Scenario-based electricity market price forecasting method based on multi-source data fusion

Abstract

The present application relates to the technical field of power market prediction, and more particularly to a scenario-based power market price prediction method based on multi-source data fusion, comprising: obtaining historical price, historical weather and future weather forecast data, aligning to the same time granularity to form a standardized time series; extracting temperature, wind speed, cloud cover, solar radiation intensity and day-night identification weather features, combining price features to construct a multi-dimensional feature matrix; counting the amount of historical price data, automatically selecting a multi-decision tree combination or a recurrent neural network model according to a threshold interval; adapting the multi-dimensional feature matrix; training the model and inputting the weather forecast data into the model to predict future electricity value in the same adaptation mode; automatically outputting the prediction results in the form of a time series curve and / or a data report and generating a report. The present application realizes self-adaptation of the price prediction model and full-process automatic execution.

Description

Technical Field

[0001] This invention relates to the field of electricity market forecasting technology, and in particular to a scenario-based electricity market price forecasting method based on multi-source data fusion. Background Technology

[0002] Against the backdrop of continuous increases in installed capacity of new energy sources, the operating characteristics of the power system are undergoing profound changes. The output of new energy sources such as wind and solar power exhibits significant weather dependence and volatility, leading to a significant increase in supply-side uncertainty and more complex, multi-factor-coupled characteristics in electricity price fluctuations. Simultaneously, the development of artificial intelligence technology has provided new technical pathways for electricity market analysis. While some electricity trading systems currently offer forecasting services, their algorithmic logic is opaque, their models are uncontrollable, and companies struggle to develop independent core technologies. Existing forecasting methods mostly employ a single model structure, unable to adaptively adjust to historical data scales, and the forecasting process heavily relies on manual operation, requiring 2-3 days from data acquisition to report generation, which fails to meet the timeliness requirements of high-frequency fluctuations in the spot market.

[0003] Therefore, the main problem with existing technologies is the lack of a method for electricity price forecasting that can dynamically select a forecasting model based on the data scale, automate the entire process from data acquisition to report generation, and ensure that the model logic is autonomously controllable. This results in forecast accuracy being limited by a single model structure, low efficiency of manual intervention, and difficulty in forming an enterprise's independent iteration capability. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a scenario-based electricity market price forecasting method based on multi-source data fusion, aiming to improve the problems in existing electricity price forecasting methods, such as the inability of the forecasting model to adaptively select based on the scale of historical data, the inefficiency caused by the reliance on manual operation in the forecasting process, and the uncontrollable model logic.

[0005] This invention provides the following technical solution: a scenario-based electricity market price prediction method based on multi-source data fusion, comprising: S1. Obtain the historical electricity price time series of the target power plant, the historical meteorological data of the corresponding period, and the meteorological forecast data of the future prediction period, and align all data to the same time granularity to form a standardized time series; S2. Based on the standardized time series, extract meteorological features that include at least temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators, and construct a multi-dimensional feature matrix by combining it with electricity price features; S3. Calculate the amount of historical electricity price data and automatically select the prediction model based on the threshold range of the data volume; S4. Adapt the multidimensional feature matrix according to the selected model type; S5. Use the adapted data to train the selected prediction model, and use the same adaptation method to input the features corresponding to the meteorological forecast data into the trained model in order to predict the electricity value at each time point within the future set prediction period. S6. Automatically output the predicted electricity value in the form of time-series curves and / or data reports, and generate a prediction report.

[0006] The present invention has the following beneficial effects: 1. In this invention, by constructing a dynamic model selection mechanism based on the scale of historical data and designing an integrated automatic process from data acquisition to report generation, the problems of the prediction model's inability to adaptively adjust and the high degree of manual dependence in the entire process in the prior art are solved, and the autonomous adaptation of the model and the automated execution of the entire process of electricity price prediction are realized.

[0007] 2. In this invention, by integrating historical electricity price data with multi-dimensional meteorological data including temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators, and adaptively selecting a multi-decision tree combination model or a recurrent neural network model according to the data volume threshold range, the problem of limited prediction accuracy of traditional methods with single models or without integrated meteorological data is solved, thus improving the prediction accuracy.

