Electric power market day-ahead price difference prediction method and related equipment
By combining the LASSO regression model with a dynamic penalty factor generator and a proximal gradient solver, the problems of insufficient quantification capability and poor interpretability of existing electricity price prediction models for sudden events are solved, and high-precision, low-latency electricity market price difference prediction is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUNAN CLEAN ENERGY BRANCH OF HUANENG INT POWER CO LTD
- Filing Date
- 2026-01-20
- Publication Date
- 2026-05-01
AI Technical Summary
Existing electricity price forecasting models are insufficient in quantifying sudden events, rely on complex manual feature engineering, lack interpretability in deep learning models, and are computationally and parameter tuning complex in hybrid models, resulting in low prediction accuracy and poor interpretability.
We employ a LASSO regression model combined with a dynamic penalty factor generator and a proximal gradient solver. By using multi-source feature input data, we predict price spreads, dynamically adjust the penalty factor, and iteratively solve for the optimal parameters using the proximal gradient algorithm and Nesterov acceleration strategy, thereby improving the model's adaptability and interpretability.
It improves the accuracy and interpretability of day-ahead price forecasts in the electricity market, reduces computational complexity, adapts to dynamic market changes, and lowers forecast latency.
Smart Images
Figure CN121961652A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of day-ahead forecasting technology in the electricity market, specifically to a method and related equipment for forecasting day-ahead price differences in the electricity market. Background Technology
[0002] Currently, the field of electricity price forecasting primarily employs deep learning and machine learning algorithms, exhibiting a diversified development trend in technical approaches. In time series statistical methods, the ARIMA-GARCH family of models is widely used due to its solid theoretical foundation and high computational efficiency; among them, the asymmetric GARCH model effectively captures the asymmetric characteristics of electricity price fluctuations. Shallow machine learning methods (such as BP neural networks, GRNN, and ELM) compensate for the shortcomings of statistical models through nonlinear mapping capabilities, enabling them to handle noisy historical data and establish complex relationship models between electricity prices and multiple factors. Deep learning models (such as LSTM and GRU) significantly improve prediction accuracy through hierarchical feature extraction capabilities, particularly excelling in processing periodic electricity price data; combined models (such as wavelet decomposition-probabilistic regression hybrid models) further integrate signal processing techniques, optimizing prediction performance. Furthermore, the traditional LASSO model balances interpretability and prediction accuracy in electricity price forecasting through sparse constraints, enabling it to finely characterize intraday electricity price fluctuations and screen key influencing factors.
[0003] However, the aforementioned existing technologies still have several inherent limitations. First, time series models lack the ability to quantify sudden events (such as extreme weather or abrupt changes in electricity prices caused by policy adjustments), leading to prediction lags. Second, existing methods heavily rely on manual feature engineering, requiring the construction of specialized features (such as real-time electricity price lag terms and load-price elasticity coefficients), increasing model construction complexity. Furthermore, deep learning models face the challenge of insufficient data, especially in the early stages of electricity market reform or emerging markets, and their black-box nature makes variable contributions difficult to interpret, limiting their application in critical decision-making scenarios. While the LASSO model possesses interpretability, its uniform L1 penalty term applies the same compression weights to influencing factors of different orders of magnitude, potentially distorting the ranking of variable importance and affecting prediction accuracy. Simultaneously, while existing hybrid models offer high accuracy, the signal decomposition process significantly increases computational complexity, and the parameter tuning process for probabilistic prediction methods (such as quantile regression combined with kernel density estimation) is complex and requires the support of domain expert knowledge. These issues constrain the reliability, efficiency, and interpretability of electricity price prediction models in practical applications. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to provide a method and related equipment for predicting the price difference of day-ahead electricity market prices, which addresses the shortcomings of the prior art and solves the problem of low prediction accuracy caused by insufficient quantitative ability of existing electricity price prediction for sudden situations.
