Electric power transaction strategy generation method and system based on Monte Carlo simulation and deep learning

Abstract

The invention relates to the technical field of electric power, provides an electric power transaction strategy generation method and system based on Monte Carlo simulation and deep learning, and solves the problem of poor scientificity and adaptability of an optimal transaction strategy of a multi-region electric power market in the prior art. The method comprises the following steps: acquiring historical price data of a multi-region power market, decomposing and extracting trends, periods and random components, and constructing a multi-dimensional feature set; the method comprises the following steps: receiving unstructured data by using a block link, packaging the unstructured data into a structured data packet through Hash verification and a smart contract, and converting the structured data packet into a multi-dimensional semantic feature vector based on a preset mapping table; associating the two types of features through a deep learning multi-modal model, and outputting price trend prediction probability distribution data with confidence; and generating a plurality of contract power splitting schemes based on Monte Carlo simulation, combining price prediction, searching a global optimal solution by using a particle swarm optimization algorithm, and generating an optimal power transaction strategy. According to the invention, the scientificity and adaptability of the optimal transaction strategy of the multi-region power market are improved.

Description

Technical Field

[0001] The present application relates to the field of power technology based on Monte Carlo simulation and deep learning, and in particular to a method and system for generating power trading strategies based on Monte Carlo simulation and deep learning. Background Art

[0002] In multi-regional electricity markets, electricity trading must comprehensively consider price fluctuations, supply and demand changes, and cross-regional allocation restrictions across regions. Trading strategies must be developed based on the exploration of historical price patterns, accurate predictions of future price trends, and the optimization of contract volume splits across multiple scenarios. With the increasing number of participants in the electricity market and the diversification of trading products, traditional strategy generation methods that rely on a single data type are unable to cope with complex market dynamics. A technical solution is urgently needed that can integrate structured historical data with unstructured real-time information, combining predictive capabilities with multi-scheme optimization capabilities, to generate optimal trading strategies that adapt to different market scenarios and meet the real-time, accuracy, and risk controllability requirements of electricity trading.

[0003] Currently, a power trading strategy generation solution that combines time series analysis with heuristic optimization exists to address these needs. This solution uses a time series model to analyze trends and cycles in historical price data to generate price forecasts. Based on these forecasts, a simple random search method is used to generate a small number of contract power splitting schemes. Pre-defined evaluation metrics are used to select the optimal schemes, ultimately forming a trading strategy.

[0004] This existing solution has obvious flaws: on the one hand, it relies solely on a single time series model to process historical price data, making it difficult to integrate unstructured information, resulting in insufficient response of price forecast results to sudden market factors and limited forecast accuracy; on the other hand, it uses a simple random search to generate a contract electricity splitting scheme, with a limited search scope and a lack of quantitative consideration of multi-regional market uncertainties. It is difficult to find the global optimal solution in a complex market environment, and the generated trading strategy has poor adaptability. Summary of the Invention

[0005] This application provides a method and system for generating electricity trading strategies based on Monte Carlo simulation and deep learning, which is used to solve the problems of poor scientificity and adaptability of optimal trading strategies in multi-regional electricity markets caused by insufficient prediction response, limited search range, and difficulty in finding global optimality in the existing technology.

[0006] In a first aspect, the present application provides a method for generating power trading strategies based on Monte Carlo simulation and deep learning, comprising: Acquiring historical price data of multi-regional electricity markets, decomposing the historical price data to extract a trend component, a periodic component, and a random component, and constructing a multidimensional feature set based on the trend component, the periodic component, and the random component; Utilizing a blockchain network to receive unstructured data from the multi-regional electricity market, performing hash verification and smart contract packaging on the unstructured data to obtain a structured data packet, and mapping the structured data packet into a multi-dimensional semantic feature vector based on a preset text vector mapping table; Utilizing a multimodal model based on deep learning, the multidimensional feature data in the multidimensional feature set is associated with the multidimensional semantic feature vector, and price trend prediction probability distribution data associated with the confidence level is output; Based on Monte Carlo simulation, multiple contract electricity splitting schemes are generated for the multi-regional electricity market. Based on the price trend prediction probability distribution data, a particle swarm optimization algorithm is used to search for the multiple contract electricity splitting schemes to obtain a global optimal solution. Based on the global optimal solution, an optimal electricity trading strategy is generated.

[0007] Optionally, the method of using a multimodal model based on deep learning to associate the multidimensional feature data in the multidimensional feature set with the multidimensional semantic feature vector and outputting price trend prediction probability distribution data associated with the confidence level includes: Inputting the multidimensional feature data in the multidimensional feature set into a first feature processing unit in a multimodal model to obtain a first feature sequence; Inputting the multi-dimensional semantic feature vector into a second feature processing unit in the multimodal model to obtain a second feature sequence; Combining the first feature sequence and the second feature sequence through a feature association unit in a multimodal model to form a fused feature sequence; Inputting the fused feature sequence into a multi-layer processing unit in a multimodal model, and obtaining an intermediate feature sequence after multiple feature conversions; Based on the generation unit in the multimodal model, the intermediate feature sequence is processed to generate initial price trend prediction probability distribution data; A confidence level corresponding to the price trend prediction probability distribution data is generated through a feature statistics unit in a multimodal model, and the price trend prediction probability distribution data is associated with the confidence level.

[0008] Optionally, generating a confidence level corresponding to the price trend prediction probability distribution data includes: Dividing the price trend prediction probability distribution data into multiple price value intervals, and calculating the corresponding proportion of each price value interval; Calculate the difference between the proportions based on the preset measurement rules; Based on the difference degree values, confidence parameters for each corresponding price value interval are generated according to a preset mapping rule; All the confidence parameters are integrated according to a preset integration rule to obtain a confidence level corresponding to the price trend prediction probability distribution data.

[0009] Optionally, performing hash verification on the unstructured data and packaging it with a smart contract to obtain a structured data packet, and mapping the structured data packet into a multi-dimensional semantic feature vector based on a preset text vector mapping table, includes: Processing the unstructured data according to preset encoding rules to generate a corresponding unique data code, and appending the data code to the unstructured data to complete hash verification; Based on preset grouping rules, the unstructured data that has completed hash verification is classified and packaged through smart contracts to form structured data packets; Based on a preset text vector mapping table, the content of each field in the structured data packet is converted into a corresponding vector value, and all the vector values ​​are integrated according to a preset arrangement rule to obtain a multi-dimensional semantic feature vector.