[0008] 3. In this invention, a dual adaptive architecture consisting of a multi-decision tree combination model and a recurrent neural network model is adopted, which is applicable to both small-sample and large-sample scenarios. This solves the problem that a single model is prone to overfitting in small-sample scenarios and cannot fully explore the temporal patterns in large-sample scenarios, thus achieving universal coverage and highly adaptive prediction for stations at different data accumulation stages. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the scenario-based electricity market price prediction method based on multi-source data fusion proposed in this invention. Figure 2 This is a schematic diagram illustrating the multi-source data acquisition and standardization alignment process of the scenario-based electricity market price prediction method based on multi-source data fusion proposed in this invention. Figure 3 This is a schematic diagram illustrating the meteorological feature extraction and multidimensional feature matrix construction process of the scenario-based electricity market price prediction method based on multi-source data fusion proposed in this invention. Figure 4 This is a schematic diagram of the adaptive model selection process based on data volume for the scenario-based electricity market price prediction method based on multi-source data fusion proposed in this invention. Figure 5 This is a schematic diagram of the model adaptation process for the scenario-based electricity market price prediction method based on multi-source data fusion proposed in this invention. Figure 6This diagram illustrates the model training, prediction, and report generation process of the scenario-based electricity market price prediction method based on multi-source data fusion proposed in this invention. Detailed Implementation

[0010] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] This invention provides a scenario-based electricity market price forecasting method based on multi-source data fusion, such as... Figures 1-6 As shown, it includes the following steps: S1. Obtain the historical electricity price time series of the target power plant, the historical meteorological data of the corresponding period, and the meteorological forecast data of the future prediction period, and align all data to the same time granularity to form a standardized time series.

[0012] Furthermore, in S1, the specific steps of aligning all data to the same time granularity to form a standardized time series include: The historical electricity price time series data is cleaned and resampled to a preset fixed time granularity to obtain a standardized subsequence of historical electricity prices; Historical meteorological data for the corresponding time period is cleaned and resampled to a preset fixed time granularity to obtain a standardized historical meteorological subsequence; The meteorological forecast data for the future prediction period is aligned to a preset fixed time granularity using interpolation to obtain a standardized subsequence of future meteorological data. The standardized subsequences of historical electricity prices, historical meteorological conditions, and future meteorological conditions are spliced ​​together in chronological order to form a standardized time series.

[0013] Specifically, this step achieves the acquisition and standardization alignment of multi-source data. First, it acquires historical electricity price time-series data for the target power plant, historical meteorological data for the corresponding time period, and meteorological forecast data for the future forecast period. The historical electricity price time-series data may be stored at a time granularity of hours or a mix of 15 minutes, while the historical meteorological data and future meteorological forecast data are typically stored at the hourly level. To establish a unified time benchmark, all data is aligned to a preset fixed time granularity. In this embodiment, the preset fixed time granularity is 15 minutes, i.e., 96 time points per day. The alignment process includes the following steps:

[0014] Historical electricity price time series data undergoes data cleaning to remove outliers that significantly exceed reasonable ranges, such as negative electricity prices or values ​​exceeding market limits. Missing values ​​are then filled using linear interpolation between preceding and following time points. After cleaning, the data is resampled to a 15-minute granularity. The resampling method is as follows: if the original data is hourly, data from each hour is copied and used to fill four 15-minute points within that hour; if the original data is 15-minute granularity and has no missing values, it is directly retained. The resulting standardized subsequence of historical electricity prices is denoted as... Where N is the total number of time points, each Let be the electricity price for the i-th 15-minute period.

[0015] Historical meteorological data corresponding to historical electricity prices is cleaned to remove obvious errors in various meteorological elements (temperature, wind speed, cloud cover, solar radiation intensity), such as temperatures exceeding reasonable ranges. The median of adjacent time points is used for padding. Then, using the same resampling method as in sub-step one, all meteorological elements are standardized to a 15-minute granularity, resulting in a standardized historical meteorological subsequence. Let the temperature sequence be denoted as... The wind speed sequence is The cloud cover sequence is The solar radiation intensity sequence is Each sequence has a length of N.

[0016] Meteorological forecast data for the future prediction period is obtained, typically at the hourly granularity. Cubic spline interpolation is then used to interpolate this data to a 15-minute granularity, yielding a standardized subsequence of future weather conditions. The interpolation formula is: For the interval... Given the function value f and f( The cubic spline interpolation function S(x) satisfies the continuity of the second derivative. The interpolation coefficients are obtained by solving a tridiagonal system of equations, which can be implemented by those skilled in the art using standard numerical methods. Finally, a 15-minute granular sequence of meteorological elements for the future forecast period is obtained, with a length denoted as M, where M = 96 × D, and D is the number of forecast days. In this embodiment, the number of forecast days is 7 days.

