Distributed new energy power transaction method and device based on output characteristics, and electronic equipment

Abstract

The invention provides a distributed new energy power transaction method and device based on output characteristics, and electronic equipment. According to the implementation scheme, a dynamic mask matrix corresponding to each time scale is adopted, frequency domains corresponding to numerical weather forecast fragments of each time scale in a first time period in a distributed new energy area are subjected to mask processing, text vectors of corresponding text weather forecast data are spliced, and weather forecast multi-modal semantic features are obtained; encoding and decoding the deep semantic features of the weather forecast to obtain deep semantic features of the weather forecast; and processing the weather forecast deep semantic features of the distributed new energy area in the first time period by adopting a large language model to obtain predicted new energy output of the distributed new energy area in the first time period so as to plan a green electricity transaction plan of the distributed new energy area. Therefore, the new energy output can be accurately predicted, and the green electricity transaction plan can be reasonably planned.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information security, and in particular to a distributed new energy power transaction method and device based on output characteristics and an electronic device. BACKGROUND

[0002] With the rise of new energy, wind power and solar power have widely entered the market, and the use of energy by users has become increasingly diverse, requiring higher reliability in energy supply. However, wind power and photovoltaic power are limited by geographical location, weather conditions and other natural factors, and their output has great randomness and intermittency. In addition, they are widely distributed and have small individual capacity, making it difficult to directly participate in market energy transactions.

[0003] Therefore, the energy output of each distributed aggregation block is generally predicted separately, and global aggregation is performed to plan the overall energy transaction and supply. Considering the original data privacy protection and data communication cost of the power generation terminal, the distributed project aggregator uses a federated learning algorithm to jointly predict and train the distributed new energy power generation equipment in different climate characteristic regions to obtain predicted power generation, thereby planning energy transactions and supply. Therefore, the accuracy of the local power generation predicted by each distributed project aggregator is a technical problem to be solved in the field. SUMMARY

[0004] The present application provides a distributed new energy power transaction method and device based on output characteristics, an electronic device and a storage medium, which can solve at least one of the above technical problems.

[0005] According to an aspect of the present application, a distributed new energy power transaction method based on output characteristics is provided, comprising: dividing numerical weather forecast data of a distributed new energy region in a first time period according to each time scale to obtain numerical weather forecast data slices of each time scale; Using a dynamic mask matrix corresponding to each time scale, the frequency domain corresponding to each numerical weather forecast slice of each time scale is masked to obtain numerical weather forecast frequency domain mask slices of each time scale; Splicing the numerical weather forecast frequency domain mask slices of each time scale and the text vectors of the text weather forecast data corresponding to the numerical weather forecast data to obtain weather forecast multi-modal semantic features of the distributed new energy region in the first time period; Using an encoder and a decoder, the weather forecast multi-modal semantic features are sequentially encoded and decoded to obtain weather forecast deep semantic features of the distributed new energy region in the first time period; The large language model is used to process weather forecast deep semantic features of the distributed new energy region in the first time period and time sequence features corresponding to the numerical weather forecast data, to obtain predicted new energy output of the distributed new energy region in the first time period, so as to plan a green electricity transaction plan of the distributed new energy region.

[0006] According to another aspect of the present application, a distributed new energy power transaction device based on output characteristics is provided, comprising: A data division module is configured to divide numerical weather forecast data of a distributed new energy region in a first time period according to various time scales, to obtain numerical weather forecast data fragments of each time scale; A data mask module is configured to use a dynamic mask matrix corresponding to each time scale to mask a frequency domain corresponding to each numerical weather forecast fragment of each time scale, to obtain numerical weather forecast frequency domain mask fragments of each time scale; A numerical text splicing module is configured to splice numerical weather forecast frequency domain mask fragments of each time scale and text vectors of text weather forecast data corresponding to the numerical weather forecast data, to obtain weather forecast multi-modal semantic features of the distributed new energy region in the first time period; An encoding and decoding module is configured to use an encoder and a decoder to sequentially encode and decode the weather forecast multi-modal semantic features, to obtain weather forecast deep semantic features of the distributed new energy region in the first time period; A new energy output prediction module is configured to use a large language model to process weather forecast deep semantic features of the distributed new energy region in the first time period and time sequence features corresponding to the numerical weather forecast data, to obtain predicted new energy output of the distributed new energy region in the first time period, so as to plan a green electricity transaction plan of the distributed new energy region.

[0007] According to an aspect of the present application, an electronic device is provided, comprising at least one processor and a memory connected in communication with the at least one processor; The memory stores instructions executable by the processor, and the processor is configured to acquire the instructions from the memory and execute the instructions, so that the processor can execute the distributed new energy power transaction method based on output characteristics according to any of the embodiments of the present application.

[0008] According to an aspect of the present application, there is provided a non-transitory computer readable storage medium storing computer instructions for providing to a computer to instruct the computer to perform the distributed new energy power transaction method based on output characteristics according to any one of the embodiments of the present application.

