Long-period electricity price prediction method, device and equipment based on supply and demand boundary prediction

Abstract

The invention discloses a long-period electricity price prediction method, device and equipment based on supply and demand boundary prediction, and relates to the technical field of electricity price prediction, and the method comprises the steps: carrying out the screening of a historical database based on a similar day matching algorithm of multi-dimensional feature weighting, obtaining a date which is close to the weather and time sequence features of a prediction day, and carrying out the calculation of the data; obtaining market supply and demand data of the screened dates; jointly inputting a historical electricity price sequence, historical meteorological data and the market supply and demand data into a depth time sequence prediction model to obtain a market supply and demand data prediction value of each set time duration of a prediction day; and constructing an integrated learning model to predict and obtain an electricity price prediction sequence of each target time point of a prediction day in combination with the market supply and demand data prediction value. According to the method, structural contradictions in ultra-long-period high-dimensional prediction are systematically solved through staged dynamic boundary coupling and a rolling error suppression mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electricity price prediction, in particular to a long-period electricity price prediction method, device and equipment based on supply and demand boundary prediction. BACKGROUND

[0002] In the electricity market, the uniform clearing price refers to the market equilibrium price formed through centralized bidding mechanism, that is, the marginal transaction price when the total supply and demand of electricity in a specific period of time reach balance. The core mechanism is that power generation enterprises are ranked from low to high according to the bid, and power users (or power purchasers) are ranked from high to low according to the bid, and the intersection of the supply and demand curves corresponds to the uniform clearing price. The existing electricity price prediction technology is mainly based on historical market data, supply and demand relationship and external influencing factors such as weather conditions, new energy output and policy regulation, and the future electricity price is predicted through statistical, time series, machine learning and deep learning algorithm models.

[0003] Because the electricity price is influenced by multiple external factors, the existing method has difficulty in solving the following contradictions when dealing with 4320-point (45-day 15-minute level) continuity prediction: (1) High-frequency time series models (such as LSTM, Transformer) rely on dense historical data to fit short-term rules, but the structural correlation between historical electricity prices and future electricity prices in long-period samples exceeding 30 days significantly decays (non-stationarity increases), resulting in a significant decrease in the model's ability to capture fluctuation intervals; (2) Existing medium and long-term prediction often relies on coarse-grained statistical modeling (such as monthly average prediction), which cannot support daily 96-point high-resolution output, and forced dimensionality reduction will mask key peak electricity price characteristics; (3) Market boundary conditions (such as bidding space and provincial load regulation) have strong time-varying characteristics in long periods, and existing methods generally treat them as static auxiliary inputs, failing to establish a causal transmission chain of "weather → boundary conditions → electricity price", resulting in continuous accumulation of boundary drift errors in the later period of prediction, eventually forming systematic deviation of the predicted value; (4) The existing data architecture cannot support joint modeling of multi-modal heterogeneous data, and the time scales and resolutions of meteorological data, historical electricity prices and market declaration information are significantly different. Traditional alignment methods cause high-frequency features to be excessively smoothed, further weakening the prediction ability of extreme values of long-period electricity prices.

[0004] The above problems collectively result in long-period prediction results that are ambiguous in medium-term trends and distorted in short-term fluctuations. SUMMARY

[0005] The application provides a long-period electricity price prediction method, device and equipment based on supply-demand boundary prediction, which systematically resolves structural contradictions in long-period high-dimensional prediction through phased dynamic boundary coupling and rolling error suppression mechanism.

[0006] In a first aspect, the embodiments of the application provide a long-period electricity price prediction method based on supply-demand boundary prediction, which comprises: A similar day matching algorithm based on multi-dimensional feature weighting is used to filter dates with similar weather and time sequence characteristics to the prediction day from a historical database, and market supply-demand data of the filtered dates is obtained. The historical electricity price sequence, historical weather data and market supply-demand data are jointly input into a deep time sequence prediction model to obtain market supply-demand data prediction values of each set time length segment of the prediction day. An ensemble learning model is constructed to combine the market supply-demand data prediction values to predict an electricity price prediction sequence of each target time point of the prediction day.