[0017] The obtained historical electricity price standardized subsequence, historical meteorological standardized subsequence, and future meteorological standardized subsequence are concatenated in chronological order. Specifically, the historical part comes first, followed by the future part, forming a continuous standardized time series covering both historical and future forecast periods. This series uses a 15-minute time step, and each time point simultaneously includes the values ​​of electricity price and all meteorological elements; that is, each time point corresponds to a multi-dimensional vector. For future periods, the electricity price portion is temporarily set to null, pending subsequent forecast generation. At this point, all data has been aligned to the same time granularity, forming a standardized time series.

[0018] S2. Based on standardized time series, extract meteorological features that include at least temperature, wind speed, cloud cover, solar radiation intensity and day / night indicators, and construct a multidimensional feature matrix by combining electricity price features.

[0019] Furthermore, in S2, the steps for extracting meteorological features that include at least temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators specifically include: Numerical sequences for temperature, wind speed, cloud cover, and solar radiation intensity were extracted from the standardized time series. Based on the geographical location and date information corresponding to each time point in the standardized time series, calculate the local sunrise and sunset times, and generate a binary sequence of day and night identifiers based on whether each time point is between sunrise and sunset times; Meteorological features are formed by aligning numerical sequences of temperature, wind speed, cloud cover, and solar radiation intensity with binary sequences labeled day and night by time points.

[0020] Furthermore, in S2, the specific steps for constructing the multidimensional feature matrix include: Electricity price data columns are extracted from standardized time series data and used as electricity price feature vectors. Acquire meteorological features consisting of temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators; The electricity price feature vector and meteorological features are horizontally aligned and concatenated according to the same time point to generate a multi-dimensional feature matrix.

[0021] Specifically, the numerical sequences corresponding to temperature, wind speed, cloud cover, and solar radiation intensity are extracted from the standardized time series. Let N be the total number of time points in the standardized time series, and let the length of the historical period be... The length of the future forecast period is , The four sequences obtained from the analysis are as follows: Temperature sequence: The unit is degrees Celsius. Wind speed sequence: The unit is meters per second. Cloud cover sequence: The value ranges from 0 to 10, where 0 represents clear skies and 10 represents completely cloudy skies. Solar radiation intensity sequence: The unit is watts per square meter.

[0022] Based on the geographical location information corresponding to each time point in the standardized time series—namely, the longitude and latitude of the target power station—and the date information for each time point—year, month, and day—the local sunrise and sunset times are calculated. The calculation of sunrise and sunset times employs a standard astronomical algorithm: for a given longitude, latitude, and date, the solar hour angle is first calculated, and then the sunrise and sunset times are solved using spherical trigonometry formulas. The specific formulas are as follows:

[0023] ; in The angle at sunset, Geographical latitude, This is the solar declination angle. The angle at sunrise is − Convert the hour angle to local time to obtain the sunrise time. and sunset time The calculation method is common knowledge in this field and will not be elaborated here.

[0024] For each time point i, obtain its corresponding timestamp and determine whether the timestamp falls between the sunrise and sunset times of that day. If it does, then use the day / night indicator. Indicates daytime; otherwise =0 represents night. This generates a binary sequence indicating day and night. The temperature sequence T, wind speed sequence W, cloud cover sequence C, solar radiation intensity sequence R, and day / night indicator sequence D are aligned according to the same time points to form a meteorological feature matrix. The matrix is ​​an N×5 two-dimensional array, with each row corresponding to a meteorological feature vector at a given time point. .

[0025] The construction of the multidimensional feature matrix includes the following sub-steps: Extracting the electricity price data series from the standardized time series. For historical periods, the electricity price data are known values; for future prediction periods, the electricity price data is temporarily empty. Let the electricity price series be... The length is N. The electricity price sequence is used as the electricity price feature vector. Directly obtain the meteorological feature matrix generated in sub-step three above. This matrix contains all the meteorological characteristics at each point in time.

[0026] Electricity price feature vector With meteorological feature matrix The data is horizontally aligned and stitched together according to the same time point. Specifically, for each time point i, the electricity price is... With meteorological feature vector Merge into a 6-dimensional vector The combined results from all time points constitute a multidimensional feature matrix. The matrix has a size of N×6. In this matrix, the rows for historical periods contain complete electricity price and meteorological data, while the rows for future prediction periods contain only meteorological data, leaving the electricity price data blank. The blank spaces will be filled in subsequent prediction steps. This completes the extraction of meteorological features and the construction of the multidimensional feature matrix. The resulting multidimensional feature matrix will serve as the basic input for subsequent model selection and adaptation.

