Model prediction-based energy optimization distribution method and system for hybrid energy storage system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NORTH CHINA GRID MEASUREMENT CENT
- Filing Date
- 2025-12-30
- Publication Date
- 2026-06-12
Smart Images
Figure CN121840737B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of energy optimization allocation, and in particular relates to a model-based prediction-based method and system for energy optimization allocation of hybrid energy storage systems. Background Technology
[0002] Hybrid energy storage systems typically consist of two or more energy storage components, such as batteries and supercapacitors. They combine the high energy density of batteries with the high power density and long cycle life of supercapacitors, meeting the variable power demands of modern power systems, electric vehicles, and microgrids. The core of a hybrid energy storage system lies in its energy management strategy, which involves real-time and rational power allocation among different energy storage units. Existing energy management strategies mainly include rule-based control and filter-based control methods. For example, a low-pass filter can be used to decompose the total power demand into high-frequency and low-frequency components, which are handled by supercapacitors and batteries respectively. While this method is simple to implement and computationally inexpensive, the control logic and parameters are usually based on empirical settings, making it difficult to maintain optimal control performance. This often leads to overuse of batteries, accelerated battery lifespan degradation, or failure to fully utilize the power buffering function of supercapacitors. Optimization-based energy management strategies, such as Model Predictive Control (MPC), solve an optimal control problem within a finite future time domain based on a predictive model of the system in each control cycle, achieving a preset optimization objective, such as minimizing system energy consumption or extending battery life. However, a mismatch exists between the predictive model and the actual system, affecting control accuracy and optimization performance. The inability to reflect the actual cost changes in battery life loss under different health conditions and load characteristics limits the application of optimization control. Moreover, during the rolling optimization process, MPC lacks an online compensation and correction mechanism for deviations between the actual state and the predicted trajectory caused by model mismatch and prediction errors. Summary of the Invention
[0003] To address the aforementioned problems, a model-predictive-based energy optimization allocation method for hybrid energy storage systems is proposed in the first aspect of this invention, comprising the following steps:
[0004] Obtain the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and perform multi-resolution singular value decomposition to obtain power subsequences with different frequency characteristics;
[0005] In each control cycle, the Lyapunov exponent of the battery state of charge (SOC) trajectory determined in the previous control cycle, the mismatch residual of the prediction model, and the disturbance estimation state are used, combined with the currently collected energy storage unit health state (SOH) and the power subsequence kurtosis value in the current prediction time domain, to update the battery lifetime degradation cost weight and the power boundary constraints of each energy storage unit in the prediction control model respectively.
[0006] The updated predictive control model is solved in a rolling manner, with the operating cost, battery life decay cost taking into account the updated weights, and supercapacitor state maintenance cost as optimization objectives, to obtain the current power allocation baseline value and the Lyapunov exponent of the battery SOC trajectory in the current prediction time domain.
[0007] Calculate the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous cycle, and update the model mismatch observer based on the residual and the Lyapunov exponent obtained in the current cycle. The observer outputs the disturbance estimation state and compensation control quantity for the next control cycle. The compensation control quantity is superimposed with the power allocation reference value to obtain the power allocation command.
[0008] Optionally, the step of obtaining the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and performing multi-resolution singular value decomposition to obtain power sub-sequences with different frequency characteristics includes:
[0009] The total power demand prediction sequence is decomposed into a low-frequency power subsequence, a mid-frequency power subsequence, and a high-frequency power subsequence. The power fluctuations of the low-frequency power subsequence are smoothed by battery cells, the power fluctuations of the high-frequency power subsequence are smoothed by supercapacitor cells, and the power fluctuations of the mid-frequency power subsequence are smoothed by battery cells and supercapacitor cells in a proportional manner.
[0010] Optionally, the battery life degradation cost weights in the updated predictive control model include:
[0011] The weights are adjusted based on the Lyapunov exponent and the magnitude of the mismatch residual in the prediction model determined in the previous control cycle. When both the Lyapunov exponent and the mismatch residual exceed a preset safety threshold, the weight values are linearly increased to enhance the stability constraint on the battery SOC trajectory. When both the Lyapunov exponent and the mismatch residual are below the preset safety threshold, the weight values are linearly decreased.
[0012] Optionally, update the power boundary constraints of each energy storage unit, including:
[0013] Based on the current battery SOH data, the maximum charge / discharge power boundary of the battery is calculated using the following formula. : in, This refers to the battery's rated power.