[0009] According to the technical scheme of the present application, the numerical weather prediction data of the distributed new energy region in the first time period is divided according to different time scales to obtain numerical weather prediction data fragments of different time scales. Then, the frequency domain corresponding to the numerical weather prediction fragments of different time scales is masked by using the dynamic mask matrix corresponding to each time scale to obtain numerical weather prediction frequency domain mask fragments of different time scales, which can accurately filter out irrelevant data in the numerical weather prediction data fragments of different time scales. Next, the numerical weather prediction frequency domain mask fragments of different time scales and the text vectors of the text weather prediction data corresponding to the numerical weather prediction data are spliced to obtain the weather prediction multi-modal semantic features of the distributed new energy region in the first time period. In this way, the numerical features and the text features can be fused to obtain accurate weather prediction multi-modal semantic features. The weather prediction multi-modal semantic features are encoded and decoded in sequence by using an encoder and a decoder to obtain the weather prediction deep semantic features of the distributed new energy region in the first time period. In this way, the weather prediction deep semantic features can be further obtained. Furthermore, the weather prediction deep semantic features of the distributed new energy region in the first time period and the time series features corresponding to the numerical weather prediction data are processed by using a large language model to accurately predict the predicted new energy output of the distributed new energy region in the first time period. Thus, the green electricity transaction plan of the distributed new energy region can be accurately planned.

[0010] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present application, nor to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0011] The accompanying drawings are used to better understand the present application and do not limit the present application. Among them: Figure 1 is a flowchart of a distributed new energy power transaction method based on output characteristics according to an embodiment of the present application; Figure 2 is a schematic diagram of a large language hybrid prediction model according to an embodiment of the present application; Figure 3 is a schematic diagram of a large language hybrid prediction model according to another embodiment of the present application; Figure 4is a flowchart of a distributed new energy power transaction device based on output characteristics according to an embodiment of the present application Figure 5 is a block diagram of an electronic device for implementing the method according to an embodiment of the present application. DETAILED DESCRIPTION

[0012] Exemplary embodiments of the present application are described herein with reference to the accompanying drawings, which are meant to be exemplary and not limiting. Therefore, it should be recognized that many changes and modifications can be made to the embodiments described herein, without departing from the scope of the present application. Also, for the sake of brevity and clarity, descriptions of well-known functions and constructions are omitted herein.

[0013] Figure 1 is a flowchart of a distributed new energy power transaction method based on output characteristics according to an embodiment of the present application.

[0014] As shown in Figure 1 , the distributed new energy power transaction method based on output characteristics comprises: S110, dividing numerical weather forecast data of a distributed new energy region in a first time period according to each time scale to obtain numerical weather forecast data slices of each time scale; S120, using a dynamic mask matrix corresponding to each time scale to respectively perform mask processing on a frequency domain corresponding to the numerical weather forecast slice of each time scale to obtain numerical weather forecast frequency domain mask slices of each time scale; S130, splicing the numerical weather forecast frequency domain mask slices of each time scale and a text vector of text weather forecast data corresponding to the numerical weather forecast data to obtain weather forecast multi-modal semantic features of the distributed new energy region in the first time period; S140, using an encoder and a decoder to sequentially encode and decode the weather forecast multi-modal semantic features to obtain weather forecast deep semantic features of the distributed new energy region in the first time period; S150, using a large language model to process the weather forecast deep semantic features of the distributed new energy region in the first time period and time series features corresponding to the numerical weather forecast data to obtain predicted new energy output of the distributed new energy region in the first time period to plan a green electricity transaction plan of the distributed new energy region.

[0015] Exemplarily, as shown in Figure 2 , the large language hybrid prediction model provided in the present embodiment comprises a dynamic mask matrix of each time scale, an encoder and a decoder, and a large language model.

[0016] Exemplarily, the new energy power generation is mainly affected by weather factors. In this example, the numerical weather prediction (NWP) is used as the input of the prediction model, that is, wherein, are the wind speed, wind direction, humidity, temperature, air pressure and light intensity at time t, respectively.

[0017] Exemplarily, the numerical weather prediction data of the distributed new energy region in the first time period can include the wind speed, wind direction, humidity, temperature, air pressure and light intensity at each time in the first time period.

[0018] Exemplarily, the input NWP data x t is decomposed into fine-grained, medium-grained and coarse-grained paragraphs to capture features and patterns at different time scales, which can help the model better understand short-term fluctuations, medium-term trends and long-term changes in the data. For example, the specific extraction process is as follows: Fine-grained: select time steps with a time interval length of 2 to form fine-grained numerical weather prediction fragments x k，1 . Medium-grained: select time steps with a time interval length of 10 to form medium-grained numerical weather prediction fragments x k，2 . Coarse-grained: select time steps with a time interval length of 30 to form coarse-grained numerical weather prediction fragments x k，3 .

[0019] Exemplarily, the discrete cosine transform is performed on the numerical weather prediction fragments of each granularity, i.e., time scale, to obtain the frequency domain corresponding to the numerical weather prediction fragments of each time scale, which is as follows:

[0020] wherein, X k，i represents the frequency domain of the numerical weather prediction fragment of the i-th time scale, x t，i represents the numerical weather prediction fragment of the i-th time scale, T represents the total number of time points in the first time period, and k represents any time point in the first time period.