[0005] The objective of this invention is achieved through the following technical solutions: In a first aspect, the present invention provides a method for predicting the price difference of day-ahead prices in the electricity market, comprising: Acquire electricity market boundary data, meteorological observation data, and line maintenance plan data for the current target area; The electricity market boundary data, meteorological observation data, and line maintenance plan data are preprocessed to form multi-source feature input data; The multi-source feature input data is input into the LASSO regression model to predict the price difference between the day-to-day and real-time values. The LASSO regression model includes a dynamic penalty factor generator and a proximal gradient solver. The dynamic penalty factor generator is used to dynamically adjust based on multi-source feature input data and the predicted price difference of the previous time period. The proximal gradient solver is used to iteratively solve for the optimal parameters according to the proximal gradient algorithm combined with the Nesterov acceleration strategy and the convergence threshold judgment condition to obtain the price difference prediction result.
[0006] As a further improvement of the present invention, the electricity market boundary data includes at least the load and renewable energy output data of the power grid dispatch center, non-market output data, and tie-line power.
[0007] As a further improvement of the present invention, the preprocessing of the electricity market boundary data, meteorological observation data, and line maintenance plan data includes: The 3σ principle is used to remove outliers from the electricity market boundary data; Missing values in the electricity market boundary data, meteorological observation data, and line maintenance plan data after outlier processing were filled using a missing value imputation method based on time series similarity. The filled-in electricity market boundary data was cleaned and then subjected to feature standardization; meteorological observation data and line maintenance plan data were also standardized.
[0008] As a further improvement of the present invention, the multi-source feature input data includes: introducing holiday dummy variables, lagged price difference features, load-new energy output features, line maintenance features, and meteorological features; The holiday dummy variable is obtained by dividing the holidays throughout the year based on the One-Hot coding method; the lagged price difference feature is obtained by statistical analysis of price difference data for a set historical period; the load-new energy output feature is obtained based on electricity market boundary data; the line maintenance feature is obtained by representing line maintenance plan data based on three-dimensional coding; the meteorological feature includes at least temperature, humidity, equivalent temperature load index, wind speed, and irradiance.
[0009] As a further improvement to the present invention, the LASSO regression model prediction step includes: Initial linear regression is performed on the multi-source feature input data and historical price difference values based on the least squares method to obtain the initial coefficients of each feature in the multi-source feature input data; The adaptive weight vector for each feature is calculated based on the initial coefficients of each feature. The baseline penalty factor is determined based on the Bayesian information criterion of the sliding window in the dynamic penalty factor generator, and the dynamic penalty factor is determined based on the baseline penalty factor. Based on the determined adaptive weight vector and dynamic penalty factor, the objective function is solved iteratively using the proximal gradient algorithm. When the convergence threshold condition is met, the predicted price difference is output.
[0010] As a further improvement of the present invention, the dynamic penalty factor generator is as follows:
[0011] In the formula, Let be the dynamic penalty factor for time period t. As the benchmark penalty factor, For adjustment coefficients, For the first Forecast price spread for the time period.
[0012] As a further improvement of the present invention, the proximal gradient algorithm also introduces an adaptive weight vector for feature selection during the parameter update process; the adaptive weight vector is:
[0013] In the formula, Let be the adaptive weight vector for the j-th feature; The initial coefficients for the j-th feature are obtained through the least squares method. For smoothing terms, =1e -4 .
[0014] Secondly, the present invention provides a price difference prediction system for day-ahead prices in the electricity market, comprising: The data acquisition module acquires electricity market boundary data, meteorological observation data, and line maintenance plan data for the current target area. The preprocessing module preprocesses the electricity market boundary data, meteorological observation data, and line maintenance plan data to form multi-source feature input data; The prediction module inputs the multi-source feature input data into the LASSO regression model to predict the price difference between the day-to-day and real-time values. The LASSO regression model includes a dynamic penalty factor generator and a proximal gradient solver. The dynamic penalty factor generator is used to dynamically adjust based on multi-source feature input data and the predicted price difference of the previous time period. The proximal gradient solver is used to iteratively solve for the optimal parameters according to the proximal gradient algorithm combined with the Nesterov acceleration strategy and the convergence threshold judgment condition to obtain the price difference prediction result.