[0014] Based on the kurtosis value of the high-frequency power subsequence Adjust the discharge power boundary of the supercapacitor according to the following formula. : in, This refers to the rated discharge power of the supercapacitor. This serves as a reference value for kurtosis.
[0015] Optionally, the optimization objective, which includes operating cost, battery life degradation cost taking into account the updated weights, and supercapacitor state maintenance cost, includes:
[0016] The expression for the optimization objective function J is: ;
[0017] Operating costs For prediction of the time domain The sum of the power exchanged between the internal power supply and the grid, multiplied by the time-of-use electricity price: ;
[0018] Battery life degradation cost The equivalent cycle number of batteries was determined using the raindrop counting method. And multiply by the unit cycle cost of the battery get: ;
[0019] supercapacitor state maintenance cost To predict the state of charge of supercapacitors at the end of the time domain Expected reference value Quadratic deviation between: ;
[0020] in, Let k be the electricity price at time k. Let k be the grid interaction power at time k. To control the cycle, For the updated battery life degradation cost weighting, To predict the battery power sequence in the time domain, Weighting coefficients for the cost of maintaining the state of a supercapacitor.
[0021] Optionally, calculating the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous period includes:
[0022] The actual SOC value of the battery at the current time k is measured by the battery management system. The predicted battery SOC value at the current moment compared to the previous control cycle. The difference is taken, and the absolute value is used as the mismatch residual of the battery SOC prediction model. .
[0023] Optionally, the Lyapunov exponential update model mismatch observer obtained based on the residuals and the current period includes:
[0024] A Kalman filter is used as the model mismatch observer, and the predicted model mismatch residual is used as the measurement information of the Kalman filter. The perturbation estimation state is updated iteratively through the Kalman gain; the process noise covariance matrix of the Kalman filter... Lyapunov exponents obtained from solving the predictive control model in the current period To regulate:
[0025]
[0026] in, The reference process noise covariance matrix, The preset stability threshold, It is a gain coefficient greater than 1.
[0027] In a second aspect, this invention proposes a model-predictive-based energy optimization allocation system for hybrid energy storage systems, comprising the following modules:
[0028] The acquisition module is used to acquire the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and perform multi-resolution singular value decomposition to obtain power subsequences with different frequency characteristics.
[0029] The update module is used to update the battery life decay cost weight and the power boundary constraints of each energy storage unit in the predictive control model in each control cycle by using the Lyapunov exponent of the battery state of charge trajectory determined in the previous control cycle, the mismatch residual of the prediction model and the disturbance estimation state, combined with the currently collected energy storage unit health state (SOH) and the power subsequence kurtosis value in the current prediction time domain.
[0030] The module is used to solve the updated predictive control model in a rolling manner, with the optimization objectives being operating cost, battery life decay cost taking into account the updated weights, and supercapacitor state maintenance cost, to obtain the current power allocation baseline value and the Lyapunov exponent of the battery SOC trajectory in the current prediction time domain.
[0031] The forming module is used to calculate the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous cycle, and update the model mismatch observer based on the residual and the Lyapunov exponent obtained in the current cycle. The observer outputs the disturbance estimation state and compensation control quantity for the next control cycle. The compensation control quantity is superimposed with the power allocation reference value to obtain the power allocation command.
[0032] Preferably, the step of obtaining the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and performing multi-resolution singular value decomposition to obtain power sub-sequences with different frequency characteristics includes:
[0033] The total power demand prediction sequence is decomposed into a low-frequency power subsequence, a mid-frequency power subsequence, and a high-frequency power subsequence. The power fluctuations of the low-frequency power subsequence are smoothed by battery cells, the power fluctuations of the high-frequency power subsequence are smoothed by supercapacitor cells, and the power fluctuations of the mid-frequency power subsequence are smoothed by battery cells and supercapacitor cells in a proportional manner.
[0034] Preferably, the battery life degradation cost weights in the updated predictive control model include:
[0035] The weights are adjusted based on the Lyapunov exponent and the magnitude of the mismatch residual in the prediction model determined in the previous control cycle. When both the Lyapunov exponent and the mismatch residual exceed a preset safety threshold, the weight values are linearly increased to enhance the stability constraint on the battery SOC trajectory. When both the Lyapunov exponent and the mismatch residual are below the preset safety threshold, the weight values are linearly decreased.
