Settlement parameter prediction method and device, storage medium and electronic equipment

Abstract

The invention discloses a settlement parameter prediction method and device, a storage medium and electronic equipment. The method comprises the following steps: preprocessing energy attribute data of a to-be-predicted target settlement period to obtain an energy attribute data sequence meeting an input condition of a settlement parameter prediction network, and performing linear regression compression processing on the energy attribute data sequence in a linear regression processing sub-network, obtaining a selected periodic energy attribute data sequence matched with the target settlement period; in the volatility prediction sub-network, prediction is carried out based on the periodic energy attribute data sequence, and settlement parameter fluctuation characteristics matched with the target settlement period are obtained; and determining a target settlement parameter corresponding to the target settlement period according to the settlement parameter fluctuation characteristic. According to the invention, the technical problem of low prediction result accuracy caused by neglecting the fluctuation characteristics of the data is solved.

Description

Technical Field

[0001] The present invention relates to the field of power market transactions, and in particular to a settlement parameter prediction method and device, a storage medium and an electronic device. Background Art

[0002] As a key link between electricity producers and consumers, the electricity market plays an extremely important role in power operation, power consumption, grid management and dispatching. Therefore, the reasonable prediction of electricity prices in the power market becomes increasingly important.

[0003] At present, in the power market, electricity prices are often affected by many factors, thus showing the characteristics of violent fluctuations. For example, seasonal climate change, demand fluctuations and power supply structure will have a significant impact on electricity prices, making it difficult to accurately capture and predict the volatility of electricity prices. Especially in short-term forecasts, the uncertainty of electricity price fluctuations increases the difficulty of forecasting.

[0004] However, most of the settlement parameter prediction models such as electricity prices provided in the related art only focus on the average value of electricity price fluctuations, while ignoring the impact of volatility on electricity prices, resulting in a large deviation between the prediction results of settlement parameters such as electricity prices and the actual values. In other words, the settlement parameter prediction methods provided in the prior art have the problem of low accuracy of prediction results.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The embodiments of the present invention provide a settlement parameter prediction method and device, a storage medium and an electronic device to at least solve the technical problem of low accuracy of prediction results caused by ignoring the volatility characteristics of the data itself.

[0007] According to one aspect of an embodiment of the present invention, a method for predicting settlement parameters is provided, comprising: preprocessing energy attribute data of a target settlement period to be predicted to obtain an energy attribute data sequence that meets the input conditions of a settlement parameter prediction network, wherein the settlement parameter prediction network includes a linear regression processing subnetwork for feature selection, and a volatility prediction subnetwork for predicting the degree of fluctuation of sequence data; in the linear regression processing subnetwork, performing linear regression compression processing on the energy attribute data sequence to obtain a selected periodic energy attribute data sequence that matches the target settlement period; in the volatility prediction subnetwork, performing prediction based on the periodic energy attribute data sequence to obtain settlement parameter fluctuation characteristics that match the target settlement period; and determining the target settlement parameter corresponding to the target settlement period according to the settlement parameter fluctuation characteristics.

[0008] As an optional implementation, the above-mentioned preprocessing of the energy attribute data of the target settlement period to be predicted to obtain the energy attribute data sequence that meets the input conditions of the settlement parameter prediction network includes: determining the data distribution characteristics of the above-mentioned energy attribute data; transforming and scaling the above-mentioned energy attribute data according to a transformation and scaling mode matching the above-mentioned data distribution characteristics to obtain the above-mentioned energy attribute data sequence.

[0009] As an optional implementation, the above-mentioned energy attribute data is transformed and scaled according to a transformation and scaling mode matching the above-mentioned data distribution characteristics to obtain the above-mentioned energy attribute data sequence, including: when the above-mentioned data distribution characteristics indicate that the above-mentioned energy attribute data is heavily-tailed distributed, the above-mentioned energy attribute data is statistically transformed to obtain a transformed first reference data sequence, wherein heteroscedasticity has been eliminated in the above-mentioned first reference data sequence; extreme value scaling is performed on the above-mentioned first reference data sequence to obtain the above-mentioned energy attribute data sequence; when the above-mentioned data distribution characteristics indicate that the above-mentioned energy attribute data is normally distributed, extreme value scaling is directly performed on the above-mentioned energy attribute data to obtain the above-mentioned energy attribute data sequence.

[0010] As an optional implementation, the energy attribute data of the target settlement period to be predicted is preprocessed to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, including: when the above-mentioned energy attribute data is a data vector carrying directional information, the above-mentioned energy attribute data is cosine transformed to obtain a transformed second reference data sequence; and the above-mentioned second reference data sequence is extreme value scaled to obtain the above-mentioned energy attribute data sequence.

[0011] As an optional implementation, in the above-mentioned linear regression processing subnetwork, linear regression compression processing is performed on the above-mentioned energy attribute data sequence to obtain a selected periodic energy attribute data sequence that matches the above-mentioned target settlement period, including: in the above-mentioned linear regression processing subnetwork, linear regression compression processing is performed on the above-mentioned energy attribute data sequence using a preset regularization parameter, wherein the above-mentioned regularization parameter is used to indicate the degree of regression compression.

[0012] As an optional implementation, in the above-mentioned volatility prediction subnetwork, prediction is performed based on the above-mentioned periodic energy attribute data sequence to obtain the settlement parameter fluctuation characteristics matching the above-mentioned target settlement period, including: in the above-mentioned volatility prediction subnetwork, the above-mentioned periodic energy attribute data sequence is fitted to obtain a residual sequence; and the conditional variance is calculated using the above-mentioned residual sequence to obtain the above-mentioned settlement parameter fluctuation characteristics.

