Electricity price prediction method and system based on multi-dimensional power data

Abstract

The invention discloses an electricity price prediction method and system based on multi-dimensional electric power data, and the method comprises the steps: obtaining the multi-dimensional electric power data of an electric power market, and the multi-dimensional electric power data comprise historical electricity price data, electric power load data, meteorological data and user behavior characteristics; based on the multi-dimensional power data, constructing an electricity price prediction model with a dynamic learning capability; and predicting the electricity price in the future time period through the electricity price prediction model, and generating an electricity price prediction result. As the model has dynamic learning ability, self-updating can be carried out according to real-time changing data, and the model is ensured to always adapt to fluctuation of markets and demands.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to an electricity price prediction method and system based on multi-dimensional power data. Background Art

[0002] As the core economic signal of the power market, accurate power price forecasting has important guiding significance for power production scheduling, market transaction decisions and optimal allocation of energy resources. The accuracy of power price forecasting directly affects the economic benefits and market competitiveness of power companies, and is also related to the electricity costs and usage strategies of power users.

[0003] At present, the methods of electricity price forecasting mainly include traditional time series forecasting methods and machine learning methods based on artificial intelligence. Traditional time series methods, such as ARIMA model and GARCH model, mainly rely on the statistical characteristics of historical electricity price data for modeling and forecasting. These methods are effective when processing linear and stable data, but their prediction accuracy is limited in the face of complex nonlinear and uncertain factors in the electricity market. Machine learning methods based on artificial intelligence, such as artificial neural networks (ANN), support vector regression (SVR), deep learning, etc., can handle nonlinear relationships, capture complex data patterns, and improve the accuracy of electricity price forecasting. However, these methods are often limited to a single data source, mainly using historical electricity price data for training, and ignoring other factors that potentially affect electricity prices.

[0004] In fact, electricity prices are affected by many factors, including but not limited to power load, renewable energy generation, fuel prices, meteorological conditions, policies and regulations, and market supply and demand. There are complex interactions between these factors, and a single data source cannot fully reflect the volatility of electricity prices. Existing methods lack the comprehensive use of multi-dimensional power data and cannot fully explore the impact of various influencing factors on electricity prices, resulting in insufficient accuracy and reliability of prediction results. In addition, with the development of smart grid and Internet of Things technologies, a large amount of multi-dimensional and multi-type power data has been collected and stored, including power load data, power generation data, geographic information, and socio-economic data. However, these rich data resources have not been fully utilized in electricity price forecasting. Existing methods are difficult to effectively process high-dimensional and heterogeneous power data, and the redundancy and noise of the data also affect the performance of the model.

[0005] Therefore, there is an urgent need for an electricity price prediction method and system based on multi-dimensional power data. Summary of the invention

[0006] The present invention provides an electricity price prediction method and system based on multi-dimensional power data to solve the above problems existing in the prior art.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] A method for predicting electricity prices based on multi-dimensional electricity data, comprising:

[0009] S101: Acquire multi-dimensional power data of the power market, where the multi-dimensional power data includes historical power price data, power load data, meteorological data, and user behavior characteristics;

[0010] S102: Based on multi-dimensional power data, build an electricity price prediction model with dynamic learning ability;

[0011] S103: Predicting the electricity price in the future period through the electricity price prediction model to generate an electricity price prediction result.

[0012] Wherein, step S101 includes:

[0013] S1011: Collect power load data in real time through smart meter collection equipment;

[0014] S1012: extracting historical electricity price data from the electricity market trading platform or the power company database;

[0015] S1013: Acquire meteorological data related to power demand through a meteorological monitoring device or a meteorological data interface, where the meteorological data includes temperature, humidity, and wind speed;

[0016] S1014: Collect user behavior characteristics using user terminal devices, where the user behavior characteristics include power usage time distribution, power usage device type, and power usage habits.

[0017] Among them, the construction of an electricity price prediction model with dynamic learning capabilities includes:

[0018] According to the multi-dimensional power data, a classification model of power price change characteristics and a classification model of power load characteristics are trained, wherein the classification model of power price change characteristics and the classification model of power load characteristics both correspond to power feature sequences;

[0019] Determine the power causal feature set according to the power feature sequence corresponding to the power price change feature classification model and the power feature sequence corresponding to the power load feature classification model;

[0020] According to the power causal feature set and the multidimensional power data set, the initial electricity price prediction model is trained to obtain an electricity price prediction model with dynamic learning ability.

[0021] Among them, the electricity price forecast results are generated, including:

[0022] Input multi-dimensional power data into the power price prediction model;

[0023] If there is an abnormality or missing in a data dimension of the multidimensional power data, the abnormal or missing data is preprocessed, which includes data cleaning, outlier removal and missing value filling;

[0024] Based on the pre-processed multi-dimensional power data, the power price forecasting model is used to forecast the power price in the future period and generate the power price forecasting result;

[0025] Determine whether the electricity price fluctuation range in the electricity price forecast result exceeds the preset fluctuation threshold; if the electricity price fluctuation range exceeds the preset fluctuation threshold, generate electricity price warning information, which includes the time period and range of abnormal electricity price fluctuation;

[0026] After the electricity price warning information is generated, the method further includes: adjusting the power load distribution plan to balance the supply and demand relationship of the power grid and reduce the impact of electricity price fluctuations;

[0027] If the price fluctuation range in the price forecast result does not exceed the preset fluctuation threshold, the price forecast result will be provided to the power user and the grid dispatching department;

[0028] After providing the electricity price forecast results, it also includes: continuously monitoring the deviation between the actual electricity price and the predicted electricity price; if the deviation exceeds the preset error threshold, updating the parameters of the electricity price forecast model and retraining the model to improve the forecast accuracy.

[0029] Among them, according to the power causal feature set and the multidimensional power data set, the initial electricity price prediction model is trained to obtain an electricity price prediction model with dynamic learning ability, including:

[0030] Add the latest meteorological features and user behavior features to the power causal feature set to obtain an updated power causal feature set;

[0031] Generate an electricity price prediction training sample set according to the training set corresponding to the multi-dimensional electricity data and the updated electricity causal feature set, wherein each training sample in the electricity price prediction training sample set includes historical electricity price data and each feature information of the corresponding electricity causal feature set, and the feature information corresponding to the meteorological feature and the user behavior feature is the meteorological data and the user behavior feature information;

[0032] Among them, based on the electricity price prediction training sample set, the following training steps are performed:

[0033] Inputting each training sample of at least one training sample included in the electricity price prediction training sample set into an embedding layer included in the initial electricity price prediction model to obtain a word vector set corresponding to the electricity causal feature set for each training sample of the at least one training sample;

[0034] Generate a predicted electricity price corresponding to each training sample in at least one training sample according to a word vector set corresponding to each training sample in at least one training sample and an initial electricity price prediction model;

[0035] Comparing the predicted electricity price and the actual electricity price corresponding to each training sample in at least one training sample to obtain a comparison result;

[0036] In response to determining that the comparison result satisfies the target condition, the initial electricity price prediction model is determined to be an electricity price prediction model with dynamic learning capability.

