Hybrid energy storage system energy optimization distribution method and system based on model prediction

Abstract

The invention relates to a hybrid energy storage system energy optimization distribution method and system based on model prediction control, and specifically, the method comprises the steps: decomposing a total demand power prediction sequence into a multi-band sub-sequence, and then in each control period, based on a battery SOC track feature, a model mismatch residual error and a disturbance estimation state, carrying out the optimization of the multi-band sub-sequence; in combination with the current energy storage unit health state and the power subsequence kurtosis value, the battery life attenuation cost weight and the power boundary constraint are updated, the operation cost, the battery life attenuation cost and the super-capacitor state maintenance cost serve as optimization targets, the model is solved in a rolling mode, and a power distribution reference value is obtained; a real-time power instruction is obtained by calculating a model mismatch residual error of an actual operation state and a predicted value and combining a current Lyapunov exponent to update a model mismatch observer, outputting a compensation control quantity and superposing the compensation control quantity with a reference value, and energy optimization distribution is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of energy optimization allocation, and particularly relates to a model prediction-based energy optimization allocation method and system for a hybrid energy storage system. BACKGROUND

[0002] A hybrid energy storage system is usually composed of two or more energy storage elements such as batteries and supercapacitors, and has the advantages of high energy density of the battery and high power density and long cycle life of the supercapacitor, and meets the variable power demand in modern power systems, electric vehicles, microgrids and other application scenarios. The core of the hybrid energy storage system is the energy management strategy, which allocates power between different energy storage units in real time and reasonably. Existing energy management strategies mainly include rule-based control and filter-based control methods. For example, a low-pass filter is used to decompose the total demand power into high-frequency and low-frequency parts, which are respectively borne by the supercapacitor and the battery. Although this method is simple to implement and has small computational complexity, the control logic and parameters are usually set based on experience, and it is difficult to maintain optimal control effect, often leading to overuse of the battery, accelerating the life decay, or failing to fully play the power buffering role of the supercapacitor. The energy management strategy based on optimization, such as model predictive control (MPC), solves an optimal control problem in a limited future time domain based on the prediction model of the system in each control cycle, and achieves the preset optimization goal, such as minimizing the system energy consumption or prolonging the battery life. However, there is a mismatch between the prediction model and the actual system, which affects the control accuracy and optimization effect. The actual cost change of the battery life consumption under different health states and different load characteristics cannot be reflected, which limits the application of the optimization control. Moreover, the MPC lacks an online compensation and correction mechanism for the deviation between the actual state and the predicted trajectory caused by the model mismatch and the prediction error in the rolling optimization process. SUMMARY

[0003] To solve the above problems, in the first aspect of the present application, a model prediction-based energy optimization allocation method for a hybrid energy storage system is provided, comprising the following steps: Obtaining a total demand power prediction sequence of the hybrid energy storage system in a future prediction time domain, and performing multi-resolution singular value decomposition to obtain power subsequences of different frequency characteristics; In each control cycle, the Lyapunov exponent of the battery state of charge (SOC) trajectory determined in the last control cycle, the prediction model mismatch residual and the disturbance estimation state are used, combined with the currently collected energy storage unit health state (SOH) and the kurtosis value of the power subsequence in the current prediction time domain, to update the battery life decay cost weight in the prediction control model and the power boundary constraint of each energy storage unit, respectively. solving the updated predictive control model to obtain power allocation reference value at current time and Lyapunov index of battery SOC trajectory in current prediction time domain, with operation cost, battery life attenuation cost considering updated weight and super capacitor state maintenance cost as optimization objectives; calculating prediction model mismatch residual between actual operation state at current time and prediction value at last period, and updating model mismatch observer based on the residual and Lyapunov index obtained in current period, and outputting disturbance estimation state and compensation control amount for next control period from the observer; superimposing the compensation control amount and the power allocation reference value to obtain power allocation instruction.

[0004] Optionally, the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain is obtained, and multi-resolution singular value decomposition is performed to obtain power subsequences of different frequency characteristics, including: the total demand power prediction sequence is decomposed into a low-frequency power subsequence, a medium-frequency power subsequence and a high-frequency power subsequence; the power fluctuation of the low-frequency power subsequence is smoothed by the battery unit, the power fluctuation of the high-frequency power subsequence is smoothed by the super capacitor unit, and the power fluctuation of the medium-frequency power subsequence is smoothed by the battery unit and the super capacitor unit in proportion.

