Medium and long term transaction load prediction method and system based on virtual power plant

Abstract

The invention relates to the technical field of data processing, in particular to a medium and long term transaction load prediction method and system based on a virtual power plant. The method comprises the following steps: acquiring load data at the current moment in real time by using a virtual power plant; determining a first updating probability and a second updating probability of the load data at the current moment; the mean value of the first updating probability and the second updating probability is used as the degree needing to be adjusted at the current moment; according to the degree needing to be adjusted, the updating frequency adjustment amplitude at the current moment is obtained; and performing parameter updating on the ARIMA model according to the updating frequency adjustment amplitude, and predicting the future load. According to the method, the first updating probability and the second updating probability of the load data at the current moment are determined, the degree needing to be adjusted is calculated according to the first updating probability and the second updating probability, then the updating frequency adjusting amplitude is obtained, the parameter updating frequency of the ARIMA model can be adjusted in a self-adaptive mode according to the change condition of the real-time load data, and the defects caused by fixed updating frequency are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular, to a medium- and long-term trading load forecasting method and system based on a virtual power plant. Background Art

[0002] In the context of the continuous development of the current power market, as an emerging and crucial market participant, the effective operation of a virtual power plant highly depends on accurate load forecasting. Through accurate load forecasting, the virtual power plant can keenly identify and deeply evaluate various risks latent in the power market. For example, in terms of power supply and demand, when the load forecasting is accurate, the virtual power plant can anticipate potential supply-demand imbalances in advance, timely adjust its power generation or power consumption strategies, and avoid problems such as resource waste or supply shortages caused by supply-demand mismatches. At the price level, load forecasting helps the virtual power plant predict price fluctuation trends, seize market opportunities, participate in power trading at more favorable prices, and thus enhance its economic benefits.

[0003] The Autoregressive Integrated Moving Average (ARIMA) model is a statistical method for analyzing and forecasting time series data. The ARIMA model combines three parts: autoregression (AR), differencing (I), and moving average (MA). Among them, the autoregressive part assumes a linear relationship between the current value and several past values. The differencing part is used to handle non-stationarity in the time series to make the series stable. The moving average part corrects the current forecast through past forecast errors. The ARIMA model can capture trends, periodicity, and random fluctuations in time series, and is especially good at dealing with non-seasonal, stationary time series. By fitting a suitable ARIMA model and combining the medium- and long-term trading demands of the virtual power plant, the load for the next few weeks, months, or even longer can be predicted.

[0004] However, the ARIMA model itself relies on historical values of time series data for forecasting. When facing newly input data, such as the real-time changing load data of a virtual power plant, if the fluctuations of these data are large and their trend and seasonal characteristics are difficult to accurately capture by conventional means, the model may generate large forecasting errors. Especially in the case of relatively large short-term fluctuations, the model may be overly sensitive to these short-term changes, resulting in large deviations in the forecasting results. Since in a virtual power plant, the load data is changing in real time, if the parameter update frequency of the ARIMA model is not high enough, it may not be able to adapt to data changes in a timely manner, thus affecting the forecasting accuracy. If the parameter update frequency of the ARIMA model is too high, it will increase the computational cost and resource consumption, and may also lead to overfitting of the model, which also affects the forecasting effect. Summary of the invention

[0005] In order to solve the problem that in a virtual power plant, load data changes in real time, if the parameter update frequency of the ARIMA model is not high enough, it may not be able to adapt to data changes in time, thereby affecting the accuracy of the prediction; if the parameter update frequency of the ARIMA model is too high, it will increase the computing cost and resource consumption, and may also cause model overfitting, which also affects the prediction effect, the present invention provides a medium- and long-term trading load prediction method and system based on a virtual power plant.

