Virtual power plant energy storage scheduling method and system and storage medium

Abstract

The invention relates to the technical field of power system dispatching control, and discloses a virtual power plant energy storage dispatching method and system and a storage medium, and the method comprises the steps: obtaining the historical load prediction data and the historical real load data of a virtual power plant; load errors are calculated based on the load historical prediction data and historical real load data, and distribution fitting is carried out; generating probability multi-scene load data based on a fitting result; and carrying out optimization solution on the probability multi-scene load data by adopting an optimization algorithm to obtain an energy storage scheduling scheme. The optimization algorithm adopts a two-stage optimization algorithm, and an energy storage scheduling scheme is generated by drawing a box diagram according to a scheduling time step length and performing statistical analysis on an optimization solving result. According to the method, the probability multiple scenes are constructed by analyzing historical prediction errors, the two-stage optimization algorithm is combined, the adaptability and stability of an energy storage scheduling scheme are improved, the operation cost is reduced, and effective technical support is provided for long-term stable operation of the virtual power plant under the peak-valley time-of-use electricity price policy.

Description

Technical Field

[0001] The present invention relates to the technical field of power system dispatching and control, and in particular to a virtual power plant energy storage dispatching method, system and storage medium. Background Art

[0002] As a new power resource management technology, virtual power plants (VPPs) integrate dispersed energy resources and loads to form a virtual power production and dispatch system. Energy storage systems are a crucial component of VPPs, and optimizing their dispatch strategies is crucial for improving the economic efficiency and stability of the entire system.

[0003] The current virtual power plant energy storage scheduling methods have the following main technical problems: Existing energy storage scheduling methods are inadequate for handling load uncertainty. Traditional scheduling methods typically optimize based on deterministic load forecast data, failing to fully consider the impact of load forecast errors. This results in scheduling schemes being less adaptable in actual operations. When actual load deviates from the forecasted load, pre-set scheduling schemes often fail to effectively address this, impacting the operational efficiency of virtual power plants.

[0004] Existing optimization algorithms have limitations when solving energy storage scheduling problems. When using a single solver, the solver optimizes solely from a mathematical perspective, tending to fully charge energy storage equipment during periods of low electricity prices and fully discharge them during peak periods. This results in frequent deep charging and discharging of batteries, frequent starts and stops of energy storage equipment, and a single, inflexible calculation result. When using an optimization algorithm alone, the calculation process is often highly volatile, sensitive to load randomness, resulting in low-quality initial solutions and fluctuating energy storage charging and discharging choices during certain periods, impacting the stability of the scheduling scheme.

[0005] Existing methods lack in-depth analysis and utilization of historical forecast errors. Most scheduling methods ignore the regularity information contained in historical forecast errors and fail to effectively use this information to improve scheduling decisions, resulting in scheduling plans that cannot fully reflect the probability distribution characteristics of actual load.

[0006] Existing scheduling schemes lack long-term applicability. Traditional methods typically optimize for specific time periods or load scenarios, lacking comprehensive consideration of multiple possible scenarios. This results in scheduling schemes with poor robustness in the long term, requiring frequent adjustments and revisions, increasing operation and maintenance costs.

[0007] These technical problems result in the existing virtual power plant energy storage scheduling methods being ineffective when faced with load uncertainty. The economy and stability of the scheduling schemes need to be improved, which limits the further promotion and application of virtual power plant technology. Summary of the Invention

[0008] The purpose of the present invention is to provide a virtual power plant energy storage scheduling method, system and storage medium, aiming to solve the problems of insufficient adaptability of energy storage scheduling schemes and limited ability to handle load uncertainty in the existing technology.

[0009] To solve the above technical problems, the present invention provides a virtual power plant energy storage scheduling method, comprising: Obtain historical load forecast data and historical actual load data of the virtual power plant; Calculating the load error based on the load historical forecast data and the historical actual load data and performing distribution fitting; Generate probabilistic multi-scenario load data based on the fitting results; and An optimization algorithm is used to optimize and solve the probabilistic multi-scenario load data to obtain an energy storage scheduling plan.

[0010] Optionally, the calculation formula for the load error is: E load = L pred -L act ; Where, E load is the load error, L pred is the load historical forecast data, L act It is the historical real load data; The distribution fitting process includes fitting the load error with an error distribution function; The error distribution function includes at least one of normal distribution, multivariate normal distribution or mixed normal distribution.

