Multi-target balanced load coordination control method for distributed virtual power plant
By analyzing the similarity and anomaly detection of historical load data of virtual power plants, and combining prediction and optimization algorithms, the problem of inaccurate load demand prediction of virtual power plants is solved, and high-precision and robust load prediction and control are achieved.
Patent Information
- Application Number
- CN202610024813.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-09
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies for load demand forecasting in virtual power plants fail to effectively consider factors such as special holidays and sudden weather changes, resulting in inaccurate load demand forecasts and affecting the real-time performance and accuracy of load regulation.
By collecting virtual power plant load data, combining the similarity of fitted curves with historical and nearby multi-day load data and anomaly detection algorithms, the reference value of load data is analyzed. Prediction and optimization algorithms are used for load prediction and regulation. Considering time series trends and periodicity, anomaly detection and local dynamic modeling are introduced to improve prediction accuracy.
It achieves high-precision and robust load forecasting, relying on historical patterns during stable periods and enhancing real-time response during abrupt changes, significantly improving the accuracy and stability of load forecasting and supporting subsequent precise regulation.
Smart Images
Figure CN122052007A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of distributed control technology, specifically to a multi-objective load balancing coordination control method for a distributed virtual power plant. Background Technology
[0002] Virtual Power Plants (VPPs) have emerged as an advanced concept for energy aggregation and management. VPPs use advanced information and communication technologies (ICT) and energy management systems (EMS) to logically aggregate and optimize the management of various geographically dispersed and relatively small-capacity distributed resources (including power generation, energy storage, and loads), enabling them to participate as a whole in electricity market transactions and grid ancillary services.
[0003] To better coordinate and manage multiple distributed power sources within a virtual power plant, it is necessary to predict and analyze the load demand of the virtual power plant. This allows for earlier and more timely coordination based on the prediction results, thereby increasing the real-time performance of load balancing control. However, existing methods rely solely on historical load data for prediction and analysis using predictive algorithms. This approach fails to adequately consider the impact of other factors, such as special holidays and sudden weather changes, on the power plant's load demand. Consequently, the obtained load demand is inaccurate, leading to inaccurate adjustments during the load regulation of each power source within the power plant. Summary of the Invention
[0004] To address the aforementioned technical problems, this application provides a multi-objective load balancing coordination control method for distributed virtual power plants, thereby resolving the existing issues.
[0005] The multi-objective load balancing coordination control method for a distributed virtual power plant proposed in this application adopts the following technical solution: One embodiment of this application provides a multi-objective load balancing coordination control method for a distributed virtual power plant, the method comprising the following steps: Collect load data of the virtual power plant at various times; Based on the similarity of load data fitting curves between historical adjacent days, and combined with anomaly detection algorithms, outliers of each similarity are analyzed; by combining the numerical distribution of all similarities and all outliers, the referenceability of the similarity between the load data fitting curves of the current day and historical dates is analyzed; and load prediction reference values for each time point are determined based on load data at the same time point on historical days. The initial load prediction value for the next moment is obtained by predicting the load data within a preset time period before each moment; the enhancement coefficient of the initial load prediction value for each moment is determined based on the disorder of the load data at the same moment of each historical day; the first prediction value for each moment is obtained by using the prediction algorithm to predict the load data at the same moment of each historical day. Based on the enhancement coefficient and the reference value, the weights of the initial load forecast value, the load forecast reference value, and the first forecast value are determined, and the load demand forecast value at each time point is calculated so as to use an optimization algorithm to regulate the virtual power plant load.
[0006] In one embodiment, the process of obtaining the outlier is as follows: The fitted curve of the daily load data is denoted as the daily load demand curve; the similarity between any two load demand curves within a preset number of days before the current time is used as the input of the anomaly detection algorithm, and the output is the anomaly value of the similarity between the load demand curves of each of the aforementioned two arbitrary days.
[0007] In one embodiment, the process of obtaining the referenceability is as follows: Within a preset number of days prior to the current moment, the similarity between the load demand curves of the current day and the historical daily load demand curves is recorded as the first similarity; the number of the first similarities with outliers greater than a preset first threshold is recorded as the first quantity; the referenceability of the similarity between the load data fitting curves of the current day and the historical dates is negatively correlated with the first similarity and the first quantity, respectively.