[0006] In a first aspect, the present invention provides a medium- and long-term transaction load forecasting method based on a virtual power plant, which adopts the following technical solution: A medium- and long-term trading load forecasting method based on a virtual power plant, comprising: using a virtual power plant to obtain load data at the current moment in real time; determining a first update probability and a second update probability of the load data at the current moment; the first update probability is used to characterize the data changes and abnormal fluctuations of the load data at the current moment in the time period to which the current moment belongs; the second update probability is used to characterize the data fluctuations of the load data at the current moment in the time period to which the current moment belongs, and the change pattern and abnormal degree of the data fluctuations of the load data at the same moment in the first history in the corresponding historical time period; taking the mean of the first update probability and the second update probability as the required adjustment degree at the current moment; obtaining the update frequency adjustment amplitude at the current moment according to the size of the required adjustment degree; inputting the update frequency adjustment amplitude into an ARIMA model to obtain an adjusted update frequency, using the adjusted update frequency to update the parameters of the ARIMA model, and then predicting the future load.

[0007] The beneficial effects are: dynamically adjusting the update frequency of the ARIMA model according to the "degree of adjustment required" of the load data, balancing the model adaptability and computing cost, and avoiding the negative impact of too low or too high update frequency; comprehensively judging the load changes and improving the prediction accuracy through the first update probability (reflecting the data changes in the current period) and the second update probability (reflecting the fluctuation difference with the historical data); the dynamic adjustment mechanism can respond to the real-time changes of the load data in a timely manner to meet the real-time requirements of the virtual power plant; by dynamically adjusting the update frequency, the model can adapt to the medium- and long-term load change trends and maintain a higher prediction accuracy.

[0008] Furthermore, the load data is load data after filtering and denoising.

[0009] Furthermore, the filtering and denoising adopts an adaptive filtering algorithm.

[0010] Furthermore, the first update probability satisfies: ; In the formula, is the first update probability of the load data at the current moment, is the number of preset lookback windows at the current moment, is the load data at the current moment, For the current moment The average load data at all times within the lookback window, For the current moment The variance of the load data at all times within the lookback window, is the slope of the load data at the current moment, is the load data of the moment before the current moment, For the current moment The fitted slope of the lookback window, is the maximum and minimum normalized function, is the absolute value symbol.

[0011] The beneficial effects are: combining the current load data with the mean, variance and slope changes within the look-back window to comprehensively capture load dynamics and abnormal fluctuations; using neighboring data to evaluate recent trends and fluctuations to enhance model robustness; introducing slope changes to increase sensitivity to short-term trends and abnormal fluctuations; standardizing calculation results through maximum and minimum normalization functions to avoid dimensional influence and improve versatility; dynamically adapting to changes in load data, suitable for scenarios with strong real-time performance and frequent fluctuations in virtual power plants.

[0012] Furthermore, the second update probability satisfies: ; In the formula, is the second update probability of the load data at the current moment, is the standard deviation of the load data at all times in the period to which the current time belongs, is the standard deviation of the load data at all times in the historical period corresponding to the first historical moment at the current moment, is the fitting slope of the time period at the current moment, is the fitting slope of the historical period corresponding to the first historical moment at the current moment, is the maximum and minimum normalized function, is the absolute value symbol.

[0013] The beneficial effects are: by comparing the standard deviation and fitting slope of the time period to which the current moment belongs with the historical time period corresponding to the same historical moment of the current moment, the fluctuation pattern and trend change of load data can be quantified, and abnormal fluctuations can be identified; by combining historical data comparison, the limitations of relying solely on current data can be avoided, and the model's adaptability to load changes can be improved; through the maximum and minimum normalization functions, the calculation results can be standardized to ensure the comparability of data of different dimensions.

[0014] Further, the historical period is from the first historical same moment of the current moment to the period before the period to which the current moment belongs.

[0015] Further, the method for obtaining the fitting slope is as follows: Using the least squares method to fit the load data at all moments within the backtracking window / the period to which the current moment belongs / the historical period, so as to obtain the fitting slope of the backtracking window / the period to which the current moment belongs / the historical period.