[0013] As an optional implementation, before the above-mentioned preprocessing of the energy attribute data of the target settlement period to be predicted, it also includes: obtaining historical energy settlement data, wherein the above-mentioned historical energy settlement data includes the attribute data of the settlement resources consumed by using the target energy in the historical period; preprocessing the above-mentioned historical energy settlement data to obtain a historical energy attribute data sequence that meets the input conditions of the above-mentioned settlement parameter prediction network; inputting the above-mentioned historical energy attribute data sequence into at least two candidate settlement parameter prediction networks, wherein different of the above-mentioned candidate settlement parameter prediction networks are networks constructed using different network frameworks; obtaining the prediction evaluation parameters corresponding to each of the above-mentioned candidate settlement parameter networks; and determining the above-mentioned settlement parameter prediction network from the above-mentioned at least two candidate settlement parameter prediction networks based on the comparison results of the above-mentioned prediction evaluation parameters.

[0014] As an optional implementation, the above-mentioned obtaining of the prediction evaluation parameters corresponding to each of the above-mentioned candidate settlement parameter networks includes one of the following: obtaining the mean absolute error matching the above-mentioned candidate settlement parameter network; obtaining the root mean square error matching the above-mentioned candidate settlement parameter network; obtaining the mean absolute percentage error matching the above-mentioned candidate settlement parameter network.

[0015] As an optional implementation, the above-mentioned inputting of the above-mentioned historical energy attribute data sequence into at least two candidate settlement parameter prediction networks includes: determining at least two candidate linear regression processing subnetworks and at least two candidate volatility prediction subnetworks, wherein the above-mentioned candidate linear regression processing subnetworks include: minimum absolute shrinkage and selection operator network, mutual information network; the above-mentioned candidate volatility prediction subnetworks include: generalized autoregressive conditional heteroskedastic volatility network, autoregressive mean and stochastic volatility network, deep learning network; the above-mentioned candidate linear regression processing subnetwork and the above-mentioned candidate volatility prediction subnetwork are combined to obtain the above-mentioned candidate settlement parameter prediction network.

[0016] According to another aspect of an embodiment of the present invention, a settlement parameter prediction device is also provided, comprising: a processing unit, used to pre-process energy attribute data of a target settlement period to be predicted, and obtain an energy attribute data sequence that meets the input conditions of a settlement parameter prediction network, wherein the settlement parameter prediction network includes a linear regression processing subnetwork for feature selection, and a volatility prediction subnetwork for predicting the degree of fluctuation of sequence data; a compression unit, used to perform linear regression compression processing on the energy attribute data sequence in the linear regression processing subnetwork, and obtain a selected periodic energy attribute data sequence that matches the target settlement period; a prediction unit, used to perform prediction based on the periodic energy attribute data sequence in the volatility prediction subnetwork, and obtain a settlement parameter fluctuation feature that matches the target settlement period; a first determination unit, used to determine the target settlement parameter corresponding to the target settlement period according to the settlement parameter fluctuation feature.

[0017] According to another aspect of the embodiments of the present application, a storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned settlement parameter prediction method when running.

[0018] According to another aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the above settlement parameter prediction method.

[0019] According to another aspect of an embodiment of the present application, there is also provided an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the settlement parameter prediction method through the computer program.

[0020] In an embodiment of the present invention, after preprocessing the energy attribute data of the target settlement period to be predicted to obtain an energy attribute data sequence, the corresponding settlement parameter fluctuation characteristics are determined by the linear regression processing subnetwork and the volatility prediction subnetwork in the settlement parameter prediction network, and the target settlement parameters of the target settlement period are further predicted based on the settlement parameter fluctuation characteristics. This combination of optimized feature selection and fluctuation prediction utilizes the prediction ability of deep learning, and also combines statistical models and feature selection methods to form an efficient and accurate prediction framework, which can not only show significant advantages in energy settlement parameter prediction and improve the accuracy of prediction results, but also provide market participants with more reliable decision-making basis, so as to better cope with the complexity and volatility of energy market data. This solves the technical problem of low accuracy of prediction results caused by ignoring the volatility characteristics of the data itself. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0022] Figure 1 is a flow chart of an optional settlement parameter prediction method according to an embodiment of the present application;

[0023] Figure 2 is a flowchart of another optional settlement parameter prediction method according to an embodiment of the present application;

[0024] Figure 3 is a flowchart of another optional settlement parameter prediction method according to an embodiment of the present application;

[0025] Figure 4 is a schematic diagram of an optional settlement parameter prediction device according to an embodiment of the present application;

[0026] Figure 5 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention 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 interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] According to an embodiment of the present invention, a method embodiment for predicting settlement parameters is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0030] Figure 1 is a method for predicting settlement parameters according to an embodiment of the present invention, such as Figure 1 As shown, the method comprises the following steps:

[0031] S102, preprocessing the energy attribute data of the target settlement period to be predicted to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, wherein the settlement parameter prediction network includes a linear regression processing subnetwork for feature selection and a volatility prediction subnetwork for predicting the degree of volatility of the sequence data;

[0032] Optionally, in this embodiment, the target settlement period may be, but is not limited to, a statistical period for delivering an equivalent price to the entity providing natural energy for the natural energy used. For example, the natural energy here may include, but is not limited to: electric energy, water energy, gas energy, etc.