[0008] In one embodiment, the process of obtaining the load prediction reference value is as follows: Among all historical dates whose similarity to the load demand curve of the current day is less than or equal to a preset first threshold, obtain the load data fusion value of the next moment of the corresponding moment of all days at the current moment, and use it as the load prediction reference value for the current moment.
[0009] In one embodiment, the process of obtaining the initial load prediction value is as follows: The load data of all times within a preset time period before the current time is used as the input of the time prediction algorithm, and the output is the predicted load data value of the next time of the current time, which is used as the initial load prediction value of the next time of the current time.
[0010] In one embodiment, the process of obtaining the enhancement coefficient is as follows: Among all historical dates whose similarity to the load demand curve of the current day is less than or equal to a preset first threshold, a predictable time period is determined based on the degree of disorder of the load data at the same moment. The enhancement coefficient of the initial load forecast value of the time outside the predictable time period is set to a first preset value, and the enhancement coefficient of the initial load forecast value of the time belonging to the predictable time period is set to a second preset value.
[0011] In one embodiment, the process of obtaining the predictable time period is as follows: The degree of disorder of the load data at the same moment is recorded as the first disorder. The first disorder of all moments is clustered. The time period obtained by connecting the moments that are in the cluster with the smallest mean of the first disorder is recorded as the predictable time period.
[0012] In one embodiment, the difference between the first preset value and the second preset value is 1.
[0013] In one embodiment, the process of obtaining the first predicted value at each time point is as follows: For all historical dates whose similarity to the load demand curve of the current day is less than or equal to a preset first threshold, all load data at the same time are used as input to the time series forecasting algorithm to obtain the predicted value at each time, which is denoted as the first predicted value.
[0014] In one embodiment, the process of obtaining the load demand forecast value is as follows: The normalized value of the enhancement coefficient is used as the weight of the initial load prediction value, and the normalized value of the reference value is used as the weight of the load prediction reference value; the normalized value of the absolute value of the difference between the enhancement coefficient and the first preset value is used as the weight of the first prediction value; the weighted fusion value of the initial load prediction value, the load prediction reference value and the first prediction value at each time is used as the load demand prediction value at each time.
[0015] This application has at least the following beneficial effects: This application analyzes the similarity between historical daily load data and introduces an anomaly detection algorithm to assess the degree of anomaly in the similarity of historical curves. This determines the reliability of predicted values obtained from historical data at the same time in previous days, effectively identifying "spurious similarities" and avoiding erroneous references due to accidental matching, thus improving the stability of the prediction model. The initial load prediction value for the next time moment is determined by the trend of data changes before each time point, and the first predicted value for each time moment is predicted using the load data at the same time in previous days. The predictability of data at each time point is assessed by analyzing the regularity of historical data at each time point, adaptively determining the enhancement coefficient of the initial load prediction value and the availability of the first predicted value, achieving differentiated prediction. It relies on historical patterns in stable periods and strengthens real-time response in abrupt changes, preventing over-reliance on historical trends and resulting lag. Through multi-source information complementarity, the final load demand prediction value is determined, significantly improving the accuracy of virtual power plant load prediction. This application not only considers the trend and periodicity of time series but also introduces historical similarity analysis, anomaly detection, and local dynamic modeling, ultimately achieving high-precision and robust load prediction and supporting subsequent precise regulation. Attached Figure Description
[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of 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 only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 A flowchart of a multi-objective load balancing coordination control method for a distributed virtual power plant provided in this application; Figure 2 This is a schematic diagram of the acquisition process for reference. Detailed Implementation
[0018] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a multi-objective load balancing coordination control method for a distributed virtual power plant proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the multi-objective load balancing coordination control method for a distributed virtual power plant provided in this application.
[0021] One embodiment of this application provides a multi-objective load balancing coordination control method for a distributed virtual power plant.
[0022] Specifically, a multi-objective load balancing coordination control method for a distributed virtual power plant is provided below. Please refer to [link / reference]. Figure 1 The method includes the following steps: Step S1: Collect load data of the virtual power plant at each time point.
[0023] In order to better coordinate and ensure the load balance of the virtual power plant, it is also necessary to monitor and analyze the power plant load. Therefore, the load demand of the virtual power plant is continuously collected and monitored.
[0024] The obtained virtual power plant load data undergoes preprocessing operations such as noise reduction. Simultaneously, the virtual power plant load data is acquired every minute, and the load in the virtual power plant is continuously monitored 24 hours a day. It should be noted that the implementer can set the load data acquisition frequency according to actual conditions; this application does not impose specific restrictions.
