Spot market electricity price prediction method and device, storage medium and electronic device

Abstract

The invention discloses a spot market electricity price prediction method and device, a storage medium and an electronic device, and relates to the technical field of power systems, and the method comprises the steps: carrying out the analysis of an offline data set at least comprising historical electricity prices and historical market data through a first target algorithm under the condition that an electricity price prediction process for a target prediction day is initiated, obtaining similar day electricity price information corresponding to M similar days most similar to the target prediction day; performing importance degree evaluation on all the features in the offline data set by using a second target algorithm to obtain N influence features; training the deep neural network model based on the similar daily electricity price information and the N influence features to obtain an offline prediction application for predicting the electricity price; wherein M and N are positive integers; and performing prediction processing through the off-line prediction application and the current market data to obtain an estimated electricity price corresponding to the target prediction day.

Description

Technical Field

[0001] The present application relates to the technical field of power systems, and in particular to a method and device for predicting spot market electricity prices, a storage medium, and an electronic device. Background Art

[0002] With the deepening of power market reform, competition in the power spot market is becoming increasingly fierce. The power spot market is based on the real-time power supply and demand relationship. The price of power commodities is determined through direct market transactions, and price signals are used to guide the optimal allocation of resources. In the power spot market, in order to optimize power generation decisions, the accuracy and reliability of power generation decisions are improved through power price forecasts based on historical data, thereby improving power generation efficiency and meeting the needs of power users at low prices. Therefore, power price forecasts in the power spot market play an extremely important role. However, the volatility of power prices is affected by many factors, including load patterns, line restrictions, regulations, renewable energy generation, weather conditions, and holidays. Due to these complex dynamic characteristics, it is difficult to achieve accurate prediction of power prices.

[0003] In the existing related technologies, there is no effective solution to the problem of low prediction accuracy due to the influence of complex factors on electricity price fluctuations.

[0004] Therefore, it is necessary to improve the existing related technologies to overcome the above-mentioned defects in the related technologies and meet the needs of current actual production. Summary of the invention

[0005] The embodiments of the present application provide a method and device for predicting spot market electricity prices, a storage medium, and an electronic device, so as to at least solve the problem in the existing related technologies that electricity price fluctuations are affected by complex factors and result in low prediction accuracy.

[0006] According to one aspect of an embodiment of the present application, a method for predicting spot market electricity prices is provided, comprising: when initiating an electricity price prediction process for a target prediction day, using a first target algorithm to analyze an offline data set containing at least historical electricity prices and historical market data to obtain similar electric power price information corresponding to M similar days that are most similar to the target prediction day; and using a second target algorithm to evaluate the importance of all features in the offline data set to obtain N influencing features; training a deep neural network model based on the similar electric power price information and the N influencing features to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers; and performing prediction processing through the offline prediction application and current market data to obtain an estimated electricity price corresponding to the target prediction day.

[0007] In an exemplary embodiment, a first target algorithm is used to analyze an offline data set containing at least historical electricity prices and historical market data to obtain similar electricity price information corresponding to M similar days that are most similar to a target forecast day, including: identifying and extracting the offline data set according to Q types of strategies corresponding to the first target algorithm to obtain Q similar electricity price sub-information, wherein the first target algorithm includes at least: a multi-benchmark similar day algorithm; determining similar electricity price information corresponding to M similar days that are most similar to the target forecast day based on the Q similar electricity price sub-information, wherein Q is a positive integer greater than or equal to 3.

[0008] In an exemplary embodiment, the offline data set is identified and extracted according to the Q-type strategy corresponding to the first target algorithm to obtain Q similar electric price sub-information, including: when the time lag strategy in the Q-type strategy is used to extract similar days, the target day corresponding to the target prediction day in the offline data set is located, similar days are searched in the previous P days based on the target day, and the first similar electric price sub-information is generated based on the electric price information corresponding to the search result; when the feature similarity strategy in the Q-type strategy is used to extract similar days, the offline data set is searched, and multiple related dates whose target correlation coefficients with the target day are greater than a preset coefficient are identified, and the second similar electric price sub-information is generated based on the electric price information corresponding to the multiple related dates; when the weighted similarity strategy in the Q-type strategy is used to extract similar days, the correlation coefficient of the electric price corresponding to each date in the offline data set is processed using the objective function to obtain the feature weight, and the target date with the highest total weighted similarity is determined from the offline data set based on the feature weight, and the third similar electric price sub-information is generated based on the electric price information corresponding to the target date.

[0009] In an exemplary embodiment, a second objective algorithm is used to evaluate the importance of all features in an offline data set to obtain N influencing features, including: calculating the feature importance of each feature, wherein the feature importance includes at least: a gain value and a weight value; the gain value is used to reflect the impact of the feature on the prediction result, and the weight value is used to indicate the number of times the feature is called during the prediction process; all features are sorted based on feature importance; the top N features in the sorting result are selected as the most influential important features for electricity price prediction, to obtain N influencing features.