[0021] Exemplarily, for each scale of the numerical weather prediction fragment X k , an independent learnable dynamic mask matrix M i is used to filter irrelevant frequency components, and the specific process of the mask is as follows: wherein, is the element multiplication, and M i represents the dynamic mask matrix of the i-th time scale.

[0022] Exemplarily, the dynamic mask matrix M includes a plurality of rows and columns of mask elements, wherein the elements with a value of 1 are not used for masking, and the elements with a value of 0 are used for masking. The values of the mask elements in the dynamic mask matrix M can be dynamically adjusted according to the subsequent mask loss. For example, the mask loss is determined once every time the mask is executed, and the dynamic mask matrix M is dynamically adjusted according to the mask loss this time for the next mask operation.

[0023] For example, using the gradient of the mask loss and the learning rate η, the dynamic mask matrix M is updated, specifically as follows:

[0024] Exemplarily, the numerical weather forecast frequency domain mask pieces of each time scale are inversely spliced according to the original way of dividing the numerical weather forecast data to obtain a numerical weather forecast frequency domain mask vector. The numerical weather forecast frequency domain mask vector and the text vector of the text weather forecast data corresponding to the numerical weather forecast data are spliced to obtain the weather forecast multi-modal semantic feature X of the distributed new energy region in the first time period. remix The weather forecast multi-modal semantic feature contains the masked numerical features and corresponding text features of different time scales, i.e., time information and semantic information, which can accurately describe the weather of the distributed new energy region in the first time period, facilitating subsequent accurate prediction of the new energy output of the distributed new energy region in the first time period.

[0025] Exemplarily, the text weather forecast data corresponding to the numerical weather forecast data can be a text description paragraph of the weather forecast. For example, the wind speed at time t is 2.5 meters per second, blowing from the southwest. The humidity is 50%, the temperature is 18℃, the air pressure is 1012hPa, and the light intensity is 3200lux. The southwest wind brings certain water vapor, moderate humidity, comfortable temperature, stable air pressure, and good light.

[0026] Exemplarily, the BERT model can be used to extract text features of the text weather forecast data to obtain a text vector of the text weather forecast data.

[0027] It can be understood that BERT is a pre-training language model based on the Transformer architecture, which is pre-trained through a large amount of text data to learn the deep semantics and grammatical structure of the language. After pre-training is completed, BERT can be used to extract text feature vectors.

[0028] Exemplarily, after obtaining the multi-modal fused semantic feature X remix , after processing by the encoder and the decoder, a weather forecast deep semantic feature F TTThe semantic features are further captured and compressed by the encoding process of the encoder, and the core information of the input sequence data is effectively transmitted. The decoder generates the weather forecast deep semantic features F TT

[0029] Exemplarily, the weather forecast deep semantic features of the distributed new energy region in the first time period and the time sequence features corresponding to the numerical weather forecast data are spliced, and the spliced results are input into the large language model to obtain the predicted new energy output of the distributed new energy region in the first time period output by the large language model. Understandably, the predicted new energy output of the distributed new energy region in the first time period includes the new energy output at each time in the first time period, i.e., the output power or power generation power.

[0030] Exemplarily, based on the predicted new energy output of the distributed new energy region in the first time period, the sub-output of each new energy in the region in the time period can be planned, so that the green electricity trading plan of the distributed new energy region can be determined. The green electricity trading plan includes the power generation power or output power of each new energy in the region for trading in the first time.

[0031] According to the above embodiment, the numerical weather forecast data of the distributed new energy region in the first time period is divided according to each time scale to obtain numerical weather forecast data fragments of each time scale. Thus, the frequency domain corresponding to each time scale of the numerical weather forecast fragment is masked by using the dynamic mask matrix corresponding to each time scale to obtain the numerical weather forecast frequency domain mask fragment of each time scale, which can accurately filter the irrelevant data in the numerical weather forecast data fragment of each time scale. Then, the numerical weather forecast frequency domain mask fragment of each time scale and the text vector of the text weather forecast data corresponding to the numerical weather forecast data are spliced to obtain the weather forecast multi-modal semantic features of the distributed new energy region in the first time period. Thus, the numerical features and text features can be fused to obtain accurate weather forecast multi-modal semantic features. The encoder and the decoder are used to encode and decode the weather forecast multi-modal semantic features in sequence to obtain the weather forecast deep semantic features of the distributed new energy region in the first time period. Thus, the weather forecast deep semantic features can be further obtained. Further, the large language model is used to process the weather forecast deep semantic features of the distributed new energy region in the first time period and the time sequence features corresponding to the numerical weather forecast data to accurately predict the predicted new energy output of the distributed new energy region in the first time period. Thus, the green electricity trading plan of the distributed new energy region can be accurately planned.