[0036] Preferably, updating the power boundary constraints of each energy storage unit includes:
[0037] Based on the current battery SOH data, the maximum charge / discharge power boundary of the battery is calculated using the following formula. : in, This refers to the battery's rated power.
[0038] Based on the kurtosis value of the high-frequency power subsequence Adjust the discharge power boundary of the supercapacitor according to the following formula. : in, This refers to the rated discharge power of the supercapacitor. This serves as a reference value for kurtosis.
[0039] Preferably, the optimization objective, which includes operating cost, battery life degradation cost taking into account the updated weights, and supercapacitor state maintenance cost, includes:
[0040] The expression for the optimization objective function J is: ;
[0041] Operating costs For prediction of the time domain The sum of the power exchanged between the internal power supply and the grid, multiplied by the time-of-use electricity price: ;
[0042] Battery life degradation cost The equivalent cycle number of batteries was determined using the raindrop counting method. And multiply by the unit cycle cost of the battery get: ;
[0043] supercapacitor state maintenance cost To predict the state of charge of supercapacitors at the end of the time domain Expected reference value Quadratic deviation between: ;
[0044] in, Let k be the electricity price at time k. Let k be the grid interaction power at time k. To control the cycle, For the updated battery life degradation cost weighting, To predict the battery power sequence in the time domain, Weighting coefficients for the cost of maintaining the state of a supercapacitor.
[0045] Preferably, the calculation of the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous period includes:
[0046] The actual SOC value of the battery at the current time k is measured by the battery management system. The predicted battery SOC value at the current moment compared to the previous control cycle. The difference is taken, and the absolute value is used as the mismatch residual of the battery SOC prediction model. .
[0047] Preferably, the mismatched observer of the Lyapunov exponential update model obtained based on the residuals and the current period includes:
[0048] A Kalman filter is used as the model mismatch observer, and the predicted model mismatch residual is used as the measurement information of the Kalman filter. The perturbation estimation state is updated iteratively through the Kalman gain; the process noise covariance matrix of the Kalman filter... Lyapunov exponents obtained from solving the predictive control model in the current period To regulate:
[0049]
[0050] in, The reference process noise covariance matrix, The preset stability threshold, It is a gain coefficient greater than 1.
[0051] This invention achieves the separation of power components at different frequencies by performing multi-resolution singular value decomposition on the predicted power. In the predictive control model, the weight of battery life degradation cost is adjusted by combining the current health state of the battery, the impact characteristics of the predicted load, and the historical operating stability indicators of the system. This makes the optimization objective closer to the actual operating conditions of the energy storage system, thereby delaying battery performance degradation in energy optimization allocation. By constructing a model mismatch observer to estimate and compensate for prediction biases caused by model uncertainties and external disturbances, and superimposing this compensation amount onto the power benchmark of the optimization solution, the accuracy of power allocation of the control system is enhanced. While ensuring stable system operation, this achieves control over the entire life cycle cost of the energy storage system. Attached Figure Description
[0052] Figure 1 A flowchart of the first embodiment;
[0053] Figure 2 This is a schematic diagram illustrating weight and boundary updates.
[0054] Figure 3 This is a schematic diagram illustrating the calculation of mismatch residuals in the prediction model. Detailed Implementation
[0055] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0056] The term "multiple" in this application refers to two or more. Furthermore, it should be understood that the terms "first," "second," etc., used in the description of this application are used only for descriptive purposes and should not be construed as indicating or implying relative importance, nor as indicating or implying order.
[0057] Example 1 provides a model-based prediction-based method for optimizing energy allocation in a hybrid energy storage system, such as... Figure 1 This includes the following steps:
[0058] S1. Obtain the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and perform multi-resolution singular value decomposition to obtain power subsequences with different frequency characteristics.
[0059] Preferably, a Long Short-Term Memory (LSTM) network model is used to predict the total demand power sequence for the next N control cycles (e.g., N=300) based on historical power data, weather information, and electricity price information. An m-row, n-column trajectory matrix, i.e., a Hankel matrix, is constructed from this predicted power sequence, where m+n-1=N. Singular value decomposition (SVD) is performed on this Hankel matrix to obtain a series of singular values arranged in numerical order. Based on abrupt changes in the magnitude of the singular values, they are grouped, for example, into a first group representing the main low-frequency trend of the power sequence and a second group representing high-frequency fluctuations. The trajectory matrix is reconstructed using the singular values from each group, and then the time series is recovered from the reconstructed trajectory matrix using the diagonal averaging method, resulting in a low-frequency power subsequence and a high-frequency power subsequence.