[0016] Further, obtaining the adjustment amplitude of the update frequency at the current moment includes: in response to the degree to be adjusted being not less than a preset adjustment threshold, calculating the adjustment amplitude of the update frequency at the current moment based on a preset increased value of the adjustment amplitude and the difference between the degree to be adjusted and the preset adjustment threshold; otherwise, not calculating the adjustment amplitude of the update frequency at the current moment.

[0017] The beneficial effects are as follows: Calculating the adjustment amplitude only when the degree to be adjusted exceeds the preset threshold, avoiding unnecessary frequent updates, and reducing the calculation cost; Dynamically adjusting the update frequency according to the difference between the degree to be adjusted and the threshold, ensuring that the model can respond to significant load changes in a timely manner, and improving the prediction accuracy; Reducing redundant calculations through threshold judgment and improving the algorithm efficiency.

[0018] Further, the adjustment amplitude of the update frequency satisfies: ; where is the adjustment amplitude of the update frequency at the current moment, is the preset increased value of the adjustment amplitude, is the degree to be adjusted at the current moment, is the preset adjustment threshold.

[0019] The beneficial effects are as follows: Dynamically adjusting the update frequency according to the difference between the degree to be adjusted and the adjustment threshold, ensuring that the model can be updated in a timely manner when the load changes significantly, and improving the prediction accuracy; Flexibly controlling the change intensity of the update frequency through the preset increased value of the adjustment amplitude, and balancing the adaptability of the model and the calculation cost.

[0020] In the second aspect, the present invention provides a medium - and long - term trading load forecasting system based on a virtual power plant, adopting the following technical solution: A medium - and long - term trading load forecasting system based on a virtual power plant includes: a processor and a memory, where the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above - mentioned medium - and long - term trading load forecasting method based on a virtual power plant is implemented.

[0021] By adopting the above technical solution, a computer program is generated from the above medium- and long-term trading load forecasting method based on a virtual power plant and stored in a memory to be loaded and executed by a processor, thereby manufacturing a terminal device according to the memory and the processor for convenient use.

[0022] The present invention has the following technical effects: By determining the first update probability and the second update probability of the load data at the current moment, calculating the degree of adjustment required based on these probabilities, and then obtaining the adjustment amplitude of the update frequency, it is possible to adaptively adjust the parameter update frequency of the ARIMA model according to the changes in real-time load data, avoiding the drawbacks brought by a fixed update frequency. It will neither be unable to adapt to data changes in a timely manner and affect the prediction accuracy due to too low an update frequency, nor will it lead to an increase in computational costs, excessive resource consumption, and overfitting problems due to too high an update frequency. Thus, the accuracy and stability of the prediction are effectively improved; the first update probability focuses on the data changes and abnormal fluctuations of the load data at the current moment within the corresponding time period, and the second update probability considers the data fluctuation change pattern and abnormal degree of the load data at the current moment compared with the load data at the same historical moment in the corresponding historical period, comprehensively considering data characteristics from multiple dimensions, making the analysis of the load data more comprehensive, and providing rich and accurate basis for more reasonable adjustment of the ARIMA model parameter update frequency; while ensuring the prediction accuracy, unnecessary high-frequency parameter updates are avoided, reducing computational costs and resource consumption. This dynamic adjustment mechanism based on the actual situation of the data realizes the optimal utilization of resources, helps the virtual power plant operate more efficiently in medium- and long-term trading load forecasting, and improves economic benefits. Description of the Drawings

[0023] Figure 1 It is the flowchart of the method in an embodiment of the medium- and long-term trading load forecasting method based on a virtual power plant of the present invention. Detailed Embodiments

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0025] An embodiment of the present invention discloses a medium- and long-term trading load forecasting method based on a virtual power plant, with reference to Figure 1 , including steps S1 - S5: S1: Use the virtual power plant to obtain the load data at the current moment in real time.