[0007] In combination with the first aspect, in an implementation mode, the similar day matching algorithm based on multi-dimensional feature weighting is used to filter dates with similar weather and time sequence characteristics to the prediction day from a historical database, and market supply-demand data of the filtered dates is obtained, specifically comprising: Based on the weather forecast data of the prediction day in the future days and the timestamp parameters of each set target time point in the prediction day, a similar day matching algorithm based on multi-dimensional feature weighting is used to filter dates with similar weather and time sequence characteristics to the prediction day from a historical database, forming a date set. The market supply-demand data of each date in the date set is obtained, and a dynamic supply-demand boundary reference curve is constructed.

[0008] In combination with the first aspect, in an implementation mode, the market supply-demand data and the dynamic supply-demand boundary reference curve both comprise a bidding space, new energy output, provincial dispatch load and supply-demand ratio.

[0009] In combination with the first aspect, in an implementation mode, the deep time sequence prediction model adopts a Seq2Seq architecture based on an attention mechanism, and an encoder performs spatiotemporal feature extraction on long-period input features, and a decoder outputs market supply-demand data prediction values of each set time length segment of the prediction day.

[0010] In combination with the first aspect, in an implementation mode, for the ensemble learning model, specifically: XGBoost and LightGBM are used as base models, and the base models are trained using real supply-demand data currently disclosed in the electricity market and corresponding time point electricity prices. Through a stacking fusion technology, a logistic regression is used as a meta model to dynamically weight base model prediction results, a robust mapping function is formed, and an integrated learning model is constructed.

[0011] In combination with the first aspect, in an implementation, the XGBoost is used to learn features of a bidding space and a supply-demand ratio, and the LightGBM is used to learn associated features of energy output fluctuation and price spikes.

[0012] In combination with the first aspect, in an implementation, the integrated learning model is constructed to combine the market supply-demand data prediction values to predict a price prediction sequence of each target time point of a prediction day, and specifically includes: The integrated learning model is constructed, the market supply-demand data prediction values are input into the integrated learning model, and a price prediction sequence of each target time point of a prediction day is predicted.

[0013] In combination with the first aspect, in an implementation, the long-period price prediction method further includes: Periodically, a weight coefficient of a similar day matching algorithm is corrected by using the latest disclosed real supply-demand data of a power market, and a deep time series prediction model is rolled and fine-tuned.

[0014] The second aspect provides a long-period price prediction device based on supply-demand boundary prediction, which includes: A screening module is configured to screen, based on a similar day matching algorithm of multi-dimensional feature weighting, dates close to meteorological and time sequence features of a prediction day from a historical database, and obtain market supply-demand data of the screened dates. A first prediction module is configured to input historical price sequences and historical meteorological data and the market supply-demand data into a deep time series prediction model to obtain market supply-demand data prediction values of each set time length section of a prediction day. A second prediction module is configured to construct an integrated learning model to combine the market supply-demand data prediction values to predict a price prediction sequence of each target time point of a prediction day.

[0015] The third aspect provides a long-period price prediction device based on supply-demand boundary prediction, which includes a processor, a memory, and a long-period price prediction program based on supply-demand boundary prediction stored in the memory and executable by the processor.

[0016] The technical scheme provided by the embodiments of the present application brings the beneficial effects including: (1) For the historical correlation decay problem, a similar day matching algorithm is used to inject physical constraints into the deep time sequence model, and the historical supply and demand boundary benchmark is selected through meteorological-time feature weighting, which effectively anchors the long-period evolution trajectory of the non-stationary electricity price sequence and suppresses the model divergence risk; (2) For the market boundary drift problem, the causal transmission chain model of "meteorology→supply and demand boundary→electricity price" is used, and the integrated differentiation capability (linear / nonlinear division) and Stacking dynamic fusion of XGBoost / LightGBM are used to accurately capture the full-cycle influence of boundary conditions such as bidding space and new energy output on electricity price; (3) For the high-dimensional volatility distortion problem, through a daily rolling calibration mechanism, the similar day weight and model parameters are updated based on the latest disclosed data, and the boundary prediction error rate is compressed and blocked in the accumulation of 4320 consecutive predictions; (4) For the problem of heterogeneous data collaboration, a multi-resolution alignment spatio-temporal feature extraction architecture is designed to enable the Seq2Seq encoder to process heterogeneous data simultaneously while preserving key details. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flowchart of a long-period electricity price prediction method based on supply and demand boundary prediction of the present application; Figure 2 is a detailed implementation flowchart of a long-period electricity price prediction method based on supply and demand boundary prediction of the present application; Figure 3 is a functional module schematic diagram of a long-period electricity price prediction device based on supply and demand boundary prediction of the present application; Figure 4 is a hardware structure schematic diagram of a long-period electricity price prediction device based on supply and demand boundary prediction of the present application. DETAILED DESCRIPTION