[0015] Thirdly, the present invention provides a computer-readable storage medium storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the above-described method for predicting the price difference of day-ahead electricity market prices.
[0016] Fourthly, the present invention provides a computing device, comprising: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including steps for performing the above-described method for predicting the day-ahead price of the electricity market.
[0017] The beneficial effects of this invention are as follows: Attached Figure Description To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a schematic diagram of the method for predicting the price difference of the day-ahead price in the electricity market in an embodiment of the present invention; Figure 2 This is a comparison chart of prediction performance in embodiments of the present invention; Figure 3 This is a schematic diagram of the internal structure of an electronic device in an embodiment of the present invention. Detailed Implementation
[0019] To make the objectives and technical solutions of this invention clearer and easier to understand, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.
[0020] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. The described embodiments are only some embodiments of the present invention, and not all embodiments.
[0021] Example 1 Among existing methods for predicting electricity market price spreads, traditional time series models (such as ARIMA) are unable to cope with the characteristics of high-frequency fluctuations and nonlinear jumps in the market, while deep learning models (such as LSTM) have the drawbacks of poor interpretability and reliance on a large amount of labeled data. Traditional LASSO models use a fixed penalty factor, which cannot adapt to dynamic changes in the market, resulting in low prediction accuracy and insufficient practicality. To address the aforementioned issues, this embodiment provides a method for predicting the day-ahead price spread in the electricity market, comprising: acquiring electricity market boundary data, meteorological observation data, and line maintenance plan data for the current target area; preprocessing the electricity market boundary data, meteorological observation data, and line maintenance plan data to form multi-source feature input data; inputting the multi-source feature input data into a LASSO regression model to predict the day-ahead-real-time price spread; the LASSO regression model includes a dynamic penalty factor generator and a proximal gradient solver, wherein the dynamic penalty factor generator is used to dynamically adjust based on the multi-source feature input data and the predicted price spread of the previous period; and the proximal gradient solver is used to iteratively solve for the optimal parameters based on the proximal gradient algorithm combined with the Nesterov acceleration strategy and convergence threshold judgment conditions to obtain the price spread prediction result.
[0022] By applying the prediction method in this embodiment, by acquiring electricity market boundary data, meteorological observation data, and line maintenance plan data for the current target area, it is possible to comprehensively collect core basic data affecting the day-ahead-real-time price difference in the electricity market. Compared with the existing technology that relies on only a single type of data (such as only load data) for prediction, it can avoid the one-sidedness of prediction due to insufficient data dimensions and provide more complete data support for subsequent predictions. By preprocessing the above three types of data to form multi-source feature input data, noise, missing values, and dimensional differences in the original data can be eliminated, improving the quality and consistency of the input data and reducing the interference of data problems on the prediction results. By inputting the multi-source feature input data into the LASSO regression model for price spread prediction, the feature selection capability inherent in the LASSO regression model improves the interpretability of the prediction results compared to existing black-box deep learning models such as LSTM, making it easier to clarify the impact of each factor on the price spread and solving the problem of black-box models struggling to explain the contribution of variables. Furthermore, by setting a dynamic penalty factor generator in the LASSO regression model that dynamically adjusts based on the multi-source feature input data and the predicted price spread of the previous period, compared to the fixed penalty factor used in traditional LASSO models, the penalty intensity can be adjusted according to market data characteristics and prediction errors. The method employs dynamic adaptation to avoid the problem of insufficient penalty during periods of high market volatility and excessive penalty during periods of market stability caused by a fixed penalty factor, thereby improving the model's adaptability to dynamic market changes. By incorporating a proximal gradient solver in the LASSO regression model that iteratively solves for optimal parameters using a proximal gradient algorithm combined with a Nesterov acceleration strategy and a convergence threshold, the convergence speed of parameter solving is accelerated. Simultaneously, the convergence threshold ensures the stability of parameter solving, avoiding excessively high prediction delays caused by non-convergence or slow convergence. The various steps of this prediction method synergistically form a complete prediction process: "comprehensive data acquisition - high-quality preprocessing - interpretable model prediction - dynamic adaptation penalty - efficient and stable solution." This achieves a comprehensive effect of improved prediction accuracy, enhanced interpretability, improved adaptability, and increased solution efficiency, better meeting the actual needs of the electricity market for high-precision, interpretable, and low-latency price spread predictions.
