Medium and long term electricity price prediction method and system for high fluctuation scene, and medium
By combining the causal TCN-HSMM model with statistical models and calibration using historical similar day data, the accuracy and stability issues of electricity price forecasting under high volatility scenarios are solved, and high-precision medium- and long-term electricity price forecasting is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-25
- Publication Date
- 2026-04-07
AI Technical Summary
Existing electricity price forecasting schemes struggle to achieve high-precision forecasts of medium- and long-term electricity prices under highly volatile scenarios, particularly in terms of peak state capture, cross-system migration, interpretability, and uncertainty quantification.
A medium- to long-term electricity price forecasting method based on causal TCN-HSMM is adopted. Baseline forecasting is performed through a neural network model, and statistical models are used to identify and stratify fluctuation states. Historical similar day data is used for calibration and ridge regression model correction to generate high-precision electricity price forecasting results.
It achieves high-precision prediction of medium- and long-term electricity prices in highly volatile scenarios, improves the stability and interpretability of predictions, reduces errors caused by state jitter, and enhances the generalization ability and convergence stability of the model.
Smart Images
Figure CN121810329A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application generally relates to the technical field of power market analysis. More specifically, the present application relates to a method, system, medium and device for medium and long term electricity price prediction in high volatility scenarios. BACKGROUND
[0002] In the electricity spot market, with the increase of renewable energy installed capacity, the enhancement of load side flexibility, and the deepening of the coupling between clearing mechanism and network constraints, the electricity price sequence shows strong non-stationarity, sudden spikes, negative prices, frequent regime switching, and other high volatility characteristics. This characteristic is driven by multiple factors: the uncertainty and rapid transition of wind and solar power output, cross-region power flow and congestion, start-stop and standby costs, demand response and strategic bidding, holiday and extreme weather impact, etc. At the same time, the adjustment of market rules such as time-of-use auction, price cap / limit, block bidding and chain constraints will also change the spread structure and residual distribution, leading to model mismatch when migrating across days / seasons.
[0003] Existing electricity price prediction schemes can be broadly divided into three categories. The first is to use statistical models such as ARIMA, ARIMAX and GARCH and their variants for prediction. This scheme has certain effect in terms of short-term linearity and conditional heteroscedasticity, but it is insufficient in terms of non-linear mutation, long-term dependence and regime switching, and it is sensitive to outliers and distribution skewness; the second is to use a single machine learning model or deep learning model for prediction. This scheme can learn non-linear relationships and long dependencies, but it often faces problems such as sparse spike samples, training and inference exposure bias, exogenous variable leakage and mismatch, and limited explainability on the operation side, and insufficient robustness to cross-date or cross-mechanism drift; the third is to combine seasonal-trend decomposition models, quantile regression models, XGBoost models or TCN models with distribution calibration models for prediction. Although this scheme can to some extent alleviate the bias of a single model, it still easily appears over or under reaction and interval coverage distortion in high volatility scenarios such as spike frequency or regime change, leading to difficulty in adapting to continuously changing market conditions, if it lacks explicit state stratification and time length constraints, controllable retrieval enhancement of similar days in history, and online light correction mechanisms.
[0004] In summary, the existing technology still has deficiencies in spike state capture, cross-regime migration, explainability and uncertainty quantification, and cannot achieve high-precision medium and long-term electricity price prediction in high volatility scenarios. SUMMARY
[0005] To at least solve one or more of the above-mentioned technical problems, the present application proposes in multiple aspects a method, system, medium and device for medium and long term electricity price prediction based on causal TCN-HSMM time length perception.
[0006] In the first aspect, the medium- to long-term electricity price forecasting method based on causal TCN-HSMM duration perception provided in this application includes the following steps: Obtain the first multi-source data required for electricity price forecasting, and preprocess the first multi-source data. The first multi-source data includes at least the electricity price, power system load, renewable energy output, meteorological elements, and holiday indicators at each time point within the historical first time period. The first multi-source data is input into a neural network model to perform baseline prediction of electricity prices under high volatility scenarios, generating a first prediction result. Using a statistical model, the fluctuation state of the first prediction result is identified, and based on the fluctuation state, the first prediction result is stratified to generate a structured second prediction result, wherein the second prediction result includes low-level interval data, mid-level interval data, and high-level interval data.