[0011] Optionally, the step of generating probabilistic multi-scenario load data includes: Sampling within a set confidence interval based on the result of the distribution fitting; Combining the sampling results with the typical load values ​​​​to form probabilistic multi-scenario load data; The sampling method includes Monte Carlo method or Markov chain sampling; the typical load value is the historical average value or median of the load.

[0012] Optionally, the optimization algorithm is a two-stage optimization algorithm, including: In the first stage, a mathematical programming solver is used to obtain the initial optimal solution; In the second stage, a swarm intelligence optimization algorithm is used to obtain a global optimal solution based on the initial optimal solution.

[0013] Optionally, the step of obtaining the energy storage scheduling plan includes: Draw a box plot of the optimization solution results according to the scheduling time step; Determining a preliminary scheduling solution based on the statistical characteristics of the box plot; and Combined with the constraints, the smaller value of the preliminary scheduling plan at time t within the scheduling period and the future predicted load is selected for correction to obtain the final energy storage scheduling plan; The statistical features include at least one of median, maximum value, minimum value, and discrete points.

[0014] Optionally, the constraints in the optimization solution process include power balance constraints, energy storage charging and discharging power constraints, energy storage charge state constraints, and grid power transmission constraints; The objective function of the optimization solution is to minimize the operating cost of the virtual power plant; The energy storage scheduling scheme is applicable to the peak-valley time-of-use electricity price policy.

[0015] Optionally, data preprocessing steps are also included after obtaining the data: Detect and process abnormal data; Visualize historical data; The abnormal data includes at least one of missing values, duplicate values, and outliers.

[0016] Optionally, after the process of calculating the load error, the method further includes: Using the preprocessed data, and according to the scheduling step requirements of the target virtual power plant, adjusting the time step of the load historical forecast data, the historical real load data, and the load error; The step size adjustment processing method includes one or more of direct averaging method, interpolation method, data aggregation method or time series prediction method.

[0017] The present invention also provides a virtual power plant energy storage scheduling system for implementing the above method, comprising: Data acquisition module, used to obtain historical load forecast data and historical actual load data; Error analysis module, used to calculate load error and perform distribution fitting; Scenario generation module, used to generate probabilistic multi-scenario load data based on fitting results; Optimization solution module, used for solving problems using optimization algorithms; The scheduling plan generation module is used to generate an energy storage scheduling plan based on the optimization results.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the above-mentioned virtual power plant energy storage scheduling method is implemented.

[0019] Compared with the prior art, the present invention has at least the following beneficial effects: By establishing a probabilistic multi-scenario model based on historical errors, this method can more accurately reflect the uncertainty characteristics of loads and avoid the limitations of traditional deterministic scheduling methods. By combining the advantages of a two-stage optimization algorithm with a mathematical programming solver and a swarm intelligence optimization algorithm, it ensures both speed and quality of solution, effectively addressing the convergence and globalization issues inherent in single optimization methods and significantly improving the adaptability and cost-effectiveness of energy storage scheduling solutions.

[0020] This invention uses box plot statistical analysis to extract common features from multi-scenario optimization results, generating a long-term energy storage scheduling solution that avoids the computational overhead of frequent recalculations. Combined with the constraints of peak and valley time-of-use electricity pricing policies, the scheduling solution accounts for load uncertainty and fully exploits price differences, minimizing operating costs while ensuring safe and stable system operation, providing effective technical support for the actual operation of virtual power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A flowchart of a method for scheduling energy storage in a virtual power plant according to an embodiment of the present invention; Figure 2 This is a flowchart of a method for scheduling energy storage in a virtual power plant according to another embodiment of the present invention; Figure 3 This is a module diagram of a virtual power plant energy storage scheduling system in one embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following, in conjunction with schematic diagrams, provides a more detailed description of a virtual power plant energy storage scheduling method, system, and storage medium of the present invention. These diagrams illustrate preferred embodiments of the present invention. It should be understood that those skilled in the art may modify the present invention described herein while still achieving the beneficial effects of the present invention. Therefore, the following description should be understood as a general guideline for those skilled in the art and not as a limitation of the present invention.

[0023] The present invention is described in more detail in the following paragraphs by way of example with reference to the accompanying drawings. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are greatly simplified and not to exact scale, and are provided solely for the purpose of assisting in the description of the embodiments of the present invention.

