A power data multi-granularity unified modeling and prediction method for a virtual power plant

Abstract

This invention discloses a unified multi-granularity modeling and prediction method for power data in virtual power plants, belonging to the field of power prediction technology. It is used to solve the problem of inaccurate power prediction and mismatch between energy storage charging and discharging decisions and actual supply conditions due to the inability to perceive deviations in energy storage execution and changes in supply constraints. The method obtains the energy storage execution deviation by collecting planned energy storage charging and discharging power and state of charge offset through weighted fusion using the analytic hierarchy process (AHP). It detects the voltage fluctuation characteristics of the grid connection point and the power output fluctuation rate of renewable energy, and calculates the supply constraint coefficient based on a multinomial regression model. Based on the supply constraint coefficient, it modulates the power prediction baseline sequence point by point through Sigmoid mapping to generate a corrected power prediction result. Based on the corrected power prediction result, it calculates the coverage fit and triggers dynamic replanning of the energy storage strategy, realizing constraint-aware correction of power prediction and adaptive adjustment of energy storage strategy.

Description

Technical Field

[0001] This invention relates to the field of power forecasting technology, and more specifically, to a multi-granular unified modeling and forecasting method for power data of virtual power plants. Background Technology

[0002] Virtual power plants aggregate heterogeneous resources such as distributed energy storage, photovoltaics, and wind power to participate in power market dispatch and ancillary services as a whole. In actual operation, the dispatch system needs to make energy storage charging and discharging decisions based on power forecast results and continuously track the execution status within the control cycle. Existing power forecasting methods typically generate a forecast baseline based on historical data and meteorological inputs, and directly transmit this baseline to the decision-making module to drive the issuance and execution of energy storage charging and discharging commands.

[0003] The existing technology has the following shortcomings: Currently, existing technologies lack real-time perception of the actual deviation state of energy storage devices and voltage fluctuations at the grid connection point, and cannot accurately reflect the comprehensive impact of multi-source supply constraints on the prediction results within the current control cycle. This results in a mismatch between energy storage charging and discharging decisions and actual supply conditions, leading to a high supply mismatch rate. Therefore, a multi-granular unified modeling and prediction method for power data oriented towards virtual power plants is proposed.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a unified multi-granularity modeling and prediction method for power data in virtual power plants. This method addresses the problems mentioned in the background art by employing energy storage to perform deviation quantification analysis, multi-source constraint fusion to calculate supply constraint coefficients, point-by-point correction of power prediction sequences based on Sigmoid mapping, and dynamic replanning of energy storage strategies driven by coverage fit.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-granularity unified modeling and prediction method for power data in virtual power plants, comprising the following steps: Step S1: Collect the planned charging and discharging power of the virtual power plant for the current time period, detect the state of charge deviation of the energy storage device, and analyze the deviation of energy storage execution by combining the planned charging and discharging power of the energy storage and the state of charge deviation. Step S2: Detect the voltage fluctuation characteristics at the grid connection point, collect the renewable energy output fluctuation rate at the power station aggregation end, and calculate the supply constraint coefficient by combining the voltage fluctuation characteristics at the grid connection point and the energy storage execution deviation. Step S3: Read the power prediction baseline sequence output by the conventional prediction module, perform sequence correction on the power prediction baseline sequence according to the supply constraint coefficient, and generate the corrected power prediction result; Step S4: Evaluate the coverage fit of the current energy storage charging and discharging decision based on the corrected power prediction results, and determine whether to trigger dynamic replanning of the energy storage strategy based on the coverage fit.

[0007] In a preferred embodiment, in step S1, the actual state of charge of the energy storage device is read through the BMS communication bus, and the expected state of charge at the current moment is calculated by combining the planned charging and discharging power with the rated battery capacity. The absolute value of the difference between the actual state of charge and the desired state of charge is used to obtain the energy storage state of charge offset.