[0007] In some examples, after generating a structured second prediction result, the method further includes: Obtain electricity prices at various time points on similar days within the historical second time period to generate a second data source; Based on the second data source, the second prediction result is calibrated to generate a third prediction result, wherein each time point of a similar day within the first historical time period is the same as each time point of each day within the second historical time period.
[0008] In some examples, the second prediction result is calibrated based on the second data source to generate a third prediction result, including: Calculate the similarity between the electricity price at each time point of each day in the second data source and the electricity price at each time point of similar days in the second prediction result; Based on the order of similarity from highest to lowest, select K electricity prices corresponding to each preset time point from the second data source; The K electricity prices at each time point are weighted and summed to obtain the aggregated electricity price at each time point. The second prediction result is calibrated based on the aggregated electricity prices at each time point to generate the third prediction result.
[0009] In some examples, after calibrating the second prediction result based on the aggregated electricity prices at various time points to generate a third prediction result, the method further includes: The fusion weight of the second prediction result and the third prediction result is calculated based on three indicators: trend consistency, low-level interval proportion, and volatility stratification. Based on the fusion weights, the second prediction result and the third prediction result are fused to generate a fourth prediction result.
[0010] In some examples, after generating the fourth prediction result, the method further includes: A linear posterior correction is performed on the fourth prediction result using a ridge regression model; Soft limiting is applied to the fourth prediction result after linear posterior correction.
[0011] In some examples, before obtaining the electricity prices at various time points on similar days within a historical second time period and generating a second data source, the method further includes: Based on the strength of the fluctuation state, a minimum dwell time and a maximum dwell time are set for the second prediction result.
[0012] In some examples, the preprocessing of the multi-source data includes: The data in the multi-source data are time-aligned, missing data are filled, outliers are removed and normalized to generate preprocessed multi-source data.
[0013] In some examples, the neural network model is the Causal TCN (Causal Temporal Convolutional Network) model.
[0014] In some examples, the statistical model is a semi-hidden Markov model (HSMM).
[0015] In the second aspect, the medium- and long-term electricity price forecasting system based on causal TCN-HSMM duration perception provided in this application includes: The preprocessing module is configured to acquire the first multi-source data required for electricity price forecasting and preprocess the first multi-source data. The first multi-source data includes at least the electricity price, power system load, renewable energy output, meteorological elements and holiday identifiers at each time point within the historical first time period. The prediction module is configured to input the first multi-source data into a neural network model, perform baseline prediction of electricity prices under high volatility scenarios, and generate a first prediction result; The identification module is configured to use a statistical model to identify the fluctuation state of the first prediction result and, based on the fluctuation state, to stratify the first prediction result and generate a structured second prediction result, wherein the second prediction result includes low-range data, mid-range data and high-range data.
[0016] In a third aspect, this application provides a computer-readable storage medium containing program instructions that, when executed by a processor, cause the method described in the first aspect to be implemented.
[0017] In a fourth aspect, this application provides an electronic device, comprising: Processor; and A memory that stores computer instructions that, when executed by the processor, cause the electronic device to perform the method described in the first aspect above.
[0018] This application uses a neural network model to perform baseline prediction of electricity prices in high-volatility scenarios, generating a first prediction result. Through a statistical model, the fluctuation state of the first prediction result is identified, and the first prediction result is layered according to the fluctuation state to generate a structured second prediction result, thereby achieving high-precision prediction of medium- and long-term electricity prices in high-volatility scenarios. Attached Figure Description
[0019] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein: Figure 1 This paper illustrates an exemplary flowchart of a medium- to long-term electricity price forecasting method based on causal TCN-HSMM duration awareness provided in an embodiment of this application. Figure 2 This illustration shows an exemplary framework diagram of a medium- to long-term electricity price forecasting system based on causal TCN-HSMM duration awareness provided in an embodiment of this application. Figure 3 An exemplary structural block diagram of an electronic device according to some embodiments of this application is shown. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0021] It should be understood that the terms "comprising" and "including" as used in the specification of this application indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.
[0022] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.
[0023] As used in this application description, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."
[0024] The specific embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0025] Example 1 like Figure 1 As shown, the medium- to long-term electricity price forecasting method for high-fluctuation scenarios provided in this application includes the following steps: S101, Obtain the first multi-source data required for electricity price forecasting, and preprocess the first multi-source data, wherein the first multi-source data includes at least the electricity price, power system load, renewable resource output, meteorological elements and holiday identifiers at each time point within the historical first time period.
