Combined prediction method and system for electricity selling income

Abstract

The invention discloses a combined prediction method and system for electricity selling income, and the system comprises a data collection module, a model training module and an electricity selling prediction module, and the method comprises the steps: obtaining the historical electricity selling income data of a power supply enterprise in a target region and the corresponding electricity selling income influence factor data through the data collection module; historical electricity selling income data of a power supply enterprise is selectively corrected according to r value calculation of grey prediction, rolling prediction and weight combination are combined to obtain a time sequence prediction result, and a combined prediction model is connected with BP nerves in series to obtain a time sequence prediction result. And taking the time sequence and the influence factor sequence as input values to obtain a corresponding electricity selling income prediction result considering multiple influence factors. According to the method, the defects of gray prediction and BP nerve can be overcome, the electricity selling income change trend is reliably extracted from the electricity selling income time sequence data with the complex nonlinear relation, then the accuracy of the prediction result is guaranteed, and reliable guarantee is provided for normal social power supply.

Description

Technical Field

[0001] The present invention relates to the technical field of power data processing, and in particular to a combined prediction method and system for electricity sales revenue. Background Art

[0002] The electricity sales revenue forecast is an important decision-making basis for power companies to rationally plan power generation, power purchase and sales strategies. Accurate electricity sales revenue forecast can help companies rationally arrange production and supply, thereby optimizing corporate resource allocation and improving operational efficiency, reducing corporate operating risks and reducing energy waste.

[0003] There are many prediction models that can be used for the prediction of electricity sales revenue. Among them, the GM (1,1) model is a commonly used gray prediction model, which has the advantages of a small number of samples required, simple calculation, and verifiability, and is more suitable for short-term prediction. However, when the data is more discrete, the prediction accuracy of the model is worse, and it is not suitable for approximating complex nonlinear functions. And when applied to long-term predictions, problems such as too fast growth rates will arise. The BP neural network prediction model can approximate complex nonlinear function relationships, and can usually provide more accurate prediction results for complex trends or seasonal changes in electricity sales revenue data. However, it sometimes ignores the overall trend because of excessive focus on certain detailed fluctuations. In addition, the BP neural network also has the disadvantages of slow convergence, local minima, and difficulty in selecting the number of hidden layer neurons and connection weights. Summary of the invention

[0004] The purpose of the present invention is to provide a combined forecasting method and system for electricity sales revenue, which forecasts electricity sales data through the weighted combination of grey prediction and rolling prediction, and uses the above electricity sales data as the input value of BP neural network, while considering key electricity sales revenue influencing factors, and combines the electricity sales forecast data set according to the weighted combination of grey prediction and rolling prediction, and obtains the corresponding electricity sales revenue forecasting model based on this training, which is used for the forecasting and analysis of electricity sales revenue in the forecasting period, and solves the application defect that the existing electricity sales revenue forecasting method is difficult to fully and effectively grasp the changing trend of electricity sales data, resulting in insufficient accuracy of electricity sales data forecasting. It can improve the effectiveness of constructing the electricity sales revenue forecasting model based on the premise of having a small amount of data, thereby ensuring the efficiency and accuracy of the forecasting and analysis of electricity sales revenue data, reasonably arranging the power generation situation, and providing reliable guarantee for the normal power supply of the society while avoiding waste of resources.

[0005] To achieve the above object, the present invention provides the following technical solutions: A combined forecasting method for electricity sales revenue comprises the following steps: S1: Obtain the historical electricity sales revenue data of power supply enterprises in the target area and the corresponding data on factors affecting electricity sales revenue; S2: According to the historical electricity sales revenue data of the target area, rolling forecast and grey forecast are performed respectively, and the two are combined by the weight method to form a new forecast series; S3: training the BP neural network according to the predicted sequence in S2, and the output value of the BP neural network is the first positive actual value after the predicted value; S4: Obtain the corresponding data of the influencing factors of electricity sales revenue in S1 as an influencing factor sequence, and use the influencing factor sequence as another input value of the BP neural network in the prediction; S5: According to the time period to be predicted, a corresponding electricity sales revenue prediction model is obtained, and an electricity sales revenue prediction analysis is performed on the time period to be predicted based on the electricity sales revenue prediction model to obtain a corresponding electricity sales revenue prediction result.