[0010] Optionally, based on a preset text vector mapping table, the content of each field in the structured data packet is converted into a corresponding vector value, and all the vector values ​​are integrated according to a preset arrangement rule to obtain a multi-dimensional semantic feature vector, including: According to the field type, each field in the structured data packet is divided into a text field, a numerical field and a classification field; Splitting the field content of the text field into multiple word units according to a preset word segmentation rule, and converting each word unit into a corresponding basic vector value according to the correspondence between the word units and vector values ​​in the preset text vector mapping table; Mapping the field content of the numerical field to a corresponding first-dimensional vector value based on an interval mapping rule in a preset text vector mapping table; Based on the category coding rules in the preset text vector mapping table, converting the field content of the classification field into the corresponding second dimension vector value; All the basic vector values, the first dimensional vector values ​​and the second dimensional vector values ​​are integrated according to the field order and dimension arrangement rule in the preset arrangement rule to obtain a multi-dimensional semantic feature vector.

[0011] Optionally, the step of searching the plurality of contract electricity splitting schemes based on the price trend prediction probability distribution data using a particle swarm optimization algorithm to obtain a global optimal solution includes: Matching the regional electricity data in each of the contract electricity splitting schemes with the price trend prediction probability distribution data of the corresponding region to obtain correlation data, and calculating the correlation data according to a preset evaluation rule to obtain a first evaluation parameter for each of the contract electricity splitting schemes; Select all contract electricity splitting schemes whose first evaluation parameter meets the preset range as the search set; A particle swarm optimization algorithm is used to search based on the search set to obtain a global optimal solution.

[0012] Optionally, the using a particle swarm optimization algorithm to search based on the search set to obtain a global optimal solution includes: Map each of the contract electricity splitting schemes in the search set to a particle in the particle swarm, wherein the position vector of the particle represents the allocation ratio of the regional electricity data, and the velocity vector of the particle represents the adjustment direction and amplitude of the allocation ratio; Adjusting the allocation ratio of the regional electricity data according to a preset adjustment rule, and obtaining a new contract electricity splitting plan based on the adjusted allocation ratio of the regional electricity data; Calculating a second evaluation parameter of the new contract power splitting scheme, comparing the second evaluation parameter with the first evaluation parameter, and retaining the contract power splitting scheme with the better evaluation parameter as the update result; Repeat the adjustment, calculation and retention steps, and retain the updated results after each execution; From all the update results, the contract electricity splitting scheme with the best evaluation parameters is selected as the global optimal solution.

[0013] In a second aspect, the present application provides a power trading strategy generation system based on Monte Carlo simulation and deep learning, comprising: an acquisition module, configured to acquire historical price data of multi-regional electricity markets, decompose the historical price data to extract a trend component, a periodic component, and a random component, and construct a multidimensional feature set based on the trend component, the periodic component, and the random component; a mapping module for receiving unstructured data from the multi-regional power market using a blockchain network, performing hash verification and smart contract packaging on the unstructured data to obtain a structured data packet, and mapping the structured data packet into a multi-dimensional semantic feature vector based on a preset text vector mapping table; an association module, configured to associate the multidimensional feature data in the multidimensional feature set with the multidimensional semantic feature vector using a multimodal model based on deep learning, and output price trend prediction probability distribution data associated with the confidence level; A generation module is used to generate multiple contract electricity splitting schemes for the multi-regional electricity market based on Monte Carlo simulation, and based on the price trend prediction probability distribution data, use a particle swarm optimization algorithm to search for the multiple contract electricity splitting schemes to obtain a global optimal solution, and generate an optimal electricity trading strategy based on the global optimal solution.

[0014] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for generating electricity trading strategies based on Monte Carlo simulation and deep learning as described in any one of the first aspects.

[0015] In a fourth aspect, the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a method for generating electricity trading strategies based on Monte Carlo simulation and deep learning as described in any one of the first aspects.

[0016] In the present application, a method for generating an electricity trading strategy based on Monte Carlo simulation and deep learning is provided, which includes: obtaining historical price data of multi-regional electricity markets, decomposing the historical price data to extract trend components, periodic components and random components, and constructing a multidimensional feature set based on the trend components, periodic components and random components; using a blockchain network to receive unstructured data from multi-regional electricity markets, performing hash verification and smart contract packaging on the unstructured data to obtain a structured data packet, and mapping the structured data packet into a multidimensional semantic feature vector based on a preset text vector mapping table; using a multimodal model based on deep learning to associate the multidimensional feature data in the multidimensional feature set with the multidimensional semantic feature vector, and outputting price trend prediction probability distribution data associated with the confidence level; based on Monte Carlo simulation, generating multiple contract electricity splitting schemes for the multi-regional electricity market, based on the price trend prediction probability distribution data, using a particle swarm optimization algorithm to search for the multiple contract electricity splitting schemes to obtain a global optimal solution, and generating an optimal electricity trading strategy based on the global optimal solution.

[0017] This application has the following advantages: By obtaining historical price data from multi-regional electricity markets, decomposing and extracting trend components, cyclical components and random components and constructing a multi-dimensional feature set, it is possible to comprehensively capture the different characteristic patterns in historical price data and provide a structured feature basis for subsequent predictions. By using the blockchain network to receive unstructured data, obtaining structured data packets through hash verification and smart contract packaging and mapping them into multi-dimensional semantic feature vectors, it is possible to ensure the reliability of unstructured data and convert them into processable feature vectors to enrich the feature dimensions. By using a multimodal model based on deep learning to associate and process multi-dimensional feature data with multi-dimensional semantic feature vectors, and outputting price trend prediction probability distribution data associated with confidence, it is possible to integrate structured and unstructured features and improve the comprehensiveness and reliability of price trend predictions. By generating multiple contract electricity splitting schemes based on Monte Carlo simulation, and combining the predicted data with the particle swarm optimization algorithm to search for the global optimal solution and generate the optimal electricity trading strategy, it is possible to cover a variety of market scenarios and improve the optimality and adaptability of the trading strategy.

[0018] Furthermore, the multimodal model processes multidimensional feature data and multidimensional semantic feature vectors through the first and second feature processing units, respectively, to obtain a feature sequence. After combined fusion and multi-layer conversion, it generates initial price trend prediction probability distribution data. The feature statistics unit then generates corresponding confidence levels and associates them. When generating confidence levels, the price ranges are divided and the proportions are calculated. Based on the degree of difference in the proportions, confidence parameters are generated and integrated to obtain the confidence level. By refining the feature processing and confidence generation process of the multimodal model, different types of features can be more accurately integrated, improving the accuracy and credibility of the price trend prediction probability distribution data, and providing a more reliable basis for subsequent optimization strategies.

