Multi-type energy storage joint peak shaving and valley filling optimization operation method in interchange and storage integration scenario

Abstract

This application provides a method for joint peak shaving and valley filling optimization operation of multiple types of energy storage in a scenario of energy storage and transportation integration, belonging to the field of energy storage optimization technology. The method includes: performing source-load time-series decomposition prediction based on the basic dataset of multiple types of energy storage devices; using performance difference information as the basis for power allocation and the static operating boundary constraint set as the limit of the search space, performing iterative search and solution; performing decomposition prediction with adjusted prediction parameters; loading the corrected prediction results with the adjusted search space limit and optimized search direction, and performing iterative search and solution again; and executing the corrected scheduling decision instruction as the initial scheduling decision instruction for the next cycle. This application can realize differentiated joint scheduling of energy-type energy storage and power-type energy storage, enabling multiple types of energy storage devices to operate in their optimal power range, thereby improving the overall peak shaving and valley filling regulation effect and extending the overall service life of the energy storage system.

Description

Technical Field

[0001] This application relates to the field of energy storage optimization technology, and in particular to a method for optimizing the operation of multiple types of energy storage in a scenario of energy exchange and storage integration, including peak shaving and valley filling. Background Technology

[0002] With the accelerated advancement of transportation electrification, large-scale connections have been made to electric vehicle charging and swapping facilities and distributed new energy power generation equipment in transportation infrastructure such as highway service areas, bus hubs, and railway traction power supply stations. This has created a scenario where traffic flow and power flow are deeply coupled and integrated. In this scenario, the disorderly charging behavior of electric vehicles and the random fluctuations in the output of new energy power generation are superimposed, resulting in a bidirectional extreme characteristic of the load curve of local distribution networks, with a sharp increase in peak power and a significant decrease in valley power. This poses a direct threat to the safe and stable operation of the power grid.

[0003] Currently, existing peak shaving and valley filling scheduling methods typically treat various types of energy storage devices, such as lithium-ion battery energy storage, supercapacitor energy storage, and flow battery energy storage, as a unified energy storage system. They use the overall total capacity and total power constraints as scheduling boundaries and allocate power commands to all energy storage devices indiscriminately according to uniform charge and discharge efficiency parameters and uniform response time parameters. The above-mentioned homogenization method ignores the essential differences between different types of energy storage in terms of physical characteristics such as energy density, power density, cycle life, and response speed. This leads to accelerated capacity decay of energy-type energy storage during frequent short-term power fluctuations, and frequent deep charging and discharging of power-type energy storage due to insufficient capacity during long-term continuous charging and discharging. The inherent advantages of both types of energy storage cannot be effectively utilized.

[0004] In summary, existing technologies suffer from the lack of a peak-shaving and valley-filling operation method that can distinguish the physical characteristics of energy-type energy storage and power-type energy storage and perform differentiated joint scheduling accordingly. This results in multiple types of energy storage being unable to allocate power according to their respective optimal operating ranges during peak-shaving and valley-filling, further affecting the overall regulation effect of joint peak-shaving and valley-filling of multiple types of energy storage in the scenario of energy storage and energy exchange integration, as well as the comprehensive benefits of energy storage equipment throughout its entire life cycle. Summary of the Invention

[0005] The purpose of this application is to provide a method for optimizing the joint peak shaving and valley filling operation of multiple types of energy storage in the scenario of energy storage and energy exchange integration. This method addresses the technical problem in the existing technology that, due to the lack of a method that can distinguish the physical characteristics of energy-type energy storage and power-type energy storage and perform differentiated joint scheduling accordingly, multiple types of energy storage cannot allocate power according to their respective optimal operating ranges during the peak shaving and valley filling process. This further affects the overall regulation effect of joint peak shaving and valley filling of multiple types of energy storage in the scenario of energy storage and energy exchange integration and the comprehensive benefits of energy storage equipment throughout its entire life cycle.

[0006] In view of the above problems, this application provides a method for joint peak shaving and valley filling optimization operation of multiple types of energy storage in a scenario of energy storage and transportation integration, including: acquiring a basic dataset of the energy storage and transportation integration scenario; acquiring performance difference information of multiple types of energy storage devices based on the basic dataset, and constructing a static operating boundary constraint set based on the basic dataset; inputting the basic dataset into a variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor to perform source-load time series decomposition prediction, generating initial prediction results; inputting the initial prediction results into an improved butterfly optimization algorithm, using the performance difference information as the basis for power allocation and the static operating boundary constraint set as the limit of the search space, performing iterative search to generate initial scheduling decision instructions for multiple types of energy storage devices; and issuing the initial scheduling decision instructions. The system executes the operation of multiple types of energy storage devices, comparing the collected actual execution data with the initial scheduling decision command to obtain an execution deviation signal. The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor adjusts the prediction parameters based on the received execution deviation signal, and then re-decomposes and predicts using the adjusted prediction parameters to generate a corrected prediction result. The improved butterfly optimization algorithm adjusts the search space bounds and optimizes the search direction based on the received execution deviation signal, and then loads the corrected prediction result with the adjusted search space bounds and optimized search direction to perform iterative search and solve again, generating a corrected scheduling decision command. According to a preset rolling cycle, the corrected scheduling decision command is executed as the initial scheduling decision command for the next cycle, and the scheduling decision command for the rolling cycle is output.

[0007] Preferably, the method for optimizing the joint peak shaving and valley filling operation of multiple types of energy storage in the scenario of energy storage integration further includes: identifying the type identifiers of the multiple types of energy storage devices in the basic dataset; classifying the multiple types of energy storage devices into energy-type energy storage devices and power-type energy storage devices according to the type identifiers; extracting capacity decay rate parameters and state-of-charge consistency parameters for the energy-type energy storage devices; extracting power response delay parameters and power ramp-up rate upper limit parameters for the power-type energy storage devices; and outputting the capacity decay rate parameters, state-of-charge consistency parameters, power response delay parameters, and power ramp-up rate upper limit parameters together as the performance difference information.

[0008] Preferably, the method for optimizing the joint peak shaving and valley filling operation of multiple types of energy storage in the scenario of energy storage integration further includes: extracting the rated capacity, rated power, upper limit of allowable state of charge (SOC), and lower limit of allowable SOC from the basic dataset; generating SOC boundary constraints based on the rated capacity, the upper limit of allowable SOC, and the lower limit of allowable SOC; generating charge / discharge power boundary constraints based on the rated power; generating energy balance constraints based on the rated capacity to ensure that the SOC values ​​of the multiple types of energy storage devices at the start and end times are equal; and outputting the SOC boundary constraints, the charge / discharge power boundary constraints, and the energy balance constraints as the set of static operating boundary constraints.

[0009] Preferably, the method for optimizing the joint peak shaving and valley filling operation of multiple types of energy storage in the scenario of energy storage integration further includes: using historical source-load time series data in the basic dataset as training samples to perform offline pre-training on the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor, so that the differential component, harmonic component and nonlinear autoregressive component in the dynamic weight matrix of the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor obtain initial weight values; the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor performs the first decomposition prediction with the initial weight values.

[0010] Preferably, the multi-type energy storage joint peak shaving and valley filling optimization operation method in the energy storage integration scenario further includes: the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor extracts source-load time-series data from the basic dataset, and inputs differential components, harmonic components, and nonlinear autoregressive components respectively; the differential component performs a difference operation on the source-load time-series data to obtain transient change trend characteristics; the harmonic component performs Fourier decomposition on the source-load time-series data to obtain periodic fluctuation characteristics; the nonlinear autoregressive component performs autoregressive fitting on the source-load time-series data to obtain long-term dependency characteristics; the weight values ​​of the dynamic weight matrix corresponding to the differential component, the harmonic component, and the nonlinear autoregressive component are read; the transient change trend characteristics, the periodic fluctuation characteristics, and the long-term dependency characteristics are weighted and superimposed according to their corresponding weight values ​​to obtain the initial prediction result.