[0010] In an exemplary embodiment, before training the deep neural network model based on similar electric price information and N influencing features, the method further includes: establishing a data conversion process of the deep neural network model, wherein the data conversion process includes: linearly transforming the input data input into the deep neural network model through a hidden layer, and nonlinearly activating the transformation result to perform nonlinear mapping of the input data for forward propagation.

[0011] In an exemplary embodiment, after performing forecast processing through offline forecasting application and current market data to obtain the estimated electricity price corresponding to the target forecast day, the above method also includes: obtaining the actual electricity price when the target forecast day is reached; calculating the difference between the actual electricity price and the estimated electricity price; and determining whether to iteratively update the deep neural network model based on the difference.

[0012] According to another aspect of an embodiment of the present application, a device for predicting spot market electricity prices is also provided, including: an analysis module, which is used to use a first target algorithm to analyze an offline data set containing at least historical electricity prices and historical market data when initiating an electricity price prediction process for a target prediction day, to obtain similar electric electricity price information corresponding to M similar days that are most similar to the target prediction day; and use a second target algorithm to evaluate the importance of all features in the offline data set to obtain N influencing features; a training module, which is used to train a deep neural network model based on similar electric electricity price information and N influencing features to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers; and a processing module, which is used to perform prediction processing through an offline prediction application and current market data to obtain an estimated electricity price corresponding to the target prediction day.

[0013] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned method for predicting spot market electricity prices when running.

[0014] According to another aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the spot market electricity price prediction method through the computer program.

[0015] According to another aspect of the embodiments of the present application, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the above-mentioned method for predicting spot market electricity prices is implemented.

[0016] Through this application, when initiating the electricity price forecasting process for the target forecast day, the first target algorithm is used to analyze the offline data set containing at least historical electricity prices and historical market data to obtain similar electricity price information corresponding to the M similar days most similar to the target forecast day; and the second target algorithm is used to evaluate the importance of all features in the offline data set to obtain N influencing features; based on the similar electricity price information and N influencing features, the deep neural network model is trained to obtain an offline forecasting application for predicting electricity prices; wherein M and N are positive integers; forecasting processing is performed through the offline forecasting application and the current market data to obtain the estimated electricity price corresponding to the target forecast day. The above technical solution solves the problem of low prediction accuracy caused by the influence of complex factors on electricity price fluctuations. Furthermore, the accuracy and reliability of the prediction are improved by extracting similar day information and evaluating the importance of features based on historical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0019] Figure 1 It is a hardware structure block diagram of a computer terminal of a method for predicting spot market electricity price according to an embodiment of the present application;

[0020] Figure 2 is a flow chart of a method for predicting spot market electricity prices according to an embodiment of the present application;

[0021] Figure 3 It is a flow chart of a method for predicting the day-ahead price of the electricity spot market based on a hybrid framework of multiple benchmark similar days and a deep neural network according to an embodiment of the present application;

[0022] Figure 4 It is a structural block diagram of a device for predicting spot market electricity prices according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0024] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] The method embodiments provided in the embodiments of the present application can be executed in a computer terminal, a mobile terminal or a similar computing device. Taking running on a computer terminal as an example, Figure 1 1 is a hardware structure block diagram of a computer terminal for predicting spot market electricity prices according to an embodiment of the present application. Figure 1 As shown, the computer terminal may include one or N ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor (Central Processing Unit, CPU) or a programmable logic device (Field Programmable Gate Array, FPGA)) and a memory 104 for storing data, wherein the above-mentioned computer terminal may also include a transmission device 106 and an input and output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0026] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the adjustment method of the monitoring network point in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or N magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0027] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0028] In this embodiment, a method for predicting spot market electricity prices is provided. Figure 2 is a flow chart of a method for predicting spot market electricity prices according to an embodiment of the present application, such as Figure 2 As shown, the process includes the following steps S202-S206:

[0029] Step S202: when initiating the electricity price forecasting process for the target forecasting day, using the first target algorithm to analyze the offline data set at least including the historical electricity price and the historical market data, to obtain the similar electricity price information corresponding to the M similar days most similar to the target forecasting day; and using the second target algorithm to evaluate the importance of all the features in the offline data set, to obtain N influencing features;