[0005] Optionally, the weight of the battery life attenuation cost in the predictive control model is updated, including: the weight is adjusted according to the Lyapunov index determined in the last control period and the size of the prediction model mismatch residual; when the Lyapunov index and the mismatch residual both exceed the preset safety threshold, the weight value is linearly increased to enhance the stability constraint of the battery SOC trajectory; when the Lyapunov index and the mismatch residual are both lower than the preset safety threshold, the weight value is linearly reduced.

[0006] Optionally, the power boundary constraint of each energy storage unit is updated, including: the maximum charge and discharge power boundary of the battery is calculated according to the currently collected battery SOH, as follows: : wherein, Pbat is the rated power of the battery; the discharge power boundary of the super capacitor is adjusted according to the kurtosis value of the high-frequency power subsequence, as follows: : wherein, Pcap is the rated discharge power of the super capacitor, and kref is the kurtosis reference reference value.

[0007] Optionally, the optimization objectives include:​ The expression of the optimization objective function J is: ; Operating cost The prediction horizon The accumulated sum of the interaction power with the power grid multiplied by the time-of-use electricity price: ; Battery life attenuation cost The equivalent cycle number of the battery is counted by the rain flow counting method And multiplied by the unit cycle cost of the battery Get: ; Supercapacitor state maintenance cost The state of charge of the supercapacitor at the end of the prediction horizon Desired reference value The quadratic deviation between: ; Where, is the electricity price at time k, is the power grid interaction power at time k, is the control period, is the updated battery life attenuation cost weight, is the battery power sequence within the prediction horizon, is the weight coefficient of the supercapacitor state maintenance cost.

[0008] Optionally, the prediction model mismatch residual between the actual operating state at the current time and the predicted value at the last period is calculated, comprising: The actual SOC value of the battery measured by the battery management system at the current time k Difference between the predicted battery SOC value at the current time predicted in the last control period Take the absolute value as the prediction model mismatch residual of the battery SOC .

[0009] Optionally, the Lyapunov index is updated based on the residual and the current period to update the model mismatch observer, comprising: Adopting a Kalman filter as the model mismatch observer, taking the prediction model mismatch residual as the measurement innovation of the Kalman filter, and updating the disturbance estimation state through Kalman gain iteration; The process noise covariance matrix of the Kalman filter According to the Lyapunov index obtained by solving the prediction control model in the current period Regulation: Where, is the reference process noise covariance matrix, is a preset stable threshold, a gain coefficient greater than 1.

[0010] In a second aspect of the present application, a model prediction-based energy optimization distribution system for a hybrid energy storage system is provided, comprising the following modules: An acquisition module is configured to acquire a total demand power prediction sequence of the hybrid energy storage system in a future prediction time domain, and perform multi-resolution singular value decomposition to obtain power sub-sequences with different frequency characteristics. An update module is configured to, in each control period, update a battery life attenuation cost weight in a prediction control model and power boundary constraints of each energy storage unit, by using a Lyapunov exponent of a battery state of charge (SOC) trajectory determined in a previous control period, a prediction model mismatch residual, and a disturbance estimation state, in combination with a currently collected energy storage unit state of health (SOH) and a kurtosis value of the power sub-sequences in the current prediction time domain. An obtaining module is configured to solve the updated prediction control model to obtain a power distribution reference value at a current time and a Lyapunov exponent of a battery SOC trajectory in the current prediction time domain, with the optimization objectives being an operation cost, a battery life attenuation cost considering the updated weight, and a super capacitor state maintenance cost. A forming module is configured to calculate a prediction model mismatch residual between an actual operation state at the current time and a prediction value in a previous period, and update a model mismatch observer based on the residual and the Lyapunov exponent obtained in the current period, to output a disturbance estimation state and a compensation control amount for a next control period from the observer; and to obtain a power distribution instruction by superimposing the compensation control amount and the power distribution reference value.

[0011] Preferably, the acquisition of the total demand power prediction sequence of the hybrid energy storage system in the future prediction time domain and the multi-resolution singular value decomposition to obtain the power sub-sequences with different frequency characteristics comprise: The total demand power prediction sequence is decomposed into a low-frequency power sub-sequence, a medium-frequency power sub-sequence, and a high-frequency power sub-sequence; power fluctuations of the low-frequency power sub-sequence are smoothed by battery units, power fluctuations of the high-frequency power sub-sequence are smoothed by super capacitor units, and power fluctuations of the medium-frequency power sub-sequence are smoothed by the battery units and the super capacitor units in proportion.