[0006] Furthermore, the historical electricity sales revenue data of the power supply enterprise in S1 includes historical daily electricity sales revenue data of a preset period; the data of factors affecting electricity sales revenue include the monthly average temperature and average humidity of the target area.

[0007] Furthermore, the combined method in S2 determines the weight according to the deviation rate of the two, and the grey prediction adopts the improved GM (1, 1) model. The specific improvement is to adopt different improvement methods according to the data situation.

[0008] Furthermore, the weighting method of the grey prediction and rolling prediction in S2 is as follows: weighting is performed according to the deviation rate of the two. First, the weight coefficient of the rolling prediction is calculated. : ; Where A is the deviation rate of rolling prediction, and B is the deviation rate of grey prediction; Secondly, calculate the weight coefficient of gray prediction : ; Adjusted predicted value P: P= + ; in, is the rolling forecast value, is the predicted value of grey prediction.

[0009] Furthermore, the specific method for improving the grey prediction model in S2 is as follows: The exponential growth rate is used as the basis for selection. The exponential growth rate describes the speed at which data grows exponentially over a period of time. It can be obtained by calculating the growth rate of data at different time points. The calculation formula is as follows: ; Where r is the exponential growth rate, is the data value at time t, is the data value at the initial time, and t is the time interval.

[0010] Furthermore, the training in S3 is used for the BP neural network to continuously adjust the network parameters through training data, so that the network can adaptively learn the characteristics and patterns of the input data.

[0011] Furthermore, the input influence factor sequence in S4 is used to help identify the impact of input features on network output and enhance the trust and interpretability of prediction results.

[0012] The present invention provides another technical solution: a combined forecasting system for electricity sales revenue, comprising: A data collection module is used to obtain the historical electricity sales revenue data of the power supply enterprises in the target area, and the historical electricity sales revenue data of the power supply enterprises includes the historical daily electricity sales revenue data of a preset time period; The model training module is used to obtain the corresponding prediction data set according to the combined prediction model, and to construct the corresponding BP neural power sales revenue prediction model according to the power sales prediction data set; The electricity sales forecasting module is used to obtain the corresponding forecasting model according to the period to be forecasted, and to perform electricity sales revenue forecasting analysis on the period to be forecasted according to the electricity sales revenue forecasting model to obtain the corresponding electricity sales revenue forecasting result.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. The combined prediction method and system of electricity sales revenue of the present invention introduces an influencing factor sequence into the BP neural network and combines it with the time series obtained by the combined prediction. It can not only make up for the shortcomings of focusing too much on certain detailed fluctuations and ignoring the overall trend, slow convergence speed, local minimum values, and difficulty in selecting the number of hidden layer neurons and connection weights, but also help to improve the problem of poor prediction accuracy of grey prediction.

