correction method at low charge / discharge rates.

[0206] Condition 2: Medium Charge / Discharge Rate (1C): Also at 25°C, the charge / discharge cycle was performed at a 1C rate for 2 hours, with a sampling frequency of 1Hz. The RMSE of the SOC estimation error increased to 0.021, and the RMSE of the SOE estimation error was 0.025. Despite this increase in error, the SOC / SOE accuracy requirements of the energy storage system were still met, and the corrected results accurately reflected the state change trend of the energy storage system.

[0207] Condition 3: High Charge / Discharge Rate (2C): Charge and discharge cycles were performed at a 2C rate for one hour at a 25°C ambient temperature, with a sampling frequency of 1Hz. The estimated SOC error (RMSE) was 0.032, and the estimated SOE error (RMSE) was 0.038. Although the error increased further at high charge / discharge rates, the correction method was able to suppress the error growth to a certain extent, keeping the estimated SOC / SOE values ​​within a reasonable range and avoiding error divergence, demonstrating good robustness.

[0208] Condition 4: Temperature Variation (-10°C-40°C): The energy storage system was charged and discharged at different temperatures (-10°C, 0°C, 10°C, 20°C, 30°C, and 40°C), with each temperature point lasting two hours. Charge and discharge were performed at a rate of 1C with a sampling frequency of 1Hz. Battery characteristics such as internal resistance and capacity change significantly with temperature, increasing the difficulty of SOC / SOE estimation. The RMSE of the corrected SOC and SOE estimates at each temperature point fluctuated between 0.015-0.035 and 0.02-0.04, respectively. Overall, the estimation accuracy remained relatively good, demonstrating the adaptability of the correction method to temperature variations.

[0209] Condition 5: Load Fluctuation: This simulation simulates the actual operating conditions of an energy storage system in a power grid, with load power fluctuating randomly between 0 and 100 kW, at a sampling frequency of 1 Hz, for 5 hours. Under this complex condition, the estimated SOC error (RMSE) was 0.028, and the estimated SOE error (RMSE) was 0.033. The correction method effectively addresses frequent load fluctuations, adjusting the estimated SOC and SOE values ​​in real time to accurately reflect the actual state of the energy storage system and ensure its reliable operation in the power grid.

[0210] The simulation experiments under the above-mentioned different working conditions further demonstrate the good adaptability of the method of the present invention in complex dynamic environments, and its ability to meet the requirements of energy storage power stations for accurate SOC / SOE correction.

[0211] Three correction methods, namely the ampere-hour integration method, the open-circuit voltage method, and the Kalman filter alone (without multi-parameter optimization), were selected for comparative experiments with the method of the present invention under the same energy storage system model and operating conditions. The experimental conditions were the medium charge and discharge rate (1C) conditions mentioned above. The experimental results are as follows:

[0212] The ampere-hour integration method has an RMSE of 0.045 for SOC estimation and 0.062 for SOE estimation. This method fails to account for factors such as battery self-discharge, temperature effects, and the accumulation of measurement errors, resulting in large errors. These errors accumulate over time, making it unable to meet the high-precision SOC / SOE requirements of energy storage systems.

[0213] Open-circuit voltage method: The RMSE for SOC estimation is 0.038, and the RMSE for SOE estimation is 0.055. The open-circuit voltage method requires measuring the open-circuit voltage after the battery has been at rest for a period of time. However, in actual energy storage system operation, batteries are rarely at rest, so this method has limited applicability. Furthermore, it is highly dependent on the battery model, and deviations in the model parameters can directly affect the estimation results.

[0214] Using only Kalman filtering (without multi-parameter optimization): The RMSE of the SOC estimation error is 0.025, and the RMSE of the SOE estimation error is 0.031. Although Kalman filtering can suppress the influence of measurement noise and model errors to a certain extent, its correction accuracy still needs to be improved because it does not optimize the state transfer matrix and observation matrix.

[0215] The proposed method achieves an RMSE of 0.021 for SOC estimation and 0.025 for SOE estimation, significantly outperforming the other three methods mentioned above. By introducing a multi-parameter optimization strategy, the parameters of the state transfer matrix and observation matrix are further optimized, enabling the correction method to better adapt to the dynamic characteristics of the energy storage system and the multi-parameter coupling relationship, thereby obtaining more accurate SOC / SOE estimates.

[0216] From the above comparative analysis, it can be seen that the method of the present invention can effectively improve the correction accuracy of the SOC / SOE of energy storage power stations under various complex working conditions. Compared with other existing correction methods, it has significant advantages and can better meet the needs of energy storage power stations for efficient and precise management.