[0026] It is understandable that meteorological elements can include cloudy, sunny, high temperature, and severe cold. Holidays are indicated by 1 or 0, meaning that when a day is a holiday, the holiday indicator is 1, and otherwise it is 0.
[0027] In some examples, the preprocessing of multi-source data includes: The data in the multi-source data are time-aligned, missing data are filled, outliers are removed and normalized to generate preprocessed multi-source data, which improves the standardization of the multi-source data and lays a good foundation for improving prediction accuracy in the future.
[0028] S102, input the first multi-source data into the neural network model to perform baseline prediction of electricity prices under high volatility scenarios and generate the first prediction result.
[0029] Specifically, the neural network model is a Causal Temporal Convolutional Network (Causal TCN). This model is trained in two stages. The first stage trains the short-field residual convergence structure, enabling the model to quickly learn local trends. The second stage expands the long-field prediction using a warm-start approach, reducing long-field exposure bias. Simultaneously, by incorporating noise perturbation, regularization, causal masking, and dilated convolution, the model exhibits stronger generalization ability and convergence stability on highly fluctuating power data. This training mechanism provides a robust baseline prediction foundation for the overall model.
[0030] S103, using a statistical model, identify the fluctuation state of the first prediction result and, based on the fluctuation state, stratify the first prediction result to generate a structured second prediction result, wherein the second prediction result includes low-range data, mid-range data, and high-range data.
[0031] Understandably, this fluctuation range includes three categories: low, normal, and peak. The first forecast results are arranged from left to right in ascending order of electricity price. The data to the left of the 50th percentile (median) of the first forecast results are the low-range data, the data between the 50th and 95th percentiles are the median range data, and the data to the right of the 95th percentile are the high-range data.
[0032] In some examples, after layering the first prediction result to generate a structured second prediction result, the method further includes: Obtain electricity prices at various time points on similar days within the historical second time period to generate a second data source; Based on the second data source, the second prediction result is calibrated to generate a third prediction result. The time points of similar days in the first historical time period are the same as the time points of each day in the second historical time period, which can effectively avoid holiday disturbances and improve prediction accuracy.
[0033] In some examples, before generating the second data source by obtaining electricity prices at various time points on similar days within the historical second time period, the method further includes: Based on the strength of the fluctuation state, a minimum dwell time and a maximum dwell time are set for the second prediction result to suppress state jitter and regulate the duration of peak states, thereby maintaining the stability of the prediction result and reducing the accumulation of errors caused by state jitter.
[0034] In some examples, the second prediction result is calibrated based on the second data source to generate a third prediction result, including: Calculate the similarity between the electricity price at each time point of each day in the second data source and the electricity price at each time point of similar days in the second prediction result; Based on the order of similarity from highest to lowest, select K electricity prices corresponding to each preset time point from the second data source; The K electricity prices at each time point are weighted and summed to obtain the aggregated electricity price at each time point. The second prediction result is calibrated based on the aggregated electricity prices at each time point to generate the third prediction result.
[0035] The following example illustrates the specific process of calibrating the second prediction result based on the second data source to generate the third prediction result: Step 1: Build a database of historical similar dates Collect and store "historical date data with key information" as the basic operation for building a historical similar date database. Among them, "similar date" refers to a historical date that is similar to the "date to be predicted" (such as tomorrow) in terms of key influencing factors (e.g., last Wednesday: a working day like tomorrow, temperature 28℃, no rain).
[0036] The historical similar date database should contain two types of data for each historical date: (1) Holiday signs (weekdays or weekends, etc.) and meteorological elements (temperature, whether it is raining, etc.); (2) Electricity price at each time point on the historical day (the interval between each time point can be 15 minutes or 1 hour).
[0037] Example: The content of a historical similar day database is a data set that includes "the weather, holiday types, and electricity prices at various times for each day within the past year, month, or week".
[0038] The second step involves using vector similarity calculation methods (Euclidean distance method, cosine similarity algorithm) to calculate the similarity between the electricity price at each time point on the day to be predicted and the electricity price at each time point in the "historical similar day database". Similarity is not only compared with the data of the current day, but also with the similarity of a "sequence within a time window" (e.g., comparing the data of "the 3 days before the day to be predicted" with the data of "the 3 days before a certain historical day").