[0033] In addition, in this embodiment, the above-mentioned energy attribute data can be but is not limited to various data variables associated with the energy market, including but not limited to electricity prices, coal power generation costs, diesel power generation costs, reserve prices, forward prices, precipitation data, etc. These data constitute the basic characteristics and dynamics of the energy market.

[0034] Optionally, in this embodiment, the settlement parameter prediction network is a combined neural network including: a linear regression processing subnetwork and a volatility prediction subnetwork. It should be noted that the two types of subnetworks in the settlement parameter prediction network can be selected and combined from multiple neural networks with corresponding functions provided in the relevant technology, but are not limited to them. Different combinations have different performances. The settlement parameter prediction network used in this embodiment is finally determined to be applied to the settlement parameter prediction scenario after comparing the network performance evaluation parameters.

[0035] S104, in the linear regression processing sub-network, performing linear regression compression processing on the energy attribute data sequence to obtain a selected periodic energy attribute data sequence that matches the target settlement period;

[0036] Optionally, in this embodiment, the above-mentioned linear regression processing subnetwork includes: Lasso, mutual information. Among them, the Least Absolute Shrinkage and Selection Operator (Lasso for short) is a linear regression analysis method that uses L1 regularization, which makes it very effective in dealing with feature selection problems. Lasso compresses the regression coefficients by adding a penalty term in the loss function, which is the sum of the absolute values ​​of all regression coefficients, thereby achieving feature selection. Mutual Information (MI for short) is an information theory concept that measures the degree of mutual dependence between two random variables. It is based on the idea of ​​information entropy, indicating that the amount of information obtained from one variable can be used to reduce the uncertainty of another variable. Mutual information can be regarded as a measure of shared information between variables. The larger its value, the stronger the dependence between the two variables.

[0037] S106, in the volatility prediction subnetwork, prediction is performed based on the periodic energy attribute data sequence to obtain a settlement parameter fluctuation characteristic that matches the target settlement period;

[0038] Optionally, in this embodiment, the volatility prediction subnetwork may include but is not limited to: Generalized Autoregressive Conditional Heteroskedasticity (GARCH), Autoregressive Mean and Stochastic Volatility (SV) and its t-distribution variants) and deep learning models (Long Short-Term Memory, LSTM and Gated Recurrent Unit, GRU).

[0039] Among them, in this embodiment, the above-mentioned generalized autoregressive conditional heteroskedasticity volatility model (GARCH) is a statistical model, which is mainly used to analyze and predict the volatility of time series data, especially in financial and economic data, such as volatility analysis of stock prices, exchange rates, commodity prices, etc. The GARCH model is based on the ARCH model (Autoregressive Conditional Heteroskedasticity), allowing the conditional variance (volatility) to change dynamically over time, that is, the model can capture the phenomenon of "volatility clustering" in the data, that is, high volatility periods are often followed by high volatility.

[0040] Among them, in this embodiment, the above-mentioned autoregressive mean and stochastic volatility model (SV) is another statistical model for estimating and predicting the volatility of time series data. Unlike GARCH, the SV model regards volatility as a random process, usually assuming that volatility obeys a random walk process or has a more complex dynamic structure. The SV model can better handle the nonlinear changes and long memory characteristics of volatility, so it performs well in processing data with complex volatility characteristics.

[0041] Among them, in this embodiment, the above-mentioned t distribution variant is in the SV model, and the t distribution variant generally refers to the use of Student's t distribution to model the residual term, which can better capture the heavy tail phenomenon in financial time series data, that is, the probability of extreme volatility is higher than the probability under the normal distribution assumption. Student's t distribution has a thicker tail, which enables the model to more accurately estimate volatility and risk management.

[0042] Among them, in this embodiment, the above-mentioned long short-term memory network (LSTM) and gated recurrent unit (GRU) are two special types of recurrent neural networks (RNN) for processing time series data in the field of deep learning. They are designed to overcome the gradient vanishing and gradient exploding problems of traditional RNN when processing long sequences. LSTM controls the flow and storage of information by introducing "memory cells" (cell state) and three gating mechanisms (input gate, forget gate, output gate), enabling it to learn long sequence dependencies. LSTM is very suitable for processing data with long-term dependencies, such as speech recognition, natural language processing, and time series prediction. GRU is a simplified version of LSTM, which merges the forget gate and input gate into an update gate while retaining a reset gate. GRU has a simpler structure and fewer parameters, so it has advantages in training speed and model complexity. GRU is also able to learn long-term dependencies in sequences, but may not be as accurate as LSTM on some tasks.

[0043] S108, determining the target settlement parameters corresponding to the target settlement period according to the settlement parameter fluctuation characteristics.

[0044] It should be noted that, in this embodiment, after analyzing the settlement parameter fluctuation characteristics of the input data through the above-mentioned settlement parameter prediction network, the target settlement parameters can be accurately determined based on the settlement parameter fluctuation characteristics obtained here. Taking electricity as an example, the time series characteristics of electricity prices can be captured in the above-mentioned manner. In the short-term prediction of high-frequency data, the inherent fluctuation characteristics of electricity prices can be predicted, which facilitates the accurate prediction of electricity prices and the reasonable allocation of electric energy resources, thereby improving energy utilization and reducing energy waste.