[0026] It should be noted that since the load data is affected by various noises during the acquisition process, such as sensor errors, transmission interference, etc.; and in actual applications, there may be missing values in the data; if not processed, it may affect the accuracy of the model. Therefore, it is necessary to preprocess the acquired load data to improve the quality of the load data.

[0027] Specifically, the load data is the load data after filtering and denoising.

[0028] Specifically, the filtering and denoising adopt an adaptive filtering algorithm.

[0029] In another embodiment, the mean interpolation method is used for the missing values in the acquired load data. Since it is real-time acquisition, when using the mean interpolation method for interpolation, only the load data at the previous moment before the current moment is referred to for interpolation.

[0030] The implementer can set the acquisition frequency according to the specific implementation situation. For example, it is acquired once every minute, and it is ensured that the load data at the historical moment and the current moment are aligned according to the same time unit.

[0031] S2: Determine the first update probability and the second update probability of the load data at the current moment.

[0032] It should be noted that in a virtual power plant, the load data may be affected by emergencies (such as equipment failures, sudden weather changes), and these emergencies will cause abnormal fluctuations in the data. Through the first update probability of the load data at the current moment, it can not only reflect the local fluctuation characteristics of the data, but also help the model capture short-term data changes and abnormal fluctuations, enabling the model to more accurately predict short-term load changes and reduce prediction errors; by analyzing the trend similarity of the load data within multiple consecutive retrospective windows, it can reflect the long-term trend characteristics of the data, helping the model identify the long-term trend and periodic changes of the data, thereby improving the stability of long-term prediction.

[0033] The first update probability is used to characterize the data changes and abnormal fluctuations of the load data at the current moment within the time period to which the current moment belongs.

[0034] Specifically, the first update probability satisfies: ; In the formula, is the first update probability of the load data at the current moment, is the number of preset retrospective windows at the current moment, is the load data at the current moment, is the th average load data of all moments within the retrospective window at the current moment, is the variance of the load data at all times within the th backtracking window at the current moment, is the slope of the load data at the current moment, is the load data at the previous moment of the current moment, is the fitting slope of the th backtracking window at the current moment, is the maximum-minimum normalization function, is the absolute value symbol.

[0035] The backtracking window is a data sequence arranged in time series. The last data within the backtracking window is the load data at the current moment. The longest backtracking window is the same as the time period to which the current moment belongs. Implementers can set the duration and quantity of the backtracking window according to the specific implementation situation. For example, the duration is 15 minutes, 30 minutes, 45 minutes, 60 minutes; the quantity is 4.

[0036] Among them, represents the sum of the differences between the load data at the current moment and the average value of the load data within the backtracking window. Using to take its average value, and its average value can reflect the local fluctuation degree of the load data at the current moment; the greater the difference between the load data at the current moment and the average value of all load data within each backtracking window, the greater the local fluctuation degree of the load data at the current moment, and vice versa; represents the sum of the variances of the corresponding all load data within each backtracking window. Using to take its average value, the greater the average value of the variances of all load data, the greater the overall fluctuation degree of the load data at the current moment, and vice versa; represents the average value of the fitting slopes of all backtracking windows, reflecting the overall fluctuation trend of all load data within multiple backtracking windows; represents the slope at the position of the load data at the current moment, reflecting the local fluctuation trend of the load data at the current moment. When the difference between them is greater, it indicates that the fluctuation trend of the load data at the current moment is more inconsistent with the overall fluctuation trend. Therefore, the greater the data change, the more likely there is an anomaly.

[0037] It should be noted that by comparing the load data of the time period to which the current moment belongs with the load data of the corresponding historical time period of the first historical same moment, the second update probability can identify the data change pattern and anomaly degree; if the current data is similar to the historical data, the update probability can be reduced to reduce unnecessary parameter adjustment. By reducing the unnecessary parameter update frequency, the model can process data more efficiently and reduce the consumption of computing resources.

[0038] The second update probability is used to characterize the data fluctuation of the load data at the current moment in the corresponding time period, compared with the change pattern and abnormal degree of the data fluctuation of the load data at the same historical moment in the first corresponding historical time period.