[0039] For example, to predict electricity prices at various times tomorrow (Wednesday), we need to consider not only "tomorrow's conditions" but also whether "the electricity price trend from Monday to Tuesday this week and weather changes" are consistent with the historical electricity price trend of "the two days before a certain Wednesday (Monday to Tuesday)".
[0040] Step 3: Top-K search Based on the similarity calculated in the second step, select the "K most similar electricity prices at each time point of each day" from the historical similar day database (K is a preset value, for example, K=5, which means finding the 5 electricity prices with the highest similarity to each time point of each day). Avoid using "all historical data" (irrelevant data will interfere), and only select a few samples that are "most relevant to the day to be predicted", such as finding 5 electricity prices that are "also a working day, 28℃, and have the same trend as the previous 3 days" as tomorrow.
[0041] Step 4: Weighted Aggregation Based on the Top-K electricity prices for each time of day selected from the historical similar day database, weights are assigned according to the similarity level, and then weighted and aggregated.
[0042] The higher the similarity of the electricity price, the greater the weight (for example, according to the order of similarity from high to low, the weights of the 5 electricity prices corresponding to each time of day selected from the historical similar day database are set to 0.6, 0.2, 0.1, 0.05 and 0.05 respectively, and the total weight = 1).
[0043] Example: Five electricity prices corresponding to 10:00 AM each day selected from the historical similar day database are 0.50 yuan / kWh, 0.48 yuan / kWh, 0.47 yuan / kWh, 0.46 yuan / kWh, and 0.46 yuan / kWh, with weights of 0.6, 0.2, 0.1, 0.05, and 0.05 respectively. The aggregation result is: 0.50×0.6 + 0.48×0.2 + 0.47×0.1 + 0.46×0.05 + 0.46×0.05 = 0.489 yuan / kWh.
[0044] The final prediction is that the electricity price at 10:00 AM tomorrow will be 0.489 yuan / kWh. This prediction is enhanced by data from similar historical days and is more accurate than the 0.5 yuan / kWh predicted directly by a neural network model. It can achieve high-precision prediction of medium- and long-term electricity prices in highly volatile scenarios.
[0045] In some examples, after calibrating the second prediction result based on the aggregated electricity prices at each time point to generate the third prediction result, the method further includes: The fusion weight of the second prediction result and the third prediction result is calculated based on three indicators: trend consistency, low-level interval proportion, and volatility stratification. Based on the fusion weights, the second prediction result and the third prediction result are fused to generate a fourth prediction result.
[0046] Specifically, the second forecast result and the second forecast result (both are the electricity prices for the 12 hours from 10:00 to 22:00 on the forecast date, denoted as t1-t12). The second forecast result is denoted as Y1, with example data: t1:0.48, t2:0.48, t3:0.50, t4:0.48, t5:0.48, t6:0.47, t7:0.47, t8:0.48, t9:0.49, t10:0.50, t11:0.49, t10.48. The third prediction result is denoted as Y2. Example data: t1: 047, t2: 048, t3: 0.49, t4: 0.49, t5: 0.49, t6: 0.47, t7: 0.47, t8: 0.48, t9: 0.50, t10: 0.49, t11: 0.48, t12: 0.47. The next key step is to calculate the fusion weights using the three indicators, and then fuse the second and third prediction results.
[0047] The first step is to calculate three indicators, which will serve as the basis for determining which prediction result is more reliable. Specific values for these indicators need to be defined and calculated first.
[0048] Indicator 1: Trend Consistency. This indicator determines whether the time-series trends of two forecasts match, specifically the similarity of the hourly electricity price trends of Y1 and Y2 (e.g., whether the trends of "rising in the morning and falling in the afternoon" are consistent). It is typically quantified using the first-difference Pearson correlation coefficient (the closer the coefficient is to 1, the more consistent the trends). The calculation steps (example) are as follows: First, calculate the first-order difference (the change in electricity price between adjacent hours, reflecting the trend): ΔY1 (t2-t1 to t12-t11): 0, 0.02, -0.02, 0, -0.01, 0, 0.01, 0.01, 0.01, -0.01, -0.01. ΔY2 (t2-t to t12-t11): 0.01, 0.01, 0, 0, -0.02, 0, 0.01, 0.02, -0.01, -0.01, -0.01. Through calculations using the mean, covariance, and standard deviation, the Pearson correlation coefficient r is finally obtained as approximately 0.9997 (almost perfectly consistent). Significance: The trends of the second and third prediction results highly overlap, indicating a very high degree of consensus between them on the time-series change patterns.