[0060] S2, In each control cycle, using the Lyapunov exponent of the battery state of charge (SOC) trajectory determined in the previous control cycle, the mismatch residual of the prediction model, and the disturbance estimation state, combined with the currently collected energy storage unit health state (SOH) and the power subsequence kurtosis value in the current prediction time domain, the battery lifetime degradation cost weight and the power boundary constraints of each energy storage unit in the prediction control model are updated respectively.
[0061] Basis weighting of battery life degradation cost The preset values are used. The kurtosis value K of the high-frequency power subsequence is calculated; a larger K value indicates a more severe power surge. Based on the Lyapunov exponent λ, model mismatch residual e, and perturbation estimation state d calculated in the previous cycle, and the currently acquired battery health state SOH, the updated weights are... ,in to As a nonlinear gain function, for example, when λ, e, d, K increase or SOH decreases, the corresponding function value increases, thereby increasing the weight W. This allows for stricter control over battery damage during optimization. Figure 2 The model mismatch residual *e* is obtained at the beginning of the current cycle by subtracting the actual measured current battery SOC from the model's predicted current SOC value from the previous cycle. *e* serves as the input to the model mismatch observer, which processes this information and outputs an updated perturbation estimate state *d*. The power boundary is updated based on the current battery health state (SOH). For example, the battery's maximum charge / discharge power equals the factory rated power multiplied by a degradation factor positively correlated with SOH.
[0062] S3, the updated predictive control model is solved in a rolling manner, with the operating cost, the battery life decay cost taking into account the updated weights and the supercapacitor state maintenance cost as the optimization objectives, to obtain the current power allocation benchmark value and the Lyapunov exponent of the battery SOC trajectory in the current prediction time domain.
[0063] The objective function J is equal to the sum of three terms: operating cost, battery life degradation cost, and supercapacitor state maintenance cost. Operating cost is represented by the product of grid interaction power and real-time electricity price; battery life degradation cost is approximated by multiplying the updated weight W by the sum of squares of battery power changes in the predicted time domain; supercapacitor state maintenance cost is represented by the sum of squares of the deviations of the state of charge from the reference value, such as 50%. Under the conditions of satisfying power balance constraints, updated power boundary constraints for each energy storage unit, state of charge range constraints, and power change rate constraints, the optimization problem is solved using a quadratic programming or nonlinear programming solver to obtain the optimal power allocation sequence for the next N cycles. The first element of this sequence is taken as the power allocation benchmark value at the current moment. Based on the optimized battery state of charge (SOC) prediction trajectory for the next N cycles, the Lyapunov exponent of the trajectory is calculated for weight updates in the next cycle.
[0064] S4, calculate the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous cycle, and update the model mismatch observer based on the residual and the Lyapunov exponent obtained in the current cycle. The observer outputs the disturbance estimation state and compensation control quantity for the next control cycle. The compensation control quantity is superimposed with the power allocation reference value to obtain the power allocation command.
[0065] Specifically, in the current control cycle k, the actual state of charge of the battery is measured. The optimization solution for the previous period k-1 predicted the battery state of charge at period k as follows: The mismatch residual e(k) of the prediction model is... and The difference is calculated. A state observer is constructed to estimate the total system disturbance d. The state update equation of the observer is d(k+1) = A × d(k) + L × e(k), where A is the system matrix and L is the observer gain. The observer gain L is not a fixed value, but is adjusted according to the Lyapunov exponent λ calculated in the current period. When λ increases, indicating instability, the value of L is increased so that the observer can respond and correct the deviation more quickly. The observer output d(k+1) is used as the disturbance estimate state when updating the weights in the next period, and a compensation control quantity is also output. The compensation amount is equal to the gain L multiplied by the residual e(k).
[0066] The current battery power distribution baseline value obtained by solving the model predictive control optimization solution. Compensation control quantity output by the model mismatched observer By performing algebraic summation, the power command issued to the battery power conversion system is obtained. Power command of supercapacitors Equal to total power demand minus This ensures the balance of total power and corrects control deviations caused by model mismatch in real time.