[0025] Step S2: Based on the similarity of load data fitting curves between historical adjacent days, and combined with an anomaly detection algorithm, analyze the outliers of each similarity; combine the numerical distribution of all similarities and all outliers to analyze the referenceability of the similarity between the load data fitting curves of the current day and historical dates; determine the load prediction reference value for each time based on the load data of each time at the same time on historical days.
[0026] (1) Considering the specific daily virtual power plant load demand, and combining the historical daily virtual power plant load demand data for analysis, we can find daily virtual power plant load demand curves with similar trends based on the similarity of the load demand change curves. Then, we can obtain the reference value for the virtual power plant load demand forecast at the next moment based on the historical virtual power plant load demand curves, specifically: First, the load data within the previous 24 hours is used as the daily load data for the current moment. Then, the load data within the previous 24 hours of the current moment is used as the previous day's load data. This process is repeated to obtain the historical daily load data for the current moment.
[0027] The fitted curve obtained by fitting all daily load data using a fitting algorithm is used as the daily load demand curve. The load demand curve obtained using the current day's load data is used as the current day's load demand curve; similarly, the load demand curve obtained using the historical daily load data at the current moment is used as the historical daily load demand curve at the current moment.
[0028] The similarity between the load demand curves of a virtual power plant for any two days is obtained. A higher similarity between the current day's load demand curve and historical daily load demand curves indicates a greater ability to predict the current day's virtual power plant load based on historical daily virtual power plant load demand curves. In this embodiment, the similarity between the load demand curves is the absolute value of the Pearson correlation coefficient between them; a larger absolute value indicates a higher similarity between the two curves. It should be noted that this application only provides one method for calculating the similarity between daily load demand curves. Many existing similarity calculation methods exist, and implementers can use other similarity algorithms to calculate the similarity between daily load demand curves. This application does not impose specific limitations.
[0029] (2) At the same time, the reliability of the virtual power plant load demand curve on the day is analyzed. That is, in the real-time analysis process, there may be special times that cause the load demand curve on the day to be dissimilar to the historical daily load demand curve. According to correlation analysis algorithms such as Pearson correlation coefficient, it can also be found that the load demand curve on the day is correlated with the historical daily load demand curve. However, these correlations may be due to the characteristics of the load demand curve itself and cannot distinguish more detailed similarity features, resulting in low reference value of the prediction results obtained based on this correlation.
[0030] Therefore, this application needs to analyze the reliability of the similarity between the virtual power plant's daily load demand curve and historical daily load demand curves, specifically: The process first involves obtaining the load demand curves for the most recent N days prior to the current moment. Preferably, in this embodiment, the value of N is set to 30. In other embodiments, the implementer may set the value of N according to actual circumstances.
[0031] The similarity between any two load demand curves within these N days is obtained and used as input to the LOF anomaly detection algorithm. The output is the outlier value of the similarity between the load demand curves of any two days. The LOF anomaly detection algorithm is a well-known technology, and its specific process will not be elaborated further. It should be noted that this application only provides one similarity calculation method for anomaly detection between any two load demand curves. Many existing similarity calculation methods exist, and implementers may use other similarity algorithms to calculate outliers; this application does not impose any specific restrictions.
[0032] Instead of simply selecting a virtual power plant daily load demand curve with relatively high similarity as a reference to obtain a predicted reference value, the similarity between the current day's load demand curve and historical daily load demand curves is determined by anomaly detection.
[0033] The similarity between the current day's load demand curve and the historical daily load demand curve is recorded as the first similarity. All outliers of the first similarity are counted. The larger the outlier, the more unique the virtual power plant load demand curve obtained on the current day may be. The lower the reliability of the predicted reference value obtained from the relatively similar historical daily load demand curve.
[0034] The similarity between the current day's load demand curve and historical load demand curves can be used as a reference, and the specific calculation is as follows: In the formula, The reference value is indicated by the similarity between the load demand curves of the current day and historical dates; N represents the number of days selected before the current time, which is N=30 in this application; s represents the number of outliers in the similarity between the load demand curves of the current day and all historical days in the N days before the current time that are greater than a preset first threshold, which is denoted as the first number. This represents an outlier indicating the similarity between the load demand curve of the current day and the historical load demand curve of day i. Preferably, in this embodiment, the first threshold is set to 1. As other embodiments of this application, implementers can set the first threshold according to actual conditions.