[0049] Indicator 2: Low-price range percentage, used to determine whether the performance of the second and third forecast results during low electricity price periods is reasonable. Definition: First, define the "low-price range" (i.e., the 50th percentile of historical electricity prices) based on historical data. Then, calculate the percentage of hours in the low-price range for both the second and third forecast results, and quantify this by matching it with the historical actual low-price range percentage. Significance: If the percentages of the second and third forecast results during low electricity price periods are consistent with historical data, their performance is reasonable.
[0050] Indicator 3: Volatility Stratification Indicator, used to determine the reliability of the second and third prediction results across different volatility periods. Definition: First, divide the time series into different strata according to historical volatility (e.g., low volatility: 10:00-11:00; medium volatility: 12:00-18:00; high volatility: 19:00-22:00). Then calculate the matching degree between the volatility value (standard deviation) of each prediction at each stratum and the historical true volatility value (the higher the matching degree, the better the volatility stratification indicator). Calculation steps (example): Historical true standard deviation σ for each stratum: low volatility σ=5, medium volatility σ=10, high volatility σ=15. Calculate the standard deviation σ for each stratum of Y1: low volatility = 7.5, medium volatility σ= 18.27, high volatility σ= 25.5, obtaining the relative error mean E1≈0.676 for Y1 (the larger the error, the worse the matching degree). The standard deviations of Y2 at each level were calculated: σ = 7.5 for low volatility, σ = 18.27 for medium volatility, and σ = 25.5 for high volatility. The mean relative error of Y2 was then calculated to be E2 ≈ 0.676. Since E1 = E2, this indicates that the second and third prediction results are completely consistent across different volatility levels, with no significant difference in performance.
[0051] The second step is to calculate the adaptive fusion weights based on the three indicators. The core of "adaptive" is that the weights are not fixed, but dynamically allocated according to the performance of the three indicators. The specific steps are: (1) Indicator normalization (mapping all indicators to the 0-1 range) Since the three indicators have different dimensions, they need to be normalized first (in the example, the three indicators are already between 0 and 1, so no additional processing is needed). Among them, trend consistency matching degree: S1=0.9997 Low-level interval proportion matching degree: S2=1 (take the average of the two) Fluctuation layer matching degree: S3=0.597 (take the average of the two); (2) Assign weights to each indicator respectively (set according to the business scenario). For example: Trend consistency weight: α=0.5 (trend matching is the core), Low-level interval proportion weight: β=0.2 (the rationality of low electricity price period) Fluctuation layer indicator weight: γ=0.3 (the reliability of different fluctuation layers); (3) Calculate the comprehensive reliability score of the two predictions. If the indicators of the second prediction result and the third prediction result are consistent, then the comprehensive reliability scores of the two are the same; if the indicator of a certain prediction result is better (such as higher fluctuation stratification matching degree), then its score is higher. In this example, since the indicators are completely consistent, the comprehensive reliability score of the first prediction result Y1 is Score1=α×S1+β×S2+γ×S3= 0.5×0.9997+0.2×1+0.3×0.597≈0.4998+0.2+0.1791=0.8789, and the comprehensive reliability score of the second prediction result Y2 is Score2=0.8789; (4) Normalize the comprehensive reliability score to obtain the fusion weight. The fusion weights for the first prediction result Y1 and the second prediction result Y2 are W1 = S1 / (S1+S2) = 0.5. According to the formula P = W1P1 + W2P2, the first and second prediction results Y1 are fused to obtain the fused prediction result P for each time point of the predicted date. Prediction result P combines the model fitting patterns and the experience of historically similar days, and the weights are dynamically adjusted based on three indicators, making it more accurate than a simple averaging. Here, P1 represents the electricity price at each time point of the first prediction result Y1, and P2 represents the electricity price at each time point of the second prediction result Y2.
[0052] In some examples, after generating the fourth prediction result, the method further includes: A linear posterior correction is performed on the fourth prediction result using a ridge regression model; Soft limiting is applied to the fourth prediction result after linear posterior correction.