[0024] The embodiment of the present invention provides a virtual power plant energy storage scheduling method, please refer to Figure 1 , including the following steps: S11. Obtain historical load forecast data and historical actual load data of the virtual power plant; S12, calculating the load error based on the load historical prediction data and the historical actual load data and performing distribution fitting; S13. Generate probabilistic multi-scenario load data based on the fitting results; S14. Using an optimization algorithm to optimize and solve the probabilistic multi-scenario load data to obtain an energy storage scheduling solution.

[0025] The virtual power plant energy storage scheduling method provided in this embodiment not only improves the applicability of energy storage scheduling by analyzing and utilizing historical load prediction errors, but also helps to reduce operation and maintenance costs, improve decision-making efficiency, and enhance data utilization efficiency.

[0026] In step S11, the load historical forecast data L of the target virtual power plant in the historical time period is selected. pred Compared with the historical real load data L act .

[0027] The load historical forecast data can be derived from a combination of one or more machine learning models, probabilistic models, and physical models. The historical time period should be long enough to ensure the representativeness of the data, and typically 1 to 3 years of historical data is selected.

[0028] In a specific example, historical load forecast data and corresponding actual load data at every 15-minute interval in the past two years may be selected to form a complete historical data set.

[0029] In step S12, the load error is first calculated using the following formula: E load = L pred -L act ; Where, E load is the load error, L pred is the load historical forecast data, L act This is the historical real load data.

[0030] The distribution fitting process includes: performing error distribution function fitting on the load error.

[0031] Specifically, the error distribution function includes at least one of normal distribution, multivariate normal distribution or mixed normal distribution.

[0032] Taking the normal distribution as an example, its expression is: ; in, are the mean and standard deviation of the normal distribution of errors, and the distribution function is: ; in, is the active power value of the load error.

[0033] In this embodiment, through in-depth analysis of historical prediction errors, the relationship between the virtual power plant prediction model and the actual load can be captured more accurately, laying the foundation for subsequent uncertainty modeling.

[0034] In step S13, the error is sampled to obtain multiple groups of values ​​with the same distribution as the load error, which are combined with the typical load value in the scheduling period to obtain probabilistic multi-scenario for subsequent optimization.

[0035] The probabilistic multi-scenario load data is generated based on the fitting results, and the step of generating the probabilistic multi-scenario load data specifically includes: S131. Sampling within a set confidence interval based on the distribution fitting result; The error distribution function f(E load ), select the confidence interval [E lower , E upper ], multiple groups of sampling are performed within the confidence interval to obtain the load error compensation value , where superscript i represents the group.

[0036] In a specific example, the sampling method includes a Monte Carlo method or a Markov chain sampling method. Among them, the Monte Carlo method has the characteristics of simple implementation and good convergence, and is a preferred sampling method.

[0037] S132. Combining the sampling results with the typical load value to form probabilistic multi-scenario load data.

[0038] Compare the sampled value with the load typical value L typical Combined to form a probabilistic multi-scenario load matrix containing uncertainty : ; In a specific example, the load typical value is a historical load average or median. The historical load average can reflect the long-term load level, while the median is more robust to abnormal values.

[0039] In this embodiment, by introducing historical forecast errors for sampling, various deviations that may occur in reality are actually simulated, so that the model can better cope with future uncertainties, rather than just the situation on a specific date.

[0040] Furthermore, an optimization algorithm is used to optimize and solve the probabilistic multi-scenario load data to obtain an energy storage scheduling solution.

[0041] In step S14, the optimization algorithm is a two-stage optimization algorithm, including: In the first stage, a mathematical programming solver is used to obtain the initial optimal solution; Will As the initial input of the optimization algorithm, it is solved. Specifically: the probability multi-scenario load matrix Use mathematical programming solvers in groups to solve the optimal initial scheduling solution for energy storage in each scenario with the optimal objective function under constraints .

[0042] In a specific example, the mathematical programming solver may be a solver such as Cplex or Gurobi, or an open source linear programming or mixed integer programming solver.

[0043] In a specific example, the Cplex solver is used to solve the initial optimal solution matrix of the target scheduling under the constraints. Cplex is a commercial optimization engine developed by IBM. It is mainly used to solve dynamic scheduling problems in optimal control and is used to quickly calculate the initial solution for energy storage charging and discharging in virtual power plant energy storage scheduling.

[0044] Furthermore, in the second stage, a swarm intelligence optimization algorithm is used to obtain a global optimal solution based on the initial optimal solution.