[0032] ​In an embodiment, a large language model is used to process the weather forecast deep semantic features of the distributed new energy region in a first time period and the time series features corresponding to the numerical weather prediction data to obtain the predicted new energy output of the distributed new energy region in the first time period, including: using a convolutional neural network corresponding to each time scale to convolve the numerical weather prediction data slices of each time scale to obtain the numerical weather prediction data slice time series features of each time scale; splicing the numerical weather prediction data slice time series features of each time scale to obtain the numerical weather prediction time series features of the distributed new energy region in the first time period; fusing the weather forecast deep semantic features of the distributed new energy region in the first time period and the numerical weather prediction time series features to obtain the weather forecast target features of the distributed new energy region in the first time period; using a large language model to predict the weather forecast target features of the distributed new energy region in the first time period to obtain the predicted new energy output of the distributed new energy region in the first time period.

[0033] As shown in the example, Figure 3 The large language hybrid prediction model provided by the embodiment also includes a convolutional neural network corresponding to each time scale, i.e., a multi-scale convolutional neural network. For example, the multi-scale convolutional neural network is used to extract time series features in the input NWP data. Long-term and short-term time series change patterns are captured through multiple parallel convolution branches. The multi-scale convolutional neural network includes multiple parallel convolution branches, and each branch captures features of different scales through different convolution and sizes. The example uses 4 branches, and the convolution kernel size of each branch is 1, 2, 3, and 4 respectively. The output feature of each branch is represented as F i , where k i represents the convolution kernel size of each branch, and is specifically as follows: F i = Conv(x t , k i )(i = 1, 2, 3, 4).

[0034] Then, after splicing and fusing the output features of each branch, the time series change pattern F MSCNN , i.e., the weather prediction time series feature mentioned above, is obtained, and is specifically as follows: F MSCNN = Concat(F1, F2, F3, F4).

[0035] As an example, the weather prediction time series feature F MSCNN is fused with the weather forecast deep semantic feature F TT to obtain the weather forecast target feature F

[0036] Specifically, F = aF + (1-a)F MSCNN TT wherein a is an adjustable weight parameter.

[0037] In this example, by combining the weather forecast time series feature F MSCNN the detailed changes of different time scales in the original time series and the advanced semantic information about the time series contained in the weather forecast deep semantic feature F TT , the influence of weather noise contained in a single level feature on the prediction of a large language model can be avoided. By multiple feature fusion, different feature information can be introduced, so that the model learns more general feature patterns, thereby enhancing the robustness of the model.

[0038] In an embodiment, further comprising: adopting a BERT model to perform text feature extraction on the text weather forecast data corresponding to the numerical weather forecast data, to obtain a text vector of the text weather forecast data corresponding to the numerical weather forecast data; obtaining a mask loss corresponding to each time scale based on a difference between the text vector and the mask slice of the numerical weather forecast frequency domain corresponding to each time scale; and updating the dynamic mask matrix corresponding to each time scale based on the mask loss corresponding to each time scale.

[0039] Illustratively, since the text weather forecast data corresponding to the numerical weather forecast data is actually a text description of the weather values in the numerical weather forecast data, therefore, by comparing the text vector corresponding to the text weather forecast data with the mask slice of the numerical weather forecast frequency domain, the mask loss can be determined. If the difference is large, it means that the accuracy of the dynamic mask matrix is insufficient and needs to be adjusted. If the difference is small, it means that the accuracy of the dynamic mask matrix meets the requirements and can be adjusted or not.

[0040] Illustratively, in a case where the mask loss corresponding to the first time scale is greater than a preset loss value, the mask matrix corresponding to the first time scale is updated based on the gradient information of the mask loss corresponding to the first time scale.

[0041] According to the above embodiment, by comparing the text vector corresponding to the text weather forecast data with the mask slice of the numerical weather forecast frequency domain corresponding to each time scale, the mask loss is determined, and the dynamic mask matrix corresponding to each time scale is updated to ensure the accuracy of the dynamic mask matrix.

[0042] ​In an embodiment, further comprising: determining a time series prediction loss based on a difference between a predicted new energy output and an actual new energy output of the distributed new energy region in the first time period; determining a semantic feature alignment loss based on a difference between a weather forecast deep semantic feature and a weather forecast time series feature of the distributed new energy region in the first time period; determining a target loss based on the time series prediction loss and the semantic feature alignment loss; and updating the convolutional neural network, the dynamic mask matrix, the BERT model, the encoder, the decoder, and the large language model based on the target loss.

[0043] Exemplarily, the fusion feature F is input into the large language model for prediction to obtain predicted distributed new energy output data

[0044] wherein LLM is a large language model. For example, GPT-3 can be used as a specific large language model.

[0045] Exemplarily, the large language hybrid prediction model of the present example includes a convolutional neural network of each time scale, a dynamic mask matrix of each time scale, a BERT model, an encoder, a decoder, and a large language model.

[0046] Exemplarily, the loss function L of the large language hybrid prediction model, i.e., the target loss, is composed of two parts: a time series prediction loss L Time and a semantic feature alignment loss L Feat , which are as follows: L=L Time +λL Feat .

[0047] wherein λ is a weight parameter for balancing the two loss parts.