[0217] Through the implementation of the above-mentioned specific application scenarios, the method of the present invention can achieve high-precision SOC and SOE correction under complex working conditions, meeting the needs of energy storage power stations for efficient and accurate management. For example, in large-scale energy storage power stations, the number and scale of energy storage units are large, and accurate correction of SOC and SOE is crucial to improving the operating efficiency of the energy storage system. Through the method of the present invention, real-time status monitoring and correction of energy storage units can be achieved, thereby optimizing the overall performance of the energy storage system. In addition, in distributed energy storage systems, the method of the present invention can also be used to solve the complexity of multi-parameter coupling relationships and improve the operating reliability and efficiency of the energy storage system.

[0218] The technical solution of the present invention is not limited to the above-mentioned specific embodiments. Any technical variations made according to the technical solution of the present invention fall within the protection scope of the present invention.

Claims

1. A Kalman filter fusion multi-parameter optimization SOC / SOE joint correction method for energy storage power station, characterized in that: The following steps are involved: S1: Constructing the energy storage system state model: Based on the physical characteristics of the energy storage system, a dynamic state equation including SOC and SOE is established; S2: Design observation model: Based on the actual operating data of the energy storage system, establish observation equations to describe the relationship between the measured values ​​and the true values ​​of SOC and SOE; S3: Initialize the Kalman filter: according to the initial state and measurement value of the energy storage system, set the initial state estimate and the initial error covariance matrix; S4: Execute Kalman filter prediction: use the state equation to predict the state at the next moment and update the error covariance matrix; S5: Execute Kalman filter update: modify the predicted state based on the observed value and update the error covariance matrix; S6: Introducing a multi-parameter optimization strategy: Based on the Kalman filter, the parameters of the state transfer matrix and the observation matrix are adjusted through the optimization algorithm; S7: Correction result output: Output the final corrected SOC and SOE values ​​to the energy storage management system; S8: Dynamic environmental adaptability assessment: Evaluate the adaptability of the correction method in complex environments by simulating the operation data of the energy storage system under different working conditions; In step S6, the optimization objective function of the optimization algorithm is defined as a dual objective function: in: Represents the algorithm parameter set The joint estimation vector of SOC / SOE at the next k-th moment; The nominal values ​​given for high-precision reference systems, including laboratory calibration values; (1) measures the accuracy of the estimation, (2) ensures smooth and continuous estimation results, and (3) measures the stability of the algorithm under different working conditions; As weight, through cross-validation or experience setting, set Algorithm parameter set include: Among them: UKF weight parameter Affects the distribution of Sigma points; and Model coefficients affect observation mapping; Particle Swarm Optimization PSO: Initialize the population: randomly generate P particles within a reasonable range, each particle corresponds to a set of θ parameter vectors, and is randomly assigned an initial velocity; Fitness calculation: For each particle, run the UKF+ online update algorithm for a period of time to calculate the objective function ; Update individual and global optimality: Update the individual optimality of particles according to fitness and global optimal ; Velocity and position updates: in: is the inertia factor, which is used to decrease generation by generation to improve convergence. is the learning factor, ; Iterate until convergence: When the global optimal fitness Stop when the change is less than a threshold or the maximum number of iterations is reached.

2. The SOC / SOE joint correction method for energy storage power station based on Kalman filtering and multi-parameter optimization according to claim 1 is characterized in that: The dynamic state equation including SOC and SOE established in step S1 is: Let the state vector at time k be defined as: in: Indicates the The state vector at time t, is the state of charge, is the energy state, is the average temperature of the battery pack, is the equivalent internal resistance correction amount; The first-order RC circuit model is used to describe the state transition: in: is the current measured at the last moment, is the charge and discharge efficiency function that changes with current and temperature, is the effective capacity of the battery, is the open circuit voltage function, is the equivalent internal resistance, is the total mass of the battery pack, is the specific heat capacity, is the thermal time constant, is the continuous change rate of internal resistance caused by aging, is the process noise; For a state with dimension n=4, 2n+1 Sigma points are generated: in: is the posterior state estimate at the previous moment, is the covariance, is the scaling parameter of UKF, take Represents the matrix The lower triangular matrix obtained after Cholesky decomposition has the increase and decrease operators in its i-th column.