[0023] It should be noted that the electricity market boundary data comes from the target region's electricity trading platform (such as the official website of the provincial electricity trading center) and the database of the power grid dispatch center. The electricity market boundary data includes at least load and renewable energy output data, non-market output data, and tie-line power from the power grid dispatch center. Meteorological observation data comes from the public data interface of the provincial meteorological service center in the target region, including indicators such as temperature, humidity, wind speed, and irradiance. Line maintenance plan data comes from the operation and maintenance management department of the power grid company in the target region, including information such as the name of the line to be maintained, the maintenance period, and the maintenance capacity.
[0024] The preprocessing method in this embodiment specifically includes: outlier removal from the electricity market boundary data using the 3σ principle; missing value imputation based on time series similarity to fill in missing values in the outlier-removed electricity market boundary data, meteorological observation data, and line maintenance plan data; data cleaning and feature standardization of the imputed electricity market boundary data; and standardization of the meteorological observation data and line maintenance plan data. Specifically, firstly, outliers in the electricity market boundary data are removed using the 3σ principle to eliminate the interference of extreme noise on the model; then, missing value imputation based on time series similarity utilizes the temporal correlation and periodicity of the data to restore reasonable values of the missing data, avoiding bias caused by simple interpolation; finally, data cleaning and Z-score standardization unify the data format and eliminate dimensional differences, enabling the model to treat different types of features (such as load and temperature) fairly, ensuring the stability of parameter solutions.
[0025] Specifically, during periods of extreme weather or equipment failure, a multiple imputation method based on time series similarity is used to handle missing values. This involves calculating the weighted average of data from adjacent time periods using a sliding window, with the weights determined by both time distance and load curve similarity. Next, the original transaction data undergoes multi-dimensional cleaning, including outlier removal (using the 3σ principle to handle price fluctuations), missing value imputation (based on time series linear interpolation), and feature standardization (Z-score normalization).
[0026] The preprocessed multi-source data is only basic data and has not been transformed into features that directly reflect the influence of price differences. Therefore, this embodiment also uses multi-source data to obtain multi-source feature input data after processing. This includes introducing holiday dummy variables, lagged price difference features, load-new energy output features, line maintenance features, and meteorological features. The holiday dummy variables are obtained by dividing the year's holidays using One-Hot coding. The lagged price difference features are obtained based on historical price difference data for a set period. The load-new energy output features are obtained from electricity market boundary data, and the line maintenance features are obtained by representing line maintenance plan data using three-dimensional coding. Meteorological features include at least temperature, humidity, equivalent temperature load index, wind speed, and irradiance.
[0027] Furthermore, the holiday dummy variables are divided into four categories based on the One-Hot encoding method: weekdays, weekends, short holidays (3-5 days, such as Qingming Festival and Labor Day), and long holidays (7 days or more, such as Spring Festival and National Day). The One-Hot encoding rules are as follows: weekdays correspond to vectors [1,0,0,0], weekends correspond to [0,1,0,0], short holidays correspond to [0,0,1,0], and long holidays correspond to [0,0,0,1]. For example, May 1, 2024 (Labor Day, short holiday) is encoded as [0,0,1,0], and May 2, 2024 (weekday) is encoded as [1,0,0,0]. The encoded data is associated with the timestamp of the current prediction period and serves as the time attribute feature of the model.