[0067] In an optional embodiment, obtaining the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and performing multi-resolution singular value decomposition to obtain power sub-sequences with different frequency characteristics, includes:
[0068] The total power demand prediction sequence is decomposed into a low-frequency power subsequence, a mid-frequency power subsequence, and a high-frequency power subsequence. The power fluctuations of the low-frequency power subsequence are smoothed by battery cells, the power fluctuations of the high-frequency power subsequence are smoothed by supercapacitor cells, and the power fluctuations of the mid-frequency power subsequence are smoothed by battery cells and supercapacitor cells in a proportional manner.
[0069] Obtain a future forecast time domain, such as a 15-minute total demand power forecast sequence, which consists of a series of power data points, like the sequence... =[10,12,15,8,…]kW. A multi-resolution singular value decomposition algorithm is used to analyze this. The sequence is processed. This algorithm can separate the original sequence into three independent subsequences based on the rate of power fluctuation: a low-frequency subsequence representing gradual power changes. High-frequency subsequences representing sharp, transient power surges and the intermediate frequency subsequences between the two. After decomposition, power is allocated according to the characteristics of each energy storage unit. For example, for a given power demand, if the decomposition yields a low-frequency component of 10kW, a mid-frequency component of 2kW, and a high-frequency component of 3kW, the entire 10kW low-frequency power will be directed to the battery units. The 3kW high-frequency power will be directed to the supercapacitor units. The 2kW mid-frequency power, based on a preset allocation coefficient (e.g., 70% from the battery and 30% from the supercapacitor), will be allocated as follows: 1.4kW to the battery units and 0.6kW to the supercapacitor units. The total response power of the battery units is 11.4kW, and the total response power of the supercapacitor units is 3.6kW.
[0070] In an optional embodiment, the battery life degradation cost weights in the updated predictive control model include:
[0071] The weights are adjusted based on the Lyapunov exponent and the magnitude of the mismatch residual in the prediction model determined in the previous control cycle. When both the Lyapunov exponent and the mismatch residual exceed a preset safety threshold, the weight values are linearly increased to enhance the stability constraint on the battery SOC trajectory. When both the Lyapunov exponent and the mismatch residual are below the preset safety threshold, the weight values are linearly decreased.
[0072] At the start of each control cycle, two key indicators calculated in the previous cycle are obtained: the Lyapunov exponent and the mismatch residual of the prediction model. Assume the preset safety thresholds are a Lyapunov exponent of 0.02 and a mismatch residual of 0.5%. Under a certain operating condition, if the Lyapunov exponent of the previous cycle is 0.03 and the mismatch residual is 0.8%, both exceeding the set safety thresholds, it indicates a risk to system stability and a decrease in model prediction accuracy. The weight of the battery life degradation cost will be linearly increased, for example, from the current 1.2 to 1.3.
[0073] Increasing the weight value will make subsequent optimization calculations prioritize stable battery operation, favoring power allocation schemes that smooth the battery's state-of-charge trajectory, even if this slightly increases operating costs. Conversely, under another operating condition, if the Lyapunov exponent of the previous cycle is 0.01 and the mismatch residual is 0.2%, both below the safety threshold, it indicates stable operation and a reliable model. In this case, the weight value will be linearly reduced, for example, from 1.2 to 1.15. Reducing the weight indicates a relaxation of constraints on battery stability, allowing the optimization algorithm to consider economics more and seek the solution with the lowest total operating cost.
[0074] In an optional embodiment, updating the power boundary constraints of each energy storage unit includes:
[0075] Based on the current battery SOH data, the maximum charge / discharge power boundary of the battery is calculated using the following formula. : in, This refers to the battery's rated power.
[0076] Based on the kurtosis value of the high-frequency power subsequence Adjust the discharge power boundary of the supercapacitor according to the following formula. : in, This refers to the rated discharge power of the supercapacitor. This serves as a reference value for kurtosis.
[0077] Specifically, for the battery section, the State of Health (SOH) value of the battery is obtained in real time from the battery management system. Assuming a battery with a rated power... For a 100kW battery, the current measured SOH value is 0.9, indicating a health level of 90%. Substituting these two values into the formula, the current maximum charge / discharge power of the battery can be calculated. =90kW. The calculated 90kW will serve as the upper limit constraint for battery charging and discharging operations within this control cycle, ensuring the safe and stable operation of the aging battery.
[0078] For the supercapacitor portion, the kurtosis value of the high-frequency power subsequence is calculated. This indicates the severity of the power surge. Assume the rated discharge power of the supercapacitor... 50kW, kurtosis reference value Set to 3. If the currently calculated kurtosis value A value of 4.5 indicates that the power fluctuation is more pronounced than usual. The adjusted supercapacitor discharge power boundary is calculated. The power output is 57.5kW. This indicates a temporary increase in the supercapacitor's discharge capacity to cope with the upcoming severe power surge, thereby better protecting the battery.