[0012] Preferably, the update of the battery life attenuation cost weight in the prediction control model comprises: The weight is adjusted according to the size of the Lyapunov exponent and the prediction model mismatch residual determined in the previous control period; 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; and when both the Lyapunov exponent and the mismatch residual are lower than the preset safety threshold, the weight value is linearly decreased.

[0013] Preferably, the power boundary constraints of each energy storage unit are updated, including: According to the current collected battery SOH, the maximum charge-discharge power boundary of the battery is calculated as follows : Wherein, is the rated power of the battery; According to the kurtosis value of the high-frequency power subsequence , the discharge power boundary of the super capacitor is adjusted as follows : Wherein, is the rated discharge power of the super capacitor, is the kurtosis reference benchmark value.

[0014] Preferably, the optimization objective is to minimize the operation cost, the battery life attenuation cost considering the updated weight, and the super capacitor state maintenance cost, including: The expression of the optimization objective function J is: ; The operation cost is the cumulative sum of the interaction power with the power grid multiplied by the time-of-use electricity price in the prediction time domain: ; The battery life attenuation cost is obtained by counting the equivalent cycle number of the battery by the rain flow counting method and multiplying the battery unit cycle cost : ; The super capacitor state maintenance cost is the quadratic deviation between the state of charge of the super capacitor at the end of the prediction time domain and the expected reference value : ; Wherein, is the electricity price at time k, is the power grid interaction power at time k, is the control period, is the updated battery life attenuation cost weight, is the battery power sequence in the prediction time domain, is the weight coefficient of the super capacitor state maintenance cost.

[0015] Preferably, the prediction model mismatch residual error between the actual operating state at the current time and the predicted value at the last period is calculated, including: The actual battery SOC value measured by the battery management system at the current time k is compared with the battery SOC prediction value at the current time predicted in the last control period ​Taking the absolute value of the difference as a prediction model mismatch residual error of the battery SOC .

[0016] Preferably, the Lyapunov index obtained based on the residual error and the current period is used to update the model mismatch observer, including: The Kalman filter is used as the model mismatch observer, the prediction model mismatch residual error is used as the measurement innovation of the Kalman filter, and the disturbance estimation state is iteratively updated through the Kalman gain; the process noise covariance matrix of the Kalman filter The Lyapunov index obtained based on the current period is solved Regulation: Wherein, is a reference process noise covariance matrix, is a preset stable threshold, is a gain coefficient greater than 1.

[0017] The present application realizes the stripping of power components of different frequencies by performing multi-resolution singular value decomposition on the predicted power. In the prediction control model, the weight of the battery life attenuation cost is adjusted in combination with the current health state of the battery, the impact characteristics of the predicted load and the historical operation stability index of the system, so that the optimization target is closer to the actual working condition of the energy storage system, thereby delaying the performance degradation of the battery in energy optimization allocation. By constructing a model mismatch observer to estimate and compensate the prediction deviation caused by model uncertainty and external disturbance, and superimposing the compensation amount on the power reference obtained by optimization solving, the accuracy of power distribution of the control system is enhanced, while ensuring stable operation of the system, the control of the whole life cycle cost of the energy storage system is realized. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of the first embodiment; Figure 2 is a weight and boundary update schematic diagram; Figure 3 is a prediction model mismatch residual error calculation schematic diagram. DETAILED DESCRIPTION

[0019] To make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] The plurality referred to in the present application refers to two or more. In addition, it needs to be understood that in the description of the present application, the words "first", "second", etc. are only used for the purpose of distinguishing the description, and cannot be understood as indicating or implying relative importance, nor can it be understood as indicating or implying order.