[0045] The following examples are used to illustrate this: Assuming that the volatility prediction subnetwork selects the generalized autoregressive conditional heteroskedasticity model (GARCH), and the linear regression processing subnetwork selects the least absolute shrinkage and selection operator (Lasso), the settlement parameter prediction network constructed by the two will accurately analyze the volatility of time series data, especially for variables with significant volatility such as electricity prices. By dynamically modeling the variance of electricity price data to reflect the characteristics of volatility, it can also flexibly reflect the conditional variance that changes over time, and effectively describe the "clustered volatility" phenomenon in the market (that is, after a period of drastic fluctuations in electricity prices, electricity prices often continue to fluctuate violently). This effectively reduces the amount of calculation, avoids overfitting caused by full features, and releases memory pressure.

[0046] Through the embodiments provided in this application, by optimizing the combination of feature selection and volatility prediction, the predictive ability of deep learning is utilized, and statistical models and feature selection methods are combined to form an efficient and accurate prediction framework, which can not only show significant advantages in the prediction of energy settlement parameters and improve the accuracy of prediction results, but also provide market participants with more reliable decision-making basis, so as to better cope with the complexity and volatility of energy market data.

[0047] As an optional solution, the energy attribute data of the target settlement period to be predicted is preprocessed to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, including:

[0048] S1, determine the data distribution characteristics of energy attribute data;

[0049] S2, transforming and scaling the energy attribute data according to a transformation and scaling mode that matches the data distribution characteristics to obtain an energy attribute data sequence.

[0050] Optionally, in this embodiment, the above data distribution characteristics may be, but are not limited to, probability distribution characteristics of energy attribute data in its natural state, including whether the data presents a normal distribution, a heavy-tailed distribution (such as Student's t distribution), a skewed distribution, a multimodal distribution, etc. The above transformation scaling mode may be, but is not limited to, a transformation function for transformation processing and a scaling function for scaling processing. Among them, the transformation processing may be, but is not limited to, in the data preprocessing stage, according to the distribution characteristics of the data, adjusting the scale and shape of the data through mathematical transformations (such as Box-Cox transformation, Yeo-Johnson transformation) to improve the statistical characteristics of the data, so that it is closer to a normal distribution or meets the assumptions of the model. Scaling processing: convert the data to a uniform scale (such as between 0-1 or -1 and 1), and eliminate the scale differences between different features through standardization or normalization techniques to ensure the stability of model training and the reliability of prediction results.

[0051] Optionally, in this embodiment, the energy attribute data is transformed and scaled according to a transformation and scaling mode that matches the data distribution characteristics, and the energy attribute data sequence obtained includes:

[0052] 1) When the data distribution characteristics indicate that the energy attribute data presents a heavy-tailed distribution, statistically transform the energy attribute data to obtain a transformed first reference data sequence, wherein heteroscedasticity has been eliminated in the first reference data sequence; and perform extreme value scaling on the first reference data sequence to obtain an energy attribute data sequence;

[0053] 2) When the data distribution characteristics indicate that the energy attribute data are normally distributed, the energy attribute data are directly extreme value scaled to obtain the energy attribute data sequence.

[0054] As an optional solution, the energy attribute data of the target settlement period to be predicted is preprocessed to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, including: when the energy attribute data is a data vector carrying directional information, the energy attribute data is cosine transformed to obtain a transformed second reference data sequence; the second reference data sequence is extreme value scaled to obtain the energy attribute data sequence.

[0055] For example, still taking the electric energy data as an example, Figure 2 The following content is explained:

[0056] In electricity price forecasting, data standardization is an important step to ensure the forecasting performance of the settlement parameter forecasting network. In this implementation, different standardization methods are used for different features of data with different probability distribution characteristics.

[0057] like Figure 2As shown, after energy attribute data is acquired in S202, S204-S206 are executed to determine data distribution characteristics and distribution types.

[0058] As in S206-1, when the data distribution characteristics indicate a heavy-tailed distribution (such as electricity prices, coal-fired power generation, diesel power generation, reserve prices, forward prices and precipitation data), because the median deviates from the mean and there is a tendency for extreme values, as in S208-1, the Box-Cox transformation is used, or as in S208-2, the Yeo-Johnson transformation (depending on whether the data is a positive value) is used for standardization, and then the extreme value (min-max) standardization scaling is used. As in S206-2, when the data distribution characteristics indicate a distribution close to normal distribution, the extreme value (min-max) scaling can be used directly. In addition, as in S203, when the energy attribute data is wind direction data, as in S203-1, the cosine transform is used to process the angle, and then the extreme value (min-max) scaling is performed. After the transformation and scaling, as in S210, the results after the transformation and scaling are normalized.

[0059] Through the embodiments provided in the present application, through these transformation and scaling processes, the scale differences of different features in the energy attribute data can be effectively eliminated, and the adaptability of the model (i.e., the settlement parameter prediction network) to different features can be enhanced.

[0060] As an optional solution, in the linear regression processing sub-network, the energy attribute data sequence is subjected to linear regression compression processing, and the selected periodic energy attribute data sequence matching the target settlement period includes:

[0061] S1, in the linear regression processing subnetwork, a preset regularization parameter is used to perform linear regression compression processing on the energy attribute data sequence, wherein the regularization parameter is used to indicate the degree of regression compression.

[0062] Optionally, in this embodiment, the above-mentioned linear regression processing subnetwork. It adds an absolute value penalty term in the loss function so that some regression coefficients are compressed to zero, thereby achieving feature selection and dimensionality reduction. This method can not only avoid the overfitting problem of the model, but also effectively eliminate features that do not contribute much to the prediction, simplify the model structure, and improve the generalization ability of the model. These include the minimum absolute shrinkage and selection operator network (Lasso), and the mutual information network.