[0039] Specifically, the second update probability satisfies: ; In the formula, is the second update probability of the load data at the current moment, is the standard deviation of the load data at all moments within the time period to which the current moment belongs, is the standard deviation of the load data at all moments within the historical time period corresponding to the first historical same moment of the current moment, is the fitting slope of the time period to which the current moment belongs, is the fitting slope of the historical time period corresponding to the first historical same moment of the current moment, is the maximum-minimum normalization function, is the absolute value symbol.

[0040] Among them, represents the difference in the degree of data fluctuation of the load data at all moments within the time period to which the current moment belongs and the degree of data fluctuation of the load data at all moments within the historical time period; represents the difference in the fluctuation trend of the load data at all moments within the time period to which the current moment belongs and the fluctuation trend of the load data at all moments within the historical time period; when the above two differences are larger, it indicates that the degree of data fluctuation and the fluctuation trend of the load data at the current moment are less similar to the degree of data fluctuation and the fluctuation trend of the historical time period. Therefore, the second update probability of the load data at the current moment should be larger, and vice versa.

[0041] Specifically, the historical time period is from the first historical same moment of the current moment to the time period before the time period to which the current moment belongs.

[0042] Specifically, the method for obtaining the fitting slope is: Use the least squares method to fit the load data at all moments within the backtracking window / the time period to which the current moment belongs / the historical time period to obtain the fitting slope of the backtracking window / the time period to which the current moment belongs / the historical time period.

[0043] S3: Take the mean of the first update probability and the second update probability as the degree of adjustment required at the current moment.

[0044] It should be noted that in a virtual power plant, the load data changes in real time and fluctuates greatly. Dynamically adjusting the update frequency can effectively solve the limitations of the ARIMA model in processing such data and improve the accuracy and reliability of prediction. According to the degree of adjustment required corresponding to the current moment, the model can dynamically adjust the amplitude of the update frequency. When the data changes slowly, the adjustment amplitude of the update frequency is reduced; when the data mutates, the adjustment amplitude of the update frequency is increased. If both the first update probability and the second update probability of the load data at the current moment are larger, it indicates that the mutation of the load data at the current moment is more obvious, and then the degree of adjustment required corresponding to the current moment is larger, and vice versa.

[0045] S4: Obtain the adjustment amplitude of the update frequency at the current moment according to the size of the degree of adjustment required.

[0046] It should be noted that after obtaining the degree of adjustment required at the current moment through the above steps, a adjustment function is defined, and this function determines the adjustment amplitude of the update frequency based on the degree of adjustment required.

[0047] Specifically, the obtaining of the adjustment amplitude of the update frequency at the current moment includes: In response to the degree of adjustment required being not less than a preset adjustment threshold, calculate the adjustment amplitude of the update frequency at the current moment based on a preset increased value of the adjustment amplitude and the difference between the degree of adjustment required and the preset adjustment threshold; otherwise, do not calculate the adjustment amplitude of the update frequency at the current moment.

[0048] Specifically, the adjustment amplitude of the update frequency satisfies: ; In the formula, is the adjustment amplitude of the update frequency at the current moment, is the preset increased value of the adjustment amplitude, is the degree of adjustment required at the current moment, is the preset adjustment threshold.

[0049] Implementers can set the adjustment threshold according to the specific implementation situation. For example, 0.7.

[0050] S5: Update the parameters of the ARIMA model according to the adjustment amplitude of the update frequency, and then predict the future load.

[0051] Input the adjustment amplitude of the update frequency into the ARIMA model to obtain the adjusted update frequency. Use the adjusted update frequency to update the parameters of the ARIMA model to obtain the updated ARIMA model, and use the updated ARIMA model to predict the future load.

[0052] An embodiment of the present invention also discloses a medium- and long-term trading load forecasting system based on a virtual power plant, which includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a medium- and long-term trading load forecasting method based on the present invention is implemented.