[0011] Preferably, the multi-type energy storage joint peak shaving and valley filling optimization operation method in the energy storage integration scenario further includes: delineating a power boundary line on the coordinate axis with charge and discharge power as the search space according to the upper and lower power limits in the charge and discharge power boundary constraints; delineating a charge state boundary plane on an auxiliary dimension with charge state as the search space according to the upper and lower charge state limits in the charge state boundary constraints; delineating an equality boundary plane connecting the starting point and the ending point in the search space according to the condition that the initial charge state and the ending charge state are equal in the energy balance constraints; and using the area enclosed by the power boundary line, the charge state boundary plane, and the equality boundary plane as the search space bound when the improved butterfly optimization algorithm performs iterative search.

[0012] Preferably, the method for optimizing the joint peak shaving and valley filling operation of multiple types of energy storage in the scenario of energy storage integration further includes: determining the priority order of the energy storage devices undertaking long-term charging and discharging tasks according to the capacity decay rate parameter, and forcing the energy storage devices of the same type to keep their state of charge values ​​consistent in the same period according to the state of charge change consistency parameter; allocating the task of smoothing instantaneous power fluctuations to the power storage device with the lowest delay according to the power response delay parameter, and limiting the power change of the power storage device per unit time according to the power ramp rate upper limit parameter.

[0013] Preferably, the method for optimizing the joint peak shaving and valley filling operation of multiple types of energy storage in the scenario of energy storage integration further includes: the variable coefficient differential-harmonic-nonlinear autoregressive coupling predictor converts the received execution deviation signal into a weight correction amount; reads the first weight value of the differential component, the second weight value of the harmonic component, and the third weight value of the nonlinear autoregressive component in the dynamic weight matrix of the variable coefficient differential-harmonic-nonlinear autoregressive coupling predictor; adds the weight correction amount to the first weight value to obtain the updated differential component weight value; subtracts the weight correction amount from the third weight value to obtain the updated nonlinear autoregressive component weight value; keeps the second weight value unchanged; and constructs the updated dynamic weight matrix with the updated differential component weight value, the updated nonlinear autoregressive component weight value, and the unchanged harmonic component weight value.

[0014] Preferably, the multi-type energy storage joint peak shaving and valley filling optimization operation method in the energy storage integration scenario further includes: the improved butterfly optimization algorithm determining the sign of the received execution deviation signal; when the execution deviation signal is negative, it indicates that the actual execution power is lower than the planned power, and the improved butterfly optimization algorithm shifts the lower limit of the power boundary line in the current search space downwards; when the execution deviation signal is positive, it indicates that the actual execution power is higher than the planned power, and the improved butterfly optimization algorithm shifts the upper limit of the power boundary line in the current search space upwards; the improved butterfly optimization algorithm simultaneously performs time integration processing on the execution deviation signal to obtain the deviation accumulation amount; when the deviation accumulation amount exceeds a preset accumulation threshold, the improved butterfly optimization algorithm increases the weight value of the peak shaving and valley filling effect index in the optimization target, and decreases the weight value of the energy storage lifetime loss index.

[0015] Preferably, the method for optimizing the operation of multiple types of energy storage in the scenario of energy exchange and storage integration further includes: the preset rolling period is a rolling time domain with the current time as the starting point and the end time of the day as the ending point.

[0016] The technical solution provided in this application has at least the following technical effects or advantages: by achieving the technical goal of differentiated joint scheduling of multiple types of energy storage with a clear division of labor, enabling energy-type energy storage devices to undertake long-term charging and discharging tasks in the power range with the lowest capacity decay rate and power-type energy storage devices to undertake instantaneous fluctuation smoothing tasks in the power range with the shortest response delay, the technical effect of improving the overall peak shaving and valley filling regulation effect in the energy storage integration scenario is achieved, while extending the overall service life of multiple types of energy storage devices and reducing the comprehensive cost of the entire life cycle of the energy storage system.

[0017] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1This is a flowchart illustrating the optimized operation method for joint peak shaving and valley filling of multiple types of energy storage in the scenario of energy storage and energy exchange integration in this application.

[0020] Figure 2 This is a flowchart illustrating the process of obtaining performance difference information in the method for optimizing the operation of multiple types of energy storage in the scenario of energy storage and energy exchange integration. Detailed Implementation

[0021] This application provides a method for optimizing the joint peak shaving and valley filling operation of multiple types of energy storage in a converged energy storage and transportation scenario. It addresses the technical problem in existing technologies where the lack of a method to differentiate between energy-type and power-type energy storage and accordingly perform differentiated joint scheduling prevents different types of energy storage from allocating power according to their optimal operating ranges during peak shaving and valley filling. This further impacts the overall regulation effect and the comprehensive life-cycle benefits of energy storage devices in converged energy storage and transportation scenarios. The method achieves a clear division of labor in the differentiated joint scheduling of multiple types of energy storage, enabling energy-type energy storage devices to undertake long-term charging and discharging tasks within the power range with the lowest capacity decay rate, and enabling power-type energy storage devices to undertake instantaneous fluctuation smoothing tasks within the power range with the shortest response delay. This results in improved overall peak shaving and valley filling regulation in converged energy storage and transportation scenarios, while extending the overall lifespan of multiple types of energy storage devices and reducing the comprehensive life-cycle cost of the energy storage system.

[0022] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0023] Please see Figure 1 , Figure 2 This application provides a method for optimizing the operation of multiple types of energy storage in a scenario of energy storage and energy exchange integration, which specifically includes the following steps: S1: Obtain the basic dataset for the exchange and storage integration scenario.

[0024] Specifically, the integrated transportation and energy storage scenario refers to a comprehensive energy application environment formed by the deep integration of transportation infrastructure and power energy storage systems at the physical and informational levels. In this environment, traffic load, new energy power generation equipment, and various types of energy storage devices together constitute a dispatchable and coordinated local power network. The basic dataset refers to the sum of multi-source raw data collected in advance to support subsequent prediction, modeling, and optimization decisions. The basic dataset includes at least historical traffic load data, new energy output forecast data, peak-valley electricity price information, and rated parameter data of various types of energy storage devices. Among them, the historical traffic load data is used to reflect the pattern of electricity demand changes over time in the integrated transportation and energy storage scenario; the new energy output forecast data is used to characterize the estimated power generation of distributed power sources such as photovoltaic or wind power in future periods; the peak-valley electricity price information is used to provide a quantitative basis for the differences in grid electricity purchase costs at different times; and the rated parameter data of various types of energy storage devices is used to describe the inherent characteristics of various energy storage devices in terms of capacity, power, and efficiency.

[0025] S2: Based on the basic dataset, obtain performance difference information of multiple types of energy storage devices, and construct a set of static operating boundary constraints based on the basic dataset.

[0026] Furthermore, this application also includes: identifying the type identifiers of the multiple types of energy storage devices in the basic dataset; classifying the multiple types of energy storage devices into energy-type energy storage devices and power-type energy storage devices according to the type identifiers; extracting capacity decay rate parameters and state-of-charge consistency parameters for the energy-type energy storage devices; extracting power response delay parameters and power ramp-up rate upper limit parameters for the power-type energy storage devices; and outputting the capacity decay rate parameters, state-of-charge consistency parameters, power response delay parameters, and power ramp-up rate upper limit parameters together as the performance difference information.

[0027] Furthermore, this application also includes: extracting the rated capacity, rated power, upper limit of allowable state of charge (SOC), and lower limit of allowable SOC from the basic dataset for multiple types of energy storage devices; generating SOC boundary constraints based on the rated capacity, the upper limit of allowable SOC, and the lower limit of allowable SOC; generating charge / discharge power boundary constraints based on the rated power; generating energy balance constraints based on the rated capacity to ensure that the SOC values ​​of the multiple types of energy storage devices at the start and end times are equal; and outputting the SOC boundary constraints, the charge / discharge power boundary constraints, and the energy balance constraints as the set of static operating boundary constraints.