3. The SOC / SOE joint correction method for energy storage power station based on Kalman filtering and multi-parameter optimization according to claim 2 is characterized in that: In step S2, the observation equation is: Let the observation vector be: in: is the SOC calculated by the ampere-hour integration method, is the SOC obtained by looking up the table of instantaneous open circuit voltage, is the SOE estimated by the energy conservation method (current × voltage integral); The above three observations are combined into a comprehensive observation equation: in: is the observation fusion weight, and the sum of the SOC weights is constrained to be , and dynamically adjusts with sensor noise, temperature, and SOC range; To correct the offset, it is estimated by an online regression model; Define the observation function : in: Indicates extracting the corresponding ampere-hour integral estimate from the state; Indicates that the SOC is checked by the voltage predicted by the state; Directly corresponds to the SOE item in the state; weight Adaptive update via recursive least squares (RLS) or online Kalman regression module; The observation noise covariance is estimated online by: First calculate the residual at each moment: in, are the observed forecasts generated by the UKF forecast step; an exponentially weighted sliding window is used: in, is the observation noise covariance matrix predicted at the kth moment, is the attenuation factor, which is used to filter out the drastic impact of the mutation point.

4. The SOC / SOE joint correction method for energy storage power station using Kalman filtering and multi-parameter optimization according to claim 3 is characterized in that: The step S3 is specifically as follows: Initial state estimate: : The open circuit voltage is measured after uniform charge and discharge, and combined with the OCV curve provided by the manufacturer, after stabilization at room temperature for 10 minutes; : After the initial SOC is measured, a simple mapping is performed based on the nominal energy curve and the current SOC: in: It is the rated capacity multiplied by the rated voltage; Initial covariance matrix P0: Each component is set according to offline calibration data or empirical values; The detection value of the temperature sensor; For a priori estimation, the factory nominal value can be taken or the value calculated through short-time constant current discharge experiment; UKF weight initialization: It is the scale parameter, proportion parameter and distribution parameter used to generate Sigma points; observation fusion weight Initialize to a uniform distribution: Initial noise covariance estimate: Using experience value: Then, adaptive updates are performed based on the online mechanisms in S4 and S5.

5. The SOC / SOE joint correction method for energy storage power station using Kalman filtering and multi-parameter optimization according to claim 4 is characterized in that: The step S4 is: Sigma point propagation: The predicted Sigma point is obtained through nonlinear state equation mapping: in: For control input, including current , ambient temperature ; Predicted states and covariances: Predicted state mean: in It represents the mean value of the predicted state at the kth moment; Forecast covariance matrix: in: is the UKF weight coefficient, is the process noise covariance after online learning / correction in the previous step; Adaptive process noise update: If the current working condition is detected to exceed the preset threshold, including temperature mutation and load mutation, the small sample online learning module is started. The small sample online learning module uses a lightweight LSTM to convert the actual residual sequence of the last few moments into As input, predict the noise covariance increment at the next moment ; renew: If there is no sudden change, the update is smoothed using exponential weighting: The matrix on the right can be regarded as the one-time estimated process noise sample covariance and is updated smoothly.

6. The SOC / SOE joint correction method for energy storage power station using Kalman filtering and multi-parameter optimization according to claim 5 is characterized in that: The step S5 is specifically as follows: Predicted observed Sigma points: Will Observed function Mapping to obtain the observed Sigma point: Forecast observation mean and covariance: Observed predicted mean: Observation covariance: State-observation covariance: Observation noise covariance matrix renew: Compute the instantaneous observation residuals: Update with sliding index: If an abnormal residual increase is detected, including N consecutive residuals exceeding the threshold, the online regression model is triggered and the initial value of R_k is re-estimated using short-term LSTM or weighted regression; Calculate the Kalman gain and state update: Kalman gain: is the inverse matrix of the observation covariance at the kth moment; Posterior state estimate: Posterior covariance: Observation fusion weight renew: According to the latest residual sequence, the fusion weights are updated through the online least squares or Kalman regression algorithm so that the fused observations have the minimum variance. The current fusion model is: Then, the weighted sum of squares of the recent M-step residuals is minimized: in Approximate output from short-term high-precision calibration or reference equipment. When this is not available, EKF / SOF technology can be used to roughly estimate it as a label. Similar update To fusion SOE estimation, constraints: .

7. The SOC / SOE joint correction method for energy storage power station using Kalman filtering and multi-parameter optimization according to claim 1 is characterized in that: In step S7, the output frequency of the final corrected SOC and SOE values ​​is adaptive: when the energy storage battery is in a fast charge and discharge state, the output frequency is increased to 1 Hz; when the battery is in a low-speed operating condition, the output frequency is reduced to below 0.1 Hz to reduce computing overhead; when the temperature suddenly changes, the output frequency is temporarily increased to 5 Hz for more precise tracking.

Images