[0008] In a preferred embodiment, in step S1, the planned charging and discharging power of the energy storage plan and the energy storage state of charge deviation are Max-Min normalized respectively. A judgment matrix is ​​constructed using the analytic hierarchy process (AHP). The normalized planned charging and discharging power of the energy storage plan and the energy storage state of charge deviation are weighted to obtain a weight vector. The energy storage execution deviation is obtained by combining the normalized planned charging and discharging power of the energy storage plan and the energy storage state of charge deviation.

[0009] In a preferred embodiment, in step S2, the current control cycle is selected as the acquisition time window, the effective values ​​of the three-phase voltage are sampled and merged into a grid connection point voltage fluctuation dataset; The standard deviation of the grid connection point voltage fluctuation dataset is used as the feature of grid connection point voltage fluctuation.

[0010] In a preferred embodiment, in step S2, the collected active power is read at the start and end times of the current control cycle, and the absolute value of the difference between the collected active power at the start and end times is divided by the control cycle duration to obtain the renewable energy output volatility. The voltage fluctuation characteristics at the grid connection point, the power output fluctuation rate of renewable energy, and the deviation of energy storage execution are used as feature vectors. A multinomial regression model is used to fuse them to calculate the supply constraint coefficient.

[0011] In a preferred embodiment, in step S3, a power prediction baseline sequence is read, which covers a set future duration and is arranged at fixed time intervals. Based on the supply constraint coefficient, the modulation factor is calculated through the Sigmoid mapping. The modulation factor is then multiplied point by point in the power prediction baseline sequence to obtain the corrected power prediction result.

[0012] In a preferred embodiment, in step S3, the steepness parameter and center point parameter in the Sigmoid mapping are determined through historical calibration: using the supply constraint coefficient of the historical control cycle as the independent variable and the ratio of the actual power to the predicted baseline as the dependent variable, Sigmoid curve fitting is performed, and the optimal parameters are solved by maximum likelihood estimation and written into the parameter file.

[0013] In a preferred embodiment, in step S4, the time-sharing power interval corresponding to the current energy storage charging and discharging decision is read, and each time point of the corrected power prediction result is judged to see if it falls within the corresponding time-sharing power interval. The ratio of the number of time points falling into the interval to the total number of time points is calculated to obtain the coverage fit.

[0014] In a preferred embodiment, in step S4, when the coverage fit is lower than the coverage fit threshold, the energy storage strategy is dynamically replanned, and the corrected power prediction result is output to the decision module as the basis for a new round of decision-making. Conversely, the current energy storage strategy is maintained, and the revised power prediction results are retained as the prediction feedback baseline for the next time step.

[0015] The technical effects and advantages of this invention are as follows: This invention collects the planned charging and discharging power and state of charge offset of energy storage, and obtains the energy storage execution deviation by weighted fusion using Max-Min normalization and analytic hierarchy process. It detects the voltage fluctuation characteristics of the grid connection point and the power output fluctuation rate of renewable energy at the power station aggregation end, and calculates the supply constraint coefficient by fusing three-dimensional constraint information based on a multinomial regression model. It reads the power prediction baseline sequence, modulates it point by point according to the supply constraint coefficient through Sigmoid mapping, and generates the corrected power prediction result. Based on the corrected power prediction result, it determines the proportion of each point that falls into the time-sharing power range of the energy storage decision, calculates the coverage fit, and triggers or maintains the dynamic replanning of the energy storage strategy. This makes the power prediction result of the virtual power plant accurately reflect the real-time supply constraints and reduces the supply mismatch rate of energy storage strategy decision. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the implementation of a multi-granularity unified modeling and prediction method for power data in virtual power plants, as described in this invention.

[0017] Figure 2 This is a schematic diagram illustrating the steps of a multi-granularity unified modeling and prediction method for power data in virtual power plants according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention achieves adaptive replanning of the coverage adaptability of energy storage strategies by multi-source fusion modeling of energy storage execution deviation, grid connection point voltage fluctuation and renewable energy output fluctuation rate, driving the supply constraint perception correction of the power prediction baseline sequence.