[0014] 2. The combined prediction method and system of electricity sales revenue of the present invention can overcome the inherent shortcomings of gray prediction and BP neural network, reliably extract the changing trend of electricity sales revenue from the time series data of electricity sales revenue with complex nonlinear relationships, and thus ensure the accuracy of the prediction results, so as to provide reliable guarantee for the power department to timely adjust the power supply plan, reasonably arrange the power generation situation, and ensure the normal power supply of the society without wasting resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of a method for predicting power sales data of a power supply enterprise in an embodiment of the present invention; Figure 2 Schematic diagram of the BP neural network structure in an embodiment of the present invention; Figure 31 is a schematic diagram of a gray prediction and rolling prediction deviation analysis line diagram in an embodiment of the present invention; Figure 4 It is a schematic diagram of the BP neural performance of the present invention; Figure 5 is a BP neural training state diagram in an embodiment of the present invention; Figure 6 It is a histogram of BP neural prediction errors in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] See also Figure 1-6 , a combined prediction method for electricity sales revenue provided in an embodiment of the present invention predicts the time series of electricity sales revenue data by weighted combination of gray prediction and rolling prediction, takes the above electricity sales data as one of the input values ​​of BP neural network, and trains the BP neural network based on this. Next, the present invention selects the monthly average temperature, monthly average GDP and meter reading time of the target area as key influencing factors affecting the prediction results of electricity sales revenue in the target area, and generates an influencing factor sequence. Finally, the time series and the influencing factor sequence are used as the input of the trained BP neural network for the prediction analysis of electricity sales revenue in the prediction period, and the output value is the electricity sales revenue prediction model that takes into account multiple influencing factors such as the monthly average temperature, monthly average GDP and meter reading time in the target area. The model can improve the effectiveness of the construction of the electricity sales revenue forecasting model under the premise of having a small amount of data. By combining the rolling forecast, gray forecast weighting and BP neural network in series, the accuracy of data time series prediction is improved while considering the key influencing factors that affect the fluctuation of electricity sales revenue data. Therefore, while overcoming the uncertainty factors, it can make up for the inherent shortcomings of gray prediction and BP neural network, and reliably extract the changing trend of electricity sales revenue from the electricity sales revenue time data with complex nonlinear relationships, thereby ensuring the accuracy of the forecast results, providing reliable guarantees for the power department to adjust the power supply plan in time, reasonably arrange the power generation situation, and ensure the normal power supply of the society without wasting resources.

[0018] Based on the above method, the present invention also provides a combined forecasting system for electricity sales revenue for implementing the above method, comprising: A data collection module is used to obtain the historical electricity sales revenue data of the power supply enterprises in the target area, and the historical electricity sales revenue data of the power supply enterprises includes the historical daily electricity sales revenue data of a preset time period; The model training module is used to obtain the corresponding prediction data set according to the combined prediction model, and to construct the corresponding BP neural power sales revenue prediction model according to the power sales prediction data set; The electricity sales forecasting module is used to obtain the corresponding forecasting model according to the period to be forecasted, and to perform electricity sales revenue forecasting analysis on the period to be forecasted according to the electricity sales revenue forecasting model to obtain the corresponding electricity sales revenue forecasting result.

[0019] The following embodiments will explain in detail the method for predicting electricity sales data for power supply enterprises of the present invention.

[0020] In one embodiment, a combined forecasting method for electricity sales revenue is provided, comprising the following steps: S1: Obtain the historical sales revenue data of the power supply enterprise in the target area and the corresponding data on the factors affecting the sales revenue of electricity; the historical sales revenue data of the power supply enterprise includes the historical daily sales revenue data of the preset time period; the data on the factors affecting the sales revenue of electricity include the average temperature and average humidity of the predicted area; among them, the target area can be understood as the power supply area of ​​the power supply enterprise where the sales revenue analysis is required, and no specific limitation is made here; the preset time period principle can be selected and determined according to the actual application needs, but in order to ensure the reliability of the subsequent analysis of key sales revenue influencing factors, at least daily data including one year's time period is required; correspondingly, the historical sales revenue data of the power supply enterprise within the preset time period corresponds one to one with the data on the factors affecting the sales revenue of electricity, that is, when the historical sales revenue data of the power supply enterprise includes the sales revenue data on January 1, 2022, the corresponding data on the factors affecting the sales revenue of electricity must also include the average temperature and average humidity data on January 1, 2022.

[0021] S2: The historical electricity sales revenue data of the target area are respectively subjected to grey prediction and rolling prediction methods to obtain the forecast time series under the joint weight. This section uses the data from January to December 2022 to predict the data from January to June 2023: first, a rolling forecast is performed on the original data to obtain the deviation rate of the rolling forecast; next, the r value is calculated for the original data, and the data is selected for grey prediction or improved grey prediction according to the size of the exponential growth rate r value; the specific method is: using the exponential growth rate (r) as the basis for selection. Exponential Growth Rate describes the speed at which data grows exponentially over a period of time. It can be obtained by calculating the growth rate of data at different time points, and is usually calculated using the following formula: ; Where r is the exponential growth rate, It's time The data value at the moment, is the data value at the initial moment, is the time interval.