[0079] In an optional embodiment, the optimization objective, which includes operating cost, battery life degradation cost taking into account updated weights, and supercapacitor state maintenance cost, includes:
[0080] The expression for the optimization objective function J is: ;
[0081] Operating costs For prediction of the time domain The sum of the power exchanged between the internal power supply and the grid, multiplied by the time-of-use electricity price: ;
[0082] Battery life degradation cost The equivalent cycle number of batteries was determined using the raindrop counting method. And multiply by the unit cycle cost of the battery get: ;
[0083] supercapacitor state maintenance cost To predict the state of charge of supercapacitors at the end of the time domain Expected reference value Quadratic deviation between: ;
[0084] in, Let k be the electricity price at time k. Let k be the grid interaction power at time k. To control the cycle, For the updated battery life degradation cost weighting, To predict the battery power sequence in the time domain, Weighting coefficients for the cost of maintaining the state of a supercapacitor.
[0085] The optimization objective consists of three cost components, and the goal is to minimize the sum of these three costs through optimization. The first component is the operating cost. This represents the economic cost of exchanging energy with the grid within the predicted time domain, such as the next 15 minutes. For example, if the electricity price for a certain minute within the predicted time domain... The cost is 1.2 yuan per kilowatt-hour, and the control period Δt is 1 minute. The optimization model calculates that 10 kW of power needs to be purchased from the grid in that minute. Then the cost for that minute is 0.2 yuan. This involves summing up the costs for each minute within those 15 minutes. The second item is the battery life degradation cost after the update. By analyzing the predicted battery power sequence using the rainflow counting method, the equivalent number of cycles within a prediction time domain is estimated. For example, 0.005 cycles. If the unit cycle cost of the battery... If the cost is 800 yuan, then the cost of battery lifespan degradation is... The cost is 4 yuan. This cost is then multiplied by the updated weight. For example, in 1.3, the cost included in the overall optimization objective is 5.2 yuan. The third item is the supercapacitor state maintenance cost. It is a penalty term used to ensure the supercapacitor's state of charge at the end of the prediction time domain. The goal is to get as close as possible to the ideal reference value, such as 50%. If the predicted end state of charge is 10%, this cost will incur a positive penalty, prompting the optimization algorithm to adjust its strategy to keep the supercapacitor's energy state at an even better level.
[0086] In an optional embodiment, calculating the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous period includes:
[0087] The actual SOC value of the battery at the current time k is measured by the battery management system. The predicted battery SOC value at the current moment compared to the previous control cycle. The difference is taken, and the absolute value is used as the mismatch residual of the battery SOC prediction model. : .
[0088] Specifically, this is executed at each control time k, and is used to represent the accuracy of the model's prediction. It requires two data points: the first is the prediction result from the previous control time k-1, which predicted that the battery's state of charge (SOC) would reach a specific value at the current time k, for example... =65.5%. The second data point is the actual measured value of k at the current time. This is obtained by directly measuring the battery's current true state of charge through the battery management system. For example, the measured value is 65.2%. Subtracting the two values yields -0.3%. Taking the absolute value of the difference gives the mismatch residual of the prediction model. It is 0.3%. For example... Figure 3 The 0.3% residual value represents the magnitude of the deviation between the model and the actual situation in this prediction.
[0089] In an optional embodiment, the Lyapunov exponential update model mismatch observer obtained based on the residuals and the current period includes:
[0090] A Kalman filter is used as the model mismatch observer, and the predicted model mismatch residual is used as the measurement information of the Kalman filter. The perturbation estimation state is updated iteratively through the Kalman gain; the process noise covariance matrix of the Kalman filter... Lyapunov exponents obtained from solving the predictive control model in the current period To regulate:
[0091]
[0092] in, The reference process noise covariance matrix, The preset stability threshold, It is a gain coefficient greater than 1.