[0018] In order for those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0019] In order to make the purpose, technical scheme and advantages of the present application clearer, the embodiments of the present application will be described in further detail below with reference to the drawings.

[0020] In the first aspect, the embodiments of this application provide a long-cycle electricity price forecasting method based on supply and demand boundary forecasting. Through a phased dynamic boundary coupling and rolling error suppression mechanism, the structural contradictions in ultra-long-cycle high-dimensional forecasting are systematically resolved.

[0021] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating the long-term electricity price forecasting method based on supply and demand boundary forecasting proposed in this application. Figure 1 As shown, long-term electricity price forecasting methods based on supply and demand boundary forecasting include: S1: Based on a multi-dimensional feature weighted similar day matching algorithm, the algorithm filters out dates from the historical database that are close to the meteorological and time series characteristics of the predicted date, and obtains the market supply and demand data of the filtered dates. S2: Input the historical electricity price series, historical meteorological data, and the market supply and demand data into the deep time series forecasting model to obtain the market supply and demand data forecast values ​​for each set time period on the forecast day; S3: Construct an ensemble learning model to combine the market supply and demand data forecasts to predict the electricity price forecast sequence for each target time point on the forecast day.

[0022] Furthermore, in one embodiment, a similar day matching algorithm based on multi-dimensional feature weighting is used to filter dates from a historical database that are close to the meteorological and time-series characteristics of the predicted date, and to obtain market supply and demand data for the filtered dates, specifically including: S101: Based on the meteorological forecast data of the predicted day with the number of future days, and the timestamp parameters of each set target time point on the predicted day, a similar day matching algorithm with multi-dimensional features is used to select dates that are close to the meteorological and time series characteristics of the predicted day from the historical database to form a date set; among them, the meteorological forecast data includes temperature, wind speed, light intensity, etc. It should be noted that for the timestamp parameter, the similar day matching algorithm is a non-time series model, so we borrow the idea of ​​time series models and use past historical time points as features to construct the timestamp parameter; for meteorological and time series features, we use meteorological data and related time series parameters, such as temperature, wind speed, weekdays, holidays, etc. over a period of time. By using a similar day matching algorithm, historical data can be used to predict data for a target time period. For example, if the weather and other factors on October 20 are similar to those on September 10, the boundary conditions on October 20 can be referenced from those on September 10. S102: Obtain market supply and demand data for each date in the date set and construct a dynamic supply and demand boundary benchmark curve. Both the market supply and demand data and the dynamic supply and demand boundary benchmark curve include bidding space, renewable energy output, provincial dispatch load, and demand-supply ratio.

[0023] Specifically, for the long-period electricity price prediction method based on supply and demand boundary prediction of the application, the first stage is to construct a dynamic supply and demand boundary benchmark, for example: the input is the weather forecast data for the next 45 days and the time stamp parameters for 96 time points per day for the next 45 days, then through a multi-dimensional feature weighted similar day matching algorithm, the date set closest to the prediction day in terms of weather and time sequence characteristics is selected from the historical database, and then based on the market supply and demand data of the date set, i.e. the bidding space, new energy output, provincial load regulation, etc. The dynamic supply and demand boundary benchmark curve is constructed as the basis for subsequent prediction, the input is the dynamic supply and demand boundary benchmark curve, including bidding space, new energy output, provincial load regulation, supply-demand ratio, etc. It should be noted that the time points in the application are obtained by equal division.