[0028] The lagged price spread characteristics are obtained based on historical price spread data for a set historical period: the set historical period is specifically the period from t-1 to t-24 of the current prediction period (covering the previous 24 15-minute periods, i.e., 6 hours); the statistically obtained characteristics include: the average price spread from t-1 to t-24 (e.g., the price spreads from t-1 to t-24 are 50, 52, 48, and 49 yuan / MWh respectively, with an average of 49.5 yuan / MWh), the price spread variance (reflecting the degree of volatility, e.g., a variance of 2.3), and the price spread skewness (reflecting the symmetry of the distribution, e.g., a skewness of 0.8 indicates that the price spread is right-skewed, and there is a higher probability of a high price spread); the lagged price spread characteristics are directly related to the short-term trend of the price spread, providing historical volatility references for the model.
[0029] The load-renewable energy output characteristics are obtained based on electricity market boundary data. Specifically, they include three core indicators: 1) total system load (directly using pre-processed load data); 2) wind power / solar power output ratio (wind power output ratio = total regional wind power output / total system load, solar power output ratio = total regional solar power output / total system load; e.g., total wind power output 800MW, total system load 5000MW, wind power output ratio = 0.16); 3) net load (net load = total system load - total regional renewable energy output; e.g., total system load 5000MW, total renewable energy output 1200MW, net load = 3800MW). In addition, this embodiment also introduces the renewable energy output penetration rate mutation index ΔR, defined as:
[0030] in, The percentage of renewable energy output during the current period. This represents the percentage from the previous period, reflecting the degree of sudden changes in renewable energy output.
[0031] The characteristics of line maintenance are derived from line maintenance plan data based on three-dimensional coding. The three dimensions of the three-dimensional coding are: the proportion of the line under maintenance capacity (a continuous value, calculated as "rated capacity of the line under maintenance / total transmission capacity of the target area power grid", e.g., if the capacity of the line under maintenance is 500MW and the total transmission capacity of the area is 2500MW, the proportion = 0.2); the critical section identifier (0 / 1 dummy element, critical sections refer to 500kV and above lines connecting major power bases and load centers, 1 for critical sections, 0 otherwise); and the remaining duration of the maintenance period (a continuous value in hours, e.g., if maintenance starts at 08:00, the current predicted period is 10:00, and the maintenance plan lasts until 16:00, the remaining duration = 6 hours). The three-dimensional coding comprehensively depicts the impact of line maintenance on the power grid transmission capacity.
[0032] Meteorological characteristics include at least temperature, humidity, equivalent temperature load index, wind speed, and irradiance. All data for temperature, humidity, wind speed, and irradiance are preprocessed and standardized. The equivalent temperature load index (ETLI) is constructed using a multiple regression model.
[0033] By capturing the impact of time attributes on price spreads through holiday dummy variables (such as a decrease in load during long holidays leading to a reduction in price spreads), reflecting the short-term fluctuation trend of price spreads through lagged price spread characteristics, characterizing the core changes in supply and demand through load-renewable energy output characteristics, reflecting the impact of power grid physical constraints through line maintenance characteristics (such as transmission bottlenecks caused by maintenance of key lines, and regional differentiation of price spreads), and associating the indirect impact of meteorological factors on load and renewable energy output through meteorological characteristics, these multi-dimensional features complement each other to jointly construct a feature system that can comprehensively reflect the driving logic of price spreads, providing high-quality input for the LASSO regression model.
[0034] The LASSO regression model prediction steps include: performing initial linear regression on the multi-source feature input data and historical price difference values based on the least squares method to obtain the initial coefficients of each feature in the multi-source feature input data; calculating the adaptive weight vector of each feature based on the initial coefficients of each feature; determining the baseline penalty factor according to the Bayesian information criterion of the sliding window in the dynamic penalty factor generator; determining the dynamic penalty factor based on the baseline penalty factor; iteratively solving the objective function using the proximal gradient algorithm based on the determined adaptive weight vector and dynamic penalty factor; and outputting the price difference prediction value when the convergence threshold condition is met.