[0019] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0021] Figure 1 A flowchart of a method for generating a power trading strategy based on Monte Carlo simulation and deep learning provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a power trading strategy generation system based on Monte Carlo simulation and deep learning provided in an embodiment of the present application; Figure 3A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable people skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0023] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 11, 12, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0024] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0025] To address the problems of insufficient prediction response, limited search range, and difficulty in finding the global optimum in the existing technology, the present invention provides a method for generating power trading strategies based on Monte Carlo simulation and deep learning. The method adopts the following concept: first, historical price data of power markets in different regions are collected, and features such as trends, cycles, and random changes are extracted from them to form a feature set containing multiple information; at the same time, news, announcements, and other information without a fixed format are received through a dedicated network, and after processing, they are converted into analyzable feature vectors; then, a model that can process multiple features is used to combine the two features obtained above for analysis to obtain a price trend prediction result with a degree of reliability; finally, a variety of different regional power allocation plans are simulated and generated, and then, combined with the price trend prediction results, a dedicated method is used to find the most suitable one among these plans, thereby generating the optimal power trading strategy. This method can utilize both fixed and unfixed format information, making price predictions more accurate and responding to market emergencies faster; at the same time, it can take into account more possible situations and fully cope with the uncertainty of different regional markets, thereby finding the most suitable power allocation plan. The generated trading strategy can also better adapt to complex market environments, solving the problems of existing methods.

[0026] Figure 1 A flow chart of a method for generating power trading strategies based on Monte Carlo simulation and deep learning is provided in an embodiment of the present application, such as Figure 1 As shown, the method includes: S11. Obtain historical price data of multi-regional electricity markets, decompose the historical price data to extract trend components, periodic components, and random components, and construct a multidimensional feature set based on the trend components, periodic components, and random components.

[0027] Among them, historical price data is the record of electricity transaction prices in multi-regional electricity markets over a period of time in the past. The trend component is the overall change direction of the price over a long period of time. The cyclical component is the fluctuation pattern of the price that repeats within a fixed time period. The random component is the short-term irregular price changes caused by accidental factors. The multidimensional feature set is a set containing multiple price features formed by integrating the trend component, cyclical component and random component. This set is the generation result of S11.

[0028] In an embodiment of the present application, past price records of multi-regional electricity markets, i.e., historical price data, are first collected. For example, the daily electricity transaction prices of regions A, B, and C in the past year are collected. Then, these historical price data are processed using a data decomposition method to extract trend components, periodic components, and random components from the data. For example, the overall price in region A has shown a slow upward trend in the past year, the price in region B has rebounded slightly every 30 days or so, and the price in region C has suddenly risen due to sudden weather on certain days. Finally, the three components extracted from each region are classified and sorted by region, and combined to form a multidimensional feature set containing various price features of the three regions.

[0029] S12. Receive unstructured data from multi-regional electricity markets using a blockchain network, perform hash verification and smart contract packaging on the unstructured data to obtain a structured data packet. Map the structured data packet into a multi-dimensional semantic feature vector based on a preset text vector mapping table.

[0030] Among them, the blockchain network is a network system that can securely receive and store data. Unstructured data is information without a fixed format, such as news about the electricity market, policy notices and market announcements. Hash verification is a method to check whether the data has been modified. Smart contract packaging is to organize unstructured data into data packets with a fixed format according to preset rules. Structured data packets are data sets with a unified format after organization. The preset text vector mapping table is a corresponding rule table set in advance to convert text information into digital vectors. The multi-dimensional semantic feature vector is a digital sequence containing text meaning obtained after converting the structured data packet according to the mapping table. This vector is the generation result of S12.

[0031] In an embodiment of the present application, unstructured data of a multi-regional electricity market is first received through a blockchain network, such as news about the addition of new power generation equipment in region A and notifications of policy adjustments in region B. A hash check is then performed on the received unstructured data to confirm that the data has not been modified. The reliable unstructured data is then packaged through a smart contract to organize it into a structured data packet with a unified format, such as "region + content" in the format of "region A: new power generation equipment; region B: policy adjustments." Finally, with reference to a preset text vector mapping table, the text information in the structured data packet is converted into a multi-dimensional semantic feature vector. For example, in the mapping table, "new power generation equipment" corresponds to the vector [0.2, 0.5, -0.1], and the data packet containing the content is converted into the corresponding vector.

[0032] S13. Utilize a multimodal model based on deep learning to associate the multidimensional feature data in the multidimensional feature set with the multidimensional semantic feature vector, and output price trend prediction probability distribution data associated with the confidence level.

[0033] Among them, the multimodal model based on deep learning is a computer model that can process multiple different types of data at the same time. The multidimensional feature data is the data contained in the multidimensional feature set constructed in S11. The multidimensional semantic feature vector is the vector obtained in S12. The confidence level is an assessment of the reliability of the prediction results. The price trend prediction probability distribution data contains information about the possible direction of price changes and corresponding possibilities in the future. After the data is associated with the confidence level, it becomes the generated result of S13.

[0034] In an embodiment of the present application, the multidimensional feature data obtained by S11 is first input into the part of the multimodal model that specifically processes this type of data to obtain a first type of feature sequence. At the same time, the multidimensional semantic feature vector obtained by S12 is input into the part of the model that specifically processes this type of vector to obtain a second type of feature sequence. For example, the multidimensional feature data of area A is processed to obtain the sequence [1.2, 0.8, 0.3], and its semantic feature vector is processed to obtain the sequence [0.5, 0.2, -0.1]. Then the model combines these two types of feature sequences to form a fused feature sequence, which is converted multiple times within the model to obtain an intermediate feature sequence. Finally, the model generates predicted probability distribution data for future price trends based on the intermediate feature sequence. For example, it predicts that the probability of future price increase in area A is 55%, unchanged at 30%, and decreased at 15%. At the same time, the confidence of this prediction result is evaluated, such as 80%, and the two are associated.

[0035] S14. Based on Monte Carlo simulation, generate multiple contract electricity splitting schemes for multi-regional electricity markets. Based on price trend prediction probability distribution data, use particle swarm optimization algorithm to search for multiple contract electricity splitting schemes to obtain the global optimal solution. Based on the global optimal solution, generate the optimal electricity trading strategy.

[0036] Among them, Monte Carlo simulation is a method that covers different scenarios by randomly generating multiple possible situations. The contract power splitting plan is a specific plan to distribute the total power contract power to different areas. The price trend prediction probability distribution data is the prediction information obtained in S13. The particle swarm optimization algorithm is a method to find the best solution through multiple adjustments. The global optimal solution is the best one among all possible solutions. The optimal power trading strategy is the power trading plan formulated based on the global optimal solution. This strategy is the generated result of S14.