[0027] S3. Calculate the amount of historical electricity price data and automatically select a prediction model based on the threshold range of the data volume.

[0028] Furthermore, in S3, the step of automatically selecting the prediction model based on the threshold range of the data volume specifically includes: The time span or number of sample points included in the historical electricity price time series data is used as the statistical value of the data volume; The statistical value of the data volume is compared with a preset first threshold and a second threshold, wherein the first threshold is less than the second threshold; When the statistical value of the data volume is less than or equal to the first threshold, or between the first threshold and the second threshold, the prediction model based on the combination of multiple decision trees is selected. When the statistical value of the data volume is greater than or equal to the second threshold, a prediction model based on a recurrent neural network is selected.

[0029] Specifically, the historical electricity price time series data obtained in statistical step S1 includes the time span or number of sample points. In this embodiment, the number of sample points is used as the statistical value of the data volume. Let the total number of time points in the historical electricity price data be... Each time point corresponds to a 15-minute interval, then ,in This refers to the number of days covered by the historical data. For example, if the historical data covers approximately 3 months (90 days), then... If it covers approximately 365 days a year, then =35040. This statistical value is denoted as Q.

[0030] Preset first threshold Second threshold ,and In this embodiment, the first threshold The number of sample points corresponding to 3 months of historical data is 8640, which is the second threshold. The sample size corresponding to one year of historical data is 35040. The data volume statistic Q is then compared with... , The comparison is performed, and the following branch decision is executed:

[0031] like ,or If so, then a prediction model based on a combination of multiple decision trees is selected. This model is suitable for small or medium sample scenarios and has the advantages of fast training speed and low overfitting. If a recurrent neural network-based prediction model is chosen, this model is suitable for large-sample scenarios and can fully capture the long-term time-series dependencies of electricity price sequences.

[0032] Multi-Decision Tree Combination Model: This model employs an ensemble learning framework based on gradient boosting decision trees, such as XGBoost or LightGBM. The model is composed of multiple decision trees combined additively, with each tree iteratively trained based on the residuals of the previous tree. For a given input feature vector x, the model outputs a predicted value. Where K is the total number of trees. This is the output of the k-th decision tree. Hyperparameters such as the depth and learning rate of the decision tree are determined using cross-validation, and can be set by those skilled in the art using conventional methods. Recurrent Neural Network Model: This uses a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU). The model includes an input layer, one or more recurrent hidden layers, and an output layer. The input layer receives the sequential data at each time step, and the hidden layers maintain long-term memory states through a gating mechanism. For the input sequence... Hidden state The output layer maps the hidden state of the last time step to the predicted value through a fully connected network. In this embodiment, a two-layer LSTM is used, with 64 hidden units in each layer. The output layer is a single-neuron fully connected layer, and the activation function is a linear function.

[0033] Based on the above judgment, the system automatically outputs the type identifier of the selected model for subsequent adaptation processing in step S4. If the data volume statistics Q is exactly equal to the first threshold or the second threshold, it is assigned to the corresponding branch according to the above rules. For cases where the data volume is between the first and second thresholds, a multi-decision tree combination model is uniformly selected to avoid the risk of overfitting of the neural network under small sample size using a conservative strategy. Thus, the adaptive selection of data volume statistics and prediction model is completed.

[0034] S4. Adapt the multidimensional feature matrix according to the selected model type.

[0035] Furthermore, in S4, the steps for adapting the multidimensional feature matrix according to the selected model type specifically include: Determine the type of the selected prediction model; If it is a multi-decision tree combination model, the electricity price sequence is extracted from the multi-dimensional feature matrix, a sliding window time series feature is constructed with a preset window length, and the meteorological features corresponding to the same window are combined with the sliding window time series feature to generate an input feature set adapted to the multi-decision tree combination model. If it is a recurrent neural network model, the original time series in the multidimensional feature matrix is ​​directly used as the input feature set adapted to the recurrent neural network model.