[0022] When r is greater than 10%, the original metadata is updated with the cube root method. In other cases, the normal data value is used.

[0023] Similarly, the deviation rate of grey prediction and improved grey prediction is obtained. Next, the weights of the two methods are calculated according to the deviation rate results of rolling prediction and grey prediction, and the two are combined to form a new prediction series. The specific weight calculation method is as follows: First calculate the weight coefficient of the rolling forecast : ; Where A is the deviation rate of rolling prediction, and B is the deviation rate of grey prediction; Secondly, calculate the weight coefficient of gray prediction : ; Adjusted predicted value P: P= + ; in, is the rolling forecast value, is the predicted value of grey prediction; S3: Train the BP neural network according to the forecast time series from January to June 2023 obtained above. At this time, the output value of the BP neural network is the first positive actual value after the predicted value. The training is mainly for the BP neural network to continuously adjust the network parameters through training data so that the network can adaptively learn the characteristics and patterns of the input data. This adaptive learning ability enables the BP neural network to process various types of data and make accurate predictions after training.

[0024] S4: Obtain the influencing factor sequence and use it together with the time series as the input value of the trained BP neural network to predict the electricity sales revenue data from July to December 2023. In principle, the electricity sales revenue of power supply companies in various regions will be affected by factors such as temperature, GDP and meter reading time. Therefore, the influencing factor sequence will be used as another input value of the BP neural network in the BP neural network; the input of the influencing factor sequence can help identify which input features have a greater impact on the network output, which is very helpful for feature selection and data understanding. The most relevant and important features can be selected from the input data to improve the performance and efficiency of the network; by analyzing which input features play a decisive role in a specific prediction result, it can help users understand the prediction logic of the network and enhance the trust and interpretability of the prediction results. Finally, based on the time series data from January to June 2023, the forecast value of electricity sales revenue from July to December 2023 considering the key influencing factors was obtained.