[0024] Further, in an embodiment, the deep time series prediction model adopts a Seq2Seq architecture based on an attention mechanism, the encoder extracts spatial and temporal features from the long-period input features, and the decoder outputs the market supply and demand data prediction value of each set time length of the prediction day.

[0025] Specifically, for the long-period electricity price prediction method based on supply and demand boundary prediction of the application, the second stage is high-resolution supply and demand relationship prediction, for example: the dynamic boundary data generated in the first stage, including bidding space, new energy output, provincial load regulation, supply-demand ratio, etc., as well as historical electricity price sequence and historical weather data are jointly input into a deep time series prediction model, the deep time series prediction model is a Seq2Seq architecture based on an attention mechanism (Encoder-Decoder with Attention), the encoder extracts spatial and temporal features from the long-period input features (45 days x 96 points x N-dimensional features), and the decoder outputs high-resolution market supply and demand data prediction values (or market supply and demand relationship prediction values) for every 15 minutes (a total of 4320 points) in the future 45 days, i.e. the supply and demand relationship prediction results for 45 days and 4320 points (providing fine-grained boundary conditions for electricity price mapping).

[0026] Further, in an embodiment, for the ensemble learning model, specifically: XGBoost and LightGBM are used as base models, and the real supply and demand data currently disclosed in the electricity market and the corresponding time point electricity price are used to train the base models; through Stacking fusion technology, a meta-model is formed by dynamically weighting the base model prediction results with logistic regression to form a robust mapping function, realizing the construction of the ensemble learning model. XGBoost is used to learn the characteristics of bidding space and supply-demand ratio, and LightGBM is used to learn the correlation characteristics of energy output fluctuation and electricity price peak. XGBoost is a machine learning algorithm that focuses on computing speed and machine learning model performance; LightGBM is a distributed high-performance framework that uses decision trees for ranking, classification and regression tasks.

[0027] In this application, an ensemble learning model is constructed to combine the market supply and demand data forecast values ​​to predict the electricity price forecast sequence for each target time point on the forecast day. Specifically, this includes: constructing an ensemble learning model, inputting the market supply and demand data forecast values ​​into the ensemble learning model, and predicting the electricity price forecast sequence for each target time point on the forecast day.

[0028] Specifically, for the long-term electricity price forecasting method based on supply and demand boundary forecasting in this application, the third stage is the final forecast. For example, the input is the 45-day supply and demand boundary forecasting data output from the second stage. XGBoost and LightGBM are used as base models. The base models are trained using the real supply and demand data currently disclosed in the electricity market and the corresponding electricity prices at the time point. Among them, XGBoost focuses on learning the linearly strong correlation features such as bidding space and demand-supply ratio, while LightGBM focuses on capturing the complex nonlinear relationship between the fluctuation of new energy output and the peak of electricity price. Finally, through the Stacking fusion technology, logistic regression is used as the meta-model to dynamically weight the base model prediction results to form a robust mapping function of "dynamic boundary-electricity price". The 45-day supply and demand boundary forecasting data output from the second stage is input into the fusion model to generate a high-precision electricity price forecasting sequence for the next 45 days and 4320 points.

[0029] Furthermore, in one embodiment, the long-term electricity price forecasting method further includes: periodically using the latest disclosed real supply and demand data from the electricity market to correct the weight coefficients of the similar day matching algorithm, and performing rolling fine-tuning on the deep time series forecasting model.

[0030] Specifically, to control the propagation of errors in long-term forecasts, a feedback calibration mechanism is added. The weight coefficients of the similar-day matching algorithm are corrected daily using the latest disclosed real supply and demand data from the electricity market (i.e., timely updates to the model training set and retraining for parameter tuning). Furthermore, the deep time series forecasting model undergoes rolling fine-tuning (i.e., daily updates to the training set and batch training), thereby suppressing the accumulation of forecast bias caused by boundary condition drift. See also Figure 2 The diagram shown is a detailed implementation flowchart of the long-cycle electricity price forecasting method based on supply and demand boundary forecasting in this application.