[0037] In an embodiment of the present application, a Monte Carlo simulation is first used to generate multiple contract electricity splitting schemes, and the range of total contract electricity and regional allocation is set. For example, the total electricity is 2000 units, and the allocation amount for each region is between 200-800 units. 100 schemes are randomly generated, such as Scheme 1: Region A 500, Region B 800, and Region C 700. Then, combined with the price trend prediction probability distribution data obtained in S13, for example, the possibility of price increase in Region A is high, and these schemes are evaluated using a particle swarm optimization algorithm. By adjusting the regional allocation amount in the scheme multiple times, such as increasing the allocation for Region A and reducing the allocation for Region C, each adjusted scheme is evaluated and the better one is retained. This process is repeated until the global optimal solution is found, and finally the optimal electricity trading strategy is generated based on the global optimal solution.

[0038] This application provides the following specific examples: For the electricity markets in the three regions A, B, and C, the daily electricity transaction prices in each region for the past 12 months are first obtained. After processing, the monthly average price in region A increases by 0.5 yuan compared with the previous month (trend component), the price fluctuates slightly every 7 days (cyclical component), and the price rebounds every 30 days in region B (cyclical component), etc., which are integrated into a multidimensional feature set; unstructured data such as news about new power generation equipment in region A is received through the blockchain network, and after hash verification and packaging, it is converted into a multidimensional semantic feature vector according to the mapping table; the two types of features are input into the multimodal model, and correlation results such as the probability of price increase in region A is 60% (confidence 85%) and the probability of price increase in region B is 40% (confidence 78%) are obtained; Monte Carlo simulation is used to generate 200 splitting plans with a total electricity volume of 3,000 units. Combined with the predicted data, the global optimal solution of 1,200 units in region A, 1,000 units in region B, and 800 units in region C is found through adjustment and evaluation, and a trading strategy is generated accordingly.

[0039] By executing S11 to S14, the embodiment of the present application can comprehensively capture the historical price characteristics and unstructured information of multi-regional electricity markets, integrate multiple characteristics to predict price trends and associate them with reliability, and then find the optimal solution based on multiple possible electricity distribution scenarios. The ultimately generated electricity trading strategy can comprehensively consider market rules and real-time information, adapt to complex market changes, and improve the rationality and adaptability of transactions.

[0040] In one possible embodiment, S13 utilizes a multimodal model based on deep learning to associate the multidimensional feature data in the multidimensional feature set with the multidimensional semantic feature vector, and outputs price trend prediction probability distribution data associated with the confidence level, including: Step 131: Input the multidimensional feature data in the multidimensional feature set into the first feature processing unit in the multimodal model to obtain a first feature sequence.

[0041] Among them, the multidimensional feature set is a collection of price trends, cycles, and random change characteristics of multi-regional electricity markets, and the multidimensional feature data is the specific feature information in the collection; the multimodal model is a model that can process various types of data, the first feature processing unit is the part of the model that specifically processes multidimensional feature data, and the first feature sequence is an ordered feature combination formed after the multidimensional feature data is processed by this unit.

[0042] In an embodiment of the present application, the multidimensional feature data in the multidimensional feature set is input into the first feature processing unit of the multimodal model. The unit organizes and converts the data, and arranges the scattered feature information in a certain order to form a first feature sequence. For example, after the feature data corresponding to the trend, cycle, and random component of area A are input, the feature sequence [1.2, 0.8, 0.3, ...] arranged in chronological order is obtained through processing, where 1.2 corresponds to the trend feature value and 0.8 corresponds to the period feature value.

[0043] Step 132: Input the multi-dimensional semantic feature vector into the second feature processing unit in the multimodal model to obtain a second feature sequence.

[0044] Among them, the multidimensional semantic feature vector is a digital sequence containing textual meaning converted from unstructured data; the second feature processing unit is the part of the multimodal model that specifically processes this type of vector, and the second feature sequence is an ordered feature combination formed after the vector is processed.

[0045] In an embodiment of the present application, the multidimensional semantic feature vector is input into the second feature processing unit of the multimodal model. The unit organizes the vector and sorts it by the degree of semantic association to form a second feature sequence. For example, after the semantic vector [0.2, 0.5, -0.1] corresponding to the "new power generation equipment" in area A is input, the sequence [0.5, 0.2, -0.1, ...] arranged by semantic importance is obtained through processing.

[0046] Step 133: Combine the first feature sequence and the second feature sequence through the feature association unit in the multimodal model to form a fused feature sequence.

[0047] Among them, the first feature sequence is an ordered combination after processing of multidimensional feature data, and the second feature sequence is an ordered combination after processing of semantic feature vectors; the feature association unit is the part of the multimodal model that combines these two sequences, and the fused feature sequence is a comprehensive feature sequence formed by combining the two.

[0048] In an embodiment of the present application, the feature association unit receives the first feature sequence and the second feature sequence, and merges them according to a preset rule to form a fused feature sequence. For example, the first feature sequence [0.5, 0.11, 1.5] and the second feature sequence [0.44, 0.34] of region A are combined alternately by position to obtain [0.5, 0.44, 0.11, 0.34, 1.5, ...].

[0049] Step 134: Input the fused feature sequence into the multi-layer processing unit in the multimodal model, and after multiple feature conversions, obtain an intermediate feature sequence.

[0050] Among them, the fused feature sequence is the comprehensive sequence of the first and second feature sequences; the multi-layer processing unit is the part that performs multiple feature conversions in the multimodal model and can extract key features. The intermediate feature sequence is the core feature sequence obtained after its processing.

[0051] In an embodiment of the present application, the fused feature sequence is input into a multi-layer processing unit, which gradually extracts key features through multiple screening and reorganization to form an intermediate feature sequence. For example, the fused sequence retains important elements in the first layer, and is adjusted according to the influence weight in the second layer, and finally obtains [0.4, 0.26, 0.75].

[0052] Step 135: Based on the generation unit in the multimodal model, the intermediate feature sequence is processed to generate initial price trend prediction probability distribution data.

[0053] Among them, the intermediate feature sequence is the core feature sequence refined by multi-layer processing units; the generation unit is the part of the multimodal model that generates prediction results, and the initial price trend prediction probability distribution data is the information generated by it that contains the possibility of future price changes.

[0054] In an embodiment of the present application, the intermediate feature sequence is input into a generation unit, which analyzes the possible future price change directions (increase, remain unchanged, decrease), calculates the probability of each direction to form prediction data, for example, based on the intermediate feature sequence of area A, generates a prediction result of "increase by 60%, remain unchanged by 25%, decrease by 15%".

[0055] Step 136: Generate a confidence level corresponding to the price trend prediction probability distribution data through the feature statistical unit in the multimodal model, and associate the price trend prediction probability distribution data with the confidence level.

