Spot electricity price prediction method and device based on spatial-temporal characteristics, and storage medium

Abstract

The invention relates to a spot electricity price prediction method and device based on spatial-temporal characteristics and a storage medium, and the method comprises the following steps: obtaining market data corresponding to a plurality of historical moments, including partition load demands at different moments and clearing electricity prices at different partitions at the moments; extracting time features and space features from the market data; and taking the time feature and the spatial feature as input of a trained prediction meta-model, and obtaining the partition clearing electricity price at the prediction moment. Compared with the prior art, the method has the advantages that the method can be applied to an environment with incomplete market information, the influence of time and space factors on the clearing electricity price is comprehensively considered, and the electricity price prediction precision is high.

Description

Technical Field

[0001] The present invention relates to the technical field of electricity price forecasting, and in particular to a method, device and storage medium for forecasting spot electricity prices based on spatiotemporal characteristics. Background Art

[0002] Electricity price forecasting refers to the prediction of electricity prices in a certain period of time in the future by considering known power systems, economic, social and meteorological factors and combining historical data. From the perspective of power generation, accurate electricity price forecasting is conducive to its grasp of market trends and market opportunities, so as to construct the optimal power generation and electricity price bidding strategy to obtain maximum profit; from the perspective of electricity users, electricity prices form their electricity purchase costs. Accurate electricity price forecasting allows users to control their electricity consumption according to demand and formulate reasonable electricity consumption plans. Some ordinary users can enjoy the benefits of low-peak electricity consumption by purchasing home appliances that can automatically control time and have a high degree of intelligence, thereby reducing the cost of living and playing the role of peak shaving and valley filling. However, with the continuous progress of power market reform, the volatility of electricity prices has also increased. On the one hand, multiple power suppliers compete in the same market, resulting in a more complex and uncertain market supply and demand relationship, making electricity prices difficult to predict. On the other hand, changes in market supply and demand and market participants' expectations of market prospects will have an impact on electricity prices. At the same time, more and more people are participating in the electricity market, resulting in an increase in information asymmetry and an increase in electricity market transaction costs, which seriously affects the stability of electricity prices. Therefore, accurate electricity price forecasting plays an important role in the normal operation of the power system and the power market. High-precision electricity price forecasting helps to improve the rational allocation of resources and is one of the important branches of power technology research.

[0003] Machine learning and deep learning have strong learning capabilities for long-term time series and are widely used in electricity price forecasting. Common machine learning algorithms include K-nearest neighbor regression, support vector machine, gradient boosting decision tree, extreme learning machine and random forest. Deep learning algorithms include multi-layer perceptron, gated recurrent unit grid and long short-term memory network. These algorithms are trained by using a large amount of historical data related to electricity prices, load information, weather, holidays and energy prices to extract and map the intrinsic relationship between input and output, and have achieved good results in electricity price forecasting. However, these electricity price forecasting models are often based on time series prediction, ignoring the spatial correlation between different nodes of electricity prices, which restricts the accuracy of electricity price forecasting. Summary of the invention

[0004] The purpose of the present invention is to overcome the defect that the above-mentioned prior art often makes predictions based on time series and ignores the spatial correlation between different nodes of electricity prices, and to provide a spot electricity price prediction method, device and storage medium based on spatiotemporal characteristics with high prediction accuracy.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A spot electricity price prediction method based on spatiotemporal characteristics comprises the following steps:

[0007] Obtain market data corresponding to several historical moments, including the load demand of different zones at different moments and the clearing electricity prices of different zones at these moments;

[0008] extracting temporal features and spatial features from the market data;

[0009] The time characteristics and space characteristics are used as inputs of a trained prediction meta-model to obtain the partition clearing electricity price at the prediction moment.