[0035] The dynamic penalty factor generator is as follows:
[0036] In the formula, Let be the dynamic penalty factor for time period t. As the benchmark penalty factor, For adjustment coefficients, For the first The predicted price spread for the previous period. The dynamic penalty factor generator uses a benchmark penalty factor as a basis, and the absolute value of the predicted price spread for the previous period as a signal indicator of market volatility. When the value increases, the surface market fluctuates violently, thus increasing the dynamic penalty factor, enhancing the contraction effect on the characteristic coefficients, and filtering out noisy features. When When the index decreases, the market tends to stabilize, the index term approaches 1, the contraction effect is weakened, detailed characteristics are preserved, the adjustment coefficient controls the change range of the index term, avoids large fluctuations in the dynamic penalty factor due to small errors, and ensures the smoothness of the penalty intensity adjustment.
[0037] The proximal gradient algorithm also introduces an adaptive weight vector for feature selection during parameter updates; the adaptive weight vector is:
[0038] In the formula, Let be the adaptive weight vector for the j-th feature; The initial coefficients for the j-th feature are obtained through the least squares method. For smoothing terms, =1e -4 .
[0039] The initial coefficients of each feature are obtained by least squares, providing a reliable basis for adaptive weight calculation. The adaptive weight vector is adjusted in reverse according to the absolute value of the initial coefficients to ensure that important features (large coefficients) are subject to weak penalties and minor features (small coefficients) are subject to strong penalties. The benchmark penalty factor is determined within a sliding window using the BIC criterion to balance the model fit and complexity. The dynamic penalty factor is adjusted in real time based on the benchmark value and the error of the previous period to adapt to market fluctuations. Finally, the objective function is efficiently solved by a proximal gradient algorithm with Nesterov acceleration, and the predicted value and confidence interval are output after the convergence condition is met.
[0040] Example 2 This embodiment provides a specific implementation method for predicting the price difference of day-ahead prices in the electricity market.
[0041] The model is solved using the proximal gradient algorithm, and its pseudocode is as follows: Input: Data matrix X, response vector Y, parameters , Maximum number of iterations T initialized: =0, =1, =1 / L (L is Lipchitz constant); while t≤Tand ; do #Calculate gradient Gradient descent;
[0042] #Proximal operator t=t+1; Endwhile; Output: Coefficient vector β; Convergence threshold Terminate the iteration when the parameter change is less than this value.
[0043] To verify the reliability of convergence, a dual verification mechanism is set up: 1) the rate of decrease of the objective function changes by less than 1% over 5 consecutive iterations; 2) the L2 norm change of the coefficient vector is less than... .
[0044] The model output is continuously optimized via a rolling update module, employing a sliding window mechanism to update parameters every 15 minutes. Through incremental learning, it preserves long-term patterns while adapting to dynamic market changes. A visual interactive interface displays real-time forecast values, confidence intervals, and weight changes of key influencing factors, providing decision support for traders.
[0045] like Figure 2 As shown, this embodiment performs a statistical significance test (t-test, p < 0.05) and then checks for logical consistency with physical constraints (such as transmission capacity limitations). The adaptive LASSO model achieves differentiated compression of key variables by introducing a penalty term weighted by historical information. Specifically, the model dynamically adjusts weights using parameter estimation results within a rolling time window: lower penalty coefficients are assigned to core variables with stable historical performance (such as lagged electricity prices and load levels); stronger compression is applied to secondary variables with higher volatility (such as temporary policy interventions).
[0046] Example 3 Based on the electricity market day-ahead price spread prediction method in Example 1, this example provides an electricity market day-ahead price spread prediction system, including: The data acquisition module acquires electricity market boundary data, meteorological observation data, and line maintenance plan data for the current target area. The preprocessing module preprocesses the electricity market boundary data, meteorological observation data, and line maintenance plan data to form multi-source feature input data. The prediction module inputs the multi-source feature input data into the LASSO regression model to predict the day-ahead-real-time price spread. The LASSO regression model includes a dynamic penalty factor generator and a proximal gradient solver. The dynamic penalty factor generator is used to dynamically adjust based on the multi-source feature input data and the predicted price spread of the previous period. The proximal gradient solver is used to iteratively solve for the optimal parameters based on the proximal gradient algorithm combined with the Nesterov acceleration strategy and convergence threshold conditions to obtain the price spread prediction result.