[0093] A Kalman filter is used to estimate and compensate for unconsidered disturbances in the model, thereby improving prediction accuracy. At each control time step, the calculated prediction model mismatch residual, for example, 0.3%, is input into the Kalman filter as new measurement information. The filter combines the new information with its internal state and performs an iterative calculation using the Kalman gain to update the estimated value of the system disturbance. The updated disturbance estimate will be used in the model prediction of the next control cycle, making the prediction results closer to reality. The stability trend of the system is judged based on the Lyapunov exponent Λ obtained from the current optimization solution. A preset stability threshold is assumed. The Lyapunov exponent Λ is 0.02, and the gain coefficient α is 5. If the calculated Lyapunov exponent Λ is 0.03, which is greater than the threshold, it indicates that the system tends to be unstable. In this case, the fundamental covariance matrix is... Multiply by the gain factor α, i.e. Equal to five times This makes the Kalman filter more confident in new measurement information, thus correcting the state estimate more quickly and helping the system suppress instability. Conversely, if Λ is 0.01, which is less than the threshold, then it remains unchanged. = This allows the filter to be estimated smoothly in a steady state.
[0094] Example 2 provides a model-predictive-based hybrid energy storage system for energy optimization and allocation, comprising the following modules:
[0095] The acquisition module is used to acquire the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and perform multi-resolution singular value decomposition to obtain power subsequences with different frequency characteristics.
[0096] The update module is used to update the battery life decay cost weight and the power boundary constraints of each energy storage unit in the predictive control model in each control cycle by using the Lyapunov exponent of the battery state of charge trajectory determined in the previous control cycle, the mismatch residual of the prediction model and the disturbance estimation state, combined with the currently collected energy storage unit health state (SOH) and the power subsequence kurtosis value in the current prediction time domain.
[0097] The module is used to solve the updated predictive control model in a rolling manner, with the optimization objectives being operating cost, battery life decay cost taking into account the updated weights, and supercapacitor state maintenance cost, to obtain the current power allocation baseline value and the Lyapunov exponent of the battery SOC trajectory in the current prediction time domain.
[0098] The forming module is used to calculate the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous cycle, and update the model mismatch observer based on the residual and the Lyapunov exponent obtained in the current cycle. The observer outputs the disturbance estimation state and compensation control quantity for the next control cycle. The compensation control quantity is superimposed with the power allocation reference value to obtain the power allocation command.
[0099] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0100] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A method for optimal energy allocation in a hybrid energy storage system based on model predictive control, characterized in that, Includes the following steps: Obtain the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and perform multi-resolution singular value decomposition to obtain power subsequences with different frequency characteristics; In each control cycle, the Lyapunov exponent of the battery state of charge (SOC) trajectory determined in the previous control cycle, the mismatch residual of the prediction model, and the disturbance estimation state are used, combined with the currently collected energy storage unit health state (SOH) and the power subsequence kurtosis value in the current prediction time domain, to update the battery lifetime degradation cost weight and the power boundary constraints of each energy storage unit in the prediction control model respectively. The updated predictive control model is solved in a rolling manner, with the operating cost, battery life decay cost taking into account the updated weights, and supercapacitor state maintenance cost as optimization objectives, to obtain the current power allocation baseline value and the Lyapunov exponent of the battery SOC trajectory in the current prediction time domain. Calculate the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous cycle, and update the model mismatch observer based on the residual and the Lyapunov exponent obtained in the current cycle. The observer outputs the disturbance estimation state and compensation control quantity for the next control cycle. The compensation control quantity is superimposed with the power allocation reference value to obtain the power allocation command. The battery life degradation cost weights in the updated predictive control model include: The weights are adjusted based on the Lyapunov index determined in the previous control cycle and the magnitude of the mismatch residuals in the prediction model. When both the Lyapunov exponent and the mismatch residual exceed a preset safety threshold, the weight value is linearly increased to enhance the stability constraint on the battery SOC trajectory; when both the Lyapunov exponent and the mismatch residual are below the preset safety threshold, the weight value is linearly decreased.
2. The method according to claim 1, characterized in that, The process involves obtaining the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and performing multi-resolution singular value decomposition to obtain power subsequences with different frequency characteristics, including: The total power demand prediction sequence is decomposed into a low-frequency power subsequence, a mid-frequency power subsequence, and a high-frequency power subsequence. The power fluctuations of the low-frequency power subsequence are smoothed by battery cells, the power fluctuations of the high-frequency power subsequence are smoothed by supercapacitor cells, and the power fluctuations of the mid-frequency power subsequence are smoothed by battery cells and supercapacitor cells in a proportional manner.