[0053] The above system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface. Their settings and functions are known in the art, so they will not be elaborated here.

[0054] The above are all preferred embodiments of the present invention. The protection scope of the present invention is not limited accordingly. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A medium- and long-term trading load forecasting method based on virtual power plants, characterized in that: include: Use virtual power plants to obtain current load data in real time; Determine a first update probability and a second update probability of the load data at the current moment; The first update probability is used to characterize the data change and abnormal fluctuation of the load data at the current moment within the time period to which the current moment belongs; The second update probability is used to characterize the data fluctuation of the load data at the current moment in the corresponding time period, compared with the change pattern and abnormality of the data fluctuation of the load data at the same moment in the first history in the corresponding historical time period; Taking the average of the first update probability and the second update probability as the degree of adjustment required at the current moment; According to the size of the required adjustment degree, obtaining the adjustment amplitude of the update frequency at the current moment; The update frequency adjustment amplitude is input into the ARIMA model to obtain the adjusted update frequency, the adjusted update frequency is used to update the parameters of the ARIMA model, and then the future load is predicted.

2. A medium- and long-term transaction load forecasting method based on a virtual power plant according to claim 1, characterized in that: The load data is the load data after filtering and denoising.

3. A medium- and long-term transaction load forecasting method based on a virtual power plant according to claim 2, characterized in that: The filtering and denoising adopts an adaptive filtering algorithm.

4. A medium- and long-term transaction load forecasting method based on a virtual power plant according to claim 1, characterized in that: The first update probability satisfies: ； In the formula, is the first update probability of the load data at the current moment, is the number of preset lookback windows at the current moment, is the load data at the current moment, For the current moment The average load data at all times within the lookback window, For the current moment The variance of the load data at all times within the lookback window, is the slope of the load data at the current moment, is the load data of the moment before the current moment, For the current moment The fitted slope of the lookback window, is the maximum and minimum normalized function, is the absolute value symbol.

5. The method for predicting medium- and long-term transaction load based on a virtual power plant according to claim 1 is characterized in that: The second update probability satisfies: ； In the formula, is the second update probability of the load data at the current moment, is the standard deviation of the load data at all times in the period to which the current time belongs, is the standard deviation of the load data at all times in the historical period corresponding to the first historical moment at the current moment, is the fitting slope of the time period at the current moment, is the fitting slope of the historical period corresponding to the first historical moment at the current moment, is the maximum and minimum normalized function, is the absolute value symbol.

6. A medium- and long-term transaction load forecasting method based on a virtual power plant according to claim 5, characterized in that: The historical time period is the time period from the first historical same moment of the current moment to the time period before the current moment.

7. A medium- and long-term transaction load forecasting method based on a virtual power plant according to claim 4 or 6, characterized in that: The fitting slope is obtained as follows: The least square method is used to fit the load data of all moments in the lookback window / the time period to which the current moment belongs / the historical time period, and the fitting slope of the lookback window / the time period to which the current moment belongs / the historical time period is obtained.

8. The method for predicting medium- and long-term transaction load based on a virtual power plant according to claim 1 is characterized in that: The obtaining of the update frequency adjustment amplitude at the current moment includes: In response to the required adjustment degree being not less than the preset adjustment threshold, the update frequency adjustment amplitude at the current moment is calculated based on the preset adjustment amplitude increase value and the difference between the required adjustment degree and the preset adjustment threshold; otherwise, the update frequency adjustment amplitude at the current moment is not calculated.

9. A medium- and long-term transaction load forecasting method based on a virtual power plant according to claim 8, characterized in that: The update frequency adjustment range satisfies: ； In the formula, The update frequency adjustment amplitude at the current moment, Adds a value to the preset adjustment. is the degree of adjustment required at the current moment, is the preset adjustment threshold.

10. A medium- and long-term trading load forecasting system based on virtual power plants, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a medium- and long-term trading load forecasting method based on a virtual power plant according to any one of claims 1-9 is implemented.

Images