[0021] Embodiment one provides a model prediction-based energy optimization allocation method for a hybrid energy storage system, as shown in Figure 1 , comprising the following steps: S1, 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 of different frequency characteristics; Preferably, a long short-term memory network model is used to predict the total demand power sequence of N control periods, for example N=300, according to historical power data, weather information and electricity price information. A trajectory matrix of m rows and n columns, i.e. a Hankel matrix, is constructed from the predicted power sequence, where m+n-1=N. Singular value decomposition is performed on the Hankel matrix to obtain a series of singular values arranged in numerical size. According to the mutation points of the singular value size, the singular values are grouped, for example, the first group represents the main low-frequency trend of the power sequence and the second group represents the high-frequency fluctuation. The trajectory matrix is reconstructed using the singular values of each group, and the time sequence is recovered from the reconstructed trajectory matrix by diagonal line averaging method to obtain a low-frequency power sub-sequence and a high-frequency power sub-sequence.

[0022] S2, in each control period, the Lyapunov exponent of the battery state of charge SOC trajectory determined in the last control period, the prediction model mismatch residual and the disturbance estimation state are combined with the currently collected energy storage unit health state SOH and the kurtosis value of the power sub-sequence in the current prediction time domain to update the battery life attenuation cost weight in the prediction control model and the power boundary constraint of each energy storage unit, respectively; The basic weight of the battery life attenuation cost is a preset value. Calculate the kurtosis value K of the high-frequency power sub-sequence, the larger the K value, the more intense the power impact. Based on the Lyapunov exponent λ, the model mismatch residual e and the disturbance estimation state d calculated in the last period, and the battery health state SOH obtained at present, the updated weight , wherein to is 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, which more strictly limits the damage to the battery in the optimization, such as Figure 2The model mismatch residual e is obtained by subtracting the actual measured current battery SOC from the last cycle model predicted current SOC value at the beginning of the current cycle, and the residual e is input to the model mismatch observer, which outputs the updated disturbance estimation state d after processing the information. The power boundary is updated according to the current battery health state SOH, for example, the maximum charging and discharging power of the battery is equal to the factory rated power multiplied by a decay factor that is positively correlated with SOH.

[0023] S3, the updated prediction control model is solved by rolling to obtain the power allocation reference value at the current time and the Lyapunov index of the battery SOC trajectory in the current prediction time domain as the optimization objective; The optimization objective function J is equal to the sum of the running cost, the battery life attenuation cost and the super capacitor state maintenance cost. The running cost is represented by the product of the grid interaction power and the real-time electricity price; the battery life attenuation cost is approximated by the updated weight W multiplied by the sum of the squares of the battery power changes in the prediction time domain; and the super capacitor state maintenance cost is represented by the sum of the squares of the state of charge deviation from the reference value, such as fifty percent. Under the conditions of satisfying the power balance constraint, the updated power boundary constraint of each energy storage unit, the state of charge range constraint and the power change rate constraint, the optimization problem is solved using a quadratic programming or nonlinear programming solver to obtain the optimal power allocation sequence in the next N cycles. The first element of the sequence is taken as the power allocation reference value at the current time. According to the battery state of charge SOC prediction trajectory obtained by optimization in the next N cycles, the Lyapunov index of the trajectory is calculated for weight updating in the next cycle.

[0024] S4, the prediction model mismatch residual between the actual running state at the current time and the predicted value in the last cycle is calculated, and the model mismatch observer is updated based on the residual and the Lyapunov index obtained in the current cycle, and the disturbance estimation state and the compensation control amount for the next control cycle are output by the observer; the compensation control amount and the power allocation reference value are superimposed to obtain the power allocation instruction.

[0025] Specifically, in the current control cycle k, the actual state of charge of the battery is measured The optimization solution in the last cycle k-1 predicted that the battery state of charge at cycle k is The prediction model mismatch residual e(k) 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).

[0026] 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.

[0027] 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: 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.

[0028] 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 the decomposition is completed, power distribution is performed according to the characteristics of each energy storage unit. For example, for a power demand at a moment, if the decomposition obtains a low-frequency component of 10 kW, a medium-frequency component of 2 kW, and a high-frequency component of 3 kW. The entire 10 kW of low-frequency power is all instructed to the battery unit to respond. The 3 kW of high-frequency power is all instructed to the super capacitor unit to respond. For the 2 kW of medium-frequency power, according to a preset distribution coefficient, for example, the battery bears 70%, and the super capacitor bears 30%, 1.4 kW is distributed to the battery unit, and 0.6 kW is distributed to the super capacitor unit. The total response power of the battery unit is 11.4 kW, and the total response power of the super capacitor unit is 3.6 kW.