[0010] Furthermore, extracting the time feature specifically includes:

[0011] For a certain partition, obtain the load demand time series and clearing electricity price time series of the partition;

[0012] Standardizing the load demand time series and the clearing electricity price time series respectively to generate corresponding demand standard time series and electricity price standard time series;

[0013] According to the set time interval, the demand standard time series and the electricity price standard time series are grouped respectively, a group of demand standard time series group and electricity price standard time series group of the same time constitute a sequence pair, and a plurality of the sequence pairs form a sequence pair time series;

[0014] Based on the sequence to time series, an LSTM model is used to extract time features.

[0015] Furthermore, extracting the spatial features specifically includes:

[0016] For each partition, obtain the load demand time series and clearing electricity price time series of the partition;

[0017] Standardizing the load demand time series and the clearing electricity price time series respectively to generate corresponding demand standard time series and electricity price standard time series;

[0018] According to a set time interval, the demand standard time series and the electricity price standard time series of each partition are grouped to obtain a spatial feature sequence;

[0019] Based on the spatial feature sequence, the GAT model is used to extract spatial features.

[0020] Furthermore, the standardization process is standard deviation standardization process.

[0021] Furthermore, the prediction meta-model is constructed based on an artificial neural network.

[0022] The present invention also provides a spot electricity price prediction device based on time and space characteristics, comprising:

[0023] The market data acquisition module is used to obtain the market data corresponding to several historical moments, including the partition load demand at different moments and the clearing electricity prices of different partitions at these moments;

[0024] A time feature generation module, used for extracting time features from the market data;

[0025] A spatial feature generation module, used for extracting spatial features from the market data;

[0026] A meta-model training module, used to obtain time features and space features corresponding to a number of known electricity prices from the time feature generation module and the space feature generation module, take the time features and space features as input, take the partition clearing electricity price at the corresponding moment as output, train and verify to obtain a prediction meta-model;

[0027] The real-time prediction module is used to obtain the time characteristics and space characteristics corresponding to the market data at several moments before the prediction moment from the time characteristic generation module and the space characteristic generation module, call the trained prediction meta-model, and obtain the partition clearing electricity price at the prediction moment.

[0028] Furthermore, in the time feature generation module, extracting the time feature specifically includes:

[0029] For a certain partition, obtain the load demand time series and clearing electricity price time series of the partition;

[0030] Standardizing the load demand time series and the clearing electricity price time series respectively to generate corresponding demand standard time series and electricity price standard time series;

[0031] According to the set time interval, the demand standard time series and the electricity price standard time series are grouped respectively, a group of demand standard time series group and electricity price standard time series group of the same time constitute a sequence pair, and a plurality of the sequence pairs form a sequence pair time series;

[0032] Based on the sequence to time series, an LSTM model is used to extract time features.

[0033] Furthermore, in the spatial feature generation module, extracting the spatial features specifically includes:

[0034] For each partition, obtain the load demand time series and clearing electricity price time series of the partition;

[0035] Standardizing the load demand time series and the clearing electricity price time series respectively to generate corresponding demand standard time series and electricity price standard time series;

[0036] According to a set time interval, the demand standard time series and the electricity price standard time series of each partition are grouped to obtain a spatial feature sequence;

[0037] Based on the spatial feature sequence, the GAT model is used to extract spatial features.

[0038] The present invention also provides an electronic device, comprising a memory, a processor, and a program stored in the memory, wherein when the processor executes the program, the method for predicting spot electricity prices based on spatiotemporal characteristics as described above is implemented.

[0039] The present invention also provides a storage medium having a program stored thereon, and when the program is executed, the method for predicting spot electricity prices based on spatiotemporal characteristics as described above is implemented.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. The present invention uses a data-driven integrated learning model to implement forecasting, without knowing the specific clearing model and detailed parameters of the power system, so it can be applied in an environment with incomplete market information;

[0042] 2. The present invention utilizes the market load demand value and clearing electricity price that are publicly released in some actual electricity markets. After feature extraction, the time feature vector and the space feature vector are obtained, which are spliced ​​as input variables for predicting the partitioned electricity price. The influence of time and space factors on the clearing electricity price is comprehensively considered, and the spatial correlation between different nodes of the electricity price is considered, which effectively improves the accuracy of electricity price prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of the present invention;