[0045] Solve the initial schedule Based on the initialization conditions set by the swarm intelligence optimization algorithm, a portion of the solution space is selected. This selected solution space serves as the initial input for some individual positions in the swarm intelligence optimization algorithm, while the remaining positions are randomly generated. Bayesian optimization is used to globally optimize the solution, thereby obtaining the optimal scheduling solution for the optimal objective function under the constraints.

[0046] In a specific example, the swarm intelligence optimization algorithm may be a whale optimization algorithm (WOA), a genetic algorithm, a particle swarm optimization algorithm, or the like.

[0047] Taking the whale optimization algorithm as an example, the algorithm dimension is the same as the total number of time steps in the cycle being sought. The last dimension satisfies the virtual power plant constraint algorithm and does not exist independently. The parameters in the other dimensions exist independently. The adjustment of the parameters in each dimension can be done by using Bayesian optimization global optimization to obtain the optimal scheduling solution for the optimal objective function under the constraints. , each dimension represents the energy storage scheduling scheme in the time interval corresponding to its order.

[0048] After optimizing each dimension of the original swarm intelligence optimization algorithm, one or more of the following methods, crossover mutation and simulated annealing, can be appropriately added based on convergence speed and result satisfaction. It should be noted that crossover mutation and simulated annealing can help improve search efficiency and solution quality. Crossover mutation is primarily used to increase solution diversity and promote the combination of excellent features, and is suitable for complex problems requiring large-scale parallel search. Simulated annealing provides an effective mechanism to avoid falling into local optimal solutions, and is suitable for problems with very large solution spaces and many local extreme points. Therefore, in a specific example, one or more of these methods can be added during WOA update iterations.

[0049] In this embodiment, the core technical effect of the two-stage optimization algorithm is that it combines the fast calculation speed of the mathematical programming solver and the excellent global performance of the swarm intelligence optimization algorithm, thereby ensuring the solution efficiency and improving the quality of the solution, avoiding the limitations of a single method.

[0050] Furthermore, the calculation process needs to consider the constraints of the system constraints, thermal power plant constraints and energy storage constraints in the virtual power plant.

[0051] Constraints are set in the optimization solution process in this embodiment.

[0052] The constraints in the optimization solution process include: power balance constraints, energy storage charging and discharging power constraints, energy storage charge state constraints and grid power transmission constraints.

[0053] Power balance constraint: Power balance is the basic condition for ensuring the operation of virtual power plants: ; Among them, P pv,t is the power generation power of the photovoltaic power station at time t; P load,t is the load value at time t; P bat,t and |P batc,t | is the discharge and charge power of the energy storage at time t (energy storage discharge is defined as a positive value and charging is defined as a negative value); P mt-n,t is the output power of the nth thermal power unit at time t; N mt is the number of thermal power units; It is a Boolean variable. When it is 1, it represents discharging, and when it is 0, it represents charging.

[0054] Energy storage constraints: Energy storage constraints are mainly limited by itself and operational requirements, and their main purpose is to extend service life: SOC min ≤ SOC t ≤ SOC max ; 0 ≤ P batd,t ≤ Pbatd max,t ; 0 ≤|P batc,t | ≤ |P batc max,t |; E storage (0) = E storage (T); P batd,t ≤ P load,t ; Among them, SOC t Indicates the current state of charge of the energy storage; SOC min , SOC max The minimum and maximum state of charge set to protect energy storage; P batd max,t 、P batc max,t They represent the maximum discharge and charge power of the energy storage system; E storage (0) and E storage (T) represents the initial state and the end state of the energy storage system during the dispatch cycle.

[0055] (3) Grid power transmission constraints: P bat,t ≤ P line,max ;|P batc,t| ≤ |P line,max | Among them, P line,max It is the maximum value of the power exchanged between the virtual power plant and the distribution network tie line.

[0056] (4) The constraints of thermal power plants mainly include output constraints, ramp constraints and reserve capacity constraints.

[0057] Specifically, output constraints: ; Among them, P mt_min is the minimum power output of the thermal power unit, P mt_max is the maximum power output of the thermal power unit, U mt-n_t Indicates the operating status of the nth thermal power unit at time t. 1 indicates it is running, and 0 indicates it is stopped.

[0058] Specifically, the reserve capacity constraints of thermal power units are: , ; in, and are the lower reserve capacity and upper reserve capacity of the nth unit at time t respectively.

[0059] Specifically, the reserve capacity of thermal power units is subject to ramp restrictions as follows: ; ; Among them, r cp,down and r cp,up are the up-climbing rate and the down-climbing rate, This is the backup response time, usually 10 minutes.