[0031] The long-cycle electricity price forecasting method based on supply and demand boundary prediction in this application addresses the historical correlation decay problem by employing a similar day matching algorithm to inject physical constraints into the deep time series model. It effectively anchors the long-cycle evolution trajectory of non-stationary electricity price sequences and suppresses model divergence risk by using historical supply and demand boundary benchmarks selected through meteorological-time feature weighting. To address the market boundary drift problem, it models a causal transmission chain of "meteorology → supply and demand boundary → electricity price," utilizing the integrated differentiation capabilities (linear / nonlinear divide-and-conquer) and Stacking dynamic fusion of XGBoost / LightGBM to accurately capture the full-cycle impact of time-varying boundary conditions such as bidding space and renewable energy output on electricity prices. To address the high-dimensional volatility distortion problem, it uses a daily rolling calibration mechanism to update the similar day weights and model parameters based on the latest disclosed data, compressing the boundary prediction error rate and preventing its accumulation in continuous predictions at 4320 points. Finally, to address the heterogeneous data collaboration problem, it designs a multi-resolution aligned spatiotemporal feature extraction architecture, enabling the Seq2Seq encoder to synchronously process heterogeneous data while preserving key details.

[0032] Secondly, embodiments of this application also provide a long-cycle electricity price forecasting device based on supply and demand boundary forecasting.

[0033] In one embodiment, reference is made to Figure 3 , Figure 3 This is a schematic diagram of the functional modules of the long-cycle electricity price forecasting device based on supply and demand boundary forecasting in this application. Figure 3 As shown, the long-term electricity price forecasting device based on supply and demand boundary forecasting includes: a screening module, a first forecasting module, and a second forecasting module.

[0034] The filtering module is used to filter dates from the historical database that are close to the meteorological and time series characteristics of the prediction date based on a multi-dimensional feature weighted similar day matching algorithm, and obtain the market supply and demand data of the filtered dates; the first prediction module is used to input the historical electricity price series, historical meteorological data, and the market supply and demand data into the deep time series prediction model to obtain the predicted value of market supply and demand data for each set time length segment of the prediction date; the second prediction module is used to construct an ensemble learning model to combine the predicted value of the market supply and demand data to predict the electricity price prediction series for each target time point of the prediction date.

[0035] Thirdly, embodiments of this application provide a long-term electricity price forecasting device based on supply and demand boundary forecasting. The long-term electricity price forecasting device based on supply and demand boundary forecasting can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.

[0036] Reference Figure 4 , Figure 4This is a schematic diagram of the hardware structure of a long-cycle electricity price forecasting device based on supply and demand boundary forecasting involved in the embodiments of this application. In the embodiments of this application, the long-cycle electricity price forecasting device based on supply and demand boundary forecasting may include a processor, a memory, a communication interface, and a communication bus.

[0037] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0038] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting devices within the long-term electricity price forecasting equipment based on supply-demand boundary forecasting, as well as interfaces used for interconnecting the long-term electricity price forecasting equipment with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0039] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0040] The processor can be a general-purpose processor, which can call a long-cycle electricity price forecasting program based on supply and demand boundary forecasting stored in memory and execute the long-cycle electricity price forecasting method based on supply and demand boundary forecasting provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the long-cycle electricity price forecasting program based on supply and demand boundary forecasting is called can refer to the various embodiments of the long-cycle electricity price forecasting method based on supply and demand boundary forecasting in this application, and will not be repeated here.

[0041] Those skilled in the art will understand that Figure 4 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0042] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0043] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0044] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0045] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0046] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0047] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A long-term electricity price forecasting method based on supply and demand boundary forecasting, characterized in that, The long-term electricity price forecasting method based on supply and demand boundary prediction includes: Based on a multi-dimensional feature weighted similar day matching algorithm, dates that are close to the meteorological and time series characteristics of the predicted date are selected from the historical database, and market supply and demand data of the selected dates are obtained. The historical electricity price series, historical meteorological data, and the market supply and demand data are input into the deep time series forecasting model to obtain the market supply and demand data forecast values ​​for each set time period on the forecast day. An ensemble learning model is constructed to combine the market supply and demand data forecasts to predict the electricity price forecast sequence for each target time point on the forecast date.