[0044] Figure 2 Schematic diagram of the LSTM model used in the embodiment of the present invention;

[0045] Figure 3 Schematic diagram of the GAT model used in the embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0047] Example 1

[0048] This embodiment provides a spot electricity price prediction method based on time and space characteristics, including the following steps: obtaining market data corresponding to several historical moments, including the partition load demand at different moments and the clearing electricity prices of different partitions at these moments; extracting time characteristics and space characteristics from the market data; using the time characteristics and space characteristics as inputs of a trained prediction meta-model to obtain the partition clearing electricity price at the prediction moment.

[0049] like Figure 1 As shown in Figure 1, the training process of the prediction meta-model specifically includes:

[0050] 1) Through the market information system of the power trading center, the market data corresponding to several historical moments are obtained, including: the partition load demand at different times and the clearing electricity price of different partitions at these times, and the data is divided into training set and test set.

[0051] For example, the market data of the historical trading time T is obtained, including the load demand values ​​of different price zones (partitions) at each time t from time 1 to time T. Taking the two partitions A and B as examples, they are recorded as and It also includes the clearing electricity prices of different price zones at each time t from time 1 to time T. Taking the two zones A and B as examples, they are recorded as and In the market data at time 1-T, the corresponding data at time 1-T1 is divided into a training set, and the corresponding data at time T1+1-T is divided into a test set.

[0052] 2) Extracting time features from market data corresponding to several historical moments, specifically including the following steps:

[0053] 201) For each time t in the same price zone from time 1 to time T, the load demand value and the clearing electricity price of the price zone are sorted. Taking price zone A as an example, the sorted time series of the load demand value and the clearing electricity price of the price zone is recorded as

[0054] 202) Following step 201, standard deviation normalization is performed on the sorted time series. Take the sorted time series in price zone A as an example, where μ D is the mean value of load demand in price zone A, μ P is the mean of the clearing price in price zone A, σ D is the standard deviation of the load demand value in price zone A, σ P is the standard deviation of the clearing price in price zone A. The time series of price zone A after standard deviation standardization is recorded as

[0055] 203) Following step 202), the time series after standard deviation standardization is grouped. Taking the time series after standard deviation standardization of price zone A as an example, the time series after grouping is recorded as

[0056] 204) Continue with step 203), input the time series after grouping into the LSTM model, and obtain the time feature information extracted by the LSTM model, which is recorded as S = [s1, s2, ..., s T-23 ].

[0057] When making real-time electricity price forecasts, the above method is also used to extract time characteristic information for historical data.

[0058] The LSTM model principle used in this embodiment is as follows Figure 2 As shown in the figure, the LSTM model introduces gate units and memory units. The gate units include forget gate, input gate, and output gate. Specifically:

[0059] The output of the forget gate is f t =σ(W f ·[h t-1 ,x t ]+b f ), W f The weight matrix representing the forget gate will be continuously optimized with training. t-1 ,x t ] means to set h t-1 Vector and x t The new vector obtained by vertical concatenation of the vectors, x t represents the input at time t, obtained from the time series after grouping, h t-1 represents the output at time t-1, b f represents the bias term of the forget gate; the output of the input gate is i t =σ(W i ·[h t-1 ,x t ]+b i ), W i represents the weight matrix of the input gate, [h t-1 ,x t ] means to set h t-1 Vector and x t The new vector obtained by vertical concatenation of the vectors, b i represents the bias term of the input gate; the output of the output gate is o t =σ(W o ·[h t-1 ,x t ]+b o ), W o represents the weight matrix of the output gate, [ht-1 ,x t ] means to set h t-1 Vector and x t The new vector obtained by vertical concatenation of the vectors, b o Represents the bias term of the output gate; the output of the memory unit is C t =tanh(W c [h t-1 ,x t ]+b c ), W c represents the weight matrix of the memory unit, [h t-1 ,x t ] means to set h t-1 Vector and x t The new vector obtained by vertical concatenation of the vectors, b c Represents the bias term of the memory unit.