[0048] wherein the time series prediction loss L Time may be as follows:

[0049] wherein y t represents the actual new energy output at time t.

[0050] wherein the semantic feature alignment loss L Feat may be as follows: L Feat =||F MSCNN -F TT || 2 .

[0051] According to the above-mentioned embodiment, the semantic feature alignment loss is determined by the difference between the weather prediction deep semantic feature reflecting the numerical value and the semantic feature and the weather prediction time sequence feature reflecting the time sequence feature, the time sequence prediction loss is determined by the difference between the predicted new energy output and the real new energy output, and thus, the parameter adjustment of the entire large language mixed prediction model in combination with the two losses can improve the accuracy of the large language mixed prediction model in predicting the new energy output, and the rationality of the green electricity trading plan of the region can also be improved subsequently.

[0052] In an embodiment, further comprising: determining local model parameters of the distributed new energy region based on the updated convolutional neural network, the updated dynamic mask matrix, the updated BERT model, the updated encoder, the updated decoder and the updated large language model; uploading the local model parameters of the distributed new energy region to the global server to enable the global server to aggregate the local model parameters of each distributed new energy region to obtain global model parameters; and based on the global model parameters, updating the convolutional neural network, the dynamic mask matrix, the BERT model, the encoder, the decoder and the large language model again in the case of receiving the global model parameters from the global server.

[0053] Illustratively, the global server determines the weight according to the output of each distributed new energy region, and then aggregates the local model parameters of each distributed new energy region by using the weight of each distributed new energy region to obtain the global model parameters.

[0054] Illustratively, the local model parameters can include the model parameters of the convolutional neural network, the dynamic mask matrix, the BERT model, the encoder, the decoder and the large language model.

[0055] Illustratively, the global model parameters can include the model parameters of the convolutional neural network, the dynamic mask matrix, the BERT model, the encoder, the decoder and the large language model.

[0056] It can be understood that the embodiments of the present application can be applied to the distributed edge node devices corresponding to the distributed new energy regions.

[0057] Illustratively, after the second update, the local models used by each distributed edge node device are the same, maintaining the consistency of the prediction.

[0058] According to the above-mentioned embodiment, the model prediction accuracy of the distributed new energy region can be improved by the federated learning algorithm of the local model update, then the global model update, and then the local model update.

[0059] In an embodiment, the method further comprises: sending the predicted new energy output of the distributed new energy region in the first time period to the global server, so that the global server plans and adjusts the predicted new energy output of each distributed new energy region in the first time period; and based on the planned and adjusted predicted new energy output, determining the green electricity transaction plan of the distributed new energy region in the first time period, when the planned and adjusted predicted new energy output from the global server is received.

[0060] Exemplarily, the global server can determine the historical new energy output of each distributed new energy region in the historical output data according to the time characteristics of the first time period, and then use the historical new energy output of each distributed new energy region to plan and adjust the predicted new energy output of each distributed new energy region in the first time period. For example, if the historical new energy output corresponding to the time characteristics of the first time period is too different from the predicted new energy output in the first time period, the historical new energy output is used as the new predicted new energy output. If the difference is not large, the predicted new energy output is kept unchanged.

[0061] Exemplarily, based on the planned and adjusted predicted new energy output, the green electricity transaction plan of the distributed new energy region in the first time period can be determined by using the following game process.

[0062] Exemplarily, it is assumed that the distributed new energy project aggregators (demand side) and the distributed new energy generators (supply side) in the distributed new energy region form a system. The supply side includes distributed new energy generators such as solar energy and wind energy, and the demand side includes users or enterprises that need to purchase new energy power.

[0063] First, the model assumes the following: 1. Supply side: it is assumed that the supply side has N distributed new energy generators, each generator i has its own power generation cost c i and maximum power generation capacity P i .

[0064] 2. Demand side: it is assumed that the demand side has M users, each user j has its own power demand D i and willingness to pay w i .

[0065] 3. Pricing mechanism: the aggregator needs to set a fair unified price p for each generator, and a purchase price q for each user.

[0066] Second, the model construction process includes master-slave game modeling and cooperative game modeling. Specifically as follows: The model of master-slave game modeling includes: Supply-side model: The revenue of each generator i is p×P i c i ×P i Where p is the uniform price, P i It refers to electricity generation, c i It's the cost of electricity generation.

[0067] Demand-side model: The cost for each user j is q×D j Where q is the purchase price, and D j It refers to demand.

[0068] Models for cooperative game theory include: Supply-side cooperative game: The supply side allocates power generation capacity through cooperative game theory to minimize total cost. Assume the supply-side cost function is: Where P = [P1, P2, P3, ..., P N ] is the power generation capacity vector.

[0069] Demand-side cooperative game: The demand side allocates demand through cooperative game theory to minimize total cost. Assume the demand-side cost function is: Where D = [D1, D2, D3, ..., D M ] is the demand vector.