[0060] Specifically, the reserve capacity of thermal power units is subject to the maximum output limit: , ; The parameters have been explained above.

[0061] Further, the objective function is set: The objective function of the optimization solution is to minimize the operating cost of the virtual power plant: ; Among them, C pv is the price of electricity generated by the photovoltaic power station, C grid,t is the time-of-use electricity price, C bat is the unit price of energy storage charging and discharging operation cost, b n is the power generation coefficient, S n is the unit start-stop coefficient.

[0062] It should be noted that the above formula includes the addition of seven items, among which the first item is the power generation cost of the photovoltaic power station, the second item is the charging cost of the energy storage grid, the third and fourth items are the energy storage charging and discharging costs, the fifth item is the cost of the grid directly supplying power to the load, and the sixth and seventh items are the fuel costs and start-up and shutdown costs of the thermal power units.

[0063] The energy storage scheduling scheme is suitable for peak-valley time-of-use electricity pricing policies. Under peak-valley time-of-use electricity pricing policies, the energy storage system charges during low electricity price periods and discharges during peak electricity price periods, achieving peak-valley shifting and maximizing economic benefits for the power system.

[0064] In step S14, the step of obtaining the energy storage scheduling plan specifically includes: S141, performing statistical analysis on the optimization solution results; S142. Generate an energy storage scheduling plan based on the statistical analysis results.

[0065] The statistical analysis in the process of S141 specifically includes: S1411, plotting the optimization solution results in a box plot according to the scheduling time step; S1412. Determine a preliminary scheduling plan based on the statistical characteristics of the box plot; S1413. Modify the preliminary scheduling plan in combination with the constraint conditions to obtain a final energy storage scheduling plan.

[0066] Specifically, the optimal solution obtained by the swarm intelligence optimization algorithm A box plot is drawn according to the peak and valley time-of-use electricity price time interval and the scheduling time step. Combined with the statistical characteristics of the box plot, the preliminary energy storage scheduling plan P for the virtual power plant is obtained. opt .

[0067] In step S1412, the statistical feature includes at least one of a median, a maximum value, a minimum value, and a discrete point. In a specific example, the median of the box plot can be used as the current step target scheduling value, or the average value of the total amount of the electricity price time period can be calculated as the scheduling value for each time step.

[0068] In the preliminary scheduling plan, the energy storage charging capacity has a positive correlation in the same electricity price period. The energy storage discharging capacity should meet the constraints. Combined with the subsequent forecast load data, the preliminary scheduling plan P at time t within the scheduling period is selected. opt,t and future forecast load L future,t Smaller value: Adjustments are made to obtain a storage scheduling solution suitable for long-term operation of the virtual power plant under the peak and valley time-of-use electricity price and long-term stable load characteristics. .

[0069] Among them, the selective adjustment of electricity prices for consecutive time periods can adopt either the average method or the median method; the subsequent predicted load can be provided by a more accurate real-time intraday forecast.

[0070] It should be noted that this plan is a day-ahead dispatch plan for energy storage under peak-valley time-of-use electricity prices, where the power grid only supplies power to virtual power plants but does not accept power transmission. In the future, the predicted load power can be updated and adjusted in real time using the day-ahead load forecast power and the intraday load forecast power. More accurate intraday forecasts can better support the dispatch plan.

[0071] This method is applicable when the load characteristics are one or a mixture of Gaussian distribution, logarithmic distribution, and gamma distribution, and this distribution will not change significantly in the future due to external factors (such as technological changes, climate change, and policy intervention).

[0072] In this embodiment, the core technical effect of this step is that the energy storage scheduling of the target virtual power plant under the peak-valley time-of-use electricity price obtained by summarizing the advantages of the box diagram has long-term applicability, which is beneficial to the long-term operation of the virtual power plant. The segmented controllable scheduling scheme provides assistance for real-time adjustment scheduling within the day with higher precision requirements.

[0073] In an embodiment of the present invention, a scheme for generating probabilistic multiple scenarios is adopted to more closely approximate the daily load demand of the virtual power plant, and a two-stage Cplex-WOA optimization method is established. Combining the fast computational speed of Cplex and the excellent global performance of WOA, the energy storage scheduling of the target virtual power plant under peak and valley time-of-use electricity prices obtained through the advantages of the box plot has long-term applicability, which is beneficial to the long-term operation of the virtual power plant. The segmented controllable scheduling scheme provides assistance for real-time daily adjustment scheduling with higher precision requirements.