[0060] For the time series feature information obtained in step 203), the final output of LSTM is determined by the state of the output gate and the memory unit, that is: h t =o t tanh(C t ), the output h at multiple moments t It constitutes the time feature information S.

[0061] 3) Extracting spatial features from the market data corresponding to several historical moments, specifically including the following steps:

[0062] 301) Continue with step 202), process the time series after standard deviation normalization to obtain the spatial feature sequence H. Assuming that there are only two partitions A and B in the power system and they are connected, then H = [h A ,h B ],in,

[0063] 302) Following step 301, the spatial feature sequence H is input into the GAT model to obtain the spatial feature information extracted by the GAT model, which is recorded as R = [r1, r2, ..., r T-23 ].

[0064] When making real-time electricity price forecasts, the above method is also used to extract spatial feature information for historical data.

[0065] The GAT model principle used in this embodiment is as follows Figure 3As shown in Figure 1, the GAT model does not rely on prior access to the entire grid topology or all other nodes, can process all adjacent nodes, and does not assume any order among them. It is suitable for use in power markets that do not disclose partition location distribution maps. The steps of the GAT model to extract spatial feature information are as follows:

[0066] For the spatial feature sequence H of the load demand values ​​and clearing electricity prices in different price zones obtained in step 301), the self-attention mechanism is executed to calculate the attention coefficient e ij , represents the importance of the feature of node j to node i, e ij =a(Wh i ,Wh j ); all other nodes j are normalized using the softmax function, Calculate the linear combination of the corresponding features as the final output feature of each node. Here h′ i Represents the output of the GAT network, that is, the output of spatial feature information.

[0067] 4) Based on the training set and test set of the extracted temporal features and spatial features, the prediction meta-model is trained and verified, specifically including the following steps:

[0068] 401) The time feature sequence S and the space feature sequence R are concatenated as an input vector containing time and space features, denoted as X = [x1, x2, ..., x T-23 ], where x t =[s t ,r t ];

[0069] 402) Following step 401, the electricity price of partition A or partition B is used as the output of the meta-model, the input vector X containing the spatiotemporal features is input into the meta-model for training, and the network model is trained on the training set. Preferably, an artificial neural network is used as the meta-model;

[0070] 403) Following step 402, the effect of spot electricity price prediction is verified on the test set.

[0071] The above-mentioned prediction meta-model adopted in this embodiment can comprehensively consider the impact of time and space factors on the clearing electricity price, and effectively improve the accuracy of electricity price prediction.

[0072] If the above method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0073] Example 2

[0074] The present embodiment provides a spot electricity price prediction device based on spatiotemporal characteristics, including a market data acquisition module, a time feature generation module, a space feature generation module, a meta-model training module and a real-time prediction module, wherein the market data acquisition module is used to obtain market data corresponding to several historical moments, including the partition load demand at different moments and the clearing electricity prices of different partitions at these moments; the time feature generation module is used to extract time features from the market data; the space feature generation module is used to extract space features from the market data; the meta-model training module is used to obtain time features and space features corresponding to several known electricity prices from the time feature generation module and the space feature generation module, take the time features and space features as input, take the partition clearing electricity price at the corresponding moment as output, train and verify to obtain a prediction meta-model; the real-time prediction module is used to obtain the time features and space features corresponding to the market data at several moments before the prediction moment from the time feature generation module and the space feature generation module, call the trained prediction meta-model, and obtain the partition clearing electricity price at the prediction moment.

[0075] The rest is the same as in Example 1.

[0076] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A spot electricity price prediction method based on spatiotemporal characteristics, characterized in that: The following steps are involved: Obtain market data corresponding to several historical moments, including the load demand of different zones at different moments and the clearing electricity prices of different zones at these moments; extracting temporal features and spatial features from the market data; The time characteristics and space characteristics are used as inputs of a trained prediction meta-model to obtain the partition clearing electricity price at the prediction moment.