[0025] In order to facilitate the understanding of the above-mentioned improved grey prediction and BP neural network prediction method for electricity sales revenue, this embodiment takes the monthly electricity sales revenue data of Zhejiang Province from January 2022 to November 2023 shown in Table 1 as an example for explanation: Table 1 Raw data month Electricity sales revenue (unit: 100 million yuan) 202201 368.59 202202 206.69 202203 254.74 202204 277.5 202205 381 202206 303.19 202207 488.41 202208 536.88 202209 493.0434783 202210 337.22 202211 317.62 202212 323.86 202301 345.4433333 202302 246.12 202303 312.48 202304 319.5 202305 301.63 202306 339.9 202307 362.08 202308 414.16 202309 463.8 202310 331.7 202311 312.6 Based on the predictions in Table 1, the monthly electricity sales revenue process in Zhejiang Province from July 2023 to December 2023 considering the key influencing factors is as follows: The gray forecast and rolling forecast methods are used for the data from January to December 2022 to obtain the forecast time series from January to June 2023 under the joint weight. First, the r value of the original data from January to December 2022 is calculated, and the calculation results are as follows: ; Therefore, the traditional grey prediction method is used for this period of time data, and the prediction results and deviation rates are shown in Table 2 below: Table 2 Grey prediction results date Raw data Predicted value Deviation rate January 2023 345.4433333 413.381 19.67% February 2023 246.12 421.183 71.13% March 2023 312.48 429.025 37.30% April 2023 319.5 436.909 36.75% May 2023 301.63 444.834 47.48% June 2023 339.9 452.8 33.22% The calculated average deviation rate of grey prediction is: 40.92%; Next, the rolling forecast method is used to forecast the above time series, and the forecast results are shown in Table 3: Table 3 ARIMA prediction results date Raw data Predicted value Deviation rate January 2023 345.4433333 328.6007948 4.88% February 2023 246.12 331.5019077 34.69% March 2023 312.48 333.2772336 6.66% April 2023 319.5 334.3636382 4.65% May 2023 301.63 335.0284597 11.07% June 2023 339.9 335.4352949 1.31% Similarly, the rolling forecast average deviation rate is calculated to be: 10.54%; Based on the deviation rate of the above grey prediction and rolling prediction, the weights of the prediction results of the two are calculated respectively, so as to obtain the final joint prediction time series. The weight calculation process is as follows: First calculate the weight coefficient of the rolling forecast : ; Among them, A is the deviation rate of rolling prediction, and B is the deviation rate of gray prediction. Secondly, calculate the weight coefficient of gray prediction : ; Adjusted predicted value P: P= + ; in, is the rolling forecast value, is the predicted value of grey prediction; Finally, the adjusted time series of the two forecasting methods are shown in Table 4: Table 4 Adjusted forecast results date Predicted value Deviation rate January 2023 345.9699323 0.15% February 2023 349.8751025 42.16% March 2023 352.8933235 12.93% April 2023 355.3723688 11.23% May 2023 357.5246017 18.53% June 2023 359.4801023 5.76% The adjusted data is further processed by BP neural network: First, the data set from June 2022 to June 2023 is used as the training set, and the data set from July 2023 is used as the output value for model training. The specific implementation method is as follows: % Training set input data: electricity sales revenue from June 2022 to May 2023 inputTrain = [ 303.19; 488.41; 536.88; 493.0434783; 337.22; 317.62; 323.86; 345.9699323; 349.8751025; 352.8933235; 355.3723688; 357.5246017 ]; % Training set target output data: electricity sales revenue from July 2022 to June 2023 targetTrain = [ inputTrain(2:end); % Assume that next month's revenue is equal to this month's revenue, except that the last value is set to 362.08, 362.08 ]; hiddenLayerSize = 10; net = feedforwardnet(hiddenLayerSize); [net,tr] = train(net, inputTrain', targetTrain'); Secondly, the influencing factor sequence is introduced into the BP neural network to predict the electricity sales revenue from July to November 2023. The influencing factors of electricity sales here are the average temperature and average humidity. The specific data are shown in Table 5: Table 5 Sequence of influencing factors time Average temperature Average humidity June 2022 25 78.00% July 2022 30 70.00% August 2022 29 73.00% September 2022 25 76.00% October 2022 19 71.00% November 2022 13 77.00% December 2022 7 68.00% January 2023 5 71.00% February 2023 7 73.00% March 2023 12 69.00% April 2023 17 66.00% May 2023 22 69.00% June 2023 26 75.00% The specific prediction results are shown in Table 6: Table 6 BP neural prediction results time Electricity sales revenue Predicted value Deviation rate July 2023 362.08 358.4212 1.01% August 2023 414.16 403.3482 -2.61% September 2023 463.8 423.6874 -8.65% October 2023 331.7 359.631 8.42% November 2023 312.6 336.5819 7.67% In order to further verify the prediction accuracy of the method of the present invention, the prediction values ​​of the BP neural network with the influencing factors introduced and the BP neural network without the influencing factors introduced were compared. The comparison results are shown in Table 7: Table 7 Forecast data comparison table time Electricity sales revenue Add influencing factors prediction value Prediction value without influencing factors July 2023 362.08 358.4212 346.8875 August 2023 414.16 403.3482 353.1041 September 2023 463.8 423.6874 360.041 October 2023 331.7 359.631 366.9198 November 2023 312.6 336.5819 373.333 In summary: the combined prediction method and system of electricity sales revenue provided by the present invention first obtains the historical electricity sales revenue data of the power supply enterprise in the target area and the corresponding electricity sales revenue influencing factor data including the monthly average temperature and average humidity; wherein, the historical electricity sales revenue data of the power supply enterprise should be selectively corrected according to the r value calculation of the gray prediction, and the time series prediction result is obtained by combining the rolling prediction and its weight. Next, by connecting the combined prediction model in series with the BP neural network, the time series and the influencing factor series are used as input values ​​to obtain the corresponding electricity sales revenue prediction result considering multiple influencing factors. Compared with the prior art, the present invention can overcome the inherent shortcomings of gray prediction and BP neural network, reliably extract the change trend of electricity sales revenue from the electricity sales revenue time series data with complex nonlinear relationships, and thus ensure the accuracy of the prediction results, so as to provide reliable guarantee for the power department to adjust the power supply plan in time, reasonably arrange the power generation situation, and ensure the normal power supply of the society without wasting resources.