[0047] The data acquisition module collects raw data from multiple sources via multi-protocol interfaces and stores it in a MySQL database. The preprocessing module reads the raw data from the database, processes it in a pipeline to generate standardized multi-source feature input data, and caches it in Redis. The prediction module reads the feature data from Redis, the model training submodule updates the initial parameters periodically, and the prediction calculation submodule calls the dynamic penalty factor generator (adjusting the dynamic penalty factor based on the error of the previous time period) and the proximal gradient solver (iteratively solving for the optimal coefficients) to obtain the price difference prediction result and push it to the user terminal. All modules communicate through standardized interfaces (such as RESTful API and shared memory) to form an automated process of "data acquisition-preprocessing-prediction-result push" without manual intervention.
[0048] Example 4 In another embodiment of the present invention, a computer-readable storage medium is provided as a storage component within a terminal device, the function of which is to store programs and data. It should be noted that the computer-readable storage medium here encompasses not only the built-in storage components of the terminal device but also extended storage components supported by the device. Essentially, it is a tangible medium capable of containing or storing programs that can be invoked by or in conjunction with an instruction execution system, device, or apparatus. This storage medium provides storage areas for the terminal's operating system and stores one or more instructions suitable for processor loading and execution, which can constitute one or more computer programs containing program code.
[0049] Specifically, examples of computer-readable storage media (a non-exclusive list) include: electrical connections with one or more wires, portable disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable optical disc read-only memory, optical storage devices, magnetic storage devices, or any reasonable combination of the above types.
[0050] The storage medium may also include data signals propagated as part of a baseband portion or a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any reasonable combination of both. Furthermore, computer-readable storage medium may also refer to other readable media besides conventional readable storage media, capable of sending, propagating, or transmitting programs for use or operation by an instruction execution system, apparatus, or device. Program code on the storage medium can be transmitted via any suitable medium, including but not limited to wireless, wired, optical fiber, or any reasonable combination thereof.
[0051] The program code used to implement the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages such as Java and C++, as well as conventional procedural programming languages such as C. The execution modes of the program code include: running entirely on the user's computing device, running partially on the user's device as a standalone software package, running partially in a distributed manner on both the user's device and a remote computing device, or running entirely on a remote computing device or server. When a remote computing device is involved, the device can be connected to the user's computing device via any type of network such as a local area network (LAN) or a wide area network (WAN), or connected to an external computing device via the Internet through an Internet service provider.
[0052] The processor is capable of loading and executing one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the day-ahead price spread prediction method for the electricity market described in Example 1.
[0053] Example 5 Figure 3 This is a schematic diagram of a computer device provided according to an embodiment of the present invention.
[0054] Please see Figure 3 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the day-ahead price difference prediction method for the electricity market in this embodiment. To avoid repetition, these details are not elaborated here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the computing system that constitutes the day-ahead price difference prediction method for the electricity market in this embodiment. To avoid repetition, these details are not elaborated here.
[0055] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 3This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.
[0056] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, CPUs, graphics processing units (GPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, quantum computing-based data processing logic units, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.
[0057] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or RAM of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the computer device 60.
[0058] Furthermore, memory 62 may include both internal storage units and external storage devices of the computer device 60. Memory 62 is used to store computer programs and other programs and data required by the computer device. Memory 62 can also be used to temporarily store data that has been output or will be output.
[0059] Any references to memory, databases, or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (Read-Only Memory). Memory includes ROM, magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0060] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
Claims
1. A method for predicting the price difference of day-ahead electricity prices, characterized in that, include: Acquire electricity market boundary data, meteorological observation data, and line maintenance plan data for the current target area; The electricity market boundary data, meteorological observation data, and line maintenance plan data are preprocessed to form multi-source feature input data; The multi-source feature input data is input into the LASSO regression model to predict the price difference between the day-to-day and real-time values. The LASSO regression model includes a dynamic penalty factor generator and a proximal gradient solver. The dynamic penalty factor generator is used to dynamically adjust based on multi-source feature input data and the predicted price difference of the previous time period. The proximal gradient solver is used to iteratively solve for the optimal parameters according to the proximal gradient algorithm combined with the Nesterov acceleration strategy and the convergence threshold judgment condition to obtain the price difference prediction result.