[0029] In an optional embodiment, updating the battery life attenuation cost weight in the prediction control model comprises: According to the Lyapunov index and the size of the prediction model mismatch residual determined in the last control period, the weight is adjusted; when the Lyapunov index and the mismatch residual both exceed the preset safety threshold, the weight value is linearly increased to enhance the stability constraint on the battery SOC trajectory; when the Lyapunov index and the mismatch residual are both lower than the preset safety threshold, the weight value is linearly decreased.

[0030] At the beginning of each control period, two key indicators calculated in the last period, the Lyapunov index and the prediction model mismatch residual, are obtained. Assuming that the preset safety threshold is 0.02 for the Lyapunov index and 0.5% for the mismatch residual. In a working condition, if the Lyapunov index of the last period is 0.03 and the mismatch residual is 0.8%, both of which exceed the set safety threshold, it indicates that the stability of the system is at risk and the prediction accuracy of the model decreases. The weight value of the battery life attenuation cost is linearly increased, for example, from the current 1.2 to 1.3.

[0031] The increased weight value will make the subsequent optimization calculation pay more attention to the stable operation of the battery, and preferentially select a power distribution scheme that can make the battery state of charge trajectory more stable, even if it will slightly increase the operating cost. Conversely, in another working condition, if the Lyapunov index of the last period is 0.01 and the mismatch residual is 0.2%, both of which are lower than the safety threshold, it indicates that the operation is stable and the model is reliable. At this time, the weight value is linearly decreased, for example, from 1.2 to 1.15. The decrease of the weight indicates that the constraint on the stability of the battery is relaxed, allowing the optimization algorithm to consider economy more and seek a solution with the lowest total operating cost.

[0032] In an optional embodiment, updating the power boundary constraint of each energy storage unit comprises: According to the currently collected battery SOH, the maximum charge and discharge power boundary of the battery is calculated according to the following formula wherein,​ Battery rated power; According to the kurtosis value of the high-frequency power subsequence , adjust the discharge power boundary of the super capacitor as follows : Wherein, is the rated discharge power of the super capacitor, is the kurtosis reference benchmark value.

[0033] Specifically, for the battery part, the state of health SOH value of the battery is obtained in real time from the battery management system. Assuming that a battery with a rated power of 100 kW, the current measured SOH value is 0.9, i.e. the health degree is 90%. The two values are substituted into the formula to calculate the current maximum charge and discharge power of the battery, which is 90 kW. The calculated 90 kW will be used as an upper limit constraint for the battery charge and discharge operation in this control period, ensuring the safe and stable operation of the aging battery.

[0034] For the super capacitor part, the kurtosis value of the high-frequency power subsequence is calculated, which represents the severity of the power impact. Assuming that the rated discharge power of the super capacitor is 50 kW, and the kurtosis reference benchmark value is set to 3. If the currently calculated kurtosis value is 4.5, it means that the power fluctuation is more acute than usual. The adjusted discharge power boundary of the super capacitor is calculated as 57.5 kW. This means that the discharge capacity of the super capacitor has been temporarily increased to cope with the upcoming severe power impact, thereby better protecting the battery.

[0035] In an optional embodiment, the optimization objective of taking the running cost, the battery life attenuation cost considering the updated weight, and the super capacitor state maintenance cost as the optimization objective includes: The expression of the optimization objective function J is: ; The running cost is the cumulative sum of the interaction power with the power grid multiplied by the time-of-use electricity price in the prediction time domain: ; The battery life attenuation cost is obtained by multiplying the equivalent cycle number of the battery counted by the rain flow counting method by the unit cycle cost of the battery : ; The super capacitor state maintenance cost is the expected reference value of the state of charge of the super capacitor at the end of the prediction time domain : ​Quadratic deviation between the predicted and the actual state of charge: ; wherein, is the electricity price at time k, is the grid interaction power at time k, is the control period, is the updated battery life degradation cost weight, is the predicted battery power sequence within the prediction horizon, is the weight coefficient of the super capacitor state maintenance cost.