[0020] Example 1: Please refer to Figures 1 to 2 A unified multi-granularity modeling and prediction method for power data in virtual power plants includes the following steps: Step S1: Collect the planned charging and discharging power of the virtual power plant for the current time period, detect the state of charge deviation of the energy storage device, and analyze the deviation of energy storage execution by combining the planned charging and discharging power of the energy storage and the state of charge deviation. Step S2: Detect the voltage fluctuation characteristics at the grid connection point, collect the renewable energy output fluctuation rate at the power station aggregation end, and calculate the supply constraint coefficient by combining the voltage fluctuation characteristics at the grid connection point and the energy storage execution deviation. Step S3: Read the power prediction baseline sequence output by the conventional prediction module, perform sequence correction on the power prediction baseline sequence according to the supply constraint coefficient, and generate the corrected power prediction result; Step S4: Evaluate the coverage fit of the current energy storage charging and discharging decision based on the corrected power prediction results, and determine whether to trigger dynamic replanning of the energy storage strategy based on the coverage fit.

[0021] The specific implementation is as follows: In step S1, the operating cycle of the virtual power plant is discretized into a control cycle of 5 minutes, and the acquisition and analysis action of this step is initiated at the beginning of each control cycle. The system collects the planned charging and discharging power of the energy storage system for the current time period from the virtual power plant. Specifically, through the communication interface of the decision module deployed at the dispatch terminal, the system reads the planned charging and discharging power command that will be issued to the energy storage converter at the beginning of the control cycle, denoted as P1. A positive value of P1 indicates discharging, and a negative value indicates charging, with the unit being kilowatts.

[0022] The planned charging and discharging power of the energy storage is generated by the decision module based on the prediction results of the previous control cycle. It is read once in each control cycle as the execution intention benchmark for this step. At the same time, the planned charging and discharging power of the energy storage read in this cycle is synchronously written to the local cache for use in the next cycle when simulating the expected state of charge.

[0023] The state of charge offset of the energy storage device is detected. The energy storage device consists of a battery management system (BMS) and a power conversion system (PCS). The BMS is used to monitor and alarm the state of charge of the battery pack.

[0024] At the start of the current control cycle, the actual state of charge of the energy storage device is read via the BMS communication bus. Simultaneously, based on the planned charging and discharging power of the energy storage system and the rated battery capacity of the previous control cycle, the expected state of charge at the current moment is deduced. : ,in This represents the actual state of charge at the start of the previous cycle. The charging and discharging power of the energy storage plan for the previous cycle, To control the cycle duration, The rated capacity of the energy storage is given. The absolute value of the difference between the actual state of charge and the desired state of charge is taken to obtain the energy storage state of charge offset P2. The energy storage state of charge offset reflects whether the energy storage has the inherent ability to continue to execute as planned. The larger the value, the further the actual energy storage deviates from the decision expectation. The reading action of the BMS communication bus is triggered by the start time of the control cycle. If the communication times out, the reading of the previous cycle is reused and an abnormal flag is written. The abnormal flag is recorded within this step and does not participate in the subsequent deviation calculation. At the same time, the communication abnormality prompt is output through the log system for operation and maintenance personnel to check.

[0025] The deviation in energy storage execution is analyzed by combining the planned charge / discharge power and the energy storage state of charge (SCC) offset. Since the planned charge / discharge power is in kilowatts (kW) and the SCC offset is a percentage, which have different units, Max-Min normalization is first performed on each. The specific steps are as follows: Step 1, Data Preparation: Retrieve the planned charge / discharge power sequence and energy storage state of charge offset sequence for each control cycle over the past 30 days from the historical database, and calculate the maximum and minimum values ​​respectively, denoted as... , , , The extreme value window is updated every 24 hours to avoid seasonal drift causing the normalized benchmark to become invalid. Step 2, Normalization Mapping: Calculate the normalized values ​​p1' and p2' for the current period respectively: It should be noted that Max-Min normalization can map parameters to the [0,1] interval while preserving the original distribution shape, for use in subsequent weighting.