[0056] Among them, the price trend prediction probability distribution data contains information about the possibility of future price changes; the feature statistical unit is the part of the multimodal model that evaluates the reliability of the prediction, the confidence level is an assessment of the reliability of the prediction, and the association is the process of combining the prediction data with the confidence level.

[0057] In an embodiment of the present application, the feature statistics unit analyzes the distribution of each possibility in the price trend prediction probability distribution data, evaluates the confidence level, and then combines the two. For example, the prediction data for area A is "increase by 60%, remain unchanged by 25%, and decrease by 15%." The statistics unit evaluates its confidence level to be 85% and associates the two.

[0058] This application provides the following specific example: for area A, first input its multi-dimensional feature data (trend component price increase of 0.5 yuan per month, cycle component fluctuation every 7 days, random component abnormal increase for 5 days) into the first feature processing unit, and convert it to generate the first feature sequence [0.5, 0.11, 1.5]; at the same time, convert the unstructured data of "new power generation equipment" into a multi-dimensional semantic feature vector, and input it into the second feature processing unit to generate the second feature sequence [0.44, 0.34]; the feature association unit alternately combines the two sequences into The feature sequence is fused as follows: [0.5, 0.44, 0.11, 0.34, 1.5]; the multi-layer processing unit retains the first three important elements and calculates the intermediate feature sequence [0.4, 0.26, 0.75] according to weights of 0.8, 0.6, and 0.5; the generation unit combines the weights of each feature to obtain the initial price trend prediction probability distribution data of "increase 60%, unchanged 25%, decrease 15%"; the feature statistics unit calculates the difference between each probability and the average value, converts it to a confidence level of 85%, and associates the predicted data with the confidence level.

[0059] By executing steps 131 to 136, the embodiment of the present application realizes the fusion of structured historical information and unstructured real-time information by step-by-step processing of multi-dimensional feature data and semantic feature vectors. The price trend forecast data generated after multi-layer refinement combines multiple influencing factors, and the reliability of the forecast is clarified through confidence assessment, which provides a comprehensive and reliable quantitative basis for the formulation of subsequent power trading strategies and improves the accuracy and practicality of forecast analysis.

[0060] In one possible embodiment, step 136, generating a confidence level corresponding to the price trend prediction probability distribution data, includes: a1. Divide the price trend prediction probability distribution data into multiple price value intervals and calculate the corresponding proportion of each price value interval.

[0061] Among them, the price trend prediction probability distribution data is information about possible future price changes and corresponding possibilities. The price value interval is a price range divided according to a certain range. The corresponding proportion of each price value interval is the proportion of the price possibility contained in each interval in the total possibility. These proportions are the generated results of this step.

[0062] In an embodiment of the present application, the price trend prediction probability distribution data is first divided into multiple price value intervals according to a preset price range, and then the sum of all price possibilities in each interval is counted. Finally, the sum is divided by the total possibility to obtain the proportion of each interval. For example, the prediction data for area A includes a 40% probability that the price is between 100-110 yuan, 30% between 110-120 yuan, and 30% between 120-130 yuan. After dividing these three intervals, the total possibility is 40%+30%+30%=100%, and the proportion of each interval is 40%÷100%=40%, 30%÷100%=30%, and 30%÷100%=30%.

[0063] a2. Based on the preset measurement rules, calculate the degree of difference between each proportion.

[0064] Among them, the preset measurement rule is a method set in advance for calculating the differences between various proportions. The difference degree value is a numerical value calculated by the rule to reflect the size of the difference between various proportions. It is the generated result of this step.

[0065] In an embodiment of the present application, the proportion of each price value interval is calculated according to the preset measurement rules to obtain the degree of difference value. For example, the preset rule is to calculate the maximum difference between the proportions. The proportions of area A are 40%, 30%, and 30%. The difference between the maximum proportion of 40% and the minimum proportion of 30% is 40%-30%=10%. This 10% is the degree of difference value.

[0066] a3. Based on the difference degree values, confidence parameters for each corresponding price value range are generated according to the preset mapping rules.

[0067] Among them, the difference degree value is the quantitative result of the difference between each proportion, and the preset mapping rule is a correspondence set in advance to convert the difference degree value into a confidence parameter. The confidence parameter is a parameter that reflects the reliability of the prediction of each price value range, and it is the generated result of this step.

[0068] In an embodiment of the present application, according to the preset mapping rules, the difference degree value is converted into the confidence parameter corresponding to each price value interval. For example, the preset rule stipulates that when the difference degree value is between 5%-15%, the corresponding confidence parameter is 0.85; the difference degree value of area A is 10%, which is within this range, so the confidence parameter of each interval is 0.85.

[0069] a4. All confidence parameters are integrated according to the preset integration rules to obtain the confidence level corresponding to the price trend prediction probability distribution data.

[0070] Among them, the confidence parameter is a quantitative value reflecting the reliability of the prediction of each price range. The preset integration rule is a method set in advance to combine all confidence parameters into an overall indicator. The confidence level is an indicator obtained through this rule that reflects the overall reliability of the price trend prediction probability distribution data. It is the generated result of this step.

[0071] In an embodiment of the present application, the confidence parameters of all price value intervals are combined and calculated according to the preset integration rules to obtain the confidence level. For example, the preset rule is to calculate the average value of all confidence parameters; the confidence parameter of each interval in area A is 0.85, there are 3 intervals in total, and the average value is (0.85+0.85+0.85) ÷ 3 = 0.85, and this 0.85 is the confidence level.

[0072] This application provides the following specific example: for the price trend prediction probability distribution data of area A, first set the price value intervals to 100-110 yuan, 110-120 yuan, and 120-130 yuan, and calculate the probability of each interval as 40%, 30%, and 30% respectively. The total probability is calculated as 40%+30%+30%=100%, and then divide the probability of each interval by the total probability to obtain the proportion of each interval as 40%, 30%, and 30%; then according to the preset measurement rules (calculate all proportions and average The sum of the absolute values ​​of the differences in the proportions is obtained), the average proportion is (40+30+30)÷3≈33.3%, the absolute values ​​of the differences are 6.7%, 3.3%, and 3.3%, respectively, and the total is 13.3%, that is, the degree of difference is 13.3%; according to the preset mapping rule (10%-20% corresponds to 0.85), the confidence parameter of each interval is 0.85; finally, according to the preset integration rule (taking the average value), (0.85+0.85+0.85)÷3=0.85 is calculated, and the confidence level is 0.85.