[0028] Specifically, the type identifier for multi-type energy storage devices refers to a pre-assigned code or mark for each type of energy storage device to distinguish its technical category. This type identifier records the category information of the energy storage device in terms of chemical composition, working mechanism, and power output characteristics. Energy-type energy storage devices are characterized by high energy density and large capacity, suitable for long-term power regulation, while power-type energy storage devices are characterized by high power density and fast response, suitable for short-term power compensation. By parsing and classifying the type identifiers, multi-type energy storage devices are divided into energy-type and power-type categories. This allows the subsequent parameter extraction process to obtain physically meaningful differentiated parameters for different device types, thus avoiding the use of uniform parameter descriptions for all types of energy storage devices, which would mask their inherent physical characteristic differences.

[0029] The capacity decay rate parameter is a physical quantity used to describe how quickly the available capacity of an energy storage device gradually decreases with the increase of the number of charge-discharge cycles. The state of charge consistency parameter is a constraint indicator used to limit the state of charge values ​​of the same type of energy storage devices to remain synchronized within the same operating period. By extracting these two parameters for energy storage devices, the capacity degradation behavior of energy storage during long-term operation can be quantified and the state of charge balance among multiple devices of the same type can be forced. This avoids premature aging of some devices due to overcharging and discharging and the difficulty of coordinated scheduling caused by inconsistent state of charge among multiple devices.

[0030] The power response delay parameter is a physical quantity used to describe the time lag between receiving a power command and actually starting to output power in a power-type energy storage device. The power ramp rate upper limit parameter is a boundary value used to limit the maximum change in output power of a power-type energy storage device per unit time. By extracting these two parameters for a power-type energy storage device, the response speed and regulation capability of the power-type energy storage in the instantaneous power regulation process can be quantified, thereby avoiding the power-type energy storage device from being unable to effectively track load fluctuations or causing system oscillations due to excessively long response delays or excessively rapid power changes.

[0031] By using the capacity decay rate parameter, state of charge change consistency parameter, power response delay parameter, and power ramp rate upper limit parameter as the output of performance difference information, the subsequent iterative search and solution steps can directly call the four parameters as the calculation basis for power allocation. At the same time, based on the called parameters, different scheduling priorities and regulation responsibilities are assigned to different types of energy storage devices, thereby realizing the division of labor and collaboration of multiple types of energy storage on the time scale and power scale.

[0032] Rated capacity refers to the maximum electrical energy that various types of energy storage devices can store under standard operating conditions. Rated power refers to the maximum electrical power that various types of energy storage devices can output per unit time. The upper limit and lower limit of the state of charge are the highest and lowest boundary values ​​that various types of energy storage devices are allowed to reach in the state of charge within the safe operating range, respectively. By extracting four parameters from the basic dataset, the various boundary constraints generated subsequently can be precisely limited based on the inherent physical limits of each energy storage device, thereby avoiding the use of uniform empirical values ​​to impose unreasonable identical constraints on devices with different capacities.

[0033] State of charge (SCC) boundary constraints refer to the range of values ​​imposed on the SCC values ​​of various types of energy storage devices during different operating periods. The upper limit of this range is the allowable upper limit of SCC, and the lower limit is the allowable lower limit of SCC. By generating SCC boundary constraints based on rated capacity, allowable upper limit of SCC, and allowable lower limit of SCC, the SCC of each energy storage device during operation is forcibly limited to a safe range, thereby avoiding safety risks or accelerated lifespan degradation caused by overcharging or over-discharging of energy storage devices.

[0034] Charge and discharge power boundary constraints refer to the amplitude limits imposed on the charge and discharge power values ​​of various types of energy storage devices at different times. The upper limit of the amplitude limit is the positive value of the rated power, and the lower limit is the negative value of the rated power. By generating charge and discharge power boundary constraints based on the rated power, the power output or power absorption of each energy storage device per unit time is forcibly limited to the power handling capacity of the device itself, thereby avoiding damage to the energy storage device or system instability due to power exceeding the limit.

[0035] Energy balance constraints refer to the equality restrictions imposed on the state of charge (SOC) values ​​of various types of energy storage devices at the beginning and end of their operating cycles. By generating these equality constraints based on the rated capacity, the total amount of electricity charged and released by the energy storage devices during the entire operating cycle is kept in macroscopic balance, thereby avoiding the deviation of residual accumulated electricity after the end of each day's operation, which would affect the accuracy of the next day's scheduling plan.

[0036] By outputting the set of static operating boundary constraints, including the state of charge boundary constraints, the charge and discharge power boundary constraints, and the energy balance constraints, the subsequent iterative search and solution steps can simultaneously load all three types of constraints into the search space of the improved butterfly optimization algorithm. The improved butterfly optimization algorithm then uses this set of static operating boundary constraints as the bound of the search space to solve the problem, thereby ensuring that all generated scheduling decision instructions do not violate any type of safety boundary or power boundary.

[0037] S3: Input the basic dataset into the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor to perform source load time series decomposition prediction and generate initial prediction results.

[0038] Furthermore, this application also includes: using historical source load time-series data in the basic dataset as training samples to perform offline pre-training on the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor, so that the differential component, harmonic component and nonlinear autoregressive component in the dynamic weight matrix of the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor obtain initial weight values; the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor performs the first decomposition prediction with the initial weight values.

[0039] Furthermore, this application also includes: the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor extracts source load time series data from the basic dataset, and inputs differential components, harmonic components, and nonlinear autoregressive components respectively; the differential component performs a difference operation on the source load time series data to obtain transient change trend characteristics; the harmonic component performs Fourier decomposition on the source load time series data to obtain periodic fluctuation characteristics; the nonlinear autoregressive component performs autoregressive fitting on the source load time series data to obtain long-term dependency characteristics; the weight values ​​of the dynamic weight matrix corresponding to the differential component, the harmonic component, and the nonlinear autoregressive component are read; the transient change trend characteristics, the periodic fluctuation characteristics, and the long-term dependency characteristics are weighted and superimposed according to their corresponding weight values ​​to obtain the initial prediction result.

[0040] Specifically, historical source-load time-series data refers to the serialized values ​​of traffic load power and new energy power generation recorded in the basic dataset at continuous historical time points. These serialized values ​​are sampled and arranged at equal time intervals and cover at least one complete operating cycle. Using these serialized values ​​as training samples allows the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor to obtain an initial predictive capability obtained by fitting historical data before actual online prediction. The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor is a prediction model with a multi-component fusion structure. The differential component is used to perform difference operations on the input sequence to extract instantaneous change trends, the harmonic component is used to perform Fourier series expansion on the input sequence to extract periodic oscillation modes, and the nonlinear autoregressive component is used to perform nonlinear autoregressive fitting with external input on the input sequence to extract long-term dependencies. Each of the three components is assigned a weight coefficient, and the three weight coefficients together constitute a dynamic weight matrix. This dynamic weight matrix determines the contribution ratio of the three components in the final prediction result.

[0041] Offline pre-training refers to the process of using historical source-load time-series data as training samples and adjusting the weight coefficients of each component in the dynamic weight matrix through optimization algorithms without relying on real-time feedback data. This process allows the weight coefficients of the three components to converge from initial random values ​​to a set of values ​​that minimize the prediction error. This set of converged values ​​is used as the initial weight values, thereby ensuring that the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor has prediction parameters validated by historical data when first encountering real-time source-load time-series data, avoiding low prediction accuracy in the first prediction due to a lack of prior knowledge.

[0042] The initial weight values ​​include the initial weight values ​​for the differential component, the harmonic component, and the nonlinear autoregressive component. The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor uses these three initial weight values ​​to form a dynamic weight matrix in the initial state, and performs the first decomposition prediction on the real-time input basic dataset based on this dynamic weight matrix. The initial prediction result output by the first decomposition prediction serves as the input for subsequent iterative search solutions, and the output eigenvalues ​​of each component in the first decomposition prediction are also recorded as a reference benchmark for subsequent bias correction.