[0026] Next, the Analytic Hierarchy Process (AHP) is used to assign weights to p1' and p2'. A second-order judgment matrix M is constructed, where the matrix element mij represents the relative importance of the i-th factor to the j-th factor, with values ​​ranging from 1 to 9 and their reciprocals. The largest eigenvalue is then calculated from the judgment matrix. The corresponding feature vectors are then normalized to obtain the weight vector (w1, w2), where w1 corresponds to the planned energy storage charge / discharge power and w2 corresponds to the energy storage state of charge offset. Simultaneously, the consistency ratio is calculated. : ,when If the consistency check is passed, the judgment matrix is ​​readjusted; otherwise, it is not. The reason for choosing the analytic hierarchy process (AHP) is that the contributions of the energy storage plan's charging and discharging power and the energy storage's state of charge offset to the deviation are of different importance. AHP can explicitly transform expert judgments into reusable weight vectors and has a consistency check mechanism to avoid subjective bias caused by directly specifying weights.

[0027] The energy storage performance deviation is obtained by combining the following formula: It should be noted that the larger the value of D, the more serious the deviation of the actual energy storage operation from the original plan, and its value falls in the range of [0,1].

[0028] For example, a certain control cycle is taken ,but This indicates that the energy storage performance deviation during this cycle is at a moderately high level. The specific values ​​of relative importance in the judgment matrix and the critical value of 0.1 for the consistency threshold CR are common values ​​in this field and can be set by professionals in this field according to the actual situation. They will not be elaborated here.

[0029] In step S2, the grid connection point refers to the common coupling location where the virtual power plant connects to the main grid, which is used for external power exchange and power quality monitoring.

[0030] A three-phase power quality analyzer is deployed at the grid connection point. The current control cycle is selected as the acquisition time window, and the effective values ​​of the three-phase voltage are continuously sampled at 200-millisecond intervals. The samples are then merged into a grid connection point voltage fluctuation dataset. ,in To determine the total number of sampling points within a control period, when the control period is 5 minutes... Take 1500; then process the dataset. Calculate the standard deviation as the grid connection point voltage fluctuation characteristic P3 for this control cycle: ,in For dataset The arithmetic mean, This represents the total number of sampling points; If a sample is missing due to communication interruption, it is filled in by linear interpolation; if the number of consecutive missing points exceeds 5%, the grid connection point voltage fluctuation feature of that period is marked as invalid, and the grid connection point voltage fluctuation feature value of the previous period is directly used for subsequent calculations.

[0031] It should be added that a three-phase power quality analyzer refers to a power quality monitoring terminal deployed at the grid connection point of a virtual power plant, used to continuously and synchronously sample the effective value of three-phase voltage.

[0032] The power output volatility of renewable energy is collected at the power station aggregation point, which refers to the physical location where renewable energy sources such as photovoltaic and wind power flow into the internal bus of the virtual power plant. At the start of the current control cycle, the collected active power is read through the power acquisition unit at the collection end. At the end of the current control cycle, the collected active power is read again. Take the absolute value of the difference between the two readings and the control cycle duration. As a market maker, we obtain the volatility P4 of renewable energy output: The unit of renewable energy output volatility is kilowatt-minute, which reflects the rate of power change of renewable sources in the current cycle; the power acquisition unit at the collection end adopts a 0.5S-level power metering device, and the readings at both ends are sampled 20 times at 10-millisecond intervals and the arithmetic mean is taken as the endpoint value to suppress the interference of instantaneous spikes on volatility. If more than half of the two endpoint readings fail to be sampled, the renewable energy output volatility for this cycle will be the average of the last three cycles.