[0074] Furthermore, in another specific example, after acquiring the data in step S11, a data preprocessing step is further included: The data preprocessing process includes: Detect and process abnormal data; Visualize historical data.

[0075] The abnormal data includes at least one of missing values, duplicate values, and outliers.

[0076] The visualization processing includes one or more of a scatter plot, a histogram, a trend graph, or a probability density distribution graph.

[0077] Specific methods of data cleaning include: deleting duplicate data, detecting abnormal data using box plots, filling missing data with random forest regression, or converting date formats, etc. These are common methods used by those skilled in the art and will not be described in detail.

[0078] Furthermore, in a specific example, after the process of calculating the load error in step S12 , the process further includes: adjusting the time step.

[0079] The preprocessed data is used to adjust the time step of the load historical forecast data, the historical real load data and the load error according to the scheduling step requirement of the target virtual power plant.

[0080] Then, the adjusted load error is fitted to obtain the distribution of the load error.

[0081] The step size adjustment processing method includes one or more of direct averaging method, interpolation method, data aggregation method or time series prediction method.

[0082] For example, to adjust the 15-minute interval data to a scheduling step of 1 hour, you can use the direct averaging method to take the average of four 15-minute data, or use the interpolation method to align the time.

[0083] In summary, the embodiment of the present invention utilizes the analysis of load historical forecast errors to enable the model to better capture the relationship between the virtual power plant prediction model and the real load; through the scenario generation method, a large number of simulations are performed on the corrected real load containing uncertainty, thereby solving the maximum possible solution close to the real load situation; by simulating and solving a large number of scenarios, the operating rules of the virtual power plant under the peak and valley time-of-use electricity price are obtained, and the box plot analysis can obtain the energy storage scheduling rules for a single time step and a certain electricity price range. This method helps the model show better robustness in the face of noisy data or outliers; energy storage scheduling is crucial to the scheduling and operation of the power system. This solution summarizes the energy storage scheduling rules corresponding to the target virtual power plant through scenario generation, further helping the virtual power plant to more effectively schedule and manage resources, reduce real-time solution corrections caused by load fluctuations and uncertainties, and only need to modify specific electricity price time intervals compared to real-time adjustments within the day, saving computing power overhead; compared with single day-ahead scheduling, the long-term applicability of scheduling is improved, and the uncertainty caused by load fluctuations is greatly reduced, thereby reducing the operation and maintenance costs of the power grid and improving decision-making efficiency.

[0084] Example 2 Please refer to Figure 2 This embodiment provides a specific implementation process of a virtual power plant energy storage scheduling method based on the first embodiment: S21: Select historical load power data and corresponding historical predicted load power data within a certain time period.

[0085] For example, the load historical forecast data and corresponding historical forecast load power data of a virtual power plant for a period of 2 years from January 2022 to January 2024 are selected. The virtual power plant includes a 50MW photovoltaic power station, a 20MW energy storage system and several thermal power units, and the data collection interval is 15 minutes. The load historical forecast data L pred Derived from the load forecasting model based on LSTM neural network, historical real load data L act Actual measurements from the SCADA system.

[0086] S22: Visualize the data, check whether there are any abnormal data such as missing values ​​and duplicate values, and perform data cleaning.

[0087] For example, we first visualized two years of historical data, creating scatter plots, trend charts, and probability density distribution graphs. We found that the data contained approximately 0.3% missing values, 0.1% duplicate values, and some outliers due to sensor failures.

[0088] The data cleaning steps include: The box plot detection method was used to identify outliers, and data exceeding 1.5 times the interquartile range were marked as outliers; Delete duplicate data records; The random forest regression method is used to fill missing data, and the load data before and after the time as well as auxiliary variables such as weather and temperature are used for prediction and filling; Convert all timestamps to a standard date format.

[0089] S23: Calculate the difference between historical data and historical forecast data, adjust the time step, and fit the error distribution.

[0090] S24: Using the distribution formed by combining the error distribution and the uncertainty, a confidence interval is selected for random sampling to obtain secondary prediction sample data.

[0091] S25: Call Cplex to solve the initial solution of each group of samples, and input the multi-scenario load data into the Cplex solver respectively.

[0092] Establish an optimization model with the goal of minimizing the operating costs of virtual power plants: The objective function is: Constraints include power balance constraints, energy storage SOC constraints (20% to 90%), charging and discharging power constraints (±10MW), etc.