[0043] In subsequent steps, when the execution deviation signal is fed back to the variable-coefficient differential-harmonic-nonlinear autoregressive coupled predictor, the predictor, starting from the initial weight values, gradually adjusts the weight values ​​of the differential and nonlinear autoregressive components according to the direction and amplitude of the execution deviation signal, while the weight values ​​of the harmonic components remain unchanged based on the initial weight values ​​obtained from offline pre-training. This combined mechanism of offline pre-training and online weight correction enables the prediction parameters to possess both historical statistical regularity and real-time deviation correction adaptability.

[0044] Source-load time-series data refers to the ordered arrangement of traffic load power values ​​and new energy power generation values ​​continuously recorded at equal time intervals in the basic dataset on the time axis. This ordered arrangement constitutes the original input sequence for the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor to perform decomposition prediction. The differential component is the substructure within the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor responsible for extracting the transient change trend of the signal; the harmonic component is the substructure within the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor responsible for extracting the periodic oscillation mode of the signal; and the nonlinear autoregressive component is the substructure within the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor responsible for extracting the long-term dependency relationship of the signal. By simultaneously feeding the source-load time-series data into these three substructures for processing, the same set of input data can be analyzed in parallel from three different physical dimensions: rate of change, periodicity, and historical memory.

[0045] Differential operations are the process by which the differential component calculates the differences between adjacent time points in the source load time series data and forms a difference sequence. Transient trend characteristics refer to the quantitative description obtained through differential operations, which characterizes the instantaneous direction of change of the source load time series data within each local time window, indicating whether it is rising, falling, or remaining flat. This quantitative description exists in the form of a difference sequence and serves as the output of the differential component. Fourier decomposition is a mathematical operation that transforms the source load time series data from a time-domain representation into a superposition of a set of sine and cosine waves of different frequencies. Periodic fluctuation characteristics refer to the quantitative description obtained through Fourier decomposition, which characterizes the recurring oscillation patterns of the source load time series data on daily, weekly, or seasonal scales. This quantitative description exists in the form of the amplitude and phase of each frequency component and serves as the output of the harmonic component. Autoregressive fitting is a parameter estimation operation in which the nonlinear autoregressive component uses the observations of historical time in the source load time series data to predict the value of the current time through a nonlinear functional relationship. Long-term dependency features refer to the quantitative description of the trend components in the source load time series data that maintain significant correlation across multiple time steps, obtained through autoregressive fitting. This quantitative description exists in the form of regression coefficients and error terms and serves as the output of the nonlinear autoregressive component.

[0046] The dynamic weight matrix is ​​a numerical set storing the weights of the differential component, harmonic component, and nonlinear autoregressive component. Each weight value in this set corresponds to the contribution of a component to the final prediction result. Specifically, the differential component weight determines the proportion of transient trend features in the superposition result, the harmonic component weight determines the proportion of periodic fluctuation features, and the nonlinear autoregressive component weight determines the proportion of long-term dependency features. During offline pre-training, the three weight values ​​in the dynamic weight matrix are adjusted to their optimal initial values. Based on this, when the execution bias signal is fed back to the variable-coefficient differential-harmonic-nonlinear autoregressive coupled predictor, the differential and nonlinear autoregressive component weight values ​​are updated online according to the direction and amplitude of the execution bias signal, while the harmonic component weight value remains unchanged at its optimal initial value determined during offline pre-training.

[0047] Weighted superposition refers to the process of multiplying the three outputs—transient trend characteristics, periodic fluctuation characteristics, and long-term dependence characteristics—by their respective weight values ​​in the dynamic weight matrix and then summing them to form a comprehensive prediction result. The initial prediction result refers to the comprehensive prediction value sequence obtained through weighted superposition, which characterizes the expected values ​​of the source load time series data for future periods. This initial prediction result serves as the input data for the first iteration of the improved butterfly optimization algorithm. Simultaneously, the values ​​of the transient trend characteristics, periodic fluctuation characteristics, and long-term dependence characteristics are also recorded to track the contribution of each component to the prediction error during subsequent deviation feedback.

[0048] S4: Input the initial prediction results into the improved butterfly optimization algorithm, use the performance difference information as the basis for power allocation and the static operating boundary constraint set as the limit of the search space, perform iterative search and solve to generate initial scheduling decision instructions for multiple types of energy storage devices.

[0049] Furthermore, this application also includes: determining the priority order of the energy storage devices undertaking long-term charging and discharging tasks according to the capacity decay rate parameter, and forcing the energy storage devices of the same type to maintain consistent state of charge values ​​in the same time period according to the state of charge change consistency parameter; allocating the task of smoothing instantaneous power fluctuations to the power storage device with the lowest delay according to the power response delay parameter, and limiting the power change of the power storage device per unit time according to the power ramp-up rate upper limit parameter.

[0050] Furthermore, this application also includes: defining a power boundary line on the coordinate axis with charge / discharge power as the search space based on the upper and lower power limits in the charge / discharge power boundary constraints; defining a charge state boundary plane on an auxiliary dimension with charge state as the search space based on the upper and lower charge state limits in the charge state boundary constraints; defining an equality boundary plane connecting the start and end points in the search space based on the condition that the initial charge state and the final charge state are equal in the energy balance constraints; and using the area enclosed by the power boundary line, the charge state boundary plane, and the equality boundary plane as the search space bound when the improved butterfly optimization algorithm performs iterative search.

[0051] Specifically, the improved butterfly optimization algorithm is an iterative swarm intelligence search algorithm obtained by improving the standard butterfly optimization algorithm. The standard butterfly optimization algorithm simulates the behavior of individual butterflies in foraging by sensing the concentration of nectar through smell and gradually moving towards the optimal position. The improved butterfly optimization algorithm adds a constraint preservation operator and a discreteness adaptive switching strategy to the standard butterfly optimization algorithm. The constraint preservation operator is used to transform the static running boundary constraint set into the feasible domain geometric boundary in the search space and remove individuals that cross the boundary during the population evolution process. The discreteness adaptive switching strategy is used to automatically switch between global exploration mode and local development mode according to the dispersion of the population in the search space. The initial prediction result is the predicted value sequence of the future source load time series values ​​output by the pre-decomposition prediction step. The initial prediction result is input into the improved butterfly optimization algorithm as the target reference value for iterative search and solution, so that the improved butterfly optimization algorithm takes the net load adjustment demand represented by the predicted value sequence as the optimization direction during the search process. The capacity decay rate parameter is a physical quantity extracted from performance difference information to describe how quickly the available capacity of an energy storage device decreases with increasing charge-discharge cycles. Long-duration charge-discharge tasks refer to tasks that span multiple scheduling periods and are used to smooth out long-term trend fluctuations in the load curve. Priority order refers to the order in which energy storage devices undertake long-duration charge-discharge tasks, determined by arranging them according to their capacity decay rate parameters from smallest to largest. Tasks are allocated by prioritizing devices with smaller capacity decay rate parameters, allowing devices with slower capacity decay to undertake more charge-discharge cycles, thereby extending the overall lifespan of the energy storage device. The overall lifespan of the energy storage system; the state of charge (SCC) consistency parameter is a binding indicator extracted from performance difference information to force the SCC values ​​of the same type of energy storage devices to remain consistent across different time periods. The same type of energy storage devices refers to multiple energy storage devices with the same type identification and all classified as energy storage devices. By using the SCC consistency parameter, the SCC values ​​of multiple energy storage devices are forced to remain consistent at the same time period, so that the remaining power ratio of each energy storage device is exactly the same at any point in time. This avoids some devices from frequently entering overcharge protection state due to excessively high SCC or frequently entering over-discharge protection state due to excessively low SCC.