[0033] The supply constraint coefficient is calculated by combining the renewable energy output volatility with the grid connection point voltage volatility characteristics and the energy storage execution deviation. Considering the nonlinear coupling effect of grid connection point voltage volatility, renewable energy output volatility, and energy storage execution deviation on the virtual power plant's supply capacity, a multinomial regression model is used for fusion. ,in This is the supply constraint coefficient. These are the polynomial coefficients; the polynomial coefficients are obtained through a historical calibration phase, and the specific steps are as follows: Data preparation: Select a period of time as the retrieval period, and retrieve the (P3,P4,D) triples of control cycles that have occurred in the historical database, as well as the supply deviation labels obtained from post-event statistics. After removing samples with missing labels or parameters that are out of bounds, a calibration sample set is formed. Solving for the optimal coefficients: using (P3, P4, D, P3·D, P4·D) as the eigenvectors... The target vector is obtained by fitting using the least squares method. And save it to the method parameter file, and call it directly during the runtime; Coefficient maintenance: The fitting is rolled back every 30 days using a sliding window method to keep the coefficients consistent with the most recent operating conditions. If the maximum deviation between the two fitting results exceeds 10%, a calibration sample audit is triggered to confirm whether the sample set contains abnormal operating conditions.

[0034] It should be explained that the larger K is, the more suppressed the supply capacity of the current control cycle is. Its value range is determined by the calibration data and usually falls between [0,3]. For example, if the control cycle is P3=4.2 volts, P4=8.5 kilowatts / minute, and D=0.6, and the calibration coefficients are a0=0.05, a1=0.02, a2=0.03, a3=0.4, a4=0.08, and a5=0.06, then substituting them into the equation, we get K=0.05+0.02×4.2+0.03×8.5+0.4×0.6+0.08×4.2×0.6+0.06×8.5×0.6=1.1366, indicating that the supply constraint in this cycle is at a moderate level.

[0035] The number of days in the historical calibration window and the proportion of fitted samples to be removed can be set according to the actual situation, and will not be analyzed in detail here.

[0036] In step S3, the power prediction baseline sequence is the power prediction time series data for the next period generated by the conventional prediction module based on historical data and meteorological input, which is used as the benchmark target for subsequent corrections.

[0037] At the start of the current control cycle, the power prediction baseline sequence covering the next 4 hours at 5-minute intervals is read through the output interface of the prediction module. The unit is kilowatt.

[0038] If the interface returns a timeout, the power prediction baseline sequence from the previous control cycle is reused and shifted to the left by one time step. At the same time, a "baseline reuse" flag is added to the output of this step for the decision module to refer to. If two consecutive cycles return a timeout, a health check alarm for the prediction module is triggered, and the correction action of this method is suspended until the prediction module recovers.

[0039] The power prediction baseline sequence is corrected based on the supply constraint coefficient to generate the corrected power prediction result. The correction adopts a point-by-point ratio modulation method, that is, each point bi of the baseline sequence is multiplied by a modulation factor determined by the supply constraint coefficient K. The corresponding points of the corrected power prediction results are obtained. : , where the modulation factor Using Sigmoid mapping: ,in, For steepness parameters, The center point parameters were all determined through historical calibration. Data preparation: Retrieve control cycles that have occurred from the historical database. binary pair, This is the ratio of the actual power obtained from post-event statistics to the predicted baseline for that period; The solution method is as follows: For independent variable, Perform Sigmoid curve fitting on the dependent variable and solve using maximum likelihood estimation. and The optimal value is determined, and after fitting, it is written into the method parameter file, where it is directly called as a fixed parameter during the runtime phase; the optimal value is determined quarterly. and Perform a rolling recalibration to keep the corrected mapping consistent with the equipment aging curve.

[0040] It needs to be explained that, The bigger The closer the result is to 1, the closer the corrected power prediction result is to the baseline; smaller The closer the value is to 0, the more conservative the corrected power prediction result is relative to the baseline; for example, the control period is taken as... , If the first baseline point of this period The corrected power prediction result . and Once calibrated, the parameters are fixed and will not be adjusted with each run.