[0030] Optionally, in practical applications, the first target algorithm includes but is not limited to the similar day algorithm, and the second target algorithm includes but is not limited to the XGBoost algorithm. In the above embodiment, if the target prediction date is August 1, 2023, that day is a summer working day, the weather is sunny, and there are no special events. Through the similar day algorithm, the system can find the dates of summer working days, sunny weather and no special events in the past year, such as June 22, July 18, and August 9, 2022. The electricity price information on these dates will be used as the basis for predicting the electricity price on August 1, 2023. In addition, the collected historical data may contain features such as load forecasting, wind power forecasting, photovoltaic forecasting, weather forecasting, and holiday information. Through the feature importance analysis of XGBoost, the system finds that load forecasting, wind power forecasting, photovoltaic forecasting, and specific weather conditions (such as temperature, humidity, etc.) have a significant impact on electricity prices. Therefore, these N features (such as load, wind power, photovoltaic forecasting, and temperature) are used as model inputs, while ignoring other less relevant features such as humidity and holiday information. By selecting the N features with the greatest predictive power, the model’s input dimensions can be significantly reduced, overfitting can be avoided, and the model can focus on the variables that have the greatest impact on electricity prices. This not only simplifies the model, but also improves prediction efficiency and accuracy.

[0031] Step S204: training a deep neural network model based on the similar electricity price information and the N influencing features to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers;

[0032] Optionally, through the similar day algorithm (M=5), the historical electricity price information of 5 days similar to the target forecast day (August 1, 2023) is found, and the XGBoost feature selection algorithm selects N=3 key features that most affect the price: load forecast, wind power forecast and temperature forecast. Based on these data, the deep neural network model is trained through the following steps to preprocess the historical electricity price information of these 5 days and the data of N key features, including normalization and missing value processing. Design a deep neural network model, including an input layer, multiple hidden layers and an output layer. The input layer receives 5 days of similar electricity price information and 3 key features, and the output layer predicts the electricity price of the target day. The model is trained using the training data set, and the weights and biases are adjusted by the back propagation algorithm to minimize the prediction error. During the training process, the adaptive TPE method is used to optimize the model's hyperparameters, such as learning rate, number of hidden layer nodes, etc., to improve the generalization ability of the model. After completing the training, the performance of the model is evaluated using the validation data set. Once the model achieves satisfactory prediction accuracy, its parameters and structure are saved as an offline prediction application for subsequent online prediction. In this way, when power companies or traders need to predict the electricity price on August 1, 2023, they can input the load forecast, wind power forecast, temperature forecast and electricity price data of similar days on that day into the offline trained model to quickly obtain the electricity price forecast results and provide a scientific basis for the bidding strategy. In this way, even when market conditions change or new variables appear, the model can make relatively accurate predictions based on the patterns and feature relationships it has learned.

[0033] Step S206: Perform forecasting processing using the offline forecasting application and current market data to obtain an estimated electricity price corresponding to the target forecasting day.

[0034] In the above steps, when initiating the electricity price forecasting process for the target forecasting day, the first target algorithm is used to analyze the offline data set containing at least historical electricity prices and historical market data to obtain similar electricity price information corresponding to the M similar days most similar to the target forecasting day; and the second target algorithm is used to evaluate the importance of all features in the offline data set to obtain N influencing features; based on the similar electricity price information and the N influencing features, a deep neural network model is trained to obtain an offline forecasting application for forecasting electricity prices; wherein M and N are positive integers; forecasting processing is performed through the offline forecasting application and the current market data to obtain the estimated electricity price corresponding to the target forecasting day. The above technical solution solves the problem of low forecasting accuracy caused by the influence of complex factors on electricity price fluctuations. Furthermore, the accuracy and reliability of the forecast are improved by extracting similar day information and evaluating the importance of features based on historical data.

[0035] In an exemplary embodiment, a first target algorithm is used to analyze an offline data set containing at least historical electricity prices and historical market data to obtain similar electricity price information corresponding to M similar days that are most similar to a target forecast day, including: identifying and extracting the offline data set according to Q types of strategies corresponding to the first target algorithm to obtain Q similar electricity price sub-information, wherein the first target algorithm includes at least: a multi-benchmark similar day algorithm; determining similar electricity price information corresponding to M similar days that are most similar to the target forecast day based on the Q similar electricity price sub-information, wherein Q is a positive integer greater than or equal to 3.

[0036] It can be understood that the offline data set is analyzed according to the Q-type strategy of the first target algorithm. Here, Q can be 3, which represents the adoption of three types of similar day strategies, namely: time lag strategy, feature similarity strategy, and weighted similarity strategy. Under this Q=3-type similar day strategy, the offline data set is identified and extracted respectively to obtain three similar electricity price sub-information. Then, based on these three sub-information sets, the electricity price information of the M historical days that are most representative of the target forecast date, September 15, 2023, is determined through analysis and comparison. For example, assuming M=5, it may be determined that the electricity price data of August 14, August 7, July 31, September 15, 2022, and September 8, 2023 are the final similar electricity price information. By performing multi-angle similar day identification and extraction on the offline data set, the historical electricity price information that best matches the target forecast day conditions can be more accurately selected, thereby providing high-quality input data for the training of the deep neural network model and improving the prediction ability of the model.