[0052] Power response delay parameter is a physical quantity extracted from performance difference information to describe the time lag between receiving a power command and actually starting to output power in a power storage device. Instantaneous power fluctuation refers to the random and rapid change in load power or renewable energy generation power within a timescale of seconds to minutes. The smoothing task refers to the adjustment task of offsetting the impact of these instantaneous power fluctuations on the power grid through the rapid charging and discharging of power storage devices. The power storage device with the lowest delay parameter value is the one with the smallest instantaneous power response delay parameter. The task of smoothing instantaneous power fluctuations is assigned to the power storage device with the lowest delay, allowing the device with the fastest response speed to handle the instantaneous power fluctuations. The system adjusts its functions in a timely manner to ensure that the smoothing task can begin execution within the shortest possible time delay after a fluctuation occurs. The power ramp rate upper limit parameter is a boundary value extracted from performance difference information to limit the maximum change in output power of power-type energy storage devices per unit time. The power change refers to the absolute value of the difference in output power values ​​between two adjacent sampling times. By limiting the absolute value of the power ramp rate upper limit parameter, it is ensured that the absolute value cannot exceed the value specified by the power ramp rate upper limit parameter at any unit time, so that the output power of the power-type energy storage device always changes with a smooth slope rather than a step, thereby avoiding the introduction of new high-frequency harmonic interference due to power abrupt changes.

[0053] The charge and discharge power boundary constraints are constraints extracted from the static operating boundary constraint set to limit the upper and lower limits of the charge and discharge power amplitude of various types of energy storage devices. The upper limit of power refers to the maximum discharge power value that the energy storage device is allowed to output in any time period, and the lower limit of power refers to the absolute value of the maximum charging power value that the energy storage device is allowed to absorb in any time period. The search space refers to the multidimensional mathematical space in which all candidate solutions exist in the improved butterfly optimization algorithm. The coordinate axis refers to the dimensional direction in the search space that represents the charge and discharge power value of a certain type of energy storage device. The power boundary line refers to the line segment boundary drawn on the coordinate axis with the upper and lower limit of power as endpoints. By drawing power boundary lines on each power coordinate axis in the search space, the value of each candidate solution on any power coordinate axis in the search space is limited to the range specified by the upper and lower limit of power. State of charge (SCC) boundary constraints are constraints extracted from the static operating boundary constraint set to limit the upper and lower limits of the SCC value range for various types of energy storage devices. The upper SCC value refers to the highest SCC percentage value that the energy storage device is allowed to reach in any time period, and the lower SCC value refers to the lowest SCC percentage value that the energy storage device is allowed to reach in any time period. The auxiliary dimension refers to the additional dimension direction in the search space besides the power coordinate axis used to characterize the SCC value. The SCC boundary plane refers to the planar constraint surface defined on the auxiliary dimension with the upper SCC value and the lower SCC value as the boundary. By defining the SCC boundary plane on each auxiliary dimension of the search space, the value of each candidate solution in the search space on any auxiliary dimension is limited to the range specified by the upper SCC value and the lower SCC value. Energy balance constraints are constraints extracted from the static operating boundary constraint set to ensure that the state of charge (SOC) values ​​of various types of energy storage devices are equal at the start and end times. The starting point is the combination of the SOC and charge / discharge power values ​​of each energy storage device at the start of the operating cycle in the search space. The ending point is the combination of the SOC and charge / discharge power values ​​of each energy storage device at the end of the operating cycle in the search space. The equality boundary plane is a hyperplane constraint surface in the search space that connects the starting point and the ending point and satisfies the condition of equal SOC values. By defining the equality boundary plane in the search space, the SOC values ​​of all candidate solutions in the search space at the end of the operating cycle must be strictly equal to the SOC values ​​at the start time.

[0054] The power boundary line, the state-of-charge boundary plane, and the equality boundary plane together enclose a closed multidimensional geometric region in the search space. This multidimensional geometric region is called the search space bound. Any candidate solution inside the search space bound simultaneously satisfies the charge / discharge power boundary constraints, the state-of-charge boundary constraints, and the energy balance constraints. Any candidate solution outside the search space bound violates at least one of the above three types of constraints. The improved butterfly optimization algorithm strictly restricts the generation and evolution of all candidate solutions within the search space bound during the iterative search process, ensuring that each candidate solution in the population always represents a scheduling scheme that simultaneously satisfies all safety boundaries, power amplitude, and daily energy balance feasibility requirements. Iterative search and solution solving refers to the process by which the improved butterfly optimization algorithm gradually updates the candidate solution population through multiple iterations within the search space bound, making the optimal solution in the population continuously approach the target reference value represented by the initial prediction result. The initial scheduling decision instruction refers to the set of charge / discharge power values ​​of each type of energy storage device at each time period corresponding to the optimal candidate solution output by the improved butterfly optimization algorithm after completing the iterative search and solution solving. The transient trend characteristics, periodic fluctuation characteristics, and long-term dependence characteristics are weighted and superimposed according to their corresponding weight values ​​to obtain the coupled prediction result, i.e., the initial prediction result. To quantitatively evaluate the degree of fit between the coupled prediction result and the measured value, this application uses root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) as evaluation indicators, and performs statistical calculations on the deviations between the measured value and the coupled prediction result, differential components, harmonic components, and nonlinear autoregressive components, respectively. The specific calculation method is as follows: For example, using 96 sampling times (time 0 to time 96) as the statistical sample, the measured power value at the i-th sampling time is denoted as y. i The power value of a certain prediction result is The root mean square error is The mean absolute error is The mean absolute percentage error is Where N=96. Using the power data at 96 sampling times as statistical samples, the error statistics of each prediction result are shown in Table 1 below.

[0055] Table 1: Statistical Comparison of Errors Between Predicted Results and Measured Values

[0056] The calculation results show that the RMSE of the coupled prediction result is 28.6 kW, which is 67.7 kW, 87.1 kW, and 59.8 kW lower than that of the differential component, harmonic component, and nonlinear autoregressive component, respectively. The MAE of the coupled prediction result is 21.4 kW, which is 57.1 kW, 70.7 kW, and 48.8 kW lower than that of the three individual components, respectively. The MAPE of the coupled prediction result is 4.2%, which is 11.6, 14.4, and 9.7 percentage points lower than that of the three individual components, respectively. All three error indices of the coupled prediction result are smaller than those of the individual components, indicating that it has the smallest deviation from the measured values ​​and the highest degree of fit.

[0057] Furthermore, the coupled prediction results also exhibited minimal deviation during typical periods of significant load fluctuation. For example, at time 24, the measured power was 782.5 kW, while the coupled prediction result was 765.3 kW, with a deviation of 17.2 kW. The predicted results for the differential component, harmonic component, and nonlinear autoregressive component at this time were 712.4 kW, 685.7 kW, and 738.9 kW, respectively, with deviations of 70.1 kW, 96.8 kW, and 43.6 kW. At time 48, the measured power was 635.8 kW, while the coupled prediction result was 621.4 kW, with a deviation of 14.4 kW. The deviations for the differential component, harmonic component, and nonlinear autoregressive component were 58.2 kW, 73.5 kW, and 41.7 kW, respectively. At time 72, the measured power was 318.6 kW, while the coupled prediction result was 305.2 kW, with a deviation of 13.4 kW. The deviations for the differential component, harmonic component, and nonlinear autoregressive component were 52.8 kW, 68.9 kW, and 37.5 kW, respectively. The deviation comparison at these typical times further demonstrates that the coupled prediction result obtained by weighting and superimposing transient trend characteristics, periodic fluctuation characteristics, and long-term dependence characteristics according to their corresponding weight values ​​can most accurately track the changing trend of the measured value, exhibiting the highest degree of fit.

[0058] The initial scheduling decision command serves as input data for subsequent execution and comparison steps. Meanwhile, the power allocation schemes for various types of energy storage devices included in the initial scheduling decision command directly depend on the priority order and task allocation principles determined by the four parameters in the performance difference information. The feasibility of each power value in the initial scheduling decision command directly depends on the constraint effect of the search space limit transformed by the static operating boundary constraint set.

[0059] S5: Issue the initial scheduling decision command to the multi-type energy storage devices for execution, compare the collected actual execution data with the initial scheduling decision command, and obtain the execution deviation signal.