[0053] Specifically, while the fourth prediction result takes historical experience into account, it may still contain biases (such as consistently overestimating peak-hour electricity prices and underestimating off-peak-hour prices). The role of the ridge regression model is to train the model using historical prediction values and features influencing the bias, allowing the model to grasp the patterns of bias, and then use the trained model to correct the fourth prediction result. The fourth prediction result after linear posterior correction may exhibit extreme values (such as far exceeding the historical electricity price fluctuation range of 1.8 yuan / kWh) due to abnormal feature values (such as large predicted load deviations). However, actual electricity prices are constrained by policies and market supply and demand and will not fluctuate indefinitely. The role of soft limiting is to smooth out extreme values (rather than hard truncate) based on the fluctuation range of historical residual values, ensuring that the predicted values are reasonable. This solves the problem of low prediction accuracy and avoids the risk of outrageous values after correction, making electricity price predictions more supportive of actual power dispatch and user electricity consumption decisions.
[0054] On the other hand, embodiments of this application also provide an electronic device, see [link to relevant documentation]. Figure 3 , Figure 3 This is an exemplary structural block diagram of an electronic device according to an embodiment of this application, such as... Figure 3 As shown, the electronic device includes a processor and a memory, the memory storing computer instructions, and the processor executing the computer instructions to perform the method provided in this application.
[0055] Specifically, processor 601 may include a central processing unit (CPU) or a graphics processing unit (GPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application. Memory 602 may include memory for data or instructions. For example, memory 602 may be at least one of the following: a hard disk drive (HDD), read-only memory (ROM), random access memory (RAM), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, universal serial bus (USB) drive, or other physical / tangible memory storage device. Alternatively, memory 602 may include removable or non-removable (or fixed) media. Furthermore, memory 602 may be internal or external to the integrated gateway disaster recovery device. Memory 602 may be non-volatile solid-state memory. In other words, typically memory 602 includes a tangible (non-transitory) computer-readable storage medium (such as a memory device) encoded with executable instructions, wherein the stored executable instructions, when executed by processor 601 (e.g., by one or more processors), can implement the methods in the embodiments of this application.
[0056] In one example Figure 3The illustrated electronic device may also include a communication interface 603 and a bus 610. The processor 601, memory 602, and communication interface 603 are connected via bus 610 and communicate with each other. Communication interface 603 is primarily used to enable communication between modules, devices, units, and / or equipment within the electronic device. Bus 610, including hardware, software, or both, couples components of the online data flow metering device together. For example, the bus may include at least one of the following: Accelerated Graphics Port (AGP) or other graphics bus, Enhanced Industry Standard Architecture (EISA) bus, Front Side Bus (FSB), HyperTransport (HT) interconnect, Industry Standard Architecture (ISA) bus, Infinite Bandwidth Interconnect, Low Pin Count (LPC) bus, memory bus, Microchannel Architecture (MCA) bus, Peripheral Component Interconnect (PCI) bus, PCI-Express (PCI-X) bus, Serial Advanced Technology Attachment (SATA) bus, Video Electronics Standards Association Local (VLB) bus, or other suitable buses. Bus 610 may include one or more buses. Although specific buses are described or illustrated in the embodiments of this application, any suitable bus or interconnection method may be considered in the embodiments of this application.
[0057] In another aspect, embodiments of this application also provide a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the aforementioned method. The computer-readable storage medium may be, for example, a classic computer-readable storage medium, such as a read-only memory (ROM), random access memory (RAM), disk storage media, optical storage media, flash memory, or other electrical, optical, or other physical / tangible memory storage devices.
[0058] In another aspect, embodiments of this application also provide a computer program product, which includes computer program instructions that, when executed by a processor, implement the method provided in embodiments of this application. This computer program product may be, for example, a software installation package, a plug-in compatible with a related software system, etc.
[0059] The flowcharts and / or block diagrams of the methods and systems of embodiments of this application have been described above by way of example, and related aspects have been described. It should be understood that each block or combination thereof in the flowcharts and / or block diagrams can be implemented by computer program instructions, by dedicated hardware performing a specified function or action, or by a combination of dedicated hardware and computer instructions. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc.; when implemented in software, it is a program or code segment used to perform the required task. The program or code segment can be stored in memory or transmitted over a transmission medium or communication link via data signals carried in a carrier wave. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.