[0035] The smaller the average abnormality of the similarity between the current day's virtual power plant load demand curve and the historical daily virtual power plant load demand curve, and the fewer the number of abnormalities in the similarity between the current day's virtual power plant load demand curve and the historical virtual power plant load demand curve that exceed the threshold, the more reliable the current day's virtual power plant load demand curve is for obtaining a predictive reference value by referring to the historical daily load demand curve.
[0036] (3) The expression for the load prediction reference value for the next time step is: In the formula, The reference value for load forecasting the next moment is represented by the current moment; Ns represents the number of outliers less than or equal to the first threshold in the similarity between the load demand curves of the current day and all historical days in the N days prior to the current moment, which is denoted as the second number. This represents the load data for the next moment on day i, within all historical dates where the similarity to the load demand curve of the current day is less than or equal to a preset first threshold.
[0037] Step S3: Based on the load data within a preset time period before each time point, predict the initial load forecast value for the next time point; determine the enhancement coefficient of the initial load forecast value for each time point based on the disorder of the load data at the same time point of each historical day; use the prediction algorithm to predict the load data at the same time point of each historical day to obtain the first prediction value for each time point.
[0038] During the similarity calculation between virtual power plant load demand curves, the similarity of load change trends in most time periods with small fluctuations may lead to similarity between the calculated two load demand curves. However, there may be short time periods with large fluctuations that contribute little to the similarity analysis results. The load demand curves corresponding to these time periods should perhaps be analyzed based on the load change trends in the real-time curves rather than on historical load demand curves. Therefore, it is necessary to identify and suppress these periods of large fluctuations in load demand or correct the periods of small fluctuations in load demand to prevent over-reliance on historical trends from affecting the reliability of the actual prediction results.
[0039] This embodiment inputs the daily load data at the current moment, i.e., the load data within the 24 hours prior to the current moment, into the time prediction algorithm, and outputs the predicted load data for the next moment, denoted as the initial load prediction value Po for the next moment. In this embodiment, the time prediction algorithm used here and thereafter is the exponential smoothing algorithm. It should be noted that this application only provides one time prediction algorithm; many existing time prediction algorithms exist, and implementers may use other time prediction algorithms for data prediction. This application does not impose specific limitations.
[0040] This application further divides the daily load demand curve, as follows: Each sampling time point within the most recent day before the current moment is obtained. In this embodiment, the number of sampling time points within a day is 24*60. Among all historical dates whose similarity to the daily load demand curve is less than or equal to a preset first threshold, the variance of all load data at the same time point is calculated to assess the degree of disorder of all load data at the same time point, denoted as the first disorder. Then, the first disorder of all time points is clustered to obtain clusters. The cluster with the smallest mean first disorder is selected, and the time interval obtained by connecting the consecutive moments in this cluster is denoted as the predictable time interval. The smaller the variance of the historical virtual power plant load demand at the same moment within these time intervals, the more predictable the load is. Therefore, the moments corresponding to these time intervals can be used as a reference for analysis. The enhancement coefficient of the initial load prediction value for moments outside the predictable time interval is set to a first preset value. Conversely, the enhancement factor for the initial load forecast value at times belonging to the predictable time period is set to the second preset value. This is used for the adaptive addition of subsequent first predicted values and the weighting of initial load predicted values. In this embodiment of the application, For a small positive number greater than 1, in the embodiments of this application, In other embodiments of this application, the implementer may set the parameters according to the actual situation. The value of .
[0041] Meanwhile, considering that when predicting load demand based on a 24*60 data segment, segments with large fluctuations may affect the reliability of the prediction results, this application further determines whether the current moment is within a predictable time period. If the current moment is within a predictable time period, it means that existing time series prediction algorithms can be used to predict based on this sequence segment, and more reliable prediction results can be obtained. Specifically, among all historical dates whose similarity to the load demand curve of the current day is less than or equal to a preset first threshold, all historical load data that are at the same time as the next time of the current moment are input into the time series prediction algorithm, and the predicted value of the next time of the current moment is denoted as the first predicted value Tg. Wherein, if the current moment is within a predictable time period, the first predicted value needs to be obtained; otherwise, if the current moment is not within a predictable time period, its first predicted value does not need to be obtained.