2. The long-term electricity price forecasting method based on supply and demand boundary forecasting as described in claim 1, characterized in that, The similar day matching algorithm based on multi-dimensional feature weighting filters dates from historical databases that are close to the meteorological and time-series characteristics of the predicted date, and obtains market supply and demand data for the filtered dates, specifically including: Based on the meteorological forecast data of the predicted day in the future number of days, and the timestamp parameters of each set target time point in each day of the predicted day, a similar day matching algorithm with multi-dimensional feature weighting is used to filter out dates that are close to the meteorological and time sequence characteristics of the predicted day from the historical database, forming a date set; Obtain market supply and demand data for each date in the date set, and construct a dynamic supply and demand boundary benchmark curve.

3. The long-term electricity price forecasting method based on supply and demand boundary forecasting as described in claim 2, characterized in that: The market supply and demand data and dynamic supply and demand boundary benchmark curves mentioned above include bidding space, new energy output, provincial dispatch load, and demand-supply ratio.

4. The long-term electricity price forecasting method based on supply and demand boundary forecasting as described in claim 1, characterized in that: The deep time series prediction model adopts a Seq2Seq architecture based on an attention mechanism. The encoder extracts spatiotemporal features from long-period input features, while the decoder outputs the predicted market supply and demand data for each set time period of the prediction day.

5. The long-term electricity price forecasting method based on supply and demand boundary forecasting as described in claim 1, characterized in that, For ensemble learning models, specifically: Using XGBoost and LightGBM as base models, the base models are trained using the real supply and demand data disclosed in the current electricity market and the corresponding electricity prices at the time point. By using Stacking fusion technology, logistic regression is used as the meta-model to dynamically weight the prediction results of the base model, forming a robust mapping function, and thus realizing the construction of an ensemble learning model.

6. The long-term electricity price forecasting method based on supply and demand boundary forecasting as described in claim 5, characterized in that: The XGBoost is used to learn the characteristics of bidding space and demand-supply ratio, while the LightGBM is used to learn the correlation characteristics between energy output fluctuations and electricity price peaks.

7. The long-term electricity price forecasting method based on supply and demand boundary forecasting as described in claim 5, characterized in that, The construction of the ensemble learning model, which combines the market supply and demand data forecasts to predict the electricity price forecast sequence for each target time point on the forecast date, specifically includes: An ensemble learning model is constructed, and the predicted market supply and demand data are input into the ensemble learning model to predict the electricity price forecast sequence for each target time point on the prediction day.

8. The long-term electricity price forecasting method based on supply and demand boundary forecasting as described in claim 1, characterized in that, The long-term electricity price forecasting method also includes: The weighting coefficients of the similar day matching algorithm are periodically adjusted using the latest real supply and demand data disclosed in the electricity market, and the deep time series forecasting model is fine-tuned on a rolling basis.

9. A long-term electricity price forecasting device based on supply and demand boundary forecasting, characterized in that, The long-term electricity price forecasting device based on supply and demand boundary prediction includes: The filtering module is used to filter out dates that are close to the meteorological and time-series characteristics of the predicted date from the historical database based on a multi-dimensional feature weighted similar day matching algorithm, and obtain market supply and demand data for the filtered dates. The first forecasting module is used to input historical electricity price sequences, historical meteorological data, and the market supply and demand data into a deep time series forecasting model to obtain the market supply and demand data forecast values ​​for each set time period on the forecast day. The second prediction module is used to construct an integrated learning model to combine the market supply and demand data prediction values ​​to predict the electricity price prediction sequence for each target time point on the prediction day.

10. A long-term electricity price forecasting device based on supply and demand boundary forecasting, characterized in that, The long-cycle electricity price forecasting device based on supply and demand boundary forecasting includes a processor, a memory, and a long-cycle electricity price forecasting program based on supply and demand boundary forecasting stored in the memory and executable by the processor. When the long-cycle electricity price forecasting program based on supply and demand boundary forecasting is executed by the processor, it implements the steps of the long-cycle electricity price forecasting method based on supply and demand boundary forecasting as described in any one of claims 1 to 8.

Images