3. The method according to claim 1, characterized in that, Update the power boundary constraints for each energy storage unit, including: Based on the current battery SOH data, the maximum charge / discharge power boundary of the battery is calculated using the following formula. : ,in, This refers to the battery's rated power. Based on the kurtosis value of the high-frequency power subsequence Adjust the discharge power boundary of the supercapacitor according to the following formula. : in, This refers to the rated discharge power of the supercapacitor. This serves as a reference value for kurtosis.
4. The method according to claim 1, characterized in that, The optimization objectives, which include operating costs, battery life degradation costs taking into account updated weights, and supercapacitor state maintenance costs, include: The expression for the objective function J is: ; Operating costs For prediction of the time domain The sum of the power exchanged between the internal power supply and the grid, multiplied by the time-of-use electricity price: ; Battery life degradation cost The equivalent cycle number of batteries was determined using the raindrop counting method. And multiply by the unit cycle cost of the battery get: ; supercapacitor state maintenance cost To predict the state of charge of supercapacitors at the end of the time domain Expected reference value Quadratic deviation between: ; in, Let k be the electricity price at time k. Let k be the grid interaction power at time k. To control the cycle, For the updated battery life degradation cost weighting, To predict the battery power sequence in the time domain, Weighting coefficients for the cost of maintaining the state of a supercapacitor.
5. The method according to claim 1, characterized in that, The calculation of the prediction model mismatch residual between the current actual operating state and the predicted value of the previous period includes: The actual SOC value of the battery at the current time k is measured by the battery management system. The predicted battery SOC value at the current moment compared to the previous control cycle. The difference is taken, and the absolute value is used as the mismatch residual of the battery SOC prediction model. .
6. The method according to claim 1, characterized in that, The mismatched observer of the Lyapunov exponential update model, obtained based on the residuals and the current period, includes: A Kalman filter is used as the model mismatch observer, and the predicted model mismatch residual is used as the measurement information of the Kalman filter. The perturbation estimation state is updated iteratively through the Kalman gain; the process noise covariance matrix of the Kalman filter... Lyapunov exponents obtained from solving the predictive control model in the current period To regulate: in, The reference process noise covariance matrix, The preset stability threshold, It is a gain coefficient greater than 1.
7. An energy optimization allocation system for a hybrid energy storage system based on model predictive control, characterized in that, Includes the following modules: The acquisition module is used to acquire the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and perform multi-resolution singular value decomposition to obtain power subsequences with different frequency characteristics. The update module is used to update the battery life decay cost weight and the power boundary constraints of each energy storage unit in the predictive control model in each control cycle by using the Lyapunov exponent of the battery state of charge trajectory determined in the previous control cycle, the mismatch residual of the prediction model and the disturbance estimation state, combined with the currently collected energy storage unit health state (SOH) and the power subsequence kurtosis value in the current prediction time domain. The module is used to solve the updated predictive control model in a rolling manner, with the optimization objectives being operating cost, battery life decay cost taking into account the updated weights, and supercapacitor state maintenance cost, to obtain the current power allocation baseline value and the Lyapunov exponent of the battery SOC trajectory in the current prediction time domain. The forming module is used to calculate the prediction model mismatch residual between the actual operating state at the current moment and the predicted value of the previous cycle, and update the model mismatch observer based on the residual and the Lyapunov exponent obtained in the current cycle. The observer outputs the disturbance estimation state and compensation control quantity for the next control cycle. The compensation control quantity is superimposed with the power allocation reference value to obtain the power allocation command. The battery life degradation cost weights in the updated predictive control model include: The weights are adjusted based on the Lyapunov index determined in the previous control cycle and the magnitude of the mismatch residuals in the prediction model. When both the Lyapunov exponent and the mismatch residual exceed a preset safety threshold, the weight value is linearly increased to enhance the stability constraint on the battery SOC trajectory; when both the Lyapunov exponent and the mismatch residual are below the preset safety threshold, the weight value is linearly decreased.
8. The system according to claim 7, characterized in that, The process involves obtaining the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain, and performing multi-resolution singular value decomposition to obtain power subsequences with different frequency characteristics, including: The total power demand prediction sequence is decomposed into a low-frequency power subsequence, a mid-frequency power subsequence, and a high-frequency power subsequence. The power fluctuations of the low-frequency power subsequence are smoothed by battery cells, the power fluctuations of the high-frequency power subsequence are smoothed by supercapacitor cells, and the power fluctuations of the mid-frequency power subsequence are smoothed by battery cells and supercapacitor cells in a proportional manner.