[0037] Among them, the electricity price forecast results are generated, including:

[0038] When the electricity price forecast results show abnormal fluctuations, the dynamic abnormality repair operation is initiated, and the forecast results are smoothed by exponentially weighted moving average technology to reduce volatility;

[0039] Optimize electricity price prediction results by retraining the electricity price prediction model or updating input data.

[0040] Among them, an electricity price prediction system based on multi-dimensional power data includes:

[0041] A data acquisition unit is used to acquire multi-dimensional power data of the power market, the multi-dimensional power data includes historical power price data, power load data, meteorological data and user behavior characteristics;

[0042] Constructing an electricity price prediction model unit, which is used to construct an electricity price prediction model with dynamic learning capability based on multi-dimensional power data;

[0043] Obtain electricity price forecast results, which are used to predict electricity prices in future time periods through electricity price forecast models and generate electricity price forecast results.

[0044] The data acquisition unit includes:

[0045] The first data acquisition module is used to collect power load data in real time through a smart meter collection device;

[0046] The second data acquisition module is used to extract historical electricity price data from the power market trading platform or the power company database;

[0047] A third data acquisition module is used to acquire meteorological data related to power demand through a meteorological monitoring device or a meteorological data interface, where the meteorological data includes temperature, humidity and wind speed;

[0048] The fourth data acquisition module is used to collect user behavior characteristics using user terminal devices, and the user behavior characteristics include power consumption time distribution, power consumption device type and power consumption habits.

[0049] Among them, the construction of an electricity price prediction model with dynamic learning capabilities includes:

[0050] According to the multi-dimensional power data, a classification model of power price change characteristics and a classification model of power load characteristics are trained, wherein the classification model of power price change characteristics and the classification model of power load characteristics both correspond to power feature sequences;

[0051] Determine the power causal feature set according to the power feature sequence corresponding to the power price change feature classification model and the power feature sequence corresponding to the power load feature classification model;

[0052] According to the power causal feature set and the multidimensional power data set, the initial electricity price prediction model is trained to obtain an electricity price prediction model with dynamic learning ability.

[0053] Compared with the prior art, the present invention has the following advantages:

[0054] A method for predicting electricity prices based on multi-dimensional electricity data includes: obtaining multi-dimensional electricity data of the electricity market, the multi-dimensional electricity data including historical electricity price data, electricity load data, meteorological data and user behavior characteristics; constructing an electricity price prediction model with dynamic learning ability based on the multi-dimensional electricity data; predicting the electricity price in the future period through the electricity price prediction model to generate electricity price prediction results. Since the model has dynamic learning ability, it can update itself according to real-time changing data, ensuring that the model always adapts to market and demand fluctuations.

[0055] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.

[0056] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0058] Figure 1 This is a flow chart of a method for predicting electricity prices based on multi-dimensional electricity data in an embodiment of the present invention;

[0059] Figure 2 A flowchart of obtaining multi-dimensional power data of the power market in an embodiment of the present invention;

[0060] Figure 3 It is a structural diagram of an electricity price prediction system based on multi-dimensional power data in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0062] The embodiment of the present invention provides a method for predicting electricity prices based on multi-dimensional electricity data, comprising:

[0063] S101: Acquire multi-dimensional power data of the power market, where the multi-dimensional power data includes historical power price data, power load data, meteorological data, and user behavior characteristics;

[0064] S102: Based on multi-dimensional power data, build an electricity price prediction model with dynamic learning ability;

[0065] S103: Predicting the electricity price in the future period through the electricity price prediction model to generate an electricity price prediction result.

[0066] The working principle of the above technical solution is as follows: first, multi-dimensional power data is collected from different data sources, mainly including: Power load data: real-time collection of current power load (i.e., power demand of the system or user) through smart meters. Historical electricity price data: past electricity price data extracted from the power market or power company database. Meteorological data: weather data such as temperature, humidity, wind speed, etc. that affect power demand are obtained through meteorological monitoring equipment or data interfaces. User behavior characteristics: data such as power consumption time distribution, device type, and usage habits collected through user terminal devices. This part of the data reflects the user's demand pattern for electricity and provides richer input information for the prediction model.

[0067] Based on the collected multidimensional data, the "electricity price change feature classification model" and the "electricity load feature classification model" are first trained. The goal of these two models is to identify the change patterns of electricity prices and electricity demand from historical data. After that, these patterns are combined into an "electricity causal feature set", which contains feature sequences of electricity price changes and electricity load changes. Finally, these features are input into the "initial electricity price prediction model" together with the multidimensional electricity data set for training to achieve the dynamic learning ability of the model.

[0068] Once the prediction model is trained, it will input multidimensional data in real time and output the electricity price forecast results for a specific time period in the future. If anomalies or data missing are found in the multidimensional power data (such as missing meteorological data or user data), data preprocessing will be performed, including cleaning abnormal data and filling missing values. After generating the electricity price forecast results, the system will determine whether the fluctuation range of the electricity price exceeds the set threshold. If the fluctuation exceeds the threshold, an early warning message will be generated, and the power load distribution will be adjusted to maintain the supply and demand balance of the power grid. Otherwise, the forecast results will be provided to users and the power grid dispatching department.

[0069] The system continuously monitors the deviation between the predicted electricity price and the actual electricity price. Once the deviation exceeds the set threshold, the parameters of the prediction model will be updated and the model will be retrained to further improve the prediction accuracy. In addition, by adding the latest meteorological features and user behavior features to the "electricity causal feature set", the model can maintain dynamic learning under the input of new data.

[0070] The beneficial effects of the above technical solution are as follows: the method establishes a more comprehensive prediction model by integrating historical electricity prices, power loads, meteorological and user behavior data, which helps to improve prediction accuracy; because the model has dynamic learning capabilities, it can self-update according to real-time changing data to ensure that the model always adapts to market and demand fluctuations; using multi-dimensional information such as meteorological data and user behavior characteristics, the prediction model not only considers the impact of historical electricity prices and loads, but also includes the potential impact of external factors on electricity prices; the system has the ability to handle exceptions, issues warnings when data is abnormal or fluctuates too much, and automatically adjusts power load distribution, which helps maintain the stability and reliability of the power grid; the prediction results can provide more accurate references for power users and power grid dispatching departments, making it easier for decision makers to optimize resource allocation, control costs and improve the balance of power supply and demand.