2. The spot electricity price prediction method based on spatiotemporal characteristics according to claim 1 is characterized in that: Extracting the time feature specifically includes: For a certain partition, obtain the load demand time series and clearing electricity price time series of the partition; Standardizing the load demand time series and the clearing electricity price time series respectively to generate corresponding demand standard time series and electricity price standard time series; According to the set time interval, the demand standard time series and the electricity price standard time series are grouped respectively, a group of demand standard time series group and electricity price standard time series group of the same time constitute a sequence pair, and a plurality of the sequence pairs form a sequence pair time series; Based on the sequence to time series, an LSTM model is used to extract time features.

3. The spot electricity price prediction method based on spatiotemporal characteristics according to claim 1, characterized in that: Extracting the spatial features specifically includes: For each partition, obtain the load demand time series and clearing electricity price time series of the partition; Standardizing the load demand time series and the clearing electricity price time series respectively to generate corresponding demand standard time series and electricity price standard time series; According to a set time interval, the demand standard time series and the electricity price standard time series of each partition are grouped to obtain a spatial feature sequence; Based on the spatial feature sequence, the GAT model is used to extract spatial features.

4. The spot electricity price prediction method based on spatiotemporal characteristics according to claim 2 or 3, characterized in that: The standardization process is a standard deviation standardization process.

5. The spot electricity price prediction method based on spatiotemporal characteristics according to claim 1, characterized in that: The prediction meta-model is constructed based on an artificial neural network.

6. A spot electricity price prediction device based on spatiotemporal characteristics, characterized in that: include: The market data acquisition module is used to obtain the market data corresponding to several historical moments, including the partition load demand at different moments and the clearing electricity prices of different partitions at these moments; A time feature generation module, used for extracting time features from the market data; A spatial feature generation module, used for extracting spatial features from the market data; A meta-model training module, used to obtain time features and space features corresponding to a number of known electricity prices from the time feature generation module and the space feature generation module, take the time features and space features as input, take the partition clearing electricity price at the corresponding moment as output, train and verify to obtain a prediction meta-model; The real-time prediction module is used to obtain the time characteristics and space characteristics corresponding to the market data at several moments before the prediction moment from the time characteristic generation module and the space characteristic generation module, call the trained prediction meta-model, and obtain the partition clearing electricity price at the prediction moment.

7. The spot electricity price prediction device based on spatiotemporal characteristics according to claim 6, characterized in that: In the time feature generation module, extracting the time feature specifically includes: For a certain partition, obtain the load demand time series and clearing electricity price time series of the partition; Standardizing the load demand time series and the clearing electricity price time series respectively to generate corresponding demand standard time series and electricity price standard time series; According to the set time interval, the demand standard time series and the electricity price standard time series are grouped respectively, a group of demand standard time series group and electricity price standard time series group of the same time constitute a sequence pair, and a plurality of the sequence pairs form a sequence pair time series; Based on the sequence to time series, an LSTM model is used to extract time features.

8. The spot electricity price prediction device based on spatiotemporal characteristics according to claim 6, characterized in that: In the spatial feature generation module, extracting the spatial features specifically includes: For each partition, obtain the load demand time series and clearing electricity price time series of the partition; Standardizing the load demand time series and the clearing electricity price time series respectively to generate corresponding demand standard time series and electricity price standard time series; According to a set time interval, the demand standard time series and the electricity price standard time series of each partition are grouped to obtain a spatial feature sequence; Based on the spatial feature sequence, the GAT model is used to extract spatial features.

9. An electronic device comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the spot electricity price prediction method based on spatiotemporal characteristics as described in any one of claims 1 to 6 is implemented.

10. A storage medium having a program stored thereon, characterized in that: When the program is executed, the spot electricity price prediction method based on spatiotemporal characteristics as described in any one of claims 1 to 6 is implemented.

Images