[0036] Specifically, this step adapts the multidimensional feature matrix generated in step S2 to the type of prediction model selected in step S3, generating an input feature set suitable for that model. Let the multidimensional feature matrix... The dimensions are N×6, where Each line includes electricity price ,temperature Wind speed Cloud cover Solar radiation intensity Day and night indicators Historical periods total Okay, the future forecast period is as follows: Okay. The adaptation process only applies to labeled data for historical periods; features for future periods will be processed in the same way during prediction. Read the model type identifier output from step S3 and execute the following branch. If the selected model is a multi-decision tree combination model, perform the following operation: Extract the electricity price sequence for historical periods from the multi-dimensional feature matrix. and the corresponding meteorological feature matrix (size ×5). A sliding window timing feature is constructed using a preset window length L. In this embodiment, L=24, corresponding to a 6-hour window (since the time granularity is 15 minutes, 24 points equal 6 hours). The sliding window construction method is as follows: For the window starting position j from 1 to... Take the electricity price sub-sequence within the current window. As a temporal feature, the meteorological feature vector of the last time point within the same window is taken as the exogenous feature corresponding to that window. The model's prediction objective is the electricity price at the next time point after the window. This generates a training sample set. , where input Output All samples constitute the input feature set adapted to the multi-decision tree combination model. In this embodiment, the window length L can also be 48 (12 hours) or 96 (24 hours), which can be adjusted by the user according to the data characteristics.

[0037] If the selected model is a recurrent neural network (RNN), there is no need to construct a sliding window; the original time series from the multidimensional feature matrix can be directly used as the model input. For an RNN, the input at each time step includes the electricity price and all meteorological features at that time point, i.e., the input vector. Input sequences of historical time periods As the training input, the corresponding output sequence is That is, predicting the electricity price for the next step at each step. Specifically, for time step i, the model receives the current input. And the hidden state of the previous step Output the predicted electricity price for the current step. During training, a time-step unfolding approach is used, and the network parameters are optimized through backpropagation. This method preserves complete temporal sequence information and eliminates the need for manually constructing feature windows.

[0038] Regardless of which branch is executed, an input feature set adapted to the selected prediction model is ultimately generated. This feature set will be directly used for model training in step S5. After the adaptation process is complete, the system records the currently used window length or sequence information so that the same preprocessing method can be used for weather forecast data in the prediction stage. Thus, the adaptation process of the multidimensional feature matrix according to the selected model type is completed.

[0039] S5. Use the adapted data to train the selected prediction model, and use the same adaptation method to input the features corresponding to the meteorological forecast data into the trained model in order to predict the electricity value at each time point within the future set prediction period.

[0040] Furthermore, in S5, the steps for predicting the electricity value at each point in time within a predetermined forecast period specifically include: The adapted input feature set is used as training samples and input into the selected prediction model for iterative training until the model converges, thus obtaining the trained prediction model. Meteorological forecast data corresponding to the future forecast period are extracted from the standardized time series. The same adaptation processing method as in step S4 is used to convert the meteorological forecast data into forecast input features that are adapted to the selected forecast model. The predicted input features are fed into the trained prediction model, and the electricity value at each time point within the set prediction period is output sequentially or directly.

[0041] Furthermore, the specific steps for extracting meteorological forecast data corresponding to future forecast periods from standardized time series include: Determine the start and end times of the future forecast period; Based on the start and end times, extract subsequences within the specified time interval from the standardized time series as standardized subsequences for future periods; Data columns corresponding to temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators are extracted from the standardized subsequence of the future time period to serve as meteorological forecast data for the future prediction period.

[0042] Specifically, this step uses the data processed in step S4 to train the selected prediction model, and uses the same adaptation method to input meteorological forecast data into the trained model to predict the electricity value at each time point within the set prediction period. The input feature set adapted to the selected model generated in step S4 is used as training samples and input into the selected prediction model for iterative training until the model converges, resulting in a trained prediction model. If the selected model is a multi-decision tree combination model, the training process is as follows: The training sample set is... ,in For the input feature vector, Let M be the target electricity price, and M be the total number of samples. The model uses a gradient boosting decision tree framework, and the objective function is to minimize the mean squared error loss. In each iteration, a new decision tree is added to fit the current residual. The learning rate is set to 0.1, the maximum tree depth is set to 6, and the number of trees is set to 100. An early stopping strategy is used during training; training stops when the validation set loss does not decrease for 10 consecutive iterations. If the selected model is a recurrent neural network, the training process is as follows: The input sequence is... The target sequence is A teacher-mandated training method is used, with the actual electricity price as input at each time step to calculate the loss. The loss function uses the mean squared error: The optimizer used is Adam, with an initial learning rate of 0.001, a batch size of 32, and 100 training epochs. Training is terminated early when the validation set loss no longer decreases. Network parameters are saved after model convergence.