[0071] In another embodiment, step S101 includes:

[0072] S1011: Collect power load data in real time through smart meter collection equipment;

[0073] S1012: extracting historical electricity price data from the electricity market trading platform or the power company database;

[0074] S1013: Acquire meteorological data related to power demand through a meteorological monitoring device or a meteorological data interface, where the meteorological data includes temperature, humidity, and wind speed;

[0075] S1014: Collect user behavior characteristics using user terminal devices, where the user behavior characteristics include power usage time distribution, power usage device type, and power usage habits.

[0076] The working principle of the above technical solution is: collect multi-dimensional power data from different data sources, mainly including: Power load data: collect the current power load (i.e. the power demand of the system or user) in real time through smart meters. Historical electricity price data: past electricity price data extracted from the power market or power company database. Meteorological data: obtain weather data such as temperature, humidity, wind speed, etc. that affect power demand through meteorological monitoring equipment or data interfaces. User behavior characteristics: data such as power consumption time distribution, device type, and usage habits collected through user terminal devices. This part of the data reflects the user's demand pattern for electricity and provides richer input information for the prediction model.

[0077] The beneficial effect of the above technical solution is that the method builds a more comprehensive prediction model by integrating historical electricity prices, power load, meteorological and user behavior data, which helps to improve prediction accuracy.

[0078] In another embodiment, a power price prediction model with dynamic learning capability is constructed, including:

[0079] According to the multi-dimensional power data, a classification model of power price change characteristics and a classification model of power load characteristics are trained, wherein the classification model of power price change characteristics and the classification model of power load characteristics both correspond to power feature sequences;

[0080] Determine the power causal feature set according to the power feature sequence corresponding to the power price change feature classification model and the power feature sequence corresponding to the power load feature classification model;

[0081] According to the power causal feature set and the multidimensional power data set, the initial electricity price prediction model is trained to obtain an electricity price prediction model with dynamic learning ability.

[0082] Among them, the power causal feature set is determined, including:

[0083] Obtain the power feature sequence corresponding to the power price change feature classification model and the power load feature classification model, including:

[0084] The power characteristic series includes: electricity price change rate, power load change rate, power demand elasticity and power supply fluctuation;

[0085] The horizontal axis of the power feature sequence is set to be the timestamp, the timestamps are evenly spaced, and the vertical axis of the power feature sequence is the value of each power feature at the corresponding timestamp;

[0086] The electricity price change rate, electricity load change rate, electricity demand elasticity and electricity supply fluctuation are respectively formed into an electricity characteristic sequence according to the time series;

[0087] Based on the power feature sequence, determining the correlation between the electricity price change rate and the power load change rate, including: calculating the change rate of each power feature sequence at each time stamp to obtain a change rate set of the electricity price change rate sequence and the power load change rate sequence;

[0088] Calculate the correlation coefficient between the electricity price change rate sequence and the power load change rate sequence, and determine whether the correlation coefficient is greater than or equal to a set second threshold;

[0089] If the correlation coefficient is greater than or equal to the second threshold, a significant relationship between the electricity price change rate and the power load change rate is screened out;

[0090] The power causal feature set is constructed based on significant relationships, including:

[0091] Setting a second weight value of the electricity price change rate and the power load change rate, wherein the second weight value is determined according to the degree of influence on the power system;

[0092] Performing a weighted summation on the change rate set of the electricity price change rate and the power load change rate and the corresponding second weight value to form a causal relationship formula;

[0093] If the weighted summation result is greater than or equal to the set summation threshold, the result is included in the power causal feature set;

[0094] Use power causal feature sets to analyze and predict power systems, including:

[0095] Construct a power causal feature set database to record each causal feature and its corresponding analysis model;

[0096] Input the power feature sequence into the analysis model in the power causal feature set database and output the power system state prediction result;

[0097] The prediction results of different analysis models and their corresponding weights are combined to calculate the final power system prediction results.

[0098] The working principle of the above technical solution is to analyze the characteristic sequence of electricity prices changing over time. The specific method is to divide the electricity price data into several time windows and generate feature vectors through feature extraction (such as fluctuation amplitude, frequency, average growth rate, etc.). Based on these features, the classification model is trained to classify different electricity price change patterns (such as "increase", "decrease", "fluctuation", etc.). Similar feature extraction and classification are performed on power load data. Power load characteristics can include peak values, valley values, change cycles, etc., and different load patterns (such as "peak period" and "low period" are identified through classification models. According to the characteristic sequence generated by the electricity price changes and the power load classification model, the power causal feature set is determined. This feature set uses causal analysis techniques (such as Granger causal analysis) to identify the causal relationship between electricity prices and power loads, such as whether electricity prices have an upward trend during peak load periods. This step is critical, as it helps the model identify which features are more critical to electricity price prediction.

[0099] The initial electricity price prediction model is trained using the constructed electricity causal feature set and the multi-dimensional electricity data set. The model uses machine learning algorithms (such as LSTM or decision trees) to predict future electricity prices based on features in historical data. In order to enable the prediction model to have dynamic learning capabilities, the model is updated regularly to adapt to the latest data and electricity price fluctuation trends. This can be achieved through a "sliding window" mechanism, that is, the model is regularly retrained on the latest data, allowing it to continue learning and improve prediction accuracy.

[0100] Construct two classification models: Electricity price change characteristic classification model: This model extracts the characteristics of the electricity price change rate based on the electricity characteristic sequence, and predicts the changing trend of electricity prices by analyzing historical electricity price data. The electricity price change rate usually refers to the magnitude of change in electricity prices within a certain period of time, which may be affected by various factors such as market supply and demand, seasonal fluctuations, and policy adjustments. Electricity load characteristic classification model: This model analyzes the fluctuation characteristics of electricity load based on the characteristics of electricity load changes. The electricity load change rate reflects the speed of change of load demand, which is usually closely related to factors such as economic activities, weather changes, and social electricity demand. These two models need to identify and extract the changing characteristics of electricity prices and electricity loads through training on historical data, and classify or predict them.

[0101] The power characteristic sequence is a data sequence that describes key characteristics such as the rate of change of electricity prices, the rate of change of electricity load, the elasticity of electricity demand, and fluctuations in electricity supply. Specifically, it includes: The rate of change of electricity prices: refers to the magnitude of change in electricity prices within a time period, reflecting the volatility of electricity prices. The rate of change of electricity load: refers to the speed of change of electricity load within a certain period of time, reflecting the fluctuation of electricity demand. Electricity demand elasticity: measures the sensitivity of electricity demand to price changes, and is usually used to assess the impact of price changes on electricity demand. Electricity supply fluctuations: reflects the stability of electricity supply, including possible fluctuations in electricity production and transmission. The ordinate of these characteristic sequences is the value of each timestamp, and the abscissa is the timestamp. The time intervals are usually equally spaced (for example, every hour, every day, etc.).