[0093] S26: Set the WOA parameter range, select the initial solution as initialization according to the conditions, and select the Bayesian parameter optimization.

[0094] S27: Perform fitness calculation, dynamic mutation, etc. to obtain the optimal solution under the global optimal parameters.

[0095] The WOA algorithm calculates fitness values ​​at each iteration and uses a dynamic mutation strategy to improve the algorithm's global search capabilities. When there is no improvement over multiple generations, a simulated annealing mechanism is introduced to prevent premature convergence.

[0096] S28: Visualize the optimal solution under the global optimal parameters and draw a box plot according to the time step.

[0097] Each hour corresponds to a box plot, showing the median, quartiles, maximum, minimum, and discrete points of the energy storage dispatch power at that moment.

[0098] For example, analyzing the box plot reveals: 02:00-06:00 (low electricity price period): energy storage is mainly charged, with a median of -6.5MW; 08:00-11:00 and 18:00-21:00 (peak electricity price period): energy storage mainly discharges, with a median of +8.2MW; Other periods (flat electricity price period): Energy storage is in standby mode, with the median close to 0MW.

[0099] S29: Divide the peak and valley time-of-use electricity price time intervals into corresponding charging and discharging time intervals according to the visualized box plot.

[0100] For example, based on the local peak-valley time-of-use electricity price policy and the box plot analysis results, 24 hours are divided into three time intervals: Charging time range: 22:00-06:00 the next day (off-peak electricity price 0.35 yuan / kWh); Discharge time interval: 08:00-11:00, 18:00-23:00 (peak electricity price 0.85 yuan / kWh); Standby time range: 06:00-08:00, 11:00-18:00 (flat electricity price 0.60 yuan / kWh).

[0101] S210: Planning the charge and discharge amount in each time interval according to the selected method, and adjusting the discharge time interval to a new energy storage planning scheme according to the judgment conditions.

[0102] Formulate a preliminary scheduling plan based on the statistical characteristics of the box plot: Charging time interval: take the median of the box plot at each time as the charging power; Discharge time interval: take the median of the box plot at each moment as the discharge power reference value.

[0103] Make corrections taking into account actual operating constraints: The energy storage discharge power must not exceed the real-time load demand, ensuring that the energy storage SOC returns to its initial state at the end of the scheduling cycle and meeting the energy storage continuous charging and discharging time limit. Ultimately, a storage scheduling scheme suitable for the long-term operation of the virtual power plant was obtained. While ensuring safe and stable system operation, this scheme reduces daily operating costs compared to traditional scheduling methods and improves energy storage utilization.

[0104] Example 3 Please refer to Figure 3 An embodiment of the present invention provides a virtual power plant energy storage scheduling system for implementing the above method, including: a data acquisition module, an error analysis module, a scenario generation module, an optimization solution module and a scheduling scheme generation module.

[0105] Data Acquisition Module: This module is used to obtain historical load forecast data and historical actual load data. This module is connected to the virtual power plant's data acquisition system and extracts historical load forecast data and historical actual load data from the historical database.

[0106] Error Analysis Module: This module is used to calculate load errors and perform distribution fitting. This module is connected to the data acquisition module. After receiving load data, it calculates errors and performs statistical analysis and distribution fitting.

[0107] Scenario Generation Module: This module is used to generate probabilistic multi-scenario load data based on the fitting results. This module is connected to the Error Analysis Module and generates multiple sets of scenario data based on the fitted error distribution function.

[0108] Optimization solution module: Used to solve problems using optimization algorithms. This module includes a mathematical programming solution submodule and a swarm intelligence optimization submodule. The two submodules work together to achieve two-stage optimization.

[0109] Dispatch Plan Generation Module: This module is used to generate an energy storage dispatch plan based on the optimization results. This module is connected to the Optimization Solution Module to perform statistical analysis on the optimization results and generate the final dispatch plan.

[0110] Information is transmitted between modules through data interfaces, and the overall system adopts a modular design to facilitate maintenance and upgrades.