[0043] Extracting meteorological forecast data corresponding to the future forecast period from the standardized time series formed in step S1. Specifically, this includes the following sub-steps: Sub-step 1: Determining the start and end times of the future forecast period. In this embodiment, the start time is the first complete 15-minute interval after the current time, and the end time is the same time after adding the forecast days minus one day to the start time. The forecast days are defaulted to 7 days. Sub-step 2: Based on the start and end times, extracting a subsequence within this time interval from the standardized time series as the standardized subsequence for the future period. This subsequence contains... Each time point contains meteorological data such as temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicator. Sub-step three: Extract the data columns corresponding to temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicator from the standardized subsequence of the future time period, as the meteorological forecast data for the corresponding future prediction period. Let the future meteorological forecast data be denoted as... , where i is the index of a time point within a future time period.

[0044] Using the same adaptation process as in step S4, the aforementioned weather forecast data is converted into predictive input features adapted to the selected prediction model. If the model is a multi-decision-tree combination model, a sliding window approach is used to construct the predictive input. Let the window length used in step S4 be L. For future time periods, electricity price data from the last L time points of historical time periods are used as the initial window. Specifically, the electricity price sequence from the last L time points of historical time periods is taken. Meteorological characteristics at the first point in time in the future period The inputs are combined to form the first prediction input. The model outputs the first predicted electricity price. Then, using the predicted electricity price as new historical data, the sliding window moves forward one step and is combined with the meteorological characteristics of the next time point. This process is repeated to predict the electricity price for subsequent time points until the future is predicted. Predictions for each time point. If the model is a recurrent neural network, a sliding window is not required. The complete historical time series is used as the initial state input to the model, and then an autoregressive approach is used for prediction: the predicted electricity price output from the previous time step is used as the electricity price feature input for the next time step, along with the corresponding meteorological features, and so on, recursively generating the electricity price prediction sequence for the entire future time period. Specifically, for the first time step of the future period, the input is... , where the initial predicted value It is obtained from the forward calculation of the model; subsequent time steps are calculated in the same way.

[0045] Through the above recursive or direct multi-step prediction process, the electricity value sequence at each time point within the set prediction period is obtained. Each value in this sequence corresponds to a 15-minute time point in the electricity price forecast. This completes the forecast of electricity prices at each time point in the future forecast period. The forecast results will then be passed to step S6 for output and report generation.

[0046] S6. Automatically output the predicted electricity value in the form of time-series curves and / or data reports, and generate a prediction report.

[0047] Furthermore, in S6, the steps for generating the forecast report specifically include: The predicted electricity values ​​at each time point are plotted as a time series curve in chronological order. The predicted electricity value at each time point and the corresponding timestamp are compiled into a structured data table as a data report; Integrate time series graphs and / or data reports into a preset report template to generate a forecast report containing the forecast results, and automatically output it to a specified storage path or display interface.

[0048] Specifically, this step automatically outputs the electricity value at each time point within the future forecast period obtained in step S5 as a time-series curve and data report, and generates a forecast report. The electricity value sequence at each time point within the future forecast period output in step S5 is obtained. Each electricity value corresponds to a 15-minute time point. Simultaneously, a timestamp corresponding to each time point is obtained, extracted from a standardized time series. Using the timestamp as the x-axis and the electricity value as the y-axis, points are connected sequentially in a two-dimensional coordinate system to generate an electricity price time-series curve. The horizontal axis of the curve is labeled with date and time, and the vertical axis is labeled with the electricity price unit, such as yuan / megawatt-hour. The curve can be captioned with something like "Electricity Price Forecast Curve for the Next 7 Days," and optionally, historical electricity price curves can be added for reference. In this embodiment, the plotting is implemented using Python's Matplotlib library or the front-end ECharts library, generating a PNG image file.

[0049] The predicted electricity prices at each time point and their corresponding timestamps are compiled into a structured data table. Each row of the table represents a time point and contains three columns: timestamp, predicted electricity price, and confidence interval. The confidence interval is calculated based on the standard deviation of the prediction error on the training set. Using ±1.96 Construct a 95% confidence interval. Specifically, for each predicted electricity value... The lower limit of the confidence interval is The upper limit is If the model is a combination of multiple decision trees, This is obtained by performing five-fold cross-validation on the training set; if the model is a recurrent neural network model. The standard deviation of the predicted residuals is obtained from the validation set. This table is stored in comma-separated CSV or Excel worksheet format for easy subsequent analysis.