[0063] For example, taking electric energy data as an example, the settlement parameter is electricity price as an example, and the regression coefficient of Lasso feature extraction is as follows:

[0064]

[0065] Here, y is the dependent variable (i.e., electricity price), x is the feature, β is the coefficient of the corresponding feature, and t is the horizontal axis time of the time series.

[0066] The formula for mutual information is as follows:

[0067]

[0068] Here, X and Y are random variables.

[0069] Through the linear regression processing subnetwork provided in this application, irrelevant or redundant features can be eliminated, and only features that have an important impact on electricity price prediction are retained, thereby significantly reducing the complexity of the model.

[0070] As an optional solution, in the volatility prediction subnetwork, prediction is performed based on the periodic energy attribute data series to obtain the settlement parameter fluctuation characteristics that match the target settlement period, including:

[0071] S1, in the volatility prediction subnetwork, the periodic energy attribute data sequence is fitted to obtain the residual sequence;

[0072] S2, use the residual sequence to calculate the conditional variance and obtain the volatility characteristics of the settlement parameters.

[0073] Optionally, in this embodiment, the volatility prediction subnetwork includes: a generalized autoregressive conditional heteroskedastic volatility network (GARCH), an autoregressive mean and stochastic volatility network (SV), and a deep learning network (LSTM and GRU).

[0074] For example, taking electric energy data as an example, the settlement parameter is electricity price as an example, and the equation of the generalized autoregressive conditional heteroskedastic volatility model (GARCH) is as follows:

[0075] y t =μ t +∈ t (3)

[0076]

[0077] Here, y t is the electricity price at time t, μ t is the mean electricity price at time t, ∈ t is the error at time t, is the variance at time t.

[0078] For example, taking electric energy data as an example, the settlement parameter is electricity price as an example, and the equation of the autoregressive mean and stochastic volatility model (SV) is as follows:

[0079] y t =μ t +∈ t(7)

[0080]

[0081] ∈ t ~N(0,e ht ) (9)

[0082]

[0083] Here, y t is the electricity price at time t, μ t is the mean electricity price at time t, ∈ t is the error at time t, is the variance at time t, h t is the logarithm of the fluctuation.

[0084] Through the embodiments provided in the present application, by dynamically predicting the fluctuation characteristics of the settlement parameters that change over time, the changing trends of the settlement parameters can be accurately captured, thereby improving the accuracy of the prediction results of the settlement parameters.

[0085] As an optional solution, before preprocessing the energy attribute data of the target settlement period to be predicted, it also includes:

[0086] S1, obtaining historical energy settlement data, wherein the historical energy settlement data includes attribute data of settlement resources consumed by using target energy in a historical period;

[0087] S2, preprocessing the historical energy settlement data to obtain a historical energy attribute data sequence that meets the input conditions of the settlement parameter prediction network;

[0088] S3, inputting the historical energy attribute data sequence into at least two candidate settlement parameter prediction networks, wherein different candidate settlement parameter prediction networks are networks constructed using different network frameworks;

[0089] S4, obtaining prediction evaluation parameters corresponding to each candidate settlement parameter network;

[0090] S5, determining a settlement parameter prediction network from at least two candidate settlement parameter prediction networks according to the comparison result of the prediction evaluation parameters.

[0091] Optionally, in this embodiment, obtaining the prediction evaluation parameter corresponding to each candidate settlement parameter network includes one of the following:

[0092] 1) Obtaining the mean absolute error matching the candidate settlement parameter network;

[0093] For example, the mean absolute error can be calculated as:

[0094]

[0095] 2) obtaining a root mean square error matching the candidate settlement parameter network;

[0096] For example, the root mean square error is:

[0097]

[0098] 3) Obtain the mean absolute percentage error that matches the candidate settlement parameter network.

[0099] For example, the root mean square error and mean absolute percentage error are:

[0100]

[0101] Optionally, in this embodiment, inputting the historical energy attribute data sequence into at least two candidate settlement parameter prediction networks includes:

[0102] S1, determining at least two candidate linear regression processing subnetworks and at least two candidate volatility prediction subnetworks, wherein the candidate linear regression processing subnetworks include: a minimum absolute shrinkage and selection operator network, and a mutual information network; the candidate volatility prediction subnetworks include: a generalized autoregressive conditional heteroskedastic volatility network, an autoregressive mean and stochastic volatility network, and a deep learning network;

[0103] S2, combining the candidate linear regression processing sub-network and the candidate volatility prediction sub-network to obtain a candidate settlement parameter prediction network.

[0104] Specific combination Figure 3 The example shown is for illustration:

[0105] After obtaining the historical energy settlement data, as in S302, S304 is executed to preprocess the training data based on the historical energy settlement data to obtain standardized training data. Then, as in S306, a network model architecture for predicting settlement parameters is constructed: a deep neural network is used as a benchmark model, and a network model architecture is constructed in combination with Lasso. And as in S308, the standardized training data preprocessed in the above S304 is used to train the network model architecture constructed in the above S306. Finally, as in S310, the trained network model is used to predict settlement parameters.

[0106] It should be noted that the above Figure 3The GARCH and Lasso selected in the above is an optional combination selection. In addition, in this embodiment, it is also possible but not limited to the autoregressive mean and stochastic volatility model (SV) and its t-distribution variant) and the deep learning model (LSTM and GRU), respectively, to conduct training evaluation in combination with the following two different feature selection methods 1: Lasso, 2: mutual information. Then, the performance of the various combined network models after the above combination is evaluated, such as using at least one prediction evaluation parameter of mean absolute error, root mean square error and mean absolute percentage error for evaluation. And according to the comparison results of the prediction evaluation parameters, the optimal network architecture for settlement is determined from the various combined network models mentioned above.