[0036] The optimization objective is composed of three parts of cost, and the sum of the three costs is minimized by optimization solution. The first part is the operation cost , which represents the economic cost of energy exchange with the grid within the prediction horizon, for example, the next 15 minutes. For example, if the electricity price of a certain minute within the prediction horizon is 1.2 yuan per kilowatt-hour, the control period Δt is 1 minute, and the optimization model calculates that 10 kW of power needs to be purchased from the grid in that minute, then the cost of that minute is 0.2 yuan. The sum of the costs of each minute within the 15 minutes is the operation cost. The second part is the battery life degradation cost after updating the weight. By rain flow counting method, the equivalent cycle number within a prediction horizon is estimated , for example, 0.005 times. If the unit cycle cost of the battery is 800 yuan, then the life degradation cost of the battery is 4 yuan. Multiply this cost by the updated weight , for example, 1.3, to get the cost of 5.2 yuan, which is included in the total optimization objective. The third part is the super capacitor state maintenance cost , which is a penalty term to ensure that the state of charge of the super capacitor at the end of the prediction horizon is as close as possible to the ideal reference value, for example, 50%. If the predicted end-of-life state is 10%, this cost will produce a positive penalty, prompting the optimization algorithm to adjust the strategy to keep the energy state of the super capacitor at a more optimal level.

[0037] In an optional embodiment, the calculation of the prediction model mismatch residual between the actual running state at the current time and the predicted value at the last period includes: Subtracting the actual SOC value of the battery measured by the battery management system at the current time k from the predicted SOC value of the battery at the current time predicted in the last control period , and taking the absolute value as the prediction model mismatch residual of the battery SOC : .

[0038] Specifically, at each control time k, to indicate the accuracy of model prediction, two data are needed, the first data is the prediction result at the last control time k-1, when it is predicted that the state of charge SOC of the battery at the current time k will reach a certain value, for example =65.5%. The second data is the actual measured value at the current time k. The real state of charge of the battery is directly measured by the battery management system, and the measured value is obtained , for example, the measured value is 65.2%. Subtracting the two values, the prediction model mismatch residual is obtained 0.3%. As Figure 3 , the residual value of 0.3% indicates the deviation between the model and the actual situation in this prediction.

[0039] In an optional embodiment, the Lyapunov index obtained based on the residual and the current period is used to update the model mismatch observer, comprising: Using a Kalman filter as the model mismatch observer, taking the prediction model mismatch residual as the measurement innovation of the Kalman filter, and updating the disturbance estimation state through Kalman gain iteration; the process noise covariance matrix of the Kalman filter is regulated according to the Lyapunov index Λ obtained by solving the prediction control model in the current period: wherein, is the reference process noise covariance matrix, is a preset stable threshold, and the gain coefficient α is greater than 1.

[0040] The Kalman filter is used to estimate and compensate the disturbance not considered in the model, so as to improve the prediction accuracy. At each control time, the prediction model mismatch residual calculated above, for example 0.3%, is input into the Kalman filter as new measurement information. The filter combines the new information with its own internal state, and performs one iteration calculation through Kalman gain to update the estimated value of the system disturbance. The updated disturbance estimation value will be used in the model prediction of the next control period, so that the prediction result is closer to the actual situation. According to the Lyapunov index Λ obtained by the current optimization solution, the stability trend of the system is judged. Assuming that the preset stable threshold is 0.02, and the gain coefficient α is 5. If the calculated Lyapunov index Λ is 0.03, which is greater than the threshold, it indicates that the system tends to be unstable, and the basic covariance matrix is multiplied by the gain coefficient α, that is, which is equal to five times The Kalman filter is made to believe the new measurement information more, so as to correct the state estimation at a faster speed, and help the system to suppress the unstable trend. = The filter is kept to estimate smoothly in the stable state.

[0041] Embodiment two provides a model prediction-based hybrid energy storage system energy optimization distribution system, comprising the following modules: The acquisition module is used for acquiring a total demand power prediction sequence of the hybrid energy storage system in a future prediction time domain, and performing multi-resolution singular value decomposition to obtain power subsequences of different frequency characteristics; The update module is used for updating, in each control period, a battery state of charge (SOC) trajectory Lyapunov exponent determined in the last control period, a prediction model mismatch residual, and a disturbance estimation state, in combination with a currently collected energy storage unit state of health (SOH) and a power subsequence kurtosis value in the current prediction time domain, to update a battery life attenuation cost weight in the prediction control model and power boundary constraints of each energy storage unit, respectively; The obtaining module is used for rolling solving the updated prediction control model to obtain a power distribution reference value at the current time and a battery SOC trajectory Lyapunov exponent in the current prediction time domain, with the running cost, the battery life attenuation cost considering the updated weight, and the super capacitor state maintenance cost as the optimization objectives; The forming module is used for calculating a prediction model mismatch residual between an actual running state at the current time and a last period prediction value, and updating a model mismatch observer based on the residual and the Lyapunov exponent obtained in the current period, and outputting, from the observer, a disturbance estimation state and a compensation control amount for a next control period; and superimposing the compensation control amount and the power distribution reference value to obtain a power distribution instruction.