[0042] Step S4: Based on the enhancement coefficient and the referenceability, determine the weights of the initial load prediction value, the load prediction reference value, and the first prediction value, calculate the load demand prediction value at each time point, and use an optimization algorithm to regulate the virtual power plant load.
[0043] Based on the above information, the power plant load demand obtained at the next time step is corrected. The specific calculation method is as follows: In the formula, This represents the final virtual power plant load demand forecast for the next time step from the current time step, where U represents the enhancement factor of the initial forecast result for the next time step from the current time step. This represents the initial load forecast value for the next time step from the current time step. This indicates the reliability of the virtual power plant load curve for the current day compared to historical daily load-demand curves. This represents the load forecast reference value for the next time step from the current time step. This represents the first predicted value for the next moment from the current moment. Specifically, through the adaptive selection of U, when a first predicted value exists, it is included in the calculation of the prediction correction value; when a first predicted value does not exist, it is not necessary to include it in the calculation.
[0044] The stronger the reference value, the greater the weight of the predicted value obtained based on historical data at the same time should be; the greater the enhancement coefficient of the initial prediction result, the greater the weight of the predicted value predicted based on changes in data at nearby times should be.
[0045] The predicted load demand of the virtual power plant at the next moment is obtained through the above prediction method. Based on the obtained predicted load demand, the load value of each power load in the virtual power plant is adjusted by an optimization algorithm. In this embodiment, the power load in the virtual power plant includes thermal power (diesel power generation), photovoltaic power (photovoltaic power generation), wind power (wind power generation) and energy storage system. In actual application scenarios, the implementation of setting the multi-target power load of the virtual power plant is determined by the implementer according to the actual situation.
[0046] Preferably, in this embodiment, the SGA genetic algorithm is used to regulate the load values of each power source load in the virtual power plant load. In this embodiment, the objective function and constraints of the SGA algorithm are as follows: First, a multi-objective load balancing objective function is constructed. Preferably, in this embodiment, the objective function of the SGA algorithm is: in, The price at which electricity is purchased from or sold to the power grid; This represents the power exchanged with the power grid (positive for purchasing electricity, negative for selling electricity). Powering the photovoltaic system; The operation and maintenance cost per unit of electricity generated by a photovoltaic system; For the output of wind power generation; The operation and maintenance cost per unit of electricity generated by wind power generation; The output of the energy storage system (positive value indicates discharge, negative value indicates charging). Maintenance costs for energy storage systems; The unit power generation cost of a diesel generator; This is for the output of the diesel generator.
[0047] Furthermore, to ensure that the power plant's load requirements are met, constraints are constructed based on the output capacity of each power source load. The constraints are as follows: in, For the output of the diesel generator; Powering the photovoltaic system; For the output of wind power generation; This represents the output of the energy storage system (positive values indicate discharging, and negative values indicate charging).
[0048] Based on the objective function and corresponding constraints, the SGA algorithm is used to obtain the load value for each power system at the next time step. Adjustment is then performed based on the corresponding load value.
[0049] It should be noted that in the actual virtual power plant load value regulation, implementers can also use other existing optimization algorithms to regulate the load value of each power source load in the virtual power plant. The construction of objective functions and constraints are all existing technologies. In actual application, implementers can set the objective functions and constraints of the optimization algorithm themselves. This embodiment does not impose any special restrictions on this.
[0050] A schematic diagram of the acquisition process for reference is shown below. Figure 2 As shown.
[0051] In summary, this application's embodiments analyze the similarity between historical daily load data, introduce anomaly detection algorithms to assess the degree of anomaly in historical curve similarity, and determine the reference value of predicted values obtained from historical data at the same time in previous days. This effectively identifies "spurious similarity" situations, avoids erroneous references caused by accidental matching, and improves the stability of the prediction model. The initial load prediction value for the next time moment is determined by the data change trends before each time point, and the first prediction value for each time moment is predicted using the load data at the same time in previous days. By analyzing the regularity of historical data at each time point, the predictability of the data at each time moment is assessed, and the enhancement coefficient of the initial load prediction value and the availability of the first prediction value are adaptively determined, achieving differentiated prediction. It relies on historical patterns in stable periods and strengthens real-time response in abrupt changes, preventing over-reliance on historical trends and resulting lag. Through multi-source information complementarity, the final load demand prediction value is determined, significantly improving the accuracy of virtual power plant load prediction. This application not only considers the trend and periodicity of time series but also introduces historical similarity analysis, anomaly detection, and local dynamic modeling, ultimately achieving high-precision and highly robust load prediction and supporting subsequent precise regulation.