[0037] In an exemplary embodiment, the offline data set is identified and extracted according to the Q-type strategy corresponding to the first target algorithm to obtain Q similar electric price sub-information, including: when the time lag strategy in the Q-type strategy is used to extract similar days, the target day corresponding to the target prediction day in the offline data set is located, similar days are searched in the previous P days based on the target day, and the first similar electric price sub-information is generated based on the electric price information corresponding to the search result; when the feature similarity strategy in the Q-type strategy is used to extract similar days, the offline data set is searched, and multiple related dates whose target correlation coefficients with the target day are greater than a preset coefficient are identified, and the second similar electric price sub-information is generated based on the electric price information corresponding to the multiple related dates; when the weighted similarity strategy in the Q-type strategy is used to extract similar days, the correlation coefficient of the electric price corresponding to each date in the offline data set is processed using the objective function to obtain the feature weight, and the target date with the highest total weighted similarity is determined from the offline data set based on the feature weight, and the third similar electric price sub-information is generated based on the electric price information corresponding to the target date.

[0038] Optionally, in the time lag strategy, the electricity price information of March 31, 2024, March 24, 2024, and March 17, 2024 before the target forecast date (April 1, 2024) is located to form the first similar electricity price sub-information set. Under the feature similarity strategy, the dates with high correlation with the target forecast date in the past two years are identified, such as April 1, 2023 (working day, spring, similar weather conditions) and April 4, 2022 (with similar load and weather characteristics), and the second similar electricity price sub-information set is generated based on the electricity price information of these dates. In the weighted similarity strategy, by calculating the feature weights, the date with the highest total weighted similarity with the target forecast date is determined, such as March 27, 2024 (taking into account the influence of multiple factors such as load, weather, market supply and demand), and the third similar electricity price sub-information set is generated based on the electricity price information of this date. By integrating and further analyzing these Q=3 types of similar electricity price sub-information sets, more comprehensive and accurate training data can be provided for the deep neural network model, thereby improving the ability to predict the day-ahead price in the electricity spot market.

[0039] In an exemplary embodiment, a second objective algorithm is used to evaluate the importance of all features in an offline data set to obtain N influencing features, including: calculating the feature importance of each feature, wherein the feature importance includes at least: a gain value and a weight value; the gain value is used to reflect the impact of the feature on the prediction result, and the weight value is used to indicate the number of times the feature is called during the prediction process; all features are sorted based on feature importance; the top N features in the sorting result are selected as the most influential important features for electricity price prediction, to obtain N influencing features.

[0040] Optionally, if the offline dataset includes electricity price data for every day and every hour from 2019 to 2023 and corresponding market, weather, and holiday features. The goal is to predict the electricity price on a certain day in 2024. Using the XGBoost algorithm, the gain and weight values ​​of all features are calculated. For example, the gain value of load forecast is 0.25 and the weight value is 0.15; the gain value of average temperature is 0.12 and the weight value is 0.10; the gain value of holiday identification is 0.05 and the weight value is 0.01. Based on the feature importance sorting, the top N most influential features are determined. Assuming N = 5, the final selected features may be: load forecast, wind power forecast, photovoltaic forecast, average temperature and weekday identification. These features will be used to construct the input of the deep neural network model to predict the electricity price on the target day. Through the above steps, not only can a more efficient and accurate prediction model be built, but also a more reliable and well-founded day-ahead electricity price forecast can be provided to participants in the electricity spot market, thereby optimizing their market strategies and operational decisions.

[0041] In an exemplary embodiment, before training the deep neural network model based on similar electric price information and N influencing features, the method further includes: establishing a data conversion process of the deep neural network model, wherein the data conversion process includes: linearly transforming the input data input into the deep neural network model through a hidden layer, and nonlinearly activating the transformation result to perform nonlinear mapping of the input data for forward propagation.

[0042] Optionally, if the input data of the neural network model contains 10 features, after the linear transformation of the hidden layer, the input of each hidden node may be the weighted sum of the 10 features plus a bias term. If tanh is used as the activation function, this will make the output range of each node between -1 and 1. Through this nonlinear transformation, the model can identify that certain feature combinations are particularly important for predicting electricity prices, while some may be less relevant, thereby automatically selecting features and adjusting weights. Through this series of data transformations and nonlinear mappings, the deep neural network model can effectively handle the complexity of electricity price forecasting in the power spot market, including the interaction of multiple factors and nonlinear dynamic changes, thereby providing power companies or market participants with more accurate and reliable electricity price forecasts and helping them make better decisions.

[0043] In an exemplary embodiment, after performing forecast processing through offline forecasting application and current market data to obtain the estimated electricity price corresponding to the target forecast day, the above method also includes: obtaining the actual electricity price when the target forecast day is reached; calculating the difference between the actual electricity price and the estimated electricity price; and determining whether to iteratively update the deep neural network model based on the difference.