[0102] Based on the above power characteristic sequence, the correlation analysis between the electricity price change rate and the power load change rate is carried out. The specific steps are as follows:

[0103] Calculate the rate of change: First, calculate the rate of change of each power characteristic sequence at each time point. This can be achieved by solving the difference between adjacent time points, that is:

[0104] ΔX t =X t -X t-1

[0105] Among them, X t is the eigenvalue at time point t.

[0106] Correlation calculation: Calculate the correlation coefficient between the electricity price change rate series and the power load change rate series, using the Pearson correlation coefficient or other appropriate correlation measurement methods. The strength of the relationship between the two is determined by the correlation coefficient, and the correlation coefficient value range is [-1,1], where 1 indicates a complete positive correlation, -1 indicates a complete negative correlation, and 0 indicates no correlation. Screening for significant relationships: If the correlation coefficient is greater than or equal to the set second threshold (for example, 0.7), it is considered that there is a significant relationship between the electricity price change rate and the power load change rate.

[0107] The power causal feature set consists of significant relationships that have been analyzed and screened. In this process, it is necessary to:

[0108] Second weight value setting: Each significant relationship between the electricity price change rate and the power load change rate will be assigned a "second weight value", which is determined based on the degree of impact of the relationship on the electricity price or load in the power system. This weight value can be set through historical data analysis or expert experience. Weighted summation: The set of change rates of the electricity price change rate and the power load change rate is weighted and summed with the corresponding second weight value to obtain a comprehensive causal relationship value. If the weighted summation result is greater than or equal to the set summation threshold, it is considered that the impact of the causal relationship on the power system is significant and worthy of being included in the power causal feature set.

[0109] Use the power causal feature set to analyze and predict the power system. Once the power causal feature set is established, this feature set can be used to analyze and predict the power system: Power causal feature set database: Each causal feature and its corresponding analysis model are recorded in the database to form a causal analysis framework for the power system. Power system state prediction: Input the power feature sequence into the analysis model in the database to obtain the prediction results of the power system state. For example, it is possible to predict fluctuations in electricity prices, changes in load demand, system stability, etc. Comprehensive prediction results: In order to improve the prediction accuracy, the prediction results of different analysis models can be combined, and the final power system prediction results can be weighted according to the importance of each model (for example, through weight values).

[0110] The beneficial effects of the above technical solution are: by accurately capturing the different characteristic sequences of electricity prices and power loads through the classification model, the prediction model can more clearly identify the changing patterns and driving factors of electricity prices, thereby improving the accuracy of predictions. The introduction of dynamic learning capabilities enables the model to have the function of self-updating, and can quickly adjust the prediction capabilities when the electricity market environment or policies change, ensuring the timeliness and accuracy of the prediction results. Predicting more accurate changes in electricity prices will help power supply companies and electricity users to formulate energy procurement and use plans more reasonably and reduce operating costs. At the same time, predicting electricity price fluctuations during peak and valley periods can also effectively alleviate the pressure of electricity supply and demand and improve energy efficiency. Through the construction of causal feature sets, the model better utilizes the correlation of multidimensional data, extracts the features that have the greatest impact on electricity price fluctuations, reduces redundant information, and improves the efficiency of model training and prediction effects.

[0111] In another embodiment, generating an electricity price prediction result includes:

[0112] Input multi-dimensional power data into the power price prediction model;

[0113] If there is an abnormality or missing in a data dimension of the multidimensional power data, the abnormal or missing data is preprocessed, which includes data cleaning, outlier removal and missing value filling;

[0114] Based on the pre-processed multi-dimensional power data, the power price forecasting model is used to forecast the power price in the future period and generate the power price forecasting result;

[0115] Determine whether the electricity price fluctuation range in the electricity price forecast result exceeds the preset fluctuation threshold; if the electricity price fluctuation range exceeds the preset fluctuation threshold, generate electricity price warning information, which includes the time period and range of abnormal electricity price fluctuation;

[0116] After generating electricity price warning information, it also includes: adjusting the power load distribution plan, balancing the supply and demand relationship of the power grid, and reducing the impact of electricity price fluctuations;

[0117] If the price fluctuation range in the price forecast result does not exceed the preset fluctuation threshold, the price forecast result will be provided to the power user and the grid dispatching department;

[0118] After providing the electricity price forecast results, it also includes: continuously monitoring the deviation between the actual electricity price and the predicted electricity price; if the deviation exceeds the preset error threshold, updating the parameters of the electricity price forecast model and retraining the model to improve the forecast accuracy.

[0119] The working principle of the above technical solution is: input multi-dimensional power data into the electricity price prediction model. Multi-dimensional data contains multiple factors that affect electricity prices, such as power load, weather data, historical electricity prices, etc. If there is an abnormality or missing in one of the dimensions (such as missing data caused by sensor failure), the system will trigger the data preprocessing process.

[0120] Data preprocessing, data cleaning: remove invalid or obviously erroneous data points, such as extreme outliers in electricity price records. Outlier removal: use statistical methods (such as the three sigma method) to identify and remove abnormal data points. For example, if the power load in a certain period of time is much higher than the historical average, it can be considered an outlier. Missing value filling: use interpolation, average filling or machine learning models to fill in missing values. For example, if the electricity price record of a certain hour is missing, it can be filled based on the average or trend of the previous and next hours.

[0121] The pre-processed multi-dimensional data is used to predict electricity prices. The model (such as LSTM neural network) predicts the changes in electricity prices in a certain period of time in the future based on the trained pattern and input data, and generates electricity price prediction results. For example, predict the fluctuation of electricity prices in the next 24 hours. The system determines whether the fluctuation range of electricity prices exceeds the preset threshold in the electricity price prediction results. For example, if the preset fluctuation threshold is 5%, the fluctuation of electricity prices exceeding 5% in a short period of time will trigger an early warning. If the fluctuation range exceeds the threshold, the system generates electricity price early warning information, indicating the time period and fluctuation range of the fluctuation, and reminds relevant departments to prepare countermeasures. For example, if the forecast shows that the electricity price will rise by 10% in a certain period of time in the future, the system will issue an early warning to indicate the time period and range of the increase. According to the early warning information, the power management system can automatically or manually adjust the power load distribution plan. For example, during the period of rising electricity prices, the system can reduce the use of high-power consumption equipment, increase the supply of clean energy, or supplement electricity from energy storage equipment to balance the supply and demand relationship of the power grid and reduce the impact of fluctuations on electricity prices. If the fluctuation range of electricity prices does not exceed the threshold, the prediction results are directly passed to the power users and the power grid dispatching department for their decision-making. The system continuously monitors the deviation between the actual electricity price and the predicted electricity price. If the deviation exceeds the preset error threshold (such as the actual electricity price deviation exceeds 5%), the system will trigger a model update and retrain the model to adapt to the latest data and improve the prediction accuracy.