[0111] This embodiment also provides a computer-readable storage medium, which, when executed by a processor, implements the various method embodiments of the present application. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape, disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0112] By analyzing and utilizing historical load forecast errors, this model better captures the relationship between the virtual power plant's forecast model and the actual load, demonstrating enhanced robustness in the face of noisy data or outliers. Scenario generation allows for a large number of simulations of modified actual loads containing uncertainty, resulting in the most likely solution close to the actual load. This helps virtual power plants more effectively schedule and manage resources, reducing grid operation and maintenance costs. Energy storage scheduling is crucial to the scheduling and operation of power systems. This solution uses scenario generation to summarize the energy storage scheduling rules for the target virtual power plant, reducing the need for real-time schedule adjustments due to load fluctuations and uncertainty. Compared to intraday real-time adjustments, only specific price intervals need to be modified, saving computing power. Compared to single day-ahead scheduling, this solution improves the long-term adaptability of scheduling, significantly reduces the uncertainty caused by load fluctuations, and helps improve decision-making efficiency and enhance data utilization. Combining the fast computational speed of a mathematical programming solver with the excellent global performance of a swarm intelligence optimization algorithm, the energy storage scheduling of the target virtual power plant under peak and valley time-of-use electricity prices, summarized through the advantages of box plots, demonstrates long-term adaptability.

[0113] In summary, the present invention has important practical significance for promoting the development of virtual power plant energy storage and electricity market.

[0114] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A virtual power plant energy storage scheduling method, characterized in that: include: Obtain historical load forecast data and historical actual load data of the virtual power plant; Calculating the load error based on the load historical forecast data and the historical actual load data and performing distribution fitting; Generate probabilistic multi-scenario load data based on the fitting results; as well as An optimization algorithm is used to optimize and solve the probabilistic multi-scenario load data to obtain an energy storage scheduling plan.

2. The virtual power plant energy storage scheduling method according to claim 1, characterized in that: The calculation formula of the load error is: AND load = L pred - L act ; Where, E load is the load error, L pred is the load historical forecast data, L act It is the historical real load data; The distribution fitting process includes fitting the load error with an error distribution function; The error distribution function includes at least one of normal distribution, multivariate normal distribution or mixed normal distribution.

3. The virtual power plant energy storage scheduling method according to claim 1, characterized in that: The step of generating probabilistic multi-scenario load data includes: Sampling within a set confidence interval based on the result of the distribution fitting; Combining the sampling results with the typical load values ​​​​to form probabilistic multi-scenario load data; The sampling method includes Monte Carlo method or Markov chain sampling; the typical load value is the historical average value or median of the load.

4. The virtual power plant energy storage scheduling method according to claim 1, characterized in that: The optimization algorithm is a two-stage optimization algorithm, including: In the first stage, a mathematical programming solver is used to obtain the initial optimal solution; In the second stage, a swarm intelligence optimization algorithm is used to obtain a global optimal solution based on the initial optimal solution.

5. The virtual power plant energy storage scheduling method according to claim 1, characterized in that: The step of obtaining the energy storage scheduling plan includes: Draw a box plot of the optimization solution results according to the scheduling time step; Determining a preliminary scheduling solution based on the statistical characteristics of the box plot; and Combined with the constraints, the smaller value of the preliminary scheduling plan at time t within the scheduling period and the future predicted load is selected for correction to obtain the final energy storage scheduling plan; The statistical features include at least one of median, maximum value, minimum value, and discrete points.

6. The virtual power plant energy storage scheduling method according to claim 1, characterized in that: The constraints in the optimization solution process include power balance constraints, energy storage charging and discharging power constraints, energy storage charge state constraints and grid power transmission constraints; The objective function of the optimization solution is to minimize the operating cost of the virtual power plant; The energy storage scheduling scheme is applicable to the peak-valley time-of-use electricity price policy.

7. The virtual power plant energy storage scheduling method according to claim 1, characterized in that: After acquiring the data, the data preprocessing steps are also included: Detect and process abnormal data; Visualize historical data; The abnormal data includes at least one of missing values, duplicate values, and outliers.

8. The virtual power plant energy storage scheduling method according to claim 7, characterized in that: After the process of calculating the load error, the method further includes: Using the preprocessed data, and according to the scheduling step requirements of the target virtual power plant, adjusting the time step of the load historical forecast data, the historical real load data, and the load error; The step size adjustment processing method includes one or more of direct averaging method, interpolation method, data aggregation method or time series prediction method.

9. A virtual power plant energy storage scheduling system for implementing the method according to any one of claims 1 to 8, characterized in that: include: Data acquisition module, used to obtain historical load forecast data and historical actual load data; Error analysis module, used to calculate load error and perform distribution fitting; Scenario generation module, used to generate probabilistic multi-scenario load data based on fitting results; Optimization solution module, used for solving problems using optimization algorithms; The scheduling plan generation module is used to generate an energy storage scheduling plan based on the optimization results.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images