[0070] Next, based on the game theory model described above, the following pricing model is obtained: 1. Supply-side pricing: The goal of supply-side pricing is to maximize total revenue while ensuring that the revenue of each power generator is non-negative. The supply-side optimization problem can be expressed as:

[0071] 2. Demand-Side Pricing: The goal of demand-side pricing is to minimize total cost while ensuring that each user's cost does not exceed their willingness to pay. The demand-side optimization problem can be expressed as:

[0072] Finally, the above model is solved, and the specific process is as follows: 1. The process of solving master-slave game theory Phase 1: On the demand side, the price (p) is determined by the supply side and the consumer's willingness to pay (w). j Select demand quantity D j .

[0073] Phase Two: The supply side adjusts its supply based on the demand demand D from the demand side. j and its own cost c i Select power generation P i .

[0074] Iterative process: repeat the above two stages until Nash equilibrium state.

[0075] 2. Process of cooperative game solution Supply-side cooperative game: solve the optimal generation capacity allocation P of the supply side using linear programming or nonlinear programming method.

[0076] Demand-side cooperative game: solve the optimal demand quantity allocation D of the demand side using linear programming or nonlinear programming method.

[0077] 3. Process of pricing solution Supply-side pricing: solve the optimization problem of the supply side to determine the uniform price

[0078] Demand-side pricing: solve the optimization problem of the demand side to determine the purchase price

[0079] According to the above embodiment, after local new energy output is locally planned, global output planning is performed, and the local new energy output is updated using the global output planning result. In this way, the local new energy output is modeled by master-slave game modeling, cooperative game modeling and pricing modeling, and finally, the modeled model is solved to obtain the final green electricity trading plan, thereby optimizing the overall energy utilization efficiency.

[0080] In this way, subsequent contract signing and trading are performed according to the final green electricity trading plan.

[0081] Figure 4 is a structural block diagram of a distributed new energy power trading device based on output characteristics according to an embodiment of the present application.

[0082] As shown in Figure 4 the distributed new energy power trading device based on output characteristics includes: A data division module 410 is configured to divide numerical weather forecast data of a distributed new energy region in a first time period according to each time scale to obtain numerical weather forecast data fragments of each time scale. A data mask module 420 is configured to mask a frequency domain corresponding to each numerical weather forecast fragment of each time scale using a dynamic mask matrix corresponding to each time scale to obtain numerical weather forecast frequency domain mask fragments of each time scale. A numerical text splicing module 430 is configured to splice numerical weather forecast frequency domain mask fragments of each time scale and text vectors of text weather forecast data corresponding to the numerical weather forecast data to obtain weather forecast multi-modal semantic features of the distributed new energy region in the first time period. The coding and decoding module 440 is configured to sequentially encode and decode the weather forecast multi-modal semantic features by using an encoder and a decoder to obtain weather forecast deep semantic features of the distributed new energy region in the first time period; and the new energy output prediction module 450 is configured to process the weather forecast deep semantic features of the distributed new energy region in the first time period and time sequence features corresponding to the numerical weather forecast data by using a large language model to obtain predicted new energy output of the distributed new energy region in the first time period, so as to plan a green electricity transaction plan of the distributed new energy region.

[0083] In an embodiment, the new energy output prediction module 450 comprises: The convolution unit is configured to perform convolution on the numerical weather forecast data slices of each time scale by using a convolutional neural network corresponding to each time scale to obtain numerical weather forecast data slice time sequence features of each time scale; and the time sequence feature splicing unit is configured to splice the numerical weather forecast data slice time sequence features of each time scale to obtain numerical weather forecast time sequence features of the distributed new energy region in the first time period. The feature fusion unit is configured to fuse the weather forecast deep semantic features of the distributed new energy region in the first time period and the numerical weather forecast time sequence features to obtain weather forecast target features of the distributed new energy region in the first time period. The output prediction unit is configured to predict the weather forecast target features of the distributed new energy region in the first time period by using a large language model to obtain predicted new energy output of the distributed new energy region in the first time period.

[0084] In an embodiment, the apparatus further comprises: The text vector determination module is configured to extract text features from the text weather forecast data corresponding to the numerical weather forecast data by using a BERT model to obtain a text vector of the text weather forecast data corresponding to the numerical weather forecast data. The mask loss determination module is configured to obtain a mask loss corresponding to each time scale based on a difference between each numerical weather forecast frequency domain mask slice of each time scale and the text vector. The mask matrix update module is configured to update a dynamic mask matrix corresponding to each time scale based on the mask loss corresponding to each time scale.

[0085] In an embodiment, the apparatus further comprises: a time series prediction loss determination module configured to determine a time series prediction loss based on a difference between a predicted new energy output and an actual new energy output of the distributed new energy region in the first time period; a semantic feature alignment loss determination module configured to determine a semantic feature alignment loss based on a difference between a weather forecast deep semantic feature and a weather forecast time series feature of the distributed new energy region in the first time period; a target loss determination module configured to determine a target loss based on the time series prediction loss and the semantic feature alignment loss; a model updating module configured to update the convolutional neural network, the dynamic mask matrix, the BERT model, the encoder, the decoder, and the large language model based on the target loss.