[0122] The above technical scheme has the following beneficial effects: through data preprocessing and anomaly detection, the integrity and accuracy of the data are ensured, thereby improving the reliability of the electricity price prediction model. Through deviation monitoring and model updating, the prediction model can be adjusted in time to adapt to new data changes and reduce prediction errors. When there are large fluctuations in the electricity price forecast, the early warning mechanism can help users and management departments to respond in advance and reduce the impact of sudden fluctuations on operating costs and grid stability. In periods when electricity prices may rise sharply, early warnings can be used to remind users to use electricity at off-peak times, which helps to reduce the pressure on the grid. After the early warning, the system automatically or recommends adjusting the power load distribution, which helps to balance the supply and demand relationship of the grid, optimize the allocation of power resources, and reduce the pressure on electricity prices during peak hours. Through reasonable load management and cost control, the overall power operation cost is reduced and the energy efficiency of the grid is improved. When the electricity price forecast results deviate from the actual situation, the parameters of the model are dynamically adjusted and retrained to enable the model to adapt to the new data, thereby maintaining the accuracy and stability of the forecast in the ever-changing electricity market environment. The electricity price prediction model and its early warning mechanism provide intelligent decision-making support for power grids and electricity users, reduce ineffective electricity use, improve energy utilization, and meet the goals of energy conservation and emission reduction.

[0123] In another embodiment, the initial electricity price prediction model is trained according to the electricity causal feature set and the multi-dimensional electricity data set to obtain an electricity price prediction model with dynamic learning capability, including:

[0124] Add the latest meteorological features and user behavior features to the power causal feature set to obtain an updated power causal feature set;

[0125] According to the training set corresponding to the multi-dimensional power data and the updated power causal feature set, an electricity price prediction training sample set is generated, wherein each training sample in the electricity price prediction training sample set includes historical electricity price data and each feature information of the corresponding power causal feature set, and the feature information corresponding to the meteorological characteristics and user behavior characteristics is the meteorological data and the user behavior feature information.

[0126] The working principle of the above technical solution is as follows: In the initial stage, the electricity price prediction model is trained based on historical electricity price data and the electricity causal feature set. The electricity causal feature set includes a variety of factors that affect electricity prices, such as weather conditions (temperature, humidity, etc.), user electricity consumption behavior (power consumption patterns, load changes, etc.), and grid operation status. Over time, the external environment of the power system changes, especially meteorological conditions and user behavior. Therefore, the system regularly adds the latest meteorological features (such as temperature forecasts for the next few days) and user behavior features (such as changes in peak electricity consumption hours) to the electricity causal feature set to obtain an updated causal feature set.

[0127] By combining the updated electricity causal feature set with the multidimensional electricity data set, a new electricity price prediction training sample set is generated. Each training sample includes the following:

[0128] Historical electricity price data: reflects the past trend of electricity price changes. The individual feature information of the electricity causal feature set: covers historical and current meteorological data (such as temperature, precipitation, wind speed, etc.) and user behavior characteristics (such as daily electricity consumption patterns, peak and valley load distribution, etc.), which directly affect the balance of electricity supply and demand. For example, in summer, rising temperatures lead to increased use of air conditioners, so the meteorological characteristics and user behavior characteristics are more closely related at this time. The model generates targeted training samples through a combination of these features.

[0129] After a new training sample set is generated, the electricity price prediction model is retrained. By continuously introducing the latest meteorological features and user behavior features, the model can dynamically adjust its parameters and learn new relationships between electricity demand and electricity price changes. For example, if the weather forecast shows that the temperature will rise significantly in the next few days, the system will add this information to the new training set. After learning these features, the model can better predict future fluctuations in electricity prices.

[0130] Since the electricity price forecasting model has dynamic learning capabilities, as the external environment (such as weather and user behavior) changes, the model can adjust according to the latest data. Even if meteorological conditions change suddenly or user behavior patterns change significantly, the model can still adapt quickly and provide accurate electricity price forecasts.

[0131] The above technical solution has the following beneficial effects: the dynamically updated power causal feature set can reflect the real-time changing meteorological conditions and user behaviors. By adding these latest feature information to the training sample set, the electricity price prediction model can better predict future electricity price changes, especially in the face of sudden meteorological changes, such as extreme high temperatures or heavy rain. The electricity price prediction model has the ability to dynamically adjust and self-update by continuously introducing the latest external feature data. This dynamic learning ability can keep the model always highly accurate and avoid prediction failure due to changes in the external environment. Improving the accuracy of electricity price prediction not only helps the power grid balance the supply and demand of electricity, but also helps users adjust their electricity consumption plans according to the prediction information, avoid excessive electricity consumption during peak hours, and reduce electricity costs. The power grid dispatching department can adjust the power generation and load distribution strategies according to accurate electricity price predictions, reduce unnecessary electricity waste, and improve the economic benefits of system operation. The dynamically learned electricity price prediction model can help the power grid realize intelligent dispatching and management. For example, when the sudden change in weather causes a significant increase in electricity consumption, the system can issue early warnings through accurate electricity price predictions, adjust the power generation plan, and ensure the safe and stable operation of the power grid. When the electricity price prediction model takes into account meteorological characteristics, the power grid can optimize the dispatching of renewable energy according to meteorological conditions. For example, when the weather forecast shows that there will be plenty of sunshine in the coming week, the system can predict the increase in photovoltaic power generation in advance and adjust the power generation plan accordingly, reducing the use of traditional energy and improving the utilization rate of new energy.

[0132] In another embodiment, based on the electricity price prediction training sample set, the following training steps are performed:

[0133] Inputting each training sample of at least one training sample included in the electricity price prediction training sample set into an embedding layer included in the initial electricity price prediction model to obtain a word vector set corresponding to the electricity causal feature set for each training sample of the at least one training sample;

[0134] Generate a predicted electricity price corresponding to each training sample in at least one training sample according to a word vector set corresponding to each training sample in at least one training sample and an initial electricity price prediction model;

[0135] Comparing the predicted electricity price and the actual electricity price corresponding to each training sample in at least one training sample to obtain a comparison result;

[0136] In response to determining that the comparison result satisfies the target condition, the initial electricity price prediction model is determined to be an electricity price prediction model with dynamic learning capability.