[0042] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or one block or multiple blocks. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or one block or multiple blocks.

[0043] The above merely provides an example of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application should be included in the scope of claims of the present application.

Claims

1. A model prediction based energy optimization allocation method for hybrid energy storage system, characterized in that, The method comprises the following steps: obtaining a total demand power prediction sequence of the hybrid energy storage system in a future prediction time domain, and performing multi-resolution singular value decomposition to obtain power subsequences of different frequency characteristics; in each control period, using a Lyapunov index of a battery state of charge (SOC) trajectory determined in the last control period, a prediction model mismatch residual and a disturbance estimation state, in combination with a currently collected energy storage unit state of health (SOH) and a kurtosis value of the power subsequence in the current prediction time domain, to update a battery life attenuation cost weight in the prediction control model and power boundary constraints of each energy storage unit, respectively; rolling solving the updated prediction control model to obtain a power allocation reference value at the current time and a Lyapunov index of the battery SOC trajectory in the current prediction time domain, with an operating cost, a battery life attenuation cost considering the updated weight and a super capacitor state maintenance cost as optimization objectives; calculating a prediction model mismatch residual between an actual operating state at the current time and a prediction value in the last period, and updating a model mismatch observer based on the residual and the Lyapunov index obtained in the current period, and outputting, from the observer, a disturbance estimation state and a compensation control amount for the next control period; and superimposing the compensation control amount and the power allocation reference value to obtain a power allocation instruction.

2. The method of claim 1, wherein, The method of obtaining a total demand power prediction sequence of the hybrid energy storage system in a future prediction time domain, and performing multi-resolution singular value decomposition to obtain power subsequences of different frequency characteristics comprises: decomposing the total demand power prediction sequence into a low-frequency power subsequence, a medium-frequency power subsequence and a high-frequency power subsequence; power fluctuations of the low-frequency power subsequence are smoothed by battery units, power fluctuations of the high-frequency power subsequence are smoothed by super capacitor units, and power fluctuations of the medium-frequency power subsequence are smoothed by battery units and super capacitor units in proportion.

3. The method of claim 1, wherein, The method of updating a battery life attenuation cost weight in the prediction control model comprises: adjusting the weight according to a Lyapunov index and a size of a prediction model mismatch residual determined in the last control period; when the Lyapunov index and the mismatch residual both exceed a preset safety threshold, linearly increasing the weight value to enhance the stability constraint on the battery SOC trajectory; and when the Lyapunov index and the mismatch residual are both lower than the preset safety threshold, linearly decreasing the weight value.

4. The method of claim 1, wherein, The method of updating power boundary constraints of each energy storage unit comprises: Based on the currently collected battery SOH, the maximum charge-discharge power boundary of the battery is calculated as follows : , wherein Pbat is the battery rated power; Adjusting the discharge power boundary of the super capacitor according to the kurtosis value of the high-frequency power subsequence , the discharge power boundary of the super capacitor is adjusted according to the kurtosis value of the high-frequency power subsequence : wherein, is the rated discharge power of the super capacitor, is the kurtosis reference base value.

5. The method of claim 1, wherein, The method of taking an operating cost, a battery life attenuation cost considering the updated weight and a super capacitor state maintenance cost as optimization objectives comprises: The expression of the optimization objective function J is: ; Operating costs To predict the time domain The accumulated sum of the power exchanged with the grid multiplied by the time-of-use tariff: ; Battery life degradation cost Counting battery equivalent cycles by rainflow counting And multiplying by battery unit cycle cost Results in: ; Super-capacitor state maintenance cost To predict the end-of-time horizon super-capacitor state-of-charge Desired reference value Quadratic deviation between: ; wherein, is the electricity price at time k, is the grid interaction power at time k, is the control period, is the updated battery life degradation cost weight, is the battery power sequence within the prediction horizon, is the weight coefficient of the supercapacitor state maintenance cost.