[0044] Optionally, the goal is to predict the day-ahead electricity price of a certain electricity spot market on October 1, 2024. At the end of October 1, the actual electricity price data for every 15 minutes of the day is obtained. The electricity price predicted by the model is compared with the actual electricity price, and the difference is calculated every 15 minutes. For example, if the model predicts that the electricity price at 14:00 on October 1 is 0.05 yuan / kWh, and the actual electricity price is 0.07 yuan / kWh, the difference is 0.02 yuan / kWh. These differences can be used to calculate indicators such as mean error and mean square error to evaluate the overall prediction performance of the model. Then, an error threshold is set, such as the mean absolute error (MAE) is greater than 0.01 yuan / kWh, then the model is considered to have poor prediction performance and needs to be iteratively updated. If the mean absolute error in the prediction on October 1 is 0.015 yuan / kWh, it exceeds the threshold. In this case, it is decided to iteratively update the model. By continuously iterating and updating the neural network model, the updated model can more accurately reflect the latest situation of the electricity spot market, thereby providing more reliable electricity price forecasts and helping power companies or market participants make better decisions, such as optimizing power generation scheduling and bidding strategies. This iterative update mechanism ensures that the deep neural network model continues to maintain high prediction performance in the dynamic environment of the electricity spot market, and is an important practice of deep learning in practical applications.

[0045] Obviously, the embodiments described above are only some embodiments of the present application, not all embodiments. In order to better understand the method, the above process is described below in conjunction with the embodiments, but it is not intended to limit the technical solutions of the embodiments of the present application, specifically:

[0046] The optional embodiment of the present application provides a method for predicting the day-ahead price of the electricity spot market based on a hybrid framework of multiple benchmark similar days and deep neural networks to solve the problem of low prediction accuracy caused by the influence of complex factors on electricity price fluctuations. The method first introduces a similar day algorithm to make full use of historical data, and combines the XGBoost algorithm to screen out important features, optimize the input data dimension, and use the adaptive TPE method to optimize the hyperparameters of the deep neural network model and train and store the model. Finally, the online day-ahead spot market data is applied to predict the day-ahead electricity price for 96 points in the future. The above method can significantly improve the accuracy and stability of the day-ahead price forecast of the electricity spot market, provide more reliable price forecasts for market participants, and help optimize market operations and decision-making.

[0047] It should be noted that the 96 o'clock day-ahead electricity price refers to the electricity market price of the day. This price is determined based on the market supply and demand situation and the auction results of the exchange on the day, and is usually calculated per kilowatt-hour (kWh). 96 points means that this price is determined at the 96th time point of the day (usually one time point per hour). Therefore, the 96 o'clock day-ahead electricity price can help electricity market participants understand the fluctuations in electricity prices at different time periods of the day, which helps to make better decisions on electricity trading.

[0048] As an optional implementation, Figure 3 It is a flowchart of a method for predicting the day-ahead price of the electricity spot market based on a hybrid framework of multiple benchmark similar days and a deep neural network according to an embodiment of the present application, which specifically includes two stages as follows:

[0049] Phase 1: Offline phase, including the following steps:

[0050] Step 3.1.1: Offline data preprocessing, collect historical electricity price data and historical market disclosure data, and perform data cleaning and processing on the original data, including normalized data processing.

[0051] Step 3.1.2: Multi-reference similar day algorithm. The similar day algorithm is used to identify the list of dates most similar to the target day, which are divided into three categories: time-lagged similar days, feature-like similar days, and weighted similar days. For the target day D, the most similar date is searched in the N days before it, and a time-lagged similar day list {D l(1) ,D l(2) ,…,D l(N)}. Feature similarity days are selected by finding dates with high absolute Pearson correlation coefficients with the target day.

[0052]

[0053]

[0054] Where n is the sample size, f a,i and f b,i are the i-th value of a given feature f on target day a and reference day b, respectively.

[0055] For a given feature f, the similarity between the target day D and the reference days D-1 to DN is expressed as {S f,(D,D-1) ,S f,(D,D-2) ,…,S f,(D,D-N)} and sort in descending order to get a list of similar days based on the feature and its corresponding similarity sequence represents the date most similar to the target date D based on feature f.

[0056] Weighted similarity takes into account the influence of all factors when calculating similarity. j The weights W,{w1,w2,…,w J} is calculated by the Pearson correlation coefficient and the softmax function based on the electricity price p on the corresponding reference day:

[0057]

[0058] Total weighted similarity S reference,target It is expressed as:

[0059]

[0060] Get a list of similar days based on total weight in descending order and its corresponding total weighted similarity sequence Indicates the date with the highest total weighted similarity to the target date D.

[0061] First, extract the electricity price of the corresponding similar day in each similar day list Where C can be 1, f or w based on different similar day search methods. Then select the price of the first M similar days Make the framework more applicable to new markets with less historical data.