[0137] The working principle of the above technical solution is as follows: each training sample consists of historical electricity price data and a set of electricity causal features. The feature information in the electricity causal feature set (such as meteorological features and user behavior features) is input into the embedding layer of the initial electricity price prediction model. The role of the embedding layer is to convert the input high-dimensional feature information into a low-dimensional word vector. This conversion helps the model to process and understand complex feature relationships more effectively. For example, meteorological features such as temperature and humidity are converted into word vectors so that the model can identify the potential relationship between these features and electricity prices.

[0138] For each training sample, the word vector set output by the embedding layer is input into the initial electricity price prediction model. The model uses these word vector sets to generate the corresponding predicted electricity prices. For example, assuming that the meteorological characteristics of a training sample show a cooling trend in the next few days, the model recognizes through the word vector that this trend may lead to a decrease in electricity prices and generates a lower predicted electricity price accordingly.

[0139] The predicted electricity price generated by the model is compared with the actual electricity price and the error is calculated. This step is used to evaluate the prediction accuracy of the model. For example, if the actual electricity price of a training sample is 0.15 yuan / kWh and the model predicts 0.14 yuan / kWh, the error is 0.01 yuan / kWh.

[0140] If the comparison result (such as the error rate) meets the preset target condition (for example, the error is less than a certain threshold), the initial electricity price forecasting model is determined to have dynamic learning ability. The setting of the target condition can be adjusted according to the needs of the specific application scenario to ensure that the model has sufficient prediction accuracy and adaptability in practical applications.

[0141] The beneficial effects of the above technical solution are: by converting complex feature information into word vectors through the embedding layer, the model can better understand and process the input data and improve the accuracy of electricity price prediction. Through dynamic adjustment and verification, the model can continuously learn and adapt to new data features and maintain efficient prediction performance. This dynamic learning ability enables the model to maintain stable prediction results under different environments and conditions. Accurate electricity price prediction helps power companies optimize resource allocation and scheduling, reduce operating costs and improve economic benefits. At the same time, users can also adjust their electricity consumption behavior based on the predicted information to reduce electricity costs. The electricity price prediction model with dynamic learning ability is an important part of the smart grid. It can help the power grid achieve more intelligent management and scheduling, and improve the reliability and efficiency of the power grid. Accurate electricity price prediction can help the power system better integrate renewable energy, optimize its scheduling and use, improve the utilization rate of renewable energy, and support sustainable development goals.

[0142] In another embodiment, generating an electricity price prediction result includes:

[0143] When the electricity price forecast results show abnormal fluctuations, the dynamic abnormality repair operation is initiated, and the forecast results are smoothed by exponentially weighted moving average technology to reduce volatility;

[0144] If the abnormal fluctuations are not eliminated, the electricity price prediction results can be optimized by retraining the electricity price prediction model or updating the input data.

[0145] The characteristic indicators of power data include: load change trend indicator, power consumption change trend indicator, and voltage fluctuation trend indicator.

[0146] The working principle of the above technical solution is: during the electricity price forecasting process, the system will continuously monitor the changes in the forecast results. If there are abnormal fluctuations in the forecast results (that is, the forecast value changes dramatically in a short period of time), the system will identify and mark these anomalies. For example, if the electricity price forecast suddenly jumps from 0.15 yuan / kWh to 0.25 yuan / kWh and then drops to 0.14 yuan / kWh in one day, this drastic change may be regarded as an abnormal fluctuation.

[0147] Once an abnormal fluctuation is detected, the system initiates a dynamic abnormality repair operation. Exponentially weighted moving average (EWMA) technology is used to smooth the forecast results. EWMA calculates the average by assigning different weights to historical data, and newer data points are assigned higher weights to reflect the latest trends more quickly. For example, assuming that the predicted electricity prices for the past three days are 0.15, 0.25, and 0.14 yuan / kWh, respectively, EWMA may calculate a smoother electricity price trend, such as 0.18 yuan / kWh, to reduce volatility.

[0148] To further optimize the electricity price forecast results, the system may retrain the electricity price forecast model or update the input data. Retraining the model involves updating the model parameters using the latest historical data and feature information to improve its forecasting ability. Updating the input data may include introducing new features or correcting errors in existing data to ensure the accuracy and completeness of the model input. For example, if it is found that errors in meteorological data lead to inaccurate forecasts, the system may update this data or introduce a more accurate meteorological model.

[0149] The beneficial effects of the above technical solution are as follows: through EWMA smoothing, the system can effectively reduce the volatility of the forecast results and provide a more stable electricity price forecast. This helps users and power companies to better plan and manage power resources. Dynamic anomaly repair operations and model retraining enable the electricity price forecast model to adapt to data changes and abnormal situations, improving the robustness and reliability of the model. Stable and accurate electricity price forecasts provide more reliable decision support for power companies and users, helping them make more informed decisions in electricity procurement, scheduling and consumption. Reducing forecast volatility and anomalies helps reduce operational risks in the power market and avoid economic losses caused by inaccurate forecasts. By continuously monitoring and optimizing the forecast model, the system can continuously improve its performance and support the long-term development and optimization of the power market.

[0150] In another embodiment, a system for predicting electricity prices based on multi-dimensional electricity data includes:

[0151] A data acquisition unit is used to acquire multi-dimensional power data of the power market, the multi-dimensional power data includes historical power price data, power load data, meteorological data and user behavior characteristics;

[0152] Constructing an electricity price prediction model unit, which is used to construct an electricity price prediction model with dynamic learning capability based on multi-dimensional power data;

[0153] Obtain electricity price forecast results, which are used to predict electricity prices in future time periods through electricity price forecast models and generate electricity price forecast results.

[0154] The working principle of the above technical solution is: connect to the power market database, smart meter, meteorological monitoring system and user terminal equipment through the interface to integrate multi-source data. Historical electricity price data: extract past electricity price information from the power market database. Power load data: obtain real-time and historical power load data from smart meters. Meteorological data: obtain relevant meteorological information such as temperature, humidity, and wind speed through meteorological monitoring equipment or data interfaces. User behavior characteristics: collect electricity usage time, device type and habit data from user terminal devices.

[0155] The collected data is stored in a data warehouse and preprocessed (such as cleaning and normalization) for subsequent analysis. Example: Collect data on electricity price changes, electricity load curves, daily meteorological conditions, and user electricity usage habits in a certain area over the past month.

[0156] Build the electricity price forecasting model unit, data preparation: Use the acquired data for feature engineering to extract features that are useful for electricity price forecasting. Use historical data to train the model so that it can identify patterns and trends in the data. Through online learning or incremental learning methods, the model can be updated and adjusted based on new data to maintain the accuracy of the forecast. Use the LSTM model to train the electricity price data of the past year, and combine the load, meteorological and user behavior characteristics to improve the model's ability to capture electricity price trends. Obtain electricity price forecast results. Operation steps:

[0157] Input data: Input the latest multi-dimensional electricity data into the trained electricity price prediction model.