[0060] While this application has shown and described numerous embodiments, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will arise for those skilled in the art without departing from the spirit and intent of this application. It should be understood that various alternatives to the embodiments of this application described herein may be employed in the practice of this application.
Claims
1. A medium- to long-term electricity price forecasting method based on causal TCN-HSMM duration perception, comprising: Obtain the first multi-source data required for electricity price forecasting, and preprocess the first multi-source data. The first multi-source data includes at least the electricity price, power system load, renewable energy output, meteorological elements, and holiday indicators at each time point within the historical first time period. The first multi-source data is input into a neural network model to perform baseline prediction of electricity prices under high volatility scenarios, generating a first prediction result. Using a statistical model, the fluctuation state of the first prediction result is identified, and based on the fluctuation state, the first prediction result is stratified to generate a structured second prediction result, wherein the second prediction result includes low-level interval data, mid-level interval data, and high-level interval data.
2. The medium- and long-term electricity price forecasting method according to claim 1, characterized in that, After generating the structured second prediction result, the method further includes: Obtain electricity prices at various time points on similar days within the historical second time period to generate a second data source; Based on the second data source, the second prediction result is calibrated to generate a third prediction result, wherein each time point of a similar day within the first historical time period is the same as each time point of each day within the second historical time period.
3. The medium- and long-term electricity price forecasting method according to claim 2, characterized in that, Based on the second data source, the second prediction result is calibrated to generate a third prediction result, including: Calculate the similarity between the electricity price at each time point of each day in the second data source and the electricity price at each time point of similar days in the second prediction result; Based on the order of similarity from highest to lowest, select K electricity prices corresponding to each preset time point from the second data source; The K electricity prices at each time point are weighted and summed to obtain the aggregated electricity price at each time point. The second prediction result is calibrated based on the aggregated electricity prices at each time point to generate the third prediction result.
4. The medium- and long-term electricity price forecasting method according to claim 3, characterized in that, After calibrating the second prediction result based on the aggregated electricity prices at each time point to generate the third prediction result, the method further includes: The fusion weight of the second prediction result and the third prediction result is calculated based on three indicators: trend consistency, low-level interval proportion, and volatility stratification. Based on the fusion weights, the second prediction result and the third prediction result are fused to generate a fourth prediction result.
5. The medium- and long-term electricity price forecasting method according to claim 4, characterized in that, After generating the fourth prediction result, the method further includes: A linear posterior correction is performed on the fourth prediction result using a ridge regression model; Soft limiting is applied to the fourth prediction result after linear posterior correction.
6. The medium- and long-term electricity price forecasting method according to claim 2, characterized in that, Before obtaining the electricity prices at various time points on similar days within the historical second time period and generating the second data source, the method further includes: Based on the strength of the fluctuation state, a minimum dwell time and a maximum dwell time are set for the second prediction result.
7. The medium- and long-term electricity price forecasting method according to claim 1, characterized in that, The preprocessing of the multi-source data includes: The data in the multi-source data are time-aligned, missing data are filled, outliers are removed and normalized to generate preprocessed multi-source data.
8. The medium- and long-term electricity price forecasting method according to claim 1, characterized in that, The neural network model is the Causal TCN (Causal Temporal Convolutional Network) model.
9. The medium- and long-term electricity price forecasting method according to claim 1, characterized in that, The statistical model is a semi-hidden Markov model (HSMM).
10. A medium- to long-term electricity price forecasting system based on causal TCN-HSMM duration perception, comprising: The preprocessing module is configured to acquire the first multi-source data required for electricity price forecasting and preprocess the first multi-source data. The first multi-source data includes at least the electricity price, power system load, renewable energy output, meteorological elements and holiday identifiers at each time point within the historical first time period. The prediction module is configured to input the first multi-source data into a neural network model, perform baseline prediction of electricity prices under high volatility scenarios, and generate a first prediction result; The identification module is configured to use a statistical model to identify the fluctuation state of the first prediction result and, based on the fluctuation state, to stratify the first prediction result and generate a structured second prediction result, wherein the second prediction result includes low-range data, mid-range data and high-range data.
11. A computer-readable storage medium, characterized in that, It includes program instructions that, when executed by a processor, cause the method according to any one of claims 1-9 to be implemented.
12. An electronic device, characterized in that, include: processor; as well as A memory storing computer instructions that, when executed by the processor, cause the electronic device to perform the method according to any one of claims 1-9.