[0062] Step 3.1.3: Feature selection. The XGBoost method is used to evaluate the importance of each input feature and select the top N features with the most predictive ability to reduce the data dimension. These features include the original features such as load forecast, wind power forecast, and photovoltaic forecast cleaned in S101, as well as the similar electricity price information extracted by the similar day algorithm in S102.

[0063] Step 3.1.4: Model training. The DNN model is used for electricity price prediction, and the hyperparameters (such as learning rate, number of hidden layer nodes, etc.) are optimized through the adaptive TPE method. The deep neural network (DNN) method receives the original input data through the input layer. The input data is linearly transformed through the hidden layer:

[0064] Z (l) =W (l) A (l-1) +b (l) ;

[0065] And the nonlinear mapping of forward propagation data is achieved through nonlinear activation:

[0066] A (l) =g(Z (l) );

[0067] The output layer finally generates the prediction results.

[0068] The DNN model learns the network parameters by minimizing the loss function using the back-propagation algorithm. The back-propagation algorithm calculates the gradient of the loss function with respect to the network parameters and propagates these gradients back from the output layer to the hidden layer and the input layer:

[0069]

[0070] Among them, Z (l) represents the linear transformation result in the hidden layer l, W (l) represents the weight matrix connecting the input layer l-1 and the hidden layer l, A (l-1) represents the activation value of the input layer l-1, b (l) represents the bias term of the hidden layer l, g is the activation function, η is the learning rate, and L is the loss function.

[0071] Step 3.1.5: Model storage: Store the trained model parameters and structure to form an offline model.

[0072] Phase 2: Online phase, including the following steps:

[0073] Step 3.2.1: Online data preprocessing: Collect historical electricity price data and market disclosure data on the operating day, and perform data cleaning and processing on the original data, including normalized data processing.

[0074] Step 3.2.2: Output the prediction results for the next 96 points. Input the trained multi-reference similar day and DNN hybrid framework in step 3.1.5 of the offline phase to get the predicted 96-point day-ahead electricity price.

[0075] It should be noted that the similar day algorithm needs to be precisely adjusted to ensure that the selected date does have similar characteristics to the target forecast day. For example, when considering time-lagged similar days, it is necessary to ensure that the selected date matches the load pattern and weather conditions of the target date. When selecting feature-similar days, the Pearson correlation coefficient needs to be correctly calculated to ensure that the features of the selected date and the target day are highly correlated, which helps to improve the accuracy of the forecast. The weight calculation should reflect the actual impact of each feature on the electricity price, not just the statistical relationship in historical data. The TPE method is used for hyperparameter optimization. The search space and evaluation indicators need to be correctly set to avoid invalid exploration during the optimization process and save computing resources. During the training process, the learning curve of the model should be closely monitored to prevent the model from overfitting the training data or underfitting, resulting in poor prediction performance. Determining when and how to update the model requires careful consideration. Frequent updates may cause model instability, while too long an update cycle may not be able to adapt to market changes in a timely manner. Online iterative model updates may involve additional computing costs. It is necessary to evaluate the impact of the update process on real-time performance and balance the update frequency and resource consumption.

[0076] In summary, the optional embodiment of the present application can capture the complex pattern of electricity price fluctuations, improve the accuracy of prediction, and help electricity market participants make more accurate decisions by combining the multi-benchmark similar day algorithm and deep neural network. The combination of XGBoost feature selection and the deep learning ability of the DNN model makes the model more adaptable and robust to the fluctuations of input data and changes in market conditions. The online prediction and iterative update mechanism enables the model to self-adjust according to the actual prediction error, improve the real-time and adaptability of the model, and ensure the continuous optimization of the prediction effect. By providing more accurate electricity price forecasts, it helps to enhance the transparency of the electricity market, promote fair trading and competition, and reduce the possibility of market manipulation. At the same time, automated feature selection and automatic feature learning of deep neural networks reduce the workload of manual data preprocessing and feature engineering, and improve the efficiency of the prediction process. For new markets with less historical data, by selecting the price information of the first M similar days, it can quickly adapt to the characteristics of the new market and make effective predictions. The use of the above technical solution can significantly improve the accuracy and stability of the day-ahead price forecast of the electricity spot market, provide more reliable price forecasts for market participants, and help optimize market operations and decision-making.

[0077] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0078] In this embodiment, a device for predicting spot market electricity prices is also provided, which is used to implement the above embodiments and preferred implementations, and will not be repeated here. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0079] Figure 4 : is a structural block diagram of a device for predicting spot market electricity prices according to an embodiment of the present application, the device comprising:

[0080] The analysis module 42 is used to analyze the offline data set containing at least historical electricity prices and historical market data using a first target algorithm to obtain similar electricity price information corresponding to M similar days that are most similar to the target forecast day when the electricity price forecast process for the target forecast day is initiated; and to evaluate the importance of all features in the offline data set using a second target algorithm to obtain N influencing features;

[0081] A training module 44 is used to train a deep neural network model based on the similar electricity price information and the N influencing features to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers;

[0082] The processing module 46 is used to perform forecasting processing through the offline forecasting application and current market data to obtain the estimated electricity price corresponding to the target forecast day.