[0073] By executing steps a1-a4, this embodiment organizes the scattered forecast data by dividing the price ranges and calculating their percentages. By quantifying the differences in the percentages of each range, this provides a basis for evaluating the stability of the forecast. These differences are converted into confidence parameters to quantify the reliability of each range. Finally, the overall confidence level is integrated to fully reflect the reliability of the forecast results. These progressive steps clearly demonstrate the credibility of the price trend forecast data, facilitating the subsequent formulation of reasonable strategies based on the forecasts.

[0074] In one possible embodiment, S12 performs hash verification on the unstructured data and packages it with the smart contract to obtain a structured data packet. Based on a preset text vector mapping table, the structured data packet is mapped into a multi-dimensional semantic feature vector, including: Step 121: Process the unstructured data according to the preset encoding rules, generate a corresponding unique data code, and append the data code to the unstructured data to complete the hash check.

[0075] Among them, unstructured data is information without a fixed format, such as news and announcements in the electricity market; the preset coding rules are methods set in advance for generating unique identifiers; the unique data code is a code generated by the coding rules that can uniquely correspond to unstructured data; hash verification is the process of confirming whether the data has been modified by adding a unique data code to the unstructured data. The data that completes the hash verification is the generated result of this step.

[0076] In an embodiment of the present application, unstructured data is processed according to a preset coding rule to generate a unique data code, and then this code is appended to the original unstructured data to confirm that the data has not been modified and complete the hash check. For example, there is a news about the increase in power supply in area A. The preset coding rule is "data content keywords + release time (hours, minutes, seconds)". The keyword is extracted as "AQGYZJ" and the release time is 143025. The combination generates a unique data code "AQGYZJ143025", which is appended to the news to complete the hash check of the news.

[0077] Step 122: Based on the preset grouping rules, the unstructured data that has completed the hash check is classified and packaged through the smart contract to form a structured data packet.

[0078] Among them, the preset grouping rules are pre-set standards for classifying data, such as division by region or information type; smart contracts are preset programs that can automatically classify and package data according to grouping rules; unstructured data that has completed hash verification is data that has been confirmed in terms of integrity; classification and packaging is to group similar data together and organize it into a fixed format; structured data packets are data sets with a unified format after classification and containing similar information. It is the generated result of this step.

[0079] In an embodiment of the present application, based on a preset grouping rule, the unstructured data that has completed the hash check is classified through a smart contract, and the data belonging to the same category is packaged and organized into a structured data packet with a unified format. For example, the preset grouping rule is "classification by information type (policy category, supply category)", and the smart contract classifies the "A region electricity price adjustment policy" and "B region environmental protection policy notice" that have completed the hash check into the policy category, and packages them into a structured data packet containing "Type: Policy; Content 1: A region electricity price adjustment; Content 2: B region environmental protection policy".

[0080] Step 123: Based on a preset text vector mapping table, convert the content of each field in the structured data packet into a corresponding vector value, and integrate all the vector values ​​according to a preset arrangement rule to obtain a multi-dimensional semantic feature vector.

[0081] Among them, the preset text vector mapping table is a pre-set correspondence table between text content and digital vectors; the field content in the structured data packet is the specific text information in the data packet, such as "electricity price reduction"; the vector value is the number obtained by converting the field content through the mapping table; the preset arrangement rule is the pre-set order standard for integrating vector values; the multi-dimensional semantic feature vector is a digital sequence containing textual meaning formed by integrating all vector values ​​according to the arrangement rule, and it is the generation result of this step.

[0082] In an embodiment of the present application, according to a preset text vector mapping table, the text content of each field in the structured data packet is converted into a corresponding vector value, and then all the vector values ​​are integrated according to a preset arrangement rule to form a multi-dimensional semantic feature vector. For example, the field content of the structured data packet is "Region: A; Content: Electricity Price Reduction", and in the mapping table, "Region: A" corresponds to [0.3, 0.1], and "Electricity Price Reduction" corresponds to [0.2, 0.4]. According to the arrangement rule of "region first, content later", the integration yields [0.3, 0.1, 0.2, 0.4].

[0083] This application provides the following specific example: for the unstructured data of "Area C will undergo power grid maintenance next week", first process it according to step 121, and the preset coding rule is "region first letter + content core word first letter + date (month and day)", the region first letter is "C", the core word first letter is "DWJX", and the date is 0615, and a unique data code "CDWJX0615" is generated and attached to the original data to complete the hash check; then in step 122, the preset grouping rule is "classified by region", and the smart contract combines this data with another "power supply adjustment during maintenance in Area C" The hash check data of "whole" are classified into one category and packaged into a structured data packet of "Area: C; Content 1: Power grid maintenance next week; Content 2: Power supply adjustment during maintenance". Finally, in step 123, in the preset text vector mapping table, "Area: C" corresponds to [0.4, 0.2], "Power grid maintenance" corresponds to [0.1, 0.5], and "Power supply adjustment" corresponds to [0.3, 0.6]. According to the arrangement rule of "Area to Content 1 to Content 2", the vector [0.4, 0.2, 0.1, 0.5, 0.3, 0.6] is calculated, which is a multi-dimensional semantic feature vector.

[0084] By executing steps 121 through 123, in this embodiment, step 121 generates a unique data code to ensure that the unstructured data has not been modified, thus guaranteeing data reliability; step 122 classifies and packages the data according to rules to achieve a unified data format and clear classification, thus improving data organization; and step 123 converts the text information into an integrable vector, allowing the unstructured data to be further analyzed and utilized. These steps are interconnected, providing high-quality semantic feature data for subsequent feature fusion and strategy generation.

[0085] In one possible embodiment, step 123 converts the content of each field in the structured data packet into a corresponding vector value based on a preset text vector mapping table, and integrates all the vector values ​​according to a preset arrangement rule to obtain a multi-dimensional semantic feature vector, including: b1. According to the field type, the fields in the structured data packet are divided into text fields, numerical fields and classification fields.

[0086] Among them, the structured data packet is a data set with a unified format and containing similar information. The field type is a classification method for each information item in the data packet. The text field is an information item composed of text descriptions, the numerical field is an information item containing specific numbers, and the classification field is an information item belonging to a fixed category option. The result generated by this step is the divided text field, numerical field and classification field.

[0087] In an embodiment of the present application, each information item in a structured data packet is divided into a text field, a numerical field, and a classification field according to the type of the field. For example, in a structured data packet containing "Area: B; Description: Equipment maintenance will be carried out next month; Transaction volume: 800", "Description: Equipment maintenance will be carried out next month" belongs to the text field, "Transaction volume: 800" belongs to the numerical field, and "Area: B" belongs to the classification field.

[0088] b2. Split the field content of the text field into multiple tokens according to the preset word segmentation rules, and convert each token into a corresponding basic vector value through the correspondence between tokens and vector values ​​in the preset text vector mapping table.