[0086] In an embodiment, the apparatus further includes: a local model parameter initial updating module configured to determine local model parameters of the distributed new energy region based on the updated convolutional neural network, the updated dynamic mask matrix, the updated BERT model, the updated encoder, the updated decoder, and the updated large language model; a global model parameter updating module configured to upload the local model parameters of the distributed new energy region to a global server, so that the global server aggregates the local model parameters of each of the distributed new energy regions to obtain global model parameters; a local model parameter secondary updating module configured to, in a case where the global model parameters from the global server are received, update the convolutional neural network, the dynamic mask matrix, the BERT model, the encoder, the decoder, and the large language model based on the global model parameters.

[0087] In an embodiment, the apparatus further includes: a global output adjustment module configured to send the predicted new energy output of the distributed new energy region in the first time period to a global server, so that the global server plans and adjusts the predicted new energy output of each of the distributed new energy regions in the first time period; a green electricity transaction plan planning module configured to, in a case where the planned and adjusted predicted new energy output from the global server is received, determine a green electricity transaction plan of the distributed new energy region in the first time period based on the planned and adjusted predicted new energy output.

[0088] The specific functions and example descriptions of the modules and sub-modules of the system of the embodiments of the present application can be referred to the related descriptions of the corresponding steps in the method embodiments described above, and will not be described here again.

[0089] In the technical solutions of the present application, the acquisition, storage and application of user personal information comply with relevant laws and regulations and do not violate public order and good customs.

[0090] According to the embodiments of the present application, the present application also provides a system and a readable storage medium.

[0091] Figure 5 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application described and / or claimed in this document.

[0092] As shown in Figure 5 The electronic device 800 includes a computing unit 801 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. Various programs and data required for the operation of the electronic device 800 can also be stored in the RAM 803. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0093] A plurality of components in the electronic device 800 are connected to the I / O interface 805, including an input unit 806, such as a keyboard, a mouse, etc., an output unit 807, such as various types of displays, a speaker, etc., the storage unit 808, such as a magnetic disk, an optical disk, etc., and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the electronic device 800 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0094] The computing unit 801 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, and the like. The computing unit 801 performs various methods and processes described above, such as the power system production simulation method considering the peaking characteristics of coal-fired units. For example, in some embodiments, the power system production simulation method considering the peaking characteristics of coal-fired units can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded onto the RAM 803 and executed by the computing unit 801, one or more steps of the power system production simulation method considering the peaking characteristics of coal-fired units described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the power system production simulation method considering the peaking characteristics of coal-fired units by any other appropriate means, such as by means of firmware.

[0095] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (PLD), a computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0096] Program code for carrying out operations of the methods of the present application can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, and partially on a remote machine or entirely on a remote machine or server.

[0097] In the context of the present application, a machine-readable medium can be a tangible medium that can contain or store program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable storage medium can include, without limitation, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the foregoing.

[0098] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0099] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0100] The computer system can include clients and servers. This relationship can be. The servers are typically remote from the clients with the interactions between them occurring over a communication network. The relationship between client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The servers can be cloud servers, servers of a distributed system, or servers incorporating blockchain.

[0101] It should be understood that the steps shown in the various forms above can be reordered, added to, or deleted from. For example, the steps recited in the present disclosure can be performed in parallel, in series, or in a different order, as long as the desired results of the technology disclosed in the present disclosure are achieved, which is not limited herein.

[0102] The specific embodiments described above are not intended to limit the scope of the present application. Those skilled in the art will understand that various modifications, combinations, sub-combinations, and alternatives can be made to the specific embodiments without departing from the principles of the present application. Any such modifications, equivalents, and alternatives are intended to fall within the scope of the present application.

Claims

1. A distributed renewable energy power trading method based on output characteristics, characterized in that, include: According to each time scale, the numerical weather forecast data of the distributed new energy area in the first time period is divided to obtain the numerical weather forecast data fragments of each time scale. Using dynamic mask matrices corresponding to each time scale, the frequency domain corresponding to the numerical weather forecast slices of each time scale is masked to obtain the numerical weather forecast frequency domain mask slices of each time scale. The numerical weather forecast frequency domain mask slices of each time scale and the text vectors of the text weather forecast data corresponding to the numerical weather forecast data are concatenated to obtain the multimodal semantic features of the weather forecast of the distributed new energy area in the first time period. An encoder and a decoder are used to encode and decode the multimodal semantic features of the weather forecast in sequence to obtain the deep semantic features of the weather forecast for the distributed new energy area in the first time period. A large language model is used to process the deep semantic features of the weather forecast for the distributed new energy region during the first time period, as well as the temporal features corresponding to the numerical weather forecast data, to obtain the predicted new energy output of the distributed new energy region during the first time period, so as to plan the green electricity trading plan for the distributed new energy region.