[0158] Generate forecasts: The model calculates the electricity price for future time periods and generates forecast results.

[0159] Result output: The prediction results are output to the user interface or power management system for decision-making.

[0160] The beneficial effects of the above technical solution are as follows: by integrating multi-dimensional data, the model can more comprehensively understand the factors affecting electricity prices and improve the accuracy and reliability of predictions. Accurate electricity price predictions help power companies optimize procurement and sales strategies and improve market competitiveness and profitability. The prediction results provide data support for power dispatch, demand response and pricing strategies, helping decision makers make more informed choices. Electricity price predictions can help users optimize electricity usage time, reduce electricity costs and improve user satisfaction. By optimizing the use of power resources and reducing unnecessary waste, the system helps promote energy conservation, emission reduction and sustainable development goals.

[0161] In another embodiment, obtaining the data unit includes:

[0162] The first data acquisition module is used to collect power load data in real time through a smart meter collection device;

[0163] The second data acquisition module is used to extract historical electricity price data from the power market trading platform or the power company database;

[0164] A third data acquisition module is used to acquire meteorological data related to power demand through a meteorological monitoring device or a meteorological data interface, where the meteorological data includes temperature, humidity and wind speed;

[0165] The fourth data acquisition module is used to collect user behavior characteristics using user terminal devices, and the user behavior characteristics include power consumption time distribution, power consumption device type and power consumption habits.

[0166] The working principle of the above technical solution is as follows: First, the data acquisition module: install the smart meter collection device at the user's power entrance. Real-time monitoring and recording of the user's power load data, that is, the user's power consumption in different time periods. The data is transmitted to the data center for storage and processing through wireless communication technology (such as Wi-Fi, Zigbee). For example, the smart meter records the power consumption of a household every hour in a day. For example, the power consumption from 8 to 9 in the morning is 2kWh, and from 6 to 7 in the evening is 5kWh.

[0167] The second data acquisition module is connected to the power market trading platform or the power company database. The electricity price data for the past period of time is extracted, including daily and hourly electricity price information. The data is stored in the data warehouse for subsequent analysis and model training. For example, the electricity price data for the past month is extracted from the power market database, for example, the electricity price on a certain day changes from 0.12 yuan / kWh to 0.15 yuan / kWh.

[0168] The third data acquisition module: deploy meteorological monitoring equipment or obtain real-time meteorological data through the meteorological data interface. Collect meteorological parameters related to power demand, such as temperature, humidity, and wind speed. Correlate and analyze meteorological data with power load data. For example, the meteorological monitoring equipment records that the temperature in a certain area is 30°C, the humidity is 60%, and the wind speed is 5m / s.

[0169] Fourth, data acquisition module: Integrate data collection functions in user terminal devices (such as smart home systems). Record the user's electricity usage time distribution, power equipment type (such as air conditioners, washing machines) and electricity usage habits (such as peak power usage hours). Upload user behavior data to the data center for analysis. For example, users use air conditioners and TVs from 7 to 9 p.m., which forms the behavior characteristics of peak power usage hours.

[0170] The beneficial effects of the above technical solution are as follows: by integrating multi-source data (load, historical electricity prices, weather, user behavior), the electricity price and electricity demand forecasting model can more accurately reflect the actual situation and improve the accuracy of the forecast; real-time monitoring and analysis of electricity load and user behavior characteristics can help power companies optimize the allocation and scheduling of power resources and reduce waste; by analyzing user behavior characteristics, power companies can provide users with personalized electricity recommendations and services to improve user satisfaction; the multi-module data acquisition mechanism enables the system to flexibly respond to different demand changes and environmental changes, improving the adaptability and flexibility of the system; by optimizing electricity usage patterns and improving electricity efficiency, the system helps to achieve the goal of energy conservation and emission reduction and promote sustainable development.

[0171] In another embodiment, a power price prediction model with dynamic learning capability is constructed, including:

[0172] According to the multi-dimensional power data, a classification model of power price change characteristics and a classification model of power load characteristics are trained, wherein the classification model of power price change characteristics and the classification model of power load characteristics both correspond to power feature sequences;

[0173] Determine the power causal feature set according to the power feature sequence corresponding to the power price change feature classification model and the power feature sequence corresponding to the power load feature classification model;

[0174] According to the power causal feature set and the multidimensional power data set, the initial electricity price prediction model is trained to obtain an electricity price prediction model with dynamic learning ability.

[0175] The working principle of the above technical solution is as follows: First, collect multi-dimensional power data, including key factors such as electricity price and power load, and the data must cover fluctuations under different time and environment. In the data preprocessing step, remove outliers, fill in missing data, and standardize the data to ensure that the data is unified and suitable for model training.

[0176] Analyze the characteristic sequence of electricity prices changing over time. The specific method is to divide the electricity price data into several time windows and generate feature vectors through feature extraction (such as fluctuation amplitude, frequency, average growth rate, etc.). Based on these features, the classification model is trained to classify different electricity price change patterns (such as "increase", "decrease", "fluctuation", etc.). Similar feature extraction and classification are performed on power load data. Power load characteristics can include peak values, valley values, change cycles, etc., and different load patterns (such as "peak period" and "low period" are identified through classification models. According to the feature sequence generated by the electricity price changes and the power load classification model, the power causal feature set is determined. This feature set uses causal analysis techniques (such as Granger causal analysis) to identify the causal relationship between electricity prices and power loads, such as whether there is an upward trend in electricity prices during peak load periods. This step is critical, as it helps the model identify which features are more critical to electricity price prediction.

[0177] The initial electricity price prediction model is trained using the constructed electricity causal feature set and the multi-dimensional electricity data set. The model uses machine learning algorithms (such as LSTM or decision trees) to predict future electricity prices based on features in historical data. In order to enable the prediction model to have dynamic learning capabilities, the model is updated regularly to adapt to the latest data and electricity price fluctuation trends. This can be achieved through a "sliding window" mechanism, that is, the model is regularly retrained on the latest data, allowing it to continue learning and improve prediction accuracy.

[0178] The beneficial effects of the above technical solution are: by accurately capturing the different characteristic sequences of electricity prices and power loads through the classification model, the prediction model can more clearly identify the changing patterns and driving factors of electricity prices, thereby improving the accuracy of predictions. The introduction of dynamic learning capabilities enables the model to have the function of self-updating, and can quickly adjust the prediction capabilities when the electricity market environment or policies change, ensuring the timeliness and accuracy of the prediction results. Predicting more accurate changes in electricity prices will help power supply companies and electricity users to formulate energy procurement and use plans more reasonably and reduce operating costs. At the same time, predicting electricity price fluctuations during peak and valley periods can also effectively alleviate the pressure of electricity supply and demand and improve energy efficiency. Through the construction of causal feature sets, the model better utilizes the correlation of multidimensional data, extracts the features that have the greatest impact on electricity price fluctuations, reduces redundant information, and improves the efficiency of model training and prediction effects.