[0083] The above-mentioned device, when initiating the electricity price prediction process for the target prediction day, uses the first target algorithm to analyze the offline data set containing at least historical electricity prices and historical market data, and obtains similar electricity price information corresponding to the M similar days that are most similar to the target prediction day; and uses the second target algorithm to evaluate the importance of all features in the offline data set to obtain N influencing features; trains a deep neural network model based on the similar electricity price information and the N influencing features to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers; performs prediction processing through the offline prediction application and the current market data to obtain the estimated electricity price corresponding to the target prediction day. The above-mentioned technical solution solves the problem of low prediction accuracy caused by the influence of complex factors on electricity price fluctuations. Furthermore, the accuracy and reliability of the prediction are improved by extracting similar day information and evaluating the importance of features based on historical data.

[0084] In an exemplary embodiment, the above-mentioned analysis module is also used to identify and extract the offline data set according to the Q-type strategies corresponding to the first target algorithm, and obtain Q similar daily price sub-information, wherein the first target algorithm at least includes: a multi-benchmark similar day algorithm; based on the Q similar daily price sub-information, determine the similar daily price information corresponding to the M similar days most similar to the target forecast day, wherein Q is a positive integer greater than or equal to 3.

[0085] In an exemplary embodiment, the above-mentioned analysis module is also used to locate the target day corresponding to the target prediction day in the offline data set when the time lag strategy in the Q-type strategy is used to extract similar days, search for similar days in the previous P days based on the target day, and generate the first similar electric price sub-information based on the electricity price information corresponding to the search result; when the feature similarity strategy in the Q-type strategy is used to extract similar days, search the offline data set, identify multiple related dates whose target correlation coefficients with the target day are greater than the preset coefficient, and generate the second similar electric price sub-information based on the electricity price information corresponding to the multiple related dates; when the weighted similarity strategy in the Q-type strategy is used to extract similar days, use the objective function to process the correlation coefficient of the electricity price corresponding to each date in the offline data set to obtain the feature weight, and determine the target date with the highest total weighted similarity from the offline data set based on the feature weight, and generate the third similar electric price sub-information based on the electricity price information corresponding to the target date.

[0086] In an exemplary embodiment, the above-mentioned analysis module is also used to calculate the feature importance of each feature, wherein the feature importance includes at least: a gain value and a weight value; the gain value is used to reflect the impact of the feature on the prediction result, and the weight value is used to indicate the number of times the feature is called during the prediction process; all features are sorted based on the feature importance; the top N features in the sorting result are selected as the most influential important features for electricity price prediction, and N influencing features are obtained.

[0087] In an exemplary embodiment, the above-mentioned device also includes: an establishment module for establishing a data conversion process of the deep neural network model before training the deep neural network model based on similar electric price information and N influencing features, wherein the data conversion process includes: linearly transforming the input data input to the deep neural network model through the hidden layer, and nonlinearly activating the transformation result to perform nonlinear mapping of the input data for forward propagation.

[0088] In an exemplary embodiment, the above-mentioned device also includes: an acquisition module, which is used to perform forecasting processing through an offline forecasting application and current market data, and after obtaining the estimated electricity price corresponding to the target forecast day, obtain the actual electricity price when the target forecast day is reached; calculate the difference between the actual electricity price and the estimated electricity price; and determine whether to iteratively update the deep neural network model based on the difference.

[0089] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0090] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0091] S1. When initiating the electricity price forecasting process for the target forecasting day, use the first target algorithm to analyze the offline data set containing at least historical electricity prices and historical market data to obtain similar electricity price information corresponding to M similar days that are most similar to the target forecasting day; and use the second target algorithm to evaluate the importance of all features in the offline data set to obtain N influencing features;

[0092] S2. Training a deep neural network model based on the similar electricity price information and the N influencing features to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers;

[0093] S3. Perform forecasting processing using the offline forecasting application and current market data to obtain an estimated electricity price corresponding to the target forecasting day.

[0094] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0095] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0096] An embodiment of the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps in any one of the above method embodiments are performed.

[0097] An embodiment of the present application further provides another computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0098] An embodiment of the present application further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0099] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0100] S1. When initiating the electricity price forecasting process for the target forecasting day, use the first target algorithm to analyze the offline data set containing at least historical electricity prices and historical market data to obtain similar electricity price information corresponding to M similar days that are most similar to the target forecasting day; and use the second target algorithm to evaluate the importance of all features in the offline data set to obtain N influencing features;

[0101] S2. Training a deep neural network model based on the similar electricity price information and the N influencing features to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers;

[0102] S3. Perform forecasting processing using the offline forecasting application and current market data to obtain an estimated electricity price corresponding to the target forecasting day.