[0107] Among them, for Figure 3 The combined network model composed of GARCH and Lasso shown in the figure can eliminate irrelevant or redundant features in the power data processing scenario through the Lasso feature selection technology, and only retain the features that have an important impact on electricity price prediction, thereby significantly reducing the complexity of the model. In addition, combining the GARCH model with the feature selection (Lasso) strategy can better capture the volatility characteristics of electricity prices, so that the model can adapt to the volatility changes of time series data. Combining the Lasso feature selection method with the generalized autoregressive conditional heteroskedasticity (GARCH) model, an electricity price prediction model with higher prediction accuracy and lower overfitting risk is constructed.

[0108] Through the embodiments provided in this application, a combination of deep learning model feature selection networks with multiple different network architectures are trained, and the optimal combination is selected through evaluation and analysis. This combined evaluation of multiple models and multiple feature selection methods will be more conducive to improving the prediction accuracy of settlement parameters. In addition, unlike traditional methods that rely solely on feature selection or a single regression model, the combination of the two also achieves a balance between information processing volume and model complexity.

[0109] According to another aspect of the embodiments of the present application, a settlement parameter prediction device for implementing the above settlement parameter prediction method is also provided. Figure 4 As shown, the device comprises:

[0110] The processing unit 402 is used to pre-process the energy attribute data of the target settlement period to be predicted, and obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, wherein the settlement parameter prediction network includes a linear regression processing sub-network for feature selection and a volatility prediction sub-network for predicting the degree of volatility of the sequence data;

[0111] A compression unit 404 is used to perform linear regression compression processing on the energy attribute data sequence in the linear regression processing subnetwork to obtain a selected periodic energy attribute data sequence that matches the target settlement period;

[0112] A prediction unit 406 is used to make predictions based on the periodic energy attribute data sequence in the volatility prediction subnetwork to obtain a settlement parameter fluctuation feature that matches the target settlement period;

[0113] The first determining unit 408 determines a target settlement parameter corresponding to a target settlement period according to a settlement parameter fluctuation characteristic.

[0114] Optionally, the solution in this embodiment may be but is not limited to referring to the above method embodiment, which will not be described in detail here.

[0115] As an optional solution, the processing unit includes:

[0116] A first determination module is used to determine data distribution characteristics of energy attribute data;

[0117] The first processing module is used to perform transformation and scaling processing on the energy attribute data according to a transformation and scaling mode that matches the data distribution characteristics to obtain an energy attribute data sequence.

[0118] Optionally, the solution in this embodiment may be but is not limited to referring to the above method embodiment, which will not be described in detail here.

[0119] As an optional solution, the first processing module includes:

[0120] A first processing submodule is used for, when the data distribution characteristics indicate that the energy attribute data presents a heavy-tailed distribution, performing statistical transformation on the energy attribute data to obtain a transformed first reference data sequence, wherein heteroscedasticity has been eliminated in the first reference data sequence; and performing extreme value scaling on the first reference data sequence to obtain an energy attribute data sequence;

[0121] The second processing submodule is used to directly perform extreme value scaling on the energy attribute data to obtain an energy attribute data sequence when the data distribution characteristics indicate that the energy attribute data is normally distributed.

[0122] As an optional solution, the processing unit includes:

[0123] The second processing module is used to perform cosine transformation on the energy attribute data to obtain a transformed second reference data sequence when the energy attribute data is a data vector carrying directional information; and perform extreme value scaling on the second reference data sequence to obtain an energy attribute data sequence.

[0124] Optionally, the solution in this embodiment may be but is not limited to referring to the above method embodiment, which will not be described in detail here.

[0125] As an optional solution, the compression unit includes:

[0126] The compression module is used to perform linear regression compression processing on the energy attribute data sequence using a preset regularization parameter in the linear regression processing subnetwork, wherein the regularization parameter is used to indicate the degree of regression compression.

[0127] Optionally, the solution in this embodiment may be but is not limited to referring to the above method embodiment, which will not be described in detail here.

[0128] As an optional solution, the prediction unit includes:

[0129] The fitting module is used to fit the periodic energy attribute data sequence in the volatility prediction subnetwork to obtain the residual sequence;

[0130] The calculation module is used to calculate the conditional variance using the residual sequence to obtain the volatility characteristics of the settlement parameters.

[0131] Optionally, the solution in this embodiment may be but is not limited to referring to the above method embodiment, which will not be described in detail here.

[0132] As an optional solution, it also includes:

[0133] An acquisition unit, used for acquiring historical energy settlement data before preprocessing the energy attribute data of the target settlement period to be predicted, wherein the historical energy settlement data includes attribute data of settlement resources consumed by using the target energy in the historical period;

[0134] A preprocessing unit, used for preprocessing the historical energy settlement data to obtain a historical energy attribute data sequence that meets the input conditions of the settlement parameter prediction network;

[0135] An input unit, used for inputting the historical energy attribute data sequence into at least two candidate settlement parameter prediction networks, wherein different candidate settlement parameter prediction networks are networks constructed using different network frameworks;

[0136] An evaluation unit, used to obtain prediction evaluation parameters corresponding to each candidate settlement parameter network;

[0137] The second determining unit is used to determine a settlement parameter prediction network from at least two candidate settlement parameter prediction networks according to the comparison result of the prediction evaluation parameters.