[0060] Specifically, the initial scheduling decision command refers to the set of charging and discharging power values ​​of various types of energy storage devices at each time period corresponding to the optimal candidate solution output by the improved butterfly optimization algorithm after iterative search and solution. Multi-type energy storage devices refer to the general term for energy storage devices including energy-type energy storage devices and power-type energy storage devices. Sending the initial scheduling decision command to multi-type energy storage devices means sending the charging and discharging power values ​​of various types of energy storage devices at each time period to the corresponding energy storage device controller in the form of control commands. The energy storage device controller then drives the energy storage device to perform actual charging or discharging operations according to the received power values.

[0061] Subsequently, during the actual charging and discharging operation of the energy storage devices according to the initial scheduling decision instructions, voltage and current sensors deployed on the DC and AC sides of various types of energy storage devices continuously collect the terminal voltage values ​​and current values ​​flowing through the energy storage devices according to a preset sampling period. The terminal voltage values ​​and current values ​​are multiplied to obtain the real-time power values. The real-time power values ​​are then processed according to the same time period division method as the initial scheduling decision instructions to obtain the actual execution data. The actual execution data refers to the set of charging and discharging power values ​​actually output or absorbed by various types of energy storage devices in each time period, obtained directly by the sensors. The numerical difference between the actual execution data and the initial scheduling decision instructions comes from the deviation between the actual new energy output and the predicted value caused by weather changes, the deviation between the actual load and the predicted load caused by sudden changes in traffic flow, and the execution deviation introduced by the response delay and adjustment error of the energy storage devices themselves.

[0062] Subsequently, the comparison refers to the operation of calculating the difference between the power value in the actual execution data within the same time period and the planned power value corresponding to the initial scheduling decision instruction. Specifically, the planned charging and discharging power values ​​of each type of energy storage device in each time period are read from the initial scheduling decision instruction, and the actual charging and discharging power values ​​of the same type of energy storage device in the same time period are read from the actual execution data. The actual charging and discharging power value is subtracted from the planned charging and discharging power value, and the resulting difference is the instantaneous deviation value of that type of energy storage device in that time period. The execution deviation signal refers to the deviation sequence composed of the instantaneous deviation values ​​of various types of energy storage devices in each time period. In this deviation sequence, a positive value indicates that the actual execution power is higher than the planned power, a negative value indicates that the actual execution power is lower than the planned power, and a zero value indicates that the actual execution power is completely consistent with the planned power.

[0063] S6: The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor adjusts the prediction parameters according to the received execution deviation signal, and re-decomposes and predicts with the adjusted prediction parameters to generate a corrected prediction result.

[0064] Furthermore, this application also includes: the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor converts the received execution deviation signal into a weight correction amount; reads the first weight value of the differential component, the second weight value of the harmonic component, and the third weight value of the nonlinear autoregressive component in the dynamic weight matrix of the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor; adds the weight correction amount to the first weight value to obtain the updated differential component weight value; subtracts the weight correction amount from the third weight value to obtain the updated nonlinear autoregressive component weight value; keeps the second weight value unchanged; and constructs the updated dynamic weight matrix with the updated differential component weight value, the updated nonlinear autoregressive component weight value, and the unchanged harmonic component weight value.

[0065] Specifically, after obtaining the execution bias signal, the execution bias signal is simultaneously distributed to two receivers: the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor and the improved butterfly optimization algorithm.

[0066] The execution deviation signal contains three types of sign information: positive, negative, and zero, as well as the magnitude information corresponding to each sign. Converting the execution deviation signal into a weight correction value means transforming the power deviation value sequence into a scalar value after proportional operation. The sign of this scalar value is consistent with the sign of the execution deviation signal, and its magnitude is proportional to the magnitude of the execution deviation signal. The resulting weight correction value is used for the directional adjustment of the weight values ​​of each component in the subsequent dynamic weight matrix.

[0067] In the dynamic weight matrix, the first weight value refers to the weight coefficient corresponding to the differential component, the second weight value refers to the weight coefficient corresponding to the harmonic component, and the third weight value refers to the weight coefficient corresponding to the nonlinear autoregressive component. After converting the execution deviation signal into a weight correction value, the current values ​​of the first, second, and third weight values ​​are read from the dynamic weight matrix respectively. This allows subsequent weight update operations to perform independent addition and subtraction operations based on the current state of the three weight values, thereby avoiding an imbalance in the weight ratio between the three components due to the use of a uniform change value.

[0068] The first weight value is supplemented with a weight correction to obtain the updated differential component weight value. This increases the contribution of the differential component when the execution bias signal is positive and decreases when the execution bias signal is negative. This allows transient trend features to have a greater weight in subsequent predictions to respond to rising real-time power trends, or a smaller weight to suppress over-response to falling trends. The third weight value is subtracted from the weight correction to obtain the updated nonlinear autoregressive component weight value. This decreases the contribution of the nonlinear autoregressive component when the execution bias signal is positive and increases when the execution bias signal is negative. This ensures that the direction of change of long-term dependency features in subsequent predictions is opposite to the direction of change of the differential component. Through this complementary weight adjustment of addition and subtraction, the prediction model achieves adaptive tracking of real-time biases while maintaining a relatively stable total weight.

[0069] Maintaining the second weight value unchanged means that no addition or subtraction is performed on the weight coefficients of the harmonic components throughout the entire online weight update process. This ensures that the contribution ratio of periodic fluctuation characteristics in subsequent predictions remains at the optimal value determined by offline pre-training, thereby avoiding the destruction of the harmonic components' ability to extract stable periodic patterns from the source load time series data due to random fluctuations in real-time bias. The updated dynamic weight matrix is ​​composed of the updated differential component weight value, the updated nonlinear autoregressive component weight value, and the unchanged harmonic component weight value. In this updated dynamic weight matrix, the weight values ​​of each of the three components have been adjusted. The weight values ​​of the differential component and the nonlinear autoregressive component have been increased or decreased in opposite directions according to the sign and amplitude of the execution bias signal, while the weight value of the harmonic component remains unchanged at the optimal value obtained from offline pre-training.

[0070] The prediction parameters refer to all the weight values ​​in the dynamic weight matrix used by the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor during decomposition prediction. The adjusted prediction parameters are the combination of the weight values ​​of the three components contained in the updated dynamic weight matrix. Re-decomposing and predicting with the adjusted prediction parameters means that the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor uses the updated dynamic weight matrix to perform differential component decomposition, harmonic component decomposition, and nonlinear autoregressive component decomposition on the source load time series data again, and re-weights and superimposes the eigenvalues ​​output by the three components according to the updated weight values, thereby obtaining the corrected prediction result. The corrected prediction result serves as the prediction boundary condition loaded when the improved butterfly optimization algorithm iteratively searches and solves the problem again. This ensures that the next prediction output of the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor gradually approaches the actual power change trend under the guidance of the execution deviation signal, while the stable extraction capability of the harmonic components for periodic patterns is not affected by the random fluctuations of real-time deviations.

[0071] S7: The improved butterfly optimization algorithm adjusts the search space bounds and optimizes the search direction based on the received execution deviation signal. With the adjusted search space bounds and optimized search direction, it loads the corrected prediction results and performs iterative search again to generate a corrected scheduling decision instruction.

[0072] Furthermore, this application also includes: the improved butterfly optimization algorithm determining the sign of the received execution deviation signal; when the execution deviation signal is negative, it indicates that the actual execution power is lower than the planned power, and the improved butterfly optimization algorithm shifts the lower limit of the power boundary line in the current search space downwards; when the execution deviation signal is positive, it indicates that the actual execution power is higher than the planned power, and the improved butterfly optimization algorithm shifts the upper limit of the power boundary line in the current search space upwards; the improved butterfly optimization algorithm simultaneously performs time integration processing on the execution deviation signal to obtain the deviation accumulation amount; when the deviation accumulation amount exceeds a preset accumulation threshold, the improved butterfly optimization algorithm increases the weight value of the peak shaving and valley filling effect index in the optimization objective, and decreases the weight value of the energy storage lifetime loss index.

[0073] Specifically, the execution deviation signal is obtained by comparing the actual execution data with the power deviation value sequence of each time period obtained from the initial scheduling decision instruction. Determining the positive or negative sign of the execution deviation signal means identifying the sign attribute of the power deviation value of each time period. When the power deviation value is greater than zero, it is determined to be a positive value, and when the power deviation value is less than zero, it is determined to be a negative value. Execution deviation signals with different sign attributes will trigger different subsequent boundary adjustment operations.