6. The method of claim 1, wherein, The method of calculating a prediction model mismatch residual between an actual operating state at the current time and a prediction value in the last period comprises: The actual SOC value of the battery measured by the battery management system at the current time k The predicted value of the battery SOC at the current time predicted in the last control cycle Difference, taking the absolute value as the predicted model mismatch residual of the battery SOC .

7. The method of claim 1, wherein, The method of updating a model mismatch observer based on the residual and a Lyapunov index obtained in the current period comprises: Adopting a Kalman filter as the model mismatch observer, taking the prediction model mismatch residual as the measurement innovation of the Kalman filter, and iteratively updating the disturbance estimation state through the Kalman gain; the process noise covariance matrix of the Kalman filter Lyapunov index obtained according to the current period solving the prediction control model Performing regulation: wherein, is a reference process noise covariance matrix, is a predetermined stability threshold, is a gain coefficient greater than 1.

8. A model prediction based hybrid energy storage system energy optimized distribution system, characterized in that, The method comprises the following modules: an obtaining module, configured to obtain a total demand power prediction sequence of the hybrid energy storage system in a future prediction time domain, and perform multi-resolution singular value decomposition to obtain power subsequences of different frequency characteristics; an updating module, configured to update a battery life attenuation cost weight in the prediction control model and power boundary constraints of each energy storage unit, respectively, in each control period, using a Lyapunov index of a battery state of charge (SOC) trajectory determined in the last control period, a prediction model mismatch residual and a disturbance estimation state, in combination with a currently collected energy storage unit state of health (SOH) and a kurtosis value of the power subsequence in the current prediction time domain; a rolling solving module, configured to rolling solve the updated prediction control model to obtain a power allocation reference value at the current time and a Lyapunov index of the battery SOC trajectory in the current prediction time domain, with an operating cost, a battery life attenuation cost considering the updated weight and a super capacitor state maintenance cost as optimization objectives; a calculating module, configured to calculate a prediction model mismatch residual between an actual operating state at the current time and a prediction value in the last period, and update a model mismatch observer based on the residual and the Lyapunov index obtained in the current period, and output, from the observer, a disturbance estimation state and a compensation control amount for the next control period; and superimpose the compensation control amount and the power allocation reference value to obtain a power allocation instruction. The updating module is configured to update, in each control cycle, a battery life attenuation cost weight in a prediction control model and power boundary constraints of each energy storage unit by using a Lyapunov index of a battery state of charge (SOC) trajectory, a prediction model mismatch residual and a disturbance estimation state determined in a previous control cycle, in combination with a currently collected energy storage unit state of health (SOH) and a kurtosis value of a power subsequence in a current prediction time domain. The obtaining module is configured to solve the updated prediction control model to obtain a power allocation reference value at a current time and a Lyapunov index of a battery SOC trajectory in a current prediction time domain, with an operation cost, a battery life attenuation cost considering the updated weight and a super capacitor state maintenance cost as optimization objectives. The forming module is configured to calculate a prediction model mismatch residual between an actual operation state at a current time and a predicted value in a previous cycle, and update a model mismatch observer based on the residual and the Lyapunov index obtained in the current cycle, and output, from the observer, a disturbance estimation state and a compensation control amount for a next control cycle; and superimpose the compensation control amount and the power allocation reference value to obtain a power allocation instruction.

9. The system of claim 8, wherein, The total demand power prediction sequence of the hybrid energy storage system in a future prediction time domain is obtained, and multi-resolution singular value decomposition is performed to obtain power subsequences of different frequency characteristics, including: The total demand power prediction sequence is decomposed into a low-frequency power subsequence, a medium-frequency power subsequence and a high-frequency power subsequence; power fluctuations of the low-frequency power subsequence are smoothed by battery units, power fluctuations of the high-frequency power subsequence are smoothed by super capacitor units, and power fluctuations of the medium-frequency power subsequence are smoothed by battery units and super capacitor units in proportion.

10. The system of claim 8, wherein, The battery life attenuation cost weight in the prediction control model is updated, including: The weight is adjusted according to the size of the Lyapunov index and the prediction model mismatch residual determined in the previous control cycle; when the Lyapunov index and the mismatch residual both exceed a preset safety threshold, the weight value is linearly increased to enhance the stability constraint on the battery SOC trajectory; and when the Lyapunov index and the mismatch residual are both lower than the preset safety threshold, the weight value is linearly decreased.

Images