2. The method for predicting the price difference of day-ahead electricity market prices according to claim 1, characterized in that, The electricity market boundary data includes at least the load and renewable energy output data from the power grid dispatch center, non-market output data, and tie-line power.
3. The method for predicting the price difference of day-ahead electricity market prices according to claim 2, characterized in that, The preprocessing of the electricity market boundary data, meteorological observation data, and line maintenance plan data includes: The 3σ principle is used to remove outliers from the electricity market boundary data; Missing values in the electricity market boundary data, meteorological observation data, and line maintenance plan data after outlier processing were filled using a missing value imputation method based on time series similarity. The filled-in electricity market boundary data was cleaned and then subjected to feature standardization; meteorological observation data and line maintenance plan data were also standardized.
4. The method for predicting the price difference of day-ahead electricity market prices according to claim 3, characterized in that, The multi-source feature input data includes: introducing holiday dummy variables, lagged price difference features, load-new energy output features, line maintenance features, and meteorological features; The holiday dummy variable is obtained by dividing the holidays throughout the year based on the One-Hot coding method; the lagged price difference feature is obtained by statistical analysis of price difference data for a set historical period; the load-new energy output feature is obtained based on electricity market boundary data; the line maintenance feature is obtained by representing line maintenance plan data based on three-dimensional coding; the meteorological feature includes at least temperature, humidity, equivalent temperature load index, wind speed, and irradiance.
5. The method for predicting the price difference of day-ahead electricity market prices according to claim 1, characterized in that, The LASSO regression model prediction steps include: Initial linear regression is performed on the multi-source feature input data and historical price difference values based on the least squares method to obtain the initial coefficients of each feature in the multi-source feature input data; The adaptive weight vector for each feature is calculated based on the initial coefficients of each feature. The baseline penalty factor is determined based on the Bayesian information criterion of the sliding window in the dynamic penalty factor generator, and the dynamic penalty factor is determined based on the baseline penalty factor. Based on the determined adaptive weight vector and dynamic penalty factor, the objective function is solved iteratively using the proximal gradient algorithm. When the convergence threshold condition is met, the predicted price difference is output.
6. The method for predicting the price difference of day-ahead electricity market prices according to claim 5, characterized in that, The dynamic penalty factor generator is: In the formula, Let be the dynamic penalty factor for time period t. As the benchmark penalty factor, For adjustment coefficients, For the first Forecast price spread for the time period.
7. The method for predicting the price difference of day-ahead electricity market prices according to claim 5, characterized in that, The proximal gradient algorithm also introduces an adaptive weight vector for feature selection during parameter update; the adaptive weight vector is: In the formula, Let be the adaptive weight vector for the j-th feature; The initial coefficients for the j-th feature are obtained through the least squares method. For smoothing terms, =1e -4 .
8. A price difference prediction system for day-ahead electricity market prices, characterized in that, include: The data acquisition module acquires electricity market boundary data, meteorological observation data, and line maintenance plan data for the current target area. The preprocessing module preprocesses the electricity market boundary data, meteorological observation data, and line maintenance plan data to form multi-source feature input data; The prediction module inputs the multi-source feature input data into the LASSO regression model to predict the price difference between the day-to-day and real-time values. The LASSO regression model includes a dynamic penalty factor generator and a proximal gradient solver. The dynamic penalty factor generator is used to dynamically adjust based on multi-source feature input data and the predicted price difference of the previous time period. The proximal gradient solver is used to iteratively solve for the optimal parameters according to the proximal gradient algorithm combined with the Nesterov acceleration strategy and the convergence threshold judgment condition to obtain the price difference prediction result.
9. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform the day-ahead price spread prediction method for the electricity market as described in any one of claims 1 to 7.
10. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including steps for performing the price spread forecasting method for day-ahead electricity market prices of any one of claims 1 to 7.