[0026] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A combined forecasting method for electricity sales revenue, characterized in that: The following steps are involved: S1: Obtain the historical electricity sales revenue data of power supply enterprises in the target area and the corresponding data on factors affecting electricity sales revenue; S2: According to the historical electricity sales revenue data of the target area, rolling forecast and grey forecast are performed respectively, and the two are combined by the weight method to form a new forecast series; S3: training the BP neural network according to the predicted sequence in S2, and the output value of the BP neural network is the first positive actual value after the predicted value; S4: Obtain the corresponding data of the influencing factors of electricity sales revenue in S1 as an influencing factor sequence, and use the influencing factor sequence as another input value of the BP neural network in the prediction; S5: According to the time period to be predicted, a corresponding electricity sales revenue prediction model is obtained, and an electricity sales revenue prediction analysis is performed on the time period to be predicted based on the electricity sales revenue prediction model to obtain a corresponding electricity sales revenue prediction result.

2. A combined forecasting method for electricity sales revenue according to claim 1, characterized in that: The historical electricity sales revenue data of the power supply enterprise in S1 includes the historical daily electricity sales revenue data for a preset period of time; the data on factors affecting electricity sales revenue include the monthly average temperature and average humidity of the target area.

3. A combined forecasting method for electricity sales revenue according to claim 1, characterized in that: The combined method in S2 determines the weight according to the deviation rate of the two. Grey prediction adopts the improved GM (1, 1) model. The specific improvement is to adopt different improvement methods according to the data situation.

4. A combined forecasting method for electricity sales revenue according to claim 3, characterized in that: The weighting method of grey prediction and rolling prediction in S2 is as follows: weighting is performed according to the deviation rate of the two. First, the weight coefficient of rolling prediction is calculated. : ； Where A is the deviation rate of rolling prediction, and B is the deviation rate of grey prediction; Secondly, calculate the weight coefficient of gray prediction : ； Adjusted predicted value P: P= + ； in, is the rolling forecast value, is the predicted value of grey prediction.

5. A combined forecasting method for electricity sales revenue according to claim 3, characterized in that: The specific method of improving the grey prediction model in S2 is as follows: The exponential growth rate is used as the basis for selection. The exponential growth rate describes the speed at which data grows exponentially over a period of time. It can be obtained by calculating the growth rate of data at different time points. The calculation formula is as follows: ； Where r is the exponential growth rate, is the data value at time t, is the data value at the initial time, and t is the time interval.

6. A combined forecasting method for electricity sales revenue according to claim 1, characterized in that: The training in S3 is used for the BP neural network to continuously adjust the network parameters through training data, so that the network can adaptively learn the characteristics and patterns of the input data.

7. A combined forecasting method for electricity sales revenue according to claim 1, characterized in that: The input influence factor sequence in S4 is used to help identify the impact of input features on network output and enhance the trust and interpretability of prediction results.

8. A combined forecasting system for electricity sales revenue, used to implement a combined forecasting method for electricity sales revenue according to any one of claims 1 to 7, characterized in that: include: A data collection module is used to obtain the historical electricity sales revenue data of the power supply enterprises in the target area, and the historical electricity sales revenue data of the power supply enterprises includes the historical daily electricity sales revenue data of a preset time period; The model training module is used to obtain the corresponding prediction data set according to the combined prediction model, and to construct the corresponding BP neural power sales revenue prediction model according to the power sales prediction data set; The electricity sales forecasting module is used to obtain the corresponding forecasting model according to the period to be forecasted, and to perform electricity sales revenue forecasting analysis on the period to be forecasted according to the electricity sales revenue forecasting model to obtain the corresponding electricity sales revenue forecasting result.

Images