[0089] Among them, the text field is an information item composed of text descriptions, the preset word segmentation rule is a method set in advance to split the text content into small units, the word unit is the smallest text unit obtained after the split, the preset text vector mapping table is the correspondence table between text units and digital vectors, the basic vector value is the digital vector obtained by converting the word unit through the mapping table, and the generated result of this step is all the basic vector values ​​after conversion.

[0090] In an embodiment of the present application, the content of a text-type field is split into multiple word-units according to a preset word segmentation rule, and each word-unit is converted into a corresponding basic vector value through the correspondence between word-units and vector values ​​in a preset text vector mapping table. For example, the text-type field "carry out equipment maintenance next month" is split into four word-units "next month", "carry out", "equipment", and "maintenance" according to the word segmentation rule. If "next month" corresponds to [0.2, 0.3], "equipment" corresponds to [0.4, 0.5], "maintenance" corresponds to [0.6, 0.7], and "carry out" corresponds to [0.8, 0.9] in the mapping table, the basic vector values ​​of these four word-units are obtained after conversion.

[0091] b3. Based on the interval mapping rules in the preset text vector mapping table, the field content of the numerical field is mapped to the corresponding first dimension vector value.

[0092] Among them, the numerical field is an information item containing specific numbers, the interval mapping rule in the preset text vector mapping table is the correspondence between the pre-set numerical range and the vector value, and the first-dimensional vector value is the digital vector obtained by converting the numerical field through the interval mapping rule. The vector value is the generation result of this step.

[0093] In an embodiment of the present application, based on the interval mapping rules in the preset text vector mapping table, the range of the numbers in the numerical field is first determined, and then the vector value corresponding to the range is found as the first dimension vector value. For example, for the numerical field "trading volume: 800", if "700-900" in the mapping table corresponds to [0.3, 0.4], 800 belongs to this range, then the first dimension vector value is [0.3, 0.4].

[0094] b4. Based on the category coding rules in the preset text vector mapping table, the field content of the classification field is converted into the corresponding second-dimensional vector value.

[0095] Among them, the classification field is an information item belonging to a fixed category option, the category coding rule in the preset text vector mapping table is the correspondence between the category and the vector value set in advance, and the second-dimensional vector value is the digital vector obtained by converting the classification field through the category coding rule. The vector value is the generation result of this step.

[0096] In an embodiment of the present application, based on the category coding rules in the preset text vector mapping table, the vector value corresponding to the category of the classification field is found as the second dimension vector value. For example, for the classification field "Region: B", if "Region B" in the mapping table corresponds to [0.1, 0.2], then the second dimension vector value is [0.1, 0.2].

[0097] b5. Integrate all basic vector values, first dimension vector values, and second dimension vector values ​​according to the field order and dimension arrangement rules in the preset arrangement rules to obtain a multi-dimensional semantic feature vector.

[0098] Among them, the basic vector value is the vector obtained after the text field is split, the first dimension vector value is the vector obtained by converting the numerical field, and the second dimension vector value is the vector obtained by converting the classification field. The preset arrangement rule is the vector integration order set in advance, including the field order and dimension arrangement method. The multi-dimensional semantic feature vector is a digital sequence formed by integrating all vectors according to the arrangement rule. This vector is the generation result of this step.

[0099] In an embodiment of the present application, all basic vector values, first dimension vector values, and second dimension vector values ​​are integrated according to the field order and dimension arrangement rules in the preset arrangement rules to form a multi-dimensional semantic feature vector. For example, the preset arrangement rule is "classification field → numerical field → text field", the classification field vector is [0.1, 0.2], the numerical field vector is [0.3, 0.4], and the basic vector values ​​of the text field are [0.2, 0.3], [0.8, 0.9], [0.4, 0.5], and [0.6, 0.7]. After integration, [0.1, 0.2, 0.3, 0.4, 0.2, 0.3, 0.8, 0.9, 0.4, 0.5, 0.6, 0.7] are obtained.

[0100] This application provides the following specific examples: a structured data packet is "Area: C; Note: Power supply increased to 700 this week; Type: Temporary Adjustment". First, in step b1, "Note: Power supply increased to 700 this week" is divided into a text field, "Power supply: 700" is divided into a numerical field, and "Area: C" and "Type: Temporary Adjustment" are divided into classification fields; in step b2, "Power supply increased to 700 this week" is split into "this week", "Power supply", "Increased to", and "700" according to the word segmentation rules. Assume that in the mapping table, "this week" corresponds to [0.1, 0.1], "Power supply" corresponds to [0.2, 0.2], "Increased to" corresponds to [0.3, 0.3], and "700" corresponds to [0.4, 0.4], and four basic vector values ​​are obtained; in step b3, the numerical field "700" is mapped according to the interval mapping rule (600-800 corresponds to [0.5, 0.5]), and the first dimension vector value [0.5, 0.5] is obtained; in step b4, the classification field "Region: C" corresponds to [0.6, 0.6], and "Type: Temporary Adjustment" corresponds to [0.7, 0.7], and two second dimension vector values ​​are obtained; in step b5, the order of "region to type to value to text" is integrated to obtain the multi-dimensional semantic feature vector [0.6, 0.6, 0.7, 0.7, 0.5, 0.5, 0.1, 0.1, 0.2, 0.2, 0.3, 0.3, 0.4, 0.4].

[0101] By executing steps b1 to b5, the embodiment of the present application processes different types of fields in the structured data packet separately, converts them into a unified vector form, and integrates them according to rules, so that text descriptions, numbers, and category information can all be presented in an analyzable digital sequence. This not only retains the original meaning of each type of information, but also allows different types of information to be processed uniformly, providing standardized basic data for subsequent comprehensive analysis.

[0102] In one possible embodiment, S14, based on the price trend prediction probability distribution data, a particle swarm optimization algorithm is used to search for multiple contract power splitting schemes to obtain a global optimal solution, including: Step 141: Match the regional electricity data in each contract electricity splitting plan with the price trend prediction probability distribution data of the corresponding region to obtain associated data, calculate the associated data according to the preset evaluation rules, and obtain the first evaluation parameter of each contract electricity splitting plan.

[0103] Among them, the contract electricity splitting plan is a specific plan to allocate the total electricity to different regions, the regional electricity data is the specific allocation amount for each region in the plan, the price trend prediction probability distribution data is information about the possibility of future price changes in each region, and the associated data is the matching result of the regional electricity data and the corresponding regional price forecast data. The preset evaluation rules are the standards set in advance for calculating the evaluation parameters. The first evaluation parameter is a value calculated by the rule to reflect the quality of the plan. This parameter is the generation result of this step.