[0074] When the execution deviation signal is determined to be negative, the physical meaning of the negative value is that the actual executed power is lower than the planned power. This indicates that the actual output power of the energy storage device does not meet the command requirements, and the power selection range needs to be expanded in subsequent scheduling to compensate for the power shortage. The power boundary line of the current search space refers to the line segment boundary defined by the upper and lower power limits in the search space. The lower power limit is the endpoint value representing the maximum allowable amplitude of charging power in the line segment boundary. Shifting the lower power limit means moving the endpoint value in the negative direction to expand the allowable amplitude range of charging power. By shifting the lower limit, a larger charging power space is provided for the energy storage device in subsequent periods to make up for the shortage accumulated due to insufficient actual power in the current period.

[0075] When the execution deviation signal is determined to be positive, the physical meaning of a positive value is that the actual executed power is higher than the planned power. This indicates that the actual output power of the energy storage device exceeds the command requirements, and the selectable power range needs to be compressed in subsequent scheduling to avoid power excess. The power boundary line of the current search space refers to the line segment boundary defined by the upper and lower power limits in the search space. The upper power limit is the endpoint value representing the maximum allowable amplitude of the discharge power in the line segment boundary. Moving the upper power limit upward means moving the endpoint value in the positive direction to expand the allowable amplitude range of the discharge power. By moving the upper power limit upward, a larger discharge power space is provided for the energy storage device in subsequent periods to release the excess power output in the current period due to the excessive actual power.

[0076] The improved butterfly optimization algorithm performs time integration processing on the execution deviation signal by multiplying the power deviation value of each time period in the execution deviation signal by the time length of the corresponding time period and then accumulating them one by one. The accumulated deviation amount is obtained through the above accumulation operation. The accumulated deviation amount is a value used to characterize the cumulative effect of the execution deviation in the time dimension. The positive or negative direction of this value reflects the overall trend of power shortage or power surplus, and the magnitude of this value reflects the severity of power shortage or power surplus.

[0077] The preset accumulation threshold is a pre-defined numerical boundary used to determine whether the accumulated deviation has reached a level requiring adjustment of the optimization target weight. When the absolute value of the accumulated deviation exceeds the preset threshold, it indicates that the cumulative effect of power deviation in each time period has reached a level that cannot be ignored. The optimization objective of the improved butterfly optimization algorithm is a comprehensive evaluation function composed of peak shaving and valley filling effect indicators and energy storage life loss indicators, each with its own weight value. The weight value of the peak shaving and valley filling effect indicator determines the importance of load peak reduction and valley filling in the optimization objective, while the weight value of the energy storage life loss indicator determines the importance of charging and discharging operations on the lifespan consumption of energy storage equipment in the optimization objective. When the accumulated deviation exceeds the preset accumulation threshold, increasing the weight value of the peak shaving and valley filling effect indicator means increasing the value of this weight value to give the peak shaving and valley filling effect a larger proportion in the comprehensive evaluation function. At the same time, decreasing the weight value of the energy storage life loss indicator means decreasing the value of this weight value to give the energy storage life loss a smaller proportion in the comprehensive evaluation function. Through the reverse adjustment of the above weight values, the improved butterfly optimization algorithm prioritizes the smoothing of the load curve rather than prioritizing the lifespan protection of energy storage equipment in subsequent iterative searches.

[0078] The search space bound refers to the closed multidimensional geometric region enclosed by the power boundary line, the charge state boundary plane, and the equation boundary plane in the search space. The optimized search direction refers to the population evolution direction followed by the improved butterfly optimization algorithm as it iteratively approaches the optimal candidate solution. The corrected prediction result is a new sequence of predicted values ​​obtained by the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor after re-decomposing and predicting using the updated dynamic weight matrix. The improved butterfly optimization algorithm redefines the power boundary line based on the adjusted lower or upper power limit value according to the sign of the execution deviation signal, and updates the weight coefficients in the optimization objective by adjusting the weight values ​​of the peak shaving and valley filling effect index and the energy storage lifetime loss index according to the accumulated deviation. Within the new search space bound formed by the redefined power boundary line, the evolution direction determined by the updated weight coefficients is used as the optimized search direction. The corrected prediction result is loaded as the target reference value, and the iterative search is re-executed to solve the problem, finally outputting the corrected scheduling decision instruction.

[0079] The revised scheduling decision instruction refers to the set of updated charging and discharging power values ​​for various types of energy storage devices in each time period, corresponding to the optimal candidate solution output by the improved butterfly optimization algorithm after completing the re-iterative search. This revised scheduling decision instruction replaces the initial scheduling decision instruction and becomes the actual control command issued to the energy storage devices in the current rolling cycle. At the same time, the charging and discharging power values ​​in this revised scheduling decision instruction have been ensured to meet all static operating boundary constraints through the adjusted search space bounds, and the adjusted optimization target weight coefficients have rebalanced the peak shaving and valley filling effect and energy storage life loss according to the current demand indicated by the accumulated deviation.

[0080] S8: According to the preset rolling cycle, the modified scheduling decision instruction is executed as the initial scheduling decision instruction for the next cycle, and the scheduling decision instruction for the rolling cycle is output.

[0081] Furthermore, this application also includes: the preset rolling period is a rolling time domain that starts at the current time and ends at the end of the day.

[0082] Specifically, the preset rolling cycle refers to a pre-defined time window of fixed length. When this time window ends, the generation and execution of the next round of scheduling decision instructions are automatically initiated, and this cycle repeats until the end of the running cycle. The current time refers to the real-time point when the current round of rolling optimization operation is performed, and the end time of the day refers to the time point at midnight on the day covered by the running cycle. The rolling time domain refers to a variable-length time interval with the current time as the starting boundary and the end time of the day as the ending boundary. As the current time moves forward, the starting boundary of the rolling time domain continues to move forward while the ending boundary remains fixed, so that the time interval covered by each rolling optimization gradually shortens until the current time reaches the end time of the day, at which point the rolling process terminates.

[0083] The physical meaning of executing according to the preset rolling cycle is that the duration of each rolling time domain is equal to the value of the preset rolling cycle. Whenever the previous preset rolling cycle ends, the current time is updated to the end time of the previous preset rolling cycle. This updated current time is used as the new starting boundary, and the end time of the current day is used as the unchanged ending boundary to reconstruct a new rolling time domain. The corrected scheduling decision instruction is the latest set of charging and discharging power values ​​generated by the improved butterfly optimization algorithm after adjusting the search space bounds and optimization direction based on the execution deviation signal and iteratively searching and solving. Executing the corrected scheduling decision instruction as the initial scheduling decision instruction for the next cycle means directly assigning the corrected scheduling decision instruction generated in the current rolling cycle as the starting input for the next rolling cycle. This ensures that the iterative search and solution process of the next rolling cycle does not start from zero but inherits the optimization results of the current cycle as the initial population, thus guaranteeing the temporal continuity and smoothness of the scheduling instructions between two adjacent rolling cycles.

[0084] The initial scheduling decision command serves as the starting input for each rolling cycle. In the first rolling cycle, it is generated by the initial prediction result through the first iterative search. In the second and subsequent rolling cycles, it is directly obtained by assigning values ​​from the revised scheduling decision command of the previous rolling cycle. Through this inheritance mechanism, the charging and discharging power values ​​in the revised scheduling decision command serve as the initial population center for the iterative search of the next rolling cycle, avoiding the waste of computational resources and instruction jumps caused by starting the search from scratch in each rolling cycle. The output rolling cycle scheduling decision command refers to recording the revised scheduling decision command generated at the end of each rolling cycle in chronological order and using it as the final control command sequence issued to various types of energy storage devices. As the rolling time domain gradually advances, the above control command sequence gradually covers the entire time period from the start of the first rolling cycle to the end of the day, forming a complete all-day scheduling scheme.