[0041] In step S4, the corrected power prediction result is received. It also reads the time-of-use power range corresponding to the current energy storage charging and discharging decision through the existing data in the decision module. ,in The lower and upper bounds of the allowable power constrained for energy storage decisions at the i-th 5-minute granularity time.

[0042] The time-sharing power range is generated by the decision module based on the prediction results of the previous control cycle and market-based trading constraints. It serves as the evaluation reference for this step. The reading action is executed in parallel with step S3 to shorten the response delay within the overall control cycle.

[0043] The coverage fit of the current energy storage charging and discharging decisions is assessed based on the corrected power prediction results. The assessment uses a point-by-point judgment + mean ratio approach. The specific steps are as follows: Step 1, point-by-point determination: for ,like Then record the overwrite mark. Otherwise remember ; Step 2, Mean Ratio Calculation: Coverage fit A is calculated using the following formula: The coverage fit A value falls within the range of [0,1]. The larger A is, the higher the proportion of the corrected power prediction result falling within the time-of-use power range constrained by the energy storage decision, that is, the more fully the current decision covers the future supply.

[0044] Then, the coverage fit is used to determine whether to trigger dynamic replanning of the energy storage strategy. A coverage fit threshold is set. Establish judgment rules: Rule 1: When When this happens, the energy storage strategy is dynamically replanned, and the corrected power prediction results are output to the decision module as the basis for a new round of decision-making. Rule 2: When At that time, the current energy storage strategy is maintained, and the revised power prediction results are retained as the prediction feedback baseline for the next time step.

[0045] It should be noted that setting a coverage adaptation threshold At that time, retrieve the (A, whether post-event supply mismatch occurred) tuples from several control cycles in the historical database. Post-event supply mismatch is labeled by whether the proportion of actual power exceeding the decision-time power range exceeds 15%. Sort A values ​​from smallest to largest, and take the quantile (e.g., 90th percentile) of the A value distribution corresponding to the "post-event supply mismatch occurred" sample as... Recalibrate once per quarter The threshold is updated according to changes in the internal resource structure of the virtual power plant; for example, the calibration is obtained. When a certain control cycle is calculated to obtain hour, The current energy storage strategy will be maintained. After dynamic replanning is triggered, the decision module will re-solve the time-of-use power range based on the revised power prediction results, and issue the new energy storage plan charging and discharging power at the beginning of the next control cycle through the decision module communication interface.

[0046] Compared to metrics such as mean square error or maximum deviation, the point-by-point judgment plus mean ratio method makes it easier for dispatchers to conduct secondary verification of the trigger results. At the same time, this indicator corresponds one-to-one with the replanning trigger signal of the subsequent decision-making module, avoiding the introduction of additional black-box links in the evaluation stage, and also facilitating post-event auditing and accountability.

[0047] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0048] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0049] In this document, the singular forms “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that terms such as “comprising / including” or “having” specify the presence of the stated features, integrals, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, integrals, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0050] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0051] The above description of the disclosed embodiments will enable those skilled in the art to make or use various modifications to these embodiments. It will be readily apparent to those skilled in the art that 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.

Claims

1. A power data multi-granularity unified modeling and prediction method for a virtual power plant, characterized in that: Includes the following steps: Step S1: Collect the planned charging and discharging power of the virtual power plant for the current time period, detect the state of charge deviation of the energy storage device, and analyze the deviation of energy storage execution by combining the planned charging and discharging power of the energy storage and the state of charge deviation. Step S2: Detect the voltage fluctuation characteristics at the grid connection point, collect the renewable energy output fluctuation rate at the power station aggregation end, and calculate the supply constraint coefficient by combining the voltage fluctuation characteristics at the grid connection point and the energy storage execution deviation. Step S3: Read the power prediction baseline sequence output by the conventional prediction module, perform sequence correction on the power prediction baseline sequence according to the supply constraint coefficient, and generate the corrected power prediction result; Step S4: Evaluate the coverage fit of the current energy storage charging and discharging decision based on the corrected power prediction results, and determine whether to trigger dynamic replanning of the energy storage strategy based on the coverage fit.