[0052] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this application. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.
[0053] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0054] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them; modifications to the technical solutions described in the foregoing embodiments, or equivalent substitutions of some of the technical features, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A multi-objective load balancing coordination control method for a distributed virtual power plant, characterized in that, The method includes the following steps: Collect load data of the virtual power plant at various times; Based on the similarity of load data fitting curves between historical adjacent days, and combined with anomaly detection algorithms, outliers of each similarity are analyzed; by combining the numerical distribution of all similarities and all outliers, the referenceability of the similarity between the load data fitting curves of the current day and historical dates is analyzed; and load prediction reference values for each time point are determined based on load data at the same time point on historical days. The initial load prediction value for the next moment is obtained by predicting the load data within a preset time period before each moment; the enhancement coefficient of the initial load prediction value for each moment is determined based on the disorder of the load data at the same moment of each historical day; the first prediction value for each moment is obtained by using the prediction algorithm to predict the load data at the same moment of each historical day. Based on the enhancement coefficient and the reference value, the weights of the initial load forecast value, the load forecast reference value, and the first forecast value are determined, and the load demand forecast value at each time point is calculated so as to use an optimization algorithm to regulate the virtual power plant load.
2. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 1, characterized in that, The process for obtaining the outlier is as follows: The fitted curve of the daily load data is denoted as the daily load demand curve; the similarity between any two load demand curves within a preset number of days before the current time is used as the input of the anomaly detection algorithm, and the output is the anomaly value of the similarity between the load demand curves of each of the aforementioned two arbitrary days.
3. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 2, characterized in that, The process for obtaining the referenceability is as follows: Within a preset number of days prior to the current moment, the similarity between the load demand curves of the current day and the historical daily load demand curves is recorded as the first similarity; the number of the first similarities with outliers greater than a preset first threshold is recorded as the first quantity; the referenceability of the similarity between the load data fitting curves of the current day and the historical dates is negatively correlated with the first similarity and the first quantity, respectively.
4. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 2, characterized in that, The process for obtaining the load prediction reference value is as follows: Among all historical dates whose similarity to the load demand curve of the current day is less than or equal to a preset first threshold, obtain the load data fusion value of the next moment of the corresponding moment of all days at the current moment, and use it as the load prediction reference value for the current moment.
5. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 1, characterized in that, The process for obtaining the initial load prediction value is as follows: The load data of all times within a preset time period before the current time is used as the input of the time prediction algorithm, and the output is the predicted load data value of the next time of the current time, which is used as the initial load prediction value of the next time of the current time.
6. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 2, characterized in that, The process of obtaining the enhancement coefficient is as follows: Among all historical dates whose similarity to the load demand curve of the current day is less than or equal to a preset first threshold, a predictable time period is determined based on the degree of disorder of the load data at the same moment. The enhancement coefficient of the initial load forecast value of the time outside the predictable time period is set to a first preset value, and the enhancement coefficient of the initial load forecast value of the time belonging to the predictable time period is set to a second preset value.
7. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 6, characterized in that, The process of obtaining the predictable time period is as follows: The degree of disorder of the load data at the same moment is recorded as the first disorder. The first disorder of all moments is clustered. The time period obtained by connecting the moments that are in the cluster with the smallest mean of the first disorder is recorded as the predictable time period.
8. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 6, characterized in that, The difference between the first preset value and the second preset value is 1.
9. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 2, characterized in that, The process for obtaining the first predicted value at each time point is as follows: For all historical dates whose similarity to the load demand curve of the current day is less than or equal to a preset first threshold, all load data at the same time are used as input to the time series forecasting algorithm to obtain the predicted value at each time, which is denoted as the first predicted value.
10. The multi-objective load balancing coordination control method for a distributed virtual power plant as described in claim 6, characterized in that, The process for obtaining the predicted load demand value is as follows: The normalized value of the enhancement coefficient is used as the weight of the initial load prediction value, and the normalized value of the referenceability is used as the weight of the load prediction reference value. The normalized value of the absolute value of the difference between the enhancement coefficient and the first preset value is used as the weight of the first predicted value; The weighted sum of the initial load forecast, the load forecast reference value, and the first forecast value at each time point is used as the load demand forecast value at each time point.