[0138] Optionally, the solution in this embodiment may be but is not limited to referring to the above method embodiment, which will not be described in detail here.

[0139] As an option, the assessment unit may include one of the following:

[0140] A first evaluation module, for obtaining a mean absolute error matching a candidate settlement parameter network;

[0141] A second evaluation module, for obtaining a root mean square error matching the candidate settlement parameter network;

[0142] The third evaluation module is used to obtain the mean absolute percentage error matching the candidate settlement parameter network.

[0143] Optionally, the solution in this embodiment may be but is not limited to referring to the above method embodiment, which will not be described in detail here.

[0144] As an optional solution, the input unit includes:

[0145] The second determination module is used to determine at least two candidate linear regression processing subnetworks and at least two candidate volatility prediction subnetworks, wherein the candidate linear regression processing subnetworks include: a minimum absolute shrinkage and selection operator network, and a mutual information network; the candidate volatility prediction subnetworks include: a generalized autoregressive conditional heteroskedastic volatility network, an autoregressive mean and stochastic volatility network, and a deep learning network;

[0146] The combination module is used to combine the candidate linear regression processing sub-network and the candidate volatility prediction sub-network to obtain a candidate settlement parameter prediction network.

[0147] Optionally, the solution in this embodiment may be but is not limited to referring to the above method embodiment, which will not be described in detail here.

[0148] According to another aspect of the embodiment of the present application, an electronic device for implementing the above settlement parameter prediction method is also provided. Figure 5 As shown, the electronic device includes a memory 502 and a processor 504. The memory 502 stores a computer program, and the processor 504 is configured to execute the steps in any of the above method embodiments through the computer program.

[0149] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0150] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0151] S1, preprocessing the energy attribute data of the target settlement period to be predicted to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, wherein the settlement parameter prediction network includes a linear regression processing subnetwork for feature selection and a volatility prediction subnetwork for predicting the degree of volatility of the sequence data;

[0152] S2, in the linear regression processing sub-network, linear regression compression processing is performed on the energy attribute data sequence to obtain a selected periodic energy attribute data sequence that matches the target settlement period;

[0153] S3, in the volatility prediction subnetwork, prediction is performed based on the periodic energy attribute data sequence to obtain the settlement parameter fluctuation characteristics that match the target settlement period;

[0154] S4, determining the target settlement parameters corresponding to the target settlement period according to the fluctuation characteristics of the settlement parameters.

[0155] Alternatively, a person skilled in the art may understand that: Figure 5 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (Mobile Internet Devices, MID), a PAD, or other terminal devices. Figure 5 The structure of the electronic device is not limited. Figure 5 More or fewer components (such as network interfaces, etc.) as shown in, or with Figure 5 Different configurations shown.

[0156] Among them, the memory 502 can be used to store software programs and modules, such as the program instructions / modules corresponding to the settlement parameter prediction method and device in the embodiment of the present application. The processor 504 executes various functional applications and data processing by running the software programs and modules stored in the memory 502, that is, realizing the above-mentioned settlement parameter prediction method. The memory 502 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 502 may further include a memory remotely located relative to the processor 504, and these remote memories may be connected to the 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. As an example, Figure 5As shown, the memory 502 may include, but is not limited to, the processing unit 402, the compression unit 404, the prediction unit 406, and the first determination unit 408 in the prediction device for the settlement parameters. In addition, other module units in the prediction device for the settlement parameters may also be included but are not limited to, which will not be described in detail in this example.

[0157] Optionally, the transmission device 506 is used to receive or send data via a network. Specific examples of the network may include wired networks and wireless networks. In one example, the transmission device 506 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers via a network cable so as to communicate with the Internet or a local area network. In one example, the transmission device 506 is a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0158] In addition, the electronic device further comprises: a display 508 for displaying input energy attribute data and output target settlement parameters; and a connection bus 510 for connecting various module components in the electronic device.

[0159] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0160] According to one aspect of the present application, a computer program product is provided. The computer program product includes a computer program / instruction. The computer program / instruction contains a program code for executing the method shown in the flowchart.

[0161] According to one aspect of the present application, a computer-readable storage medium is provided, and a processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the settlement parameter prediction method provided in the above-mentioned various optional implementations.

[0162] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0163] S1, preprocessing the energy attribute data of the target settlement period to be predicted to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, wherein the settlement parameter prediction network includes a linear regression processing subnetwork for feature selection and a volatility prediction subnetwork for predicting the degree of volatility of the sequence data;

[0164] S2, in the linear regression processing sub-network, linear regression compression processing is performed on the energy attribute data sequence to obtain a selected periodic energy attribute data sequence that matches the target settlement period;

[0165] S3, in the volatility prediction subnetwork, prediction is performed based on the periodic energy attribute data sequence to obtain the settlement parameter fluctuation characteristics that match the target settlement period;

[0166] S4, determining the target settlement parameters corresponding to the target settlement period according to the fluctuation characteristics of the settlement parameters.

[0167] Optionally, in this embodiment, a person of ordinary skill in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing hardware related to the terminal device through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0168] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling one or more computer devices (which can be personal computers, servers or network devices, etc.) to execute all or part of the steps of the above methods in each embodiment of the present application.