[0085] In summary, the method for joint peak shaving and valley filling optimization operation of multiple types of energy storage in the scenario of energy storage and energy exchange integration provided in this application has the following technical effects: by achieving the technical goal of differentiated joint scheduling of multiple types of energy storage with clear division of labor, enabling energy-type energy storage devices to undertake long-term charging and discharging tasks in the power range with the smallest capacity decay rate and power-type energy storage devices to undertake instantaneous fluctuation smoothing tasks in the power range with the shortest response delay, the method achieves the technical effects of improving the overall peak shaving and valley filling regulation effect in the scenario of energy storage and energy exchange integration, while extending the overall service life of multiple types of energy storage devices and reducing the comprehensive cost of the entire life cycle of the energy storage system.

[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0087] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for optimizing the operation of multiple types of energy storage in a scenario of integrated energy storage and transportation, characterized in that: include: Obtain the basic dataset for the integration of storage and exchange scenarios; Based on the aforementioned basic dataset, performance difference information of various types of energy storage devices is obtained, and a set of static operating boundary constraints is constructed based on the aforementioned basic dataset. The basic dataset is input into the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor to perform source load time series decomposition prediction and generate initial prediction results. The initial prediction results are input into the improved butterfly optimization algorithm. The performance difference information is used as the basis for power allocation, and the static operating boundary constraint set is used as the limit of the search space. The algorithm performs iterative search and solves the problem to generate initial scheduling decision instructions for multiple types of energy storage devices. The initial scheduling decision command is issued to various types of energy storage devices for execution, and the collected actual execution data is compared with the initial scheduling decision command to obtain the execution deviation signal. The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor adjusts the prediction parameters according to the received execution deviation signal, and then re-decomposes and predicts with the adjusted prediction parameters to generate a corrected prediction result. The improved butterfly optimization algorithm adjusts the search space bounds and optimizes the search direction based on the received execution deviation signal. With the adjusted search space bounds and optimized search direction, it loads the corrected prediction results and performs iterative search again to generate a corrected scheduling decision instruction. According to the preset rolling cycle, the modified scheduling decision instruction is executed as the initial scheduling decision instruction for the next cycle, and the scheduling decision instruction for the rolling cycle is output.

2. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 1, characterized in that, Based on the aforementioned basic dataset, performance difference information for various types of energy storage devices is obtained, including: Identify the type identifiers of the various types of energy storage devices in the basic dataset; Based on the type identifier, the various types of energy storage devices are divided into energy-type energy storage devices and power-type energy storage devices; For the aforementioned energy storage device, extract the capacity decay rate parameter and the consistency parameter of state of charge change; For the aforementioned power-type energy storage device, extract the power response delay parameter and the upper limit parameter of the power ramp rate; The capacity decay rate parameter, the state of charge change consistency parameter, the power response delay parameter, and the power ramp rate upper limit parameter are collectively output as the performance difference information.

3. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 1, characterized in that, Based on the aforementioned base dataset, a set of static runtime boundary constraints is constructed, including: Extract the rated capacity, rated power, upper limit of allowable state of charge, and lower limit of allowable state of charge for various types of energy storage devices from the aforementioned basic dataset; Based on the rated capacity, the upper limit of the allowable state of charge, and the lower limit of the allowable state of charge, generate the state of charge boundary constraints; Based on the rated power, generate charge and discharge power boundary constraints; Based on the rated capacity, an energy balance constraint is generated to ensure that the state of charge (SOC) values ​​of various types of energy storage devices are equal at the start and end times. The state of charge boundary constraints, the charge and discharge power boundary constraints, and the energy balance constraints are collectively output as the set of static operating boundary constraints.

4. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 1, characterized in that, Before inputting the aforementioned basic dataset into the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor for source-load time series decomposition prediction, the following steps are included: Using historical source load time series data from the aforementioned basic dataset as training samples, offline pre-training is performed on the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor, so that the differential component, harmonic component, and nonlinear autoregressive component in the dynamic weight matrix of the variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor obtain initial weight values. The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor performs the first decomposition prediction with the initial weight values.

5. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 1, characterized in that, The basic dataset is input into a variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor for source-load time-series decomposition prediction, generating initial prediction results, including: The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor extracts source load time series data from the basic dataset and inputs differential components, harmonic components and nonlinear autoregressive components respectively; The differential component performs a difference operation on the source load time series data to obtain transient change trend characteristics; The harmonic components are subjected to Fourier decomposition of the source load time series data to obtain periodic fluctuation characteristics; The nonlinear autoregressive component performs autoregressive fitting on the source load time series data to obtain long-term dependency features; Read the weight values ​​of the dynamic weight matrix corresponding to the differential component, the harmonic component, and the nonlinear autoregressive component; The transient trend features, the periodic fluctuation features, and the long-term dependence features are weighted and superimposed according to their corresponding weight values ​​to obtain the initial prediction result.

6. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 3, characterized in that, Using the set of static operational boundary constraints as the bound of the search space, including: Based on the upper and lower power limits in the charge and discharge power boundary constraints, power boundary lines are drawn on the coordinate axes with charge and discharge power as the search space. Based on the upper limit and lower limit of the charged state in the charged state boundary constraints, the charged state boundary plane is delineated on the auxiliary dimension with the charged state as the search space. Based on the condition that the initial charged state and the final charged state are equal in the energy balance constraint, an equality boundary plane connecting the starting point and the ending point is delineated in the search space. The region enclosed by the power boundary line, the state-of-charge boundary plane, and the equation boundary plane is used as the search space bound for iterative searching by the improved butterfly optimization algorithm.

7. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 2, characterized in that, Using the aforementioned performance difference information as the basis for power allocation includes: The priority order for energy storage devices to undertake long-term charging and discharging tasks is determined based on the capacity decay rate parameter, and the state of charge change consistency parameter is used to force energy storage devices of the same type to keep their state of charge values ​​consistent during the same period. The task of smoothing out instantaneous power fluctuations is assigned to the power storage device with the lowest latency based on the power response delay parameter, and the power change of the power storage device per unit time is limited based on the power ramp rate upper limit parameter.

8. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 4, characterized in that, The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor adjusts its prediction parameters based on the received execution bias signal, including: The variable coefficient differential-harmonic-nonlinear autoregressive coupled predictor converts the received execution bias signal into a weight correction amount; Read the first weight value of the differential component, the second weight value of the harmonic component, and the third weight value of the nonlinear autoregressive component in the dynamic weight matrix of the variable coefficient differential-harmonic-nonlinear autoregressive coupler; Add the weight correction amount to the first weight value to obtain the updated differential component weight value; Subtract the weight correction amount from the third weight value to obtain the updated nonlinear autoregressive component weight value. Keep the second weight value unchanged; The updated dynamic weight matrix is ​​constructed using the updated differential component weights, the updated nonlinear autoregressive component weights, and the unchanged harmonic component weights.

9. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 1, characterized in that, The improved butterfly optimization algorithm adjusts the search space bounds and optimizes the search direction based on the received execution deviation signal, including: The improved butterfly optimization algorithm determines the sign of the received execution deviation signal; When the execution deviation signal is negative, it indicates that the actual execution power is lower than the planned power. The improved butterfly optimization algorithm will shift the lower limit of the power boundary line in the current search space downward. When the execution deviation signal is positive, it indicates that the actual execution power is higher than the planned power. The improved butterfly optimization algorithm will move the upper limit of the power boundary line in the current search space upward. The improved butterfly optimization algorithm simultaneously performs time integration processing on the execution deviation signal to obtain the accumulated deviation amount; When the accumulated deviation exceeds the preset accumulation threshold, the improved butterfly optimization algorithm will increase the weight value of the peak shaving and valley filling effect index in the optimization target, and decrease the weight value of the energy storage life loss index.

10. The method for optimized operation of multiple types of energy storage in a scenario of energy exchange and storage integration as described in claim 1, characterized in that, The preset rolling period is a rolling time domain that starts at the current time and ends at the end of the day.

Images