2. The method according to claim 1, characterized in that, The step of using a large language model to process the deep semantic features of the weather forecast of the distributed new energy area in the first time period and the temporal features corresponding to the numerical weather forecast data to obtain the predicted new energy output of the distributed new energy area in the first time period includes: using convolutional neural networks corresponding to each of the time scales to convolve the numerical weather forecast data slices of each of the time scales to obtain the temporal features of the numerical weather forecast data slices of each of the time scales. The numerical weather forecast data at each of the aforementioned time scales are spliced ​​together to obtain the numerical weather forecast time series characteristics of the distributed new energy region within the first time period. The deep semantic features of weather forecasts and the time series features of numerical weather forecasts for the distributed new energy area within the first time period are fused to obtain the target features of weather forecasts for the distributed new energy area within the first time period. A large language model is used to predict the weather forecast target characteristics of the distributed new energy area during the first time period, thereby obtaining the predicted new energy output of the distributed new energy area during the first time period.

3. The method according to claim 2, characterized in that, Also includes: The BERT model is used to extract text features from the text weather forecast data corresponding to the numerical weather forecast data, and the text vector of the text weather forecast data corresponding to the numerical weather forecast data is obtained. Based on the difference between each numerical weather forecast frequency domain mask slice at each time scale and the corresponding text vector, the mask loss corresponding to each time scale is obtained. Based on the mask loss corresponding to each time scale, the dynamic mask matrix corresponding to each time scale is updated respectively.

4. The method according to claim 3, characterized in that, Also includes: Based on the difference between the predicted renewable energy output and the actual renewable energy output of the distributed renewable energy region in the first time period, the time series prediction loss is determined. Based on the difference between the deep semantic features and the time-series features of the weather forecast in the distributed new energy area during the first time period, the semantic feature alignment loss is determined. The target loss is determined based on the time series prediction loss and the semantic feature alignment loss. Based on the target loss, the convolutional neural network, the dynamic mask matrix, the BERT model, the encoder, the decoder, and the large language model are updated.

5. The method according to claim 4, characterized in that, Also includes: Based on the updated convolutional neural network, the updated dynamic mask matrix, the updated BERT model, the updated encoder, the updated decoder, and the updated large language model, the local model parameters of the distributed new energy region are determined. The local model parameters of the distributed new energy regions are uploaded to the global server so that the global server can aggregate the local model parameters of each distributed new energy region to obtain the global model parameters. Upon receiving the global model parameters from the global server, the convolutional neural network, the dynamic mask matrix, the BERT model, the encoder, the decoder, and the large language model are updated a second time based on the global model parameters.

6. The method according to any one of claims 1-5, characterized in that, Also includes: The predicted renewable energy output of the distributed renewable energy regions during the first time period is sent to the global server so that the global server can plan and adjust the predicted renewable energy output of each of the distributed renewable energy regions during the first time period. Upon receiving the planned and adjusted predicted renewable energy output from the global server, the green electricity trading plan for the distributed renewable energy region within the first time period is determined based on the planned and adjusted predicted renewable energy output.

7. A distributed new energy power trading device based on output characteristics, characterized in that, include: The data partitioning module is used to partition the numerical weather forecast data of the distributed new energy area in the first time period according to various time scales, so as to obtain the numerical weather forecast data fragments of each time scale. The data masking module is used to perform masking processing on the frequency domain corresponding to the numerical weather forecast slices of each time scale using dynamic masking matrices corresponding to each time scale, so as to obtain the numerical weather forecast frequency domain mask slices of each time scale. The numerical text splicing module is used to splice the frequency domain mask slices of the numerical weather forecast at each time scale and the text vectors of the text weather forecast data corresponding to the numerical weather forecast data to obtain the multimodal semantic features of the weather forecast of the distributed new energy area in the first time period. The encoding and decoding module is used to encode and decode the multimodal semantic features of the weather forecast sequentially using an encoder and a decoder to obtain the deep semantic features of the weather forecast for the distributed new energy area in the first time period. The new energy output prediction module is used to process the deep semantic features of the weather forecast of the distributed new energy area in the first time period and the temporal features corresponding to the numerical weather forecast data using a large language model, so as to obtain the predicted new energy output of the distributed new energy area in the first time period, in order to plan the green electricity trading plan of the distributed new energy area.

8. The apparatus according to claim 7, characterized in that, The new energy output prediction module includes: The convolutional unit is used to perform convolution on the numerical weather forecast data slices of each time scale using the convolutional neural network corresponding to each time scale, so as to obtain the time series features of the numerical weather forecast data slices of each time scale; the time series feature splicing unit is used to splice the time series features of the numerical weather forecast data slices of each time scale to obtain the numerical weather forecast time series features of the distributed new energy area in the first time period. The feature fusion unit is used to fuse the deep semantic features of weather forecast and the time series features of numerical weather forecast of the distributed new energy area in the first time period to obtain the weather forecast target features of the distributed new energy area in the first time period. The power output prediction unit is used to use a large language model to predict the weather forecast target characteristics of the distributed new energy area in the first time period, so as to obtain the predicted new energy power output of the distributed new energy area in the first time period.

9. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to said at least one processor; The memory stores instructions that can be executed by the processor, which retrieves the instructions from the memory and executes them to enable the processor to execute the distributed new energy power trading method based on output characteristics as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are provided to the computer to instruct the computer to execute the distributed new energy power trading method based on output characteristics according to any one of claims 1-7.

Images