[0169] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0170] In the several embodiments provided in the present application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0171] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0173] The above are preferred implementations 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 settlement parameters, characterized in that: include: Preprocessing the energy attribute data of the target settlement period to be predicted to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, wherein the settlement parameter prediction network includes a linear regression processing subnetwork for feature selection and a volatility prediction subnetwork for predicting the degree of volatility of the sequence data; In the linear regression processing subnetwork, the energy attribute data sequence is subjected to linear regression compression processing to obtain a selected periodic energy attribute data sequence that matches the target settlement period; In the volatility prediction subnetwork, prediction is performed based on the periodic energy attribute data sequence to obtain a settlement parameter fluctuation feature that matches the target settlement period; The target settlement parameter corresponding to the target settlement period is determined according to the settlement parameter fluctuation characteristics.

2. The method according to claim 1, characterized in that: The energy attribute data of the target settlement period to be predicted are preprocessed to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, including: Determining data distribution characteristics of the energy attribute data; The energy attribute data is transformed and scaled according to a transformation and scaling mode that matches the data distribution characteristics to obtain the energy attribute data sequence.

3. The method according to claim 2, characterized in that The step of performing transformation and scaling processing on the energy attribute data according to a transformation and scaling mode matching the data distribution characteristics to obtain the energy attribute data sequence includes: When the data distribution characteristics indicate that the energy attribute data presents a heavy-tailed distribution, statistically transform the energy attribute data to obtain a transformed first reference data sequence, wherein heteroscedasticity has been eliminated in the first reference data sequence; and perform extreme value scaling on the first reference data sequence to obtain the energy attribute data sequence; When the data distribution characteristics indicate that the energy attribute data is normally distributed, extreme value scaling is directly performed on the energy attribute data to obtain the energy attribute data sequence.

4. The method according to claim 2, characterized in that: The energy attribute data of the target settlement period to be predicted are preprocessed to obtain an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, including: In the case where the energy attribute data is a data vector carrying direction information, performing a cosine transform on the energy attribute data to obtain a transformed second reference data sequence; The second reference data sequence is subjected to extreme value scaling to obtain the energy attribute data sequence.

5. The method according to claim 1, characterized in that In the linear regression processing subnetwork, the energy attribute data sequence is subjected to linear regression compression processing to obtain a selected periodic energy attribute data sequence that matches the target settlement period, including: In the linear regression processing subnetwork, a preset regularization parameter is used to perform linear regression compression processing on the energy attribute data sequence, wherein the regularization parameter is used to indicate the degree of regression compression.

6. The method according to claim 1, characterized in that In the volatility prediction subnetwork, prediction is performed based on the periodic energy attribute data sequence to obtain the settlement parameter fluctuation characteristics matching the target settlement period, including: In the volatility prediction subnetwork, the periodic energy attribute data sequence is fitted to obtain a residual sequence; The conditional variance is calculated using the residual sequence to obtain the settlement parameter fluctuation characteristics.

7. The method according to claim 1, characterized in that Before the energy attribute data of the target settlement period to be predicted is preprocessed, the method further includes: Acquire historical energy settlement data, wherein the historical energy settlement data includes attribute data of settlement resources consumed by using the target energy in a historical period; Preprocessing the historical energy settlement data to obtain a historical energy attribute data sequence that meets the input conditions of the settlement parameter prediction network; Inputting the historical energy attribute data sequence into at least two candidate settlement parameter prediction networks, wherein different candidate settlement parameter prediction networks are networks constructed using different network frameworks; Obtaining prediction evaluation parameters corresponding to each of the candidate settlement parameter networks; The settlement parameter prediction network is determined from the at least two candidate settlement parameter prediction networks according to the comparison result of the prediction evaluation parameters.

8. The method according to claim 7, characterized in that The acquiring of the prediction evaluation parameters corresponding to each of the candidate settlement parameter networks comprises one of the following: Obtaining a mean absolute error matching the candidate settlement parameter network; Obtaining a root mean square error matching the candidate settlement parameter network; Obtain a mean absolute percentage error matching the candidate settlement parameter network.

9. The method according to claim 7, characterized in that: The step of inputting the historical energy attribute data sequence into at least two candidate settlement parameter prediction networks comprises: Determine at least two candidate linear regression processing subnetworks and at least two candidate volatility prediction subnetworks, wherein the candidate linear regression processing subnetworks include: a minimum absolute shrinkage and selection operator network, and a mutual information network; the candidate volatility prediction subnetworks include: a generalized autoregressive conditional heteroskedastic volatility network, an autoregressive mean and stochastic volatility network, and a deep learning network; The candidate linear regression processing subnetwork and the candidate volatility prediction subnetwork are combined to obtain the candidate settlement parameter prediction network.

10. A settlement parameter prediction device, characterized in that: include: A processing unit, used for preprocessing the energy attribute data of the target settlement period to be predicted, and obtaining an energy attribute data sequence that meets the input conditions of the settlement parameter prediction network, wherein the settlement parameter prediction network includes a linear regression processing subnetwork for feature selection and a volatility prediction subnetwork for predicting the degree of volatility of the sequence data; A compression unit, configured to perform linear regression compression processing on the energy attribute data sequence in the linear regression processing subnetwork to obtain a selected periodic energy attribute data sequence that matches the target settlement period; A prediction unit, configured to perform prediction based on the periodic energy attribute data sequence in the volatility prediction subnetwork to obtain a settlement parameter fluctuation characteristic matching the target settlement period; The first determining unit determines a target settlement parameter corresponding to the target settlement period according to the settlement parameter fluctuation characteristics.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program is executed by a processor to perform the method described in any one of claims 1 to 9.

12. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

13. 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 9 through the computer program.

Images