2. The method for multi-granularity unified modeling and prediction of power data for virtual power plants according to claim 1, characterized in that: In step S1, the actual state of charge of the energy storage device is read through the BMS communication bus, and the expected state of charge at the current moment is calculated by combining the planned charging and discharging power with the rated battery capacity. The absolute value of the difference between the actual state of charge and the desired state of charge is used to obtain the energy storage state of charge offset.

3. The method for multi-granularity unified modeling and prediction of power data for virtual power plants according to claim 1, characterized in that: In step S1, the planned charging and discharging power of the energy storage plan and the energy storage state of charge deviation are Max-Min normalized respectively. The judgment matrix is ​​constructed using the analytic hierarchy process. The normalized planned charging and discharging power of the energy storage plan and the energy storage state of charge deviation are weighted to obtain a weight vector. The energy storage execution deviation is obtained by combining the normalized planned charging and discharging power of the energy storage plan and the energy storage state of charge deviation.

4. The method for multi-granularity unified modeling and prediction of power data for virtual power plants according to claim 3, characterized in that: In step S2, the current control cycle is selected as the acquisition time window, and the effective values ​​of the three-phase voltage are sampled and merged into a grid connection point voltage fluctuation dataset. The standard deviation of the grid connection point voltage fluctuation dataset is used as the feature of grid connection point voltage fluctuation.

5. The method for multi-granularity unified modeling and prediction of power data for virtual power plants according to claim 4, characterized in that: In step S2, the collected active power is read at the start and end times of the current control cycle. The absolute value of the difference between the collected active power at the start and end times is then divided by the control cycle duration to obtain the renewable energy output volatility. The voltage fluctuation characteristics at the grid connection point, the power output fluctuation rate of renewable energy, and the deviation of energy storage execution are used as feature vectors. A multinomial regression model is used to fuse them to calculate the supply constraint coefficient.

6. The method for multi-granularity unified modeling and prediction of power data for virtual power plants according to claim 1, characterized in that: In step S3, a power prediction baseline sequence is obtained by reading data that covers a set future duration and is arranged at fixed time intervals. Based on the supply constraint coefficient, the modulation factor is calculated through the Sigmoid mapping. The modulation factor is then multiplied point by point in the power prediction baseline sequence to obtain the corrected power prediction result.

7. The method for multi-granularity unified modeling and prediction of power data for virtual power plants according to claim 6, characterized in that: In step S3, the steepness parameter and center point parameter in the Sigmoid mapping are determined through historical calibration: with the supply constraint coefficient of the historical control cycle as the independent variable and the ratio of the actual power to the predicted baseline as the dependent variable, Sigmoid curve fitting is performed, and the optimal parameters are solved by maximum likelihood estimation and written into the parameter file.

8. The method for multi-granularity unified modeling and prediction of power data for virtual power plants according to claim 1, characterized in that: In step S4, the time-sharing power interval corresponding to the current energy storage charging and discharging decision is read, and each time point of the corrected power prediction result is judged to see if it falls within the corresponding time-sharing power interval. The ratio of the number of time points falling into the interval to the total number of time points is calculated to obtain the coverage fit.

9. A method for multi-granularity unified modeling and prediction of power data for virtual power plants according to claim 8, characterized in that: In step S4, when the coverage fit is lower than the coverage fit threshold, the energy storage strategy is dynamically replanned, and the corrected power prediction result is output to the decision module as the basis for a new round of decision-making. Conversely, the current energy storage strategy is maintained, and the revised power prediction results are retained as the prediction feedback baseline for the next time step.

Images