[0103] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0104] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0105] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of N computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that here, or they can be made into individual integrated circuit modules, or N of the modules or steps can be made into a single integrated circuit module for implementation. In this way, the present application is not limited to any specific combination of hardware and software.

[0106] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting spot market electricity prices, characterized in that: include: When initiating the electricity price forecasting process for the target forecasting day, using the first target algorithm to analyze the offline data set containing at least historical electricity prices and historical market data, to obtain similar electricity price information corresponding to M similar days that are most similar to the target forecasting day; and using a second objective algorithm to evaluate the importance of all features in the offline data set to obtain N influencing features; Based on the similar electricity price information and the N influencing features, a deep neural network model is trained to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers; The offline forecasting application and current market data are used to perform forecasting processing to obtain an estimated electricity price corresponding to the target forecasting day.

2. The method for predicting spot market electricity prices according to claim 1, characterized in that: The first target algorithm is used to analyze the offline data set containing at least historical electricity prices and historical market data, and obtain similar electricity price information corresponding to the M similar days that are most similar to the target forecast day, including: According to the Q-type strategies corresponding to the first target algorithm, the offline data set is identified and extracted respectively to obtain Q similar daily price sub-information, wherein the first target algorithm at least includes: a multi-benchmark similar day algorithm; Based on the Q similar daily price sub-information, similar daily price information corresponding to M similar days that are most similar to the target forecast day is determined, wherein Q is a positive integer greater than or equal to 3.

3. The method for predicting spot market electricity prices according to claim 2, characterized in that: According to the Q-type strategies corresponding to the first target algorithm, the offline data set is identified and extracted respectively to obtain Q similar electric power price sub-information, including: In the case of using the time lag strategy in the Q-type strategy to extract similar days, locating the target day corresponding to the target prediction day in the offline data set, searching for similar days in the previous P days based on the target day, and generating the first similar electric power price sub-information based on the electric power price information corresponding to the search result; When similar days are extracted using the feature similarity strategy in the Q-type strategy, the offline data set is searched to identify multiple related dates whose target correlation coefficients with the target day are greater than a preset coefficient, and second similar electric price sub-information is generated based on the electricity price information corresponding to the multiple related dates; when similar days are extracted using the weighted similarity strategy in the Q-type strategy, the correlation coefficient of the electricity price corresponding to each date in the offline data set is processed using an objective function to obtain a feature weight, and based on the feature weight, a target date with the highest total weighted similarity is determined from the offline data set, and third similar electric price sub-information is generated based on the electricity price information corresponding to the target date.

4. The method for predicting spot market electricity prices according to claim 1, characterized in that: The second objective algorithm is used to evaluate the importance of all features in the offline data set to obtain N influencing features, including: Calculate the feature importance of each feature, wherein the feature importance includes at least: a gain value and a weight value; the gain value is used to reflect the influence of the feature on the prediction result, and the weight value is used to indicate the number of times the feature is called during the prediction process; Sorting all the features based on the feature importance; The first N features in the sorting results are selected as the most important features that have the greatest impact on electricity price prediction, and N influencing features are obtained.

5. The method for predicting spot market electricity prices according to claim 1, characterized in that: Before training the deep neural network model based on the similar electric price information and the N influencing features, the method further includes: Establish a data conversion process for the deep neural network model, wherein the data conversion process includes: linearly transforming the input data input to the deep neural network model through a hidden layer, and nonlinearly activating the transformation result to perform nonlinear mapping of the input data for forward propagation.

6. The method for predicting spot market electricity prices according to claim 1, characterized in that: After obtaining the estimated electricity price corresponding to the target forecast day by performing forecast processing through the offline forecast application and current market data, the method further includes: Obtaining the actual electricity price when the target forecast date is reached; Calculating the difference between the actual electricity price and the estimated electricity price; Determine whether to perform iterative update with the deep neural network model according to the difference.

7. A device for predicting spot market electricity prices, characterized in that: include: An analysis module is used to analyze an offline data set including at least historical electricity prices and historical market data using a first target algorithm when initiating an electricity price prediction process for a target prediction day, and obtain similar electricity price information corresponding to M similar days that are most similar to the target prediction day; and using a second objective algorithm to evaluate the importance of all features in the offline data set to obtain N influencing features; A training module, used for training a deep neural network model based on the similar electricity price information and the N influencing features to obtain an offline prediction application for predicting electricity prices; wherein M and N are positive integers; The processing module is used to perform forecasting processing through the offline forecasting application and current market data to obtain the estimated electricity price corresponding to the target forecast day.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the method according to any one of claims 1 to 6 when executed.

9. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute the method according to any one of claims 1 to 6 through the computer program.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images