[0179] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting electricity prices based on multi-dimensional electricity data, characterized in that: include: S101: Acquire multi-dimensional power data of the power market, where the multi-dimensional power data includes historical power price data, power load data, meteorological data, and user behavior characteristics; S102: Based on multi-dimensional power data, build an electricity price prediction model with dynamic learning capabilities; S103: Predicting the electricity price in the future period through the electricity price prediction model to generate an electricity price prediction result.

2. The method for predicting electricity prices based on multi-dimensional electricity data according to claim 1, characterized in that: Step S101 includes: S1011: Collect power load data in real time through smart meter collection equipment; S1012: extracting historical electricity price data from the electricity market trading platform or the power company database; S1013: Acquire meteorological data related to power demand through a meteorological monitoring device or a meteorological data interface, where the meteorological data includes temperature, humidity, and wind speed; S1014: Collect user behavior characteristics using user terminal devices, where the user behavior characteristics include power usage time distribution, power usage device type, and power usage habits.

3. The method for predicting electricity prices based on multi-dimensional electricity data according to claim 1, characterized in that: Build an electricity price prediction model with dynamic learning capabilities, including: According to the multi-dimensional power data, a classification model of power price change characteristics and a classification model of power load characteristics are trained, wherein the classification model of power price change characteristics and the classification model of power load characteristics both correspond to power feature sequences; Determine the power causal feature set according to the power feature sequence corresponding to the power price change feature classification model and the power feature sequence corresponding to the power load feature classification model; According to the power causal feature set and the multidimensional power data set, the initial electricity price prediction model is trained to obtain an electricity price prediction model with dynamic learning ability.

4. The method for predicting electricity prices based on multi-dimensional electricity data according to claim 1, characterized in that: Generate electricity price forecast results, including: Input multi-dimensional power data into the power price prediction model; If there is an abnormality or missing in a data dimension of the multidimensional power data, the abnormal or missing data is preprocessed, which includes data cleaning, outlier removal and missing value filling; Based on the pre-processed multi-dimensional power data, the power price forecasting model is used to forecast the power price in the future period and generate the power price forecasting result; Determine whether the electricity price fluctuation range in the electricity price forecast result exceeds the preset fluctuation threshold; if the electricity price fluctuation range exceeds the preset fluctuation threshold, generate electricity price warning information, which includes the time period and range of abnormal electricity price fluctuation; After generating electricity price warning information, it also includes: adjusting the power load distribution plan, balancing the supply and demand relationship of the power grid, and reducing the impact of electricity price fluctuations; If the price fluctuation range in the price forecast result does not exceed the preset fluctuation threshold, the price forecast result will be provided to the power user and the grid dispatching department; After providing the electricity price forecast results, it also includes: continuously monitoring the deviation between the actual electricity price and the predicted electricity price; if the deviation exceeds the preset error threshold, updating the parameters of the electricity price forecast model and retraining the model to improve the forecast accuracy.

5. The method for predicting electricity prices based on multi-dimensional electricity data according to claim 3, characterized in that: According to the power causal feature set and the multi-dimensional power data set, the initial power price prediction model is trained to obtain a power price prediction model with dynamic learning ability, including: Add the latest meteorological features and user behavior features to the power causal feature set to obtain an updated power causal feature set; According to the training set corresponding to the multi-dimensional power data and the updated power causal feature set, an electricity price prediction training sample set is generated, wherein each training sample in the electricity price prediction training sample set includes historical electricity price data and each feature information of the corresponding power causal feature set, and the feature information corresponding to the meteorological characteristics and user behavior characteristics is the meteorological data and the user behavior feature information.

6. The method for predicting electricity prices based on multi-dimensional electricity data according to claim 5, characterized in that: Based on the electricity price prediction training sample set, the following training steps are performed: Inputting each training sample of at least one training sample included in the electricity price prediction training sample set into an embedding layer included in the initial electricity price prediction model to obtain a word vector set corresponding to the electricity causal feature set for each training sample of the at least one training sample; Generate a predicted electricity price corresponding to each training sample in at least one training sample according to a word vector set corresponding to each training sample in at least one training sample and an initial electricity price prediction model; Comparing the predicted electricity price and the actual electricity price corresponding to each training sample in at least one training sample to obtain a comparison result; In response to determining that the comparison result satisfies the target condition, the initial electricity price prediction model is determined to be an electricity price prediction model with dynamic learning capability.

7. The method for predicting electricity prices based on multi-dimensional electricity data according to claim 4, characterized in that: Generate electricity price forecast results, including: When the electricity price forecast results show abnormal fluctuations, the dynamic abnormality repair operation is initiated, and the forecast results are smoothed by exponentially weighted moving average technology to reduce volatility; Optimize electricity price prediction results by retraining the electricity price prediction model or updating input data.

8. An electricity price prediction system based on multi-dimensional power data, characterized in that: include: A data acquisition unit is used to acquire multi-dimensional power data of the power market, the multi-dimensional power data includes historical power price data, power load data, meteorological data and user behavior characteristics; Constructing an electricity price prediction model unit, which is used to construct an electricity price prediction model with dynamic learning capability based on multi-dimensional power data; Obtain electricity price forecast results, which are used to predict electricity prices in future time periods through electricity price forecast models and generate electricity price forecast results.

9. The electricity price prediction system based on multi-dimensional power data according to claim 8, characterized in that: The data acquisition unit includes: The first data acquisition module is used to collect power load data in real time through a smart meter collection device; The second data acquisition module is used to extract historical electricity price data from the power market trading platform or the power company database; A third data acquisition module is used to acquire meteorological data related to power demand through a meteorological monitoring device or a meteorological data interface, where the meteorological data includes temperature, humidity and wind speed; The fourth data acquisition module is used to collect user behavior characteristics using user terminal devices, and the user behavior characteristics include power consumption time distribution, power consumption device type and power consumption habits.

10. The electricity price prediction system based on multi-dimensional power data according to claim 8, characterized in that: Build an electricity price prediction model with dynamic learning capabilities, including: According to the multi-dimensional power data, a classification model of power price change characteristics and a classification model of power load characteristics are trained, wherein the classification model of power price change characteristics and the classification model of power load characteristics both correspond to power feature sequences; Determine the power causal feature set according to the power feature sequence corresponding to the power price change feature classification model and the power feature sequence corresponding to the power load feature classification model; According to the power causal feature set and the multidimensional power data set, the initial electricity price prediction model is trained to obtain an electricity price prediction model with dynamic learning ability.

Images