[0050] A pre-defined report template is provided, containing fixed information such as title, generation time, and descriptions of prediction parameters. The aforementioned time-series graphs and structured data reports are integrated into this template. Specifically, the time-series graphs are embedded as images in the upper half of the report, and the structured data tables are placed in the lower half. The report is in Excel file format, with the first worksheet named "Prediction Curve" containing the plotted graph, and the second worksheet named "Prediction Data" containing the structured data tables. After the report is generated, the current system time is automatically added as a filename suffix. The system automatically outputs the generated prediction report to the specified storage path according to the preset configuration. Simultaneously, in the local visualization interface, the system automatically displays the time-series graphs in the chart area and the structured data tables in the table controls at the bottom of the interface. Users can manually save the report by clicking the "Save" button or export it as a PDF by clicking the "Export" button. The entire output process requires no manual intervention, achieving full automation from data acquisition to report generation. This completes the automatic output of prediction results and the generation of the prediction report. Users receive complete electricity price forecasts, which can be used for subsequent electricity market trading decisions.

[0051] After predicting the electricity value at each point in time within the set forecast period, but before generating the forecast report, a manual correction step is also included, specifically: The predicted electricity value at each time point is displayed in a visualization interface, and users can manually adjust some or all of the electricity value. The manual adjustment command is used to modify the electricity value at the indicated time point, and a correction log containing the adjustment time, adjustment content, and information before and after the adjustment is generated. The corrected electricity value is used as the final prediction result to generate the prediction report.

[0052] Specifically, after predicting the electricity value at each point in time within the set forecast period but before generating the forecast report, a manual correction step is also included, as follows: The electricity value sequence at each time point predicted in step S5 The corresponding timestamps are displayed in the chart area and table control of the local visualization interface. The chart area shows the distribution of electricity value over time as a time-series curve, and the table control shows the predicted electricity price for each time point in a row-list format. The system monitors user interactions in real time and receives manual adjustment commands from users for some or all of the electricity value. Users can enter new values ​​by clicking on nodes on the curve, directly edit the electricity price value in the corresponding cell in the table, or make corrections using the "batch adjustment" function provided in the interface (such as offsetting the entire value by a certain percentage). The system records the time point index targeted by the user for each adjustment and the modified electricity value.

[0053] Upon receiving the manual adjustment command, the system modifies the electricity price at the indicated time point, replacing the original forecast value with the new value entered by the user. Simultaneously, the system automatically generates a correction log, which contains three elements: adjustment time (the current system time when the user performed the correction operation, accurate to the second), adjustment details (recording the original electricity price forecast at the adjusted time point, the user-modified electricity price, and the difference between the two), and comparison information before and after the adjustment. The correction log can be appended to a local log file or database for easy traceability and auditing. If the user makes multiple corrections for the same time point, the system only retains the last correction result and records the history of each correction.

[0054] The corrected electricity value is used as the final prediction result. Specifically, for time points that are not corrected, the final predicted value is still the original prediction value output by the model; for time points that are corrected, the final predicted value is replaced with the new value input by the user. The corrected complete electricity value sequence is denoted as... This sequence will be passed to step S6 to generate the final prediction report. If the user does not perform any correction operations, the final prediction result will be completely consistent with the original model output, and the system will automatically skip the correction log generation step and proceed directly to step S6. This manual correction step is an optional post-processing module that does not change the automation attribute of the main process; users can choose whether to enable it according to their actual needs. This completes the entire process of the manual correction step.

[0055] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A scenario-based electricity market price forecasting method based on multi-source data fusion, characterized in that, include: S1. Obtain the historical electricity price time series of the target power plant, the historical meteorological data of the corresponding period, and the meteorological forecast data of the future prediction period, and align all data to the same time granularity to form a standardized time series; S2. Based on the standardized time series, extract meteorological features that include at least temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators, and construct a multi-dimensional feature matrix by combining it with electricity price features; S3. Calculate the amount of historical electricity price data and automatically select the prediction model based on the threshold range of the data volume; S4. Adapt the multidimensional feature matrix according to the selected model type; S5. Use the adapted data to train the selected prediction model, and use the same adaptation method to input the features corresponding to the meteorological forecast data into the trained model in order to predict the electricity value at each time point within the future set prediction period. S6. Automatically output the predicted electricity value in the form of time-series curves and / or data reports, and generate a prediction report.

2. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 1, characterized in that, In S1, the step of aligning all data to the same time granularity to form a standardized time series specifically includes: The historical electricity price time series data is cleaned and resampled to a preset fixed time granularity to obtain a standardized subsequence of historical electricity prices; The historical meteorological data for the corresponding time period is cleaned and resampled to the preset fixed time granularity to obtain a standardized historical meteorological subsequence. The meteorological forecast data for the future prediction period are aligned to the preset fixed time granularity using interpolation to obtain a standardized subsequence of future meteorological data. The standardized time series is formed by splicing the historical electricity price subsequence, the historical meteorological standardized subsequence, and the future meteorological standardized subsequence in chronological order.

3. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 1, characterized in that, In step S2, the step of extracting meteorological features that include at least temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators specifically includes: Numerical sequences corresponding to temperature, wind speed, cloud cover, and solar radiation intensity were extracted from the standardized time series. Based on the geographical location and date information corresponding to each time point in the standardized time series, calculate the local sunrise and sunset times, and generate a binary sequence of day and night identifiers based on whether each time point is between sunrise and sunset times; The meteorological features are formed by aligning the numerical sequences of temperature, wind speed, cloud cover, and solar radiation intensity with the binary sequence of day and night markers according to time points.

4. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 1, characterized in that, In S2, the steps for constructing the multidimensional feature matrix specifically include: The electricity price data column is parsed from the standardized time series and used as the electricity price feature vector; Acquire meteorological features consisting of temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators; The electricity price feature vector and the meteorological feature vector are horizontally aligned and concatenated according to the same time point to generate the multidimensional feature matrix.

5. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 1, characterized in that, In step S3, the step of automatically selecting the prediction model based on the threshold range of the data volume specifically includes: The time span or number of sample points included in the historical electricity price time series data is counted as the data volume statistics; The statistical value of the data volume is compared with a preset first threshold and a second threshold, wherein the first threshold is less than the second threshold; When the statistical value of the data volume is less than or equal to the first threshold, or between the first threshold and the second threshold, a prediction model based on multi-decision tree combination is selected. When the statistical value of the data volume is greater than or equal to the second threshold, a prediction model based on a recurrent neural network is selected.

6. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 5, characterized in that, In step S4, the step of adapting the multidimensional feature matrix according to the selected model type specifically includes: Determine the type of the selected prediction model; If it is a multi-decision tree combination model, the electricity price sequence is extracted from the multi-dimensional feature matrix, a sliding window time series feature is constructed with a preset window length, and the meteorological features corresponding to the same window are combined with the sliding window time series feature to generate an input feature set adapted to the multi-decision tree combination model. If it is a recurrent neural network model, the original time series in the multidimensional feature matrix is ​​directly used as the input feature set adapted to the recurrent neural network model.

7. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 1, characterized in that, In step S5, the step of predicting the electricity value at each time point within a predetermined prediction period specifically includes: The adapted input feature set is used as training samples and input into the selected prediction model for iterative training until the model converges, thus obtaining the trained prediction model. Meteorological forecast data corresponding to future prediction periods are extracted from the standardized time series, and the same adaptation processing method as in step S4 is used to convert the meteorological forecast data into prediction input features that are adapted to the selected prediction model. The predicted input features are input into the trained prediction model, and the electricity value at each time point within the future set prediction period is output sequentially or directly.

8. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 7, characterized in that, The specific steps for extracting meteorological forecast data corresponding to future prediction periods from the standardized time series include: Determine the start and end times of the future forecast period; Based on the start and end times, a subsequence within the time interval is extracted from the standardized time series and used as a standardized subsequence for the future time period; Data columns corresponding to temperature, wind speed, cloud cover, solar radiation intensity, and day / night indicators are extracted from the standardized subsequence of the future time period and used as meteorological forecast data for the future prediction period.

9. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 1, characterized in that, In step S6, the step of generating the prediction report specifically includes: The predicted electricity values ​​at each time point are plotted as a time series curve in chronological order. The predicted electricity value at each time point and the corresponding timestamp are compiled into a structured data table as a data report; The time series graph and / or the data report are integrated into a preset report template to generate a prediction report containing the prediction results, and automatically output to a specified storage path or display interface.

10. The scenario-based electricity market price forecasting method based on multi-source data fusion according to claim 1, characterized in that, After predicting the electricity value at each point in time within the set forecast period, but before generating the forecast report, a manual correction step is also included, specifically: The predicted electricity value at each time point is displayed in a visualization interface, and users can manually adjust some or all of the electricity value. The manual adjustment command is used to modify the electricity value at the indicated time point, and a correction log containing the adjustment time, adjustment content, and information before and after the adjustment is generated. The corrected electricity value is used as the final prediction result to generate the prediction report.

Images