[0104] In an embodiment of the present application, the electricity data of each region in each contract electricity splitting plan and the price trend prediction probability distribution data of the corresponding region are matched one by one to form associated data, and the associated data is calculated according to the preset evaluation rules to obtain the first evaluation parameter of each plan. For example, in a certain plan, 300 units of electricity are allocated to area A, and the probability of price increase in this area is 60%. 200 units of electricity are allocated to area B, and the probability of price increase is 40%. According to the evaluation rule "the sum of the electricity volume of each region × the probability of price increase", the first evaluation parameter of the plan is 300×60%+200×40%=180+80=260.

[0105] Step 142: Select all contract electricity splitting schemes whose first evaluation parameters fall within a preset range as a search set.

[0106] Among them, the first evaluation parameter is a numerical value reflecting the quality of the contract electricity splitting plan, the preset range is the qualified interval of the evaluation parameter set in advance, and the search set is a set of contract electricity splitting plans whose first evaluation parameter is within the preset range. This set is the generation result of this step.

[0107] In an embodiment of the present application, the first evaluation parameter of each contract electricity splitting plan is compared with a preset range, and plans with parameters within the range are selected to form a search set. For example, the preset range is "evaluation parameters between 200 and 300", and there are four plans with first evaluation parameters of 250, 180, 280, and 310, respectively. Among them, 250 and 280 are within the range, and these two plans constitute the search set.

[0108] Step 143: Use the particle swarm optimization algorithm to search based on the search set to obtain a global optimal solution.

[0109] Among them, the particle swarm optimization algorithm is a method of finding the best solution by simulating group collaboration. The search set is a set of contract power splitting solutions that meet the preset range. The global optimal solution is the best contract power splitting solution found in the search set. The optimal solution is the generated result of this step.

[0110] In an embodiment of the present application, a particle swarm optimization algorithm is used to search for contract electricity splitting schemes in a search set. By continuously adjusting the electricity distribution in each area (for example, increasing the electricity in areas with a high probability of price increases and reducing the electricity in areas with a low probability), the pros and cons of each adjusted scheme are evaluated, and the best scheme, i.e., the global optimal solution, is found after repeated optimization. For example, there are two schemes in the search set. Through adjustment, the algorithm increases the electricity in area A (with a high probability of price increases) from 300 to 350, and reduces the electricity in area B (with a lower probability) from 200 to 150. After evaluation, the adjusted scheme is determined to be the global optimal solution.

[0111] This application provides the following specific example: for the three regions A, B, and C, the total contracted electricity is 1000 units, and there are three contracted electricity splitting plans. In step 141, the allocation of plan 1 is 400 in region A, 300 in region B, and 300 in region C. The price increase probabilities of the corresponding regions are 70%, 50%, and 20%, respectively. According to the preset evaluation rule "the sum of the electricity volume of each region × the price increase probability", the first evaluation parameter is 400×70%+300×50%+300×20%=280+150+60=490; the parameter of plan 2 is 420, and the parameter of plan 3 is 350. In step 142, the preset range is "the evaluation parameter is between 380 and 500", and plans 1 and 2 meet the range and constitute the search set. In step 143, the particle swarm optimization algorithm is used to adjust the two solutions, increasing the power in area A and reducing the power in area C. Finally, a solution with 450 in area A, 300 in area B, and 250 in area C is obtained. The parameter is evaluated to be 510, which is determined to be the global optimal solution.

[0112] By executing steps 141 to 143, the embodiment of the present application calculates evaluation parameters by associating electricity distribution with price forecasts, screens out qualified solutions, and then finds the best solution through an optimization algorithm, so that the final contract electricity splitting solution can better match market price trends, thereby improving the rationality and adaptability of the solution.

[0113] In a possible embodiment, step 143, using a particle swarm optimization algorithm to search based on a search set to obtain a global optimal solution, includes: c1. Map each contract electricity splitting scheme in the search set to a particle in the particle swarm. The particle's position vector represents the distribution ratio of regional electricity data, and the particle's velocity vector represents the adjustment direction and amplitude of the distribution ratio.

[0114] The search set is a set of eligible contract electricity splitting schemes. The contract electricity splitting scheme is a specific plan for allocating total electricity to different regions. The particle swarm is a virtual group that simulates group collaboration to find the best solution. A particle is an individual in the group corresponding to a contract electricity splitting scheme. The position vector is a digital sequence representing the electricity allocation ratio of each region. The velocity vector is a digital sequence representing the direction and amplitude of the adjustment of these ratios. The result of this step is the mapped particle.

[0115] In an embodiment of the present application, each contract electricity splitting scheme in the search set is mapped to a particle in the particle swarm, the position vector is used to represent the electricity allocation ratio of each area, and the velocity vector is used to represent the adjustment direction and amplitude of these ratios. For example, there is a scheme in the search set that allocates 50% of electricity to area A, 30% to area B, and 20% to area C. It is mapped to a particle with a position vector of [0.5, 0.3, 0.2] and a velocity vector of [0.03, -0.01, -0.02], indicating that the proportion of area A may increase by 3%, area B may decrease by 1%, and area C may decrease by 2%.

[0116] c2. Adjust the allocation ratio of regional electricity data according to the preset adjustment rules, and obtain a new contract electricity splitting plan based on the adjusted allocation ratio of regional electricity data.

[0117] Among them, the preset adjustment rule is a method set in advance to adjust the distribution ratio according to the speed vector. The distribution ratio of regional electricity data is the proportion of each region in the total electricity. The new contract electricity splitting plan is the distribution plan formed after adjusting the ratio. This plan is the generated result of this step.

[0118] In an embodiment of the present application, according to a preset adjustment rule, the power distribution ratio of each area is adjusted in combination with the velocity vector of the particle, and then the specific distribution amount of each area is calculated based on the total power to form a new contract power splitting plan. For example, the position vector of a particle is [0.5, 0.3, 0.2], and the velocity vector is [0.03, -0.01, -0.02]. According to the rule of "position vector acceleration vector", the adjusted ratio is [0.53, 0.29, 0.18]. When the total power is 1000 units, the new plan is 530 units in area A, 290 units in area B, and 180 units in area C.

[0119] c3. Calculate the second evaluation parameter of the new contract power splitting plan, compare the second evaluation parameter with the first evaluation parameter, and retain the contract power splitting plan with the better evaluation parameter as the update result.

[0120] Among them, the new contract electricity splitting plan is the allocation plan formed after adjustment, the second evaluation parameter is a value reflecting the pros and cons of the new plan, the first evaluation parameter is the pros and cons value of the original plan, and the better evaluation parameter refers to a value that is more in line with the preset pros and cons standards. The updated result is the retained better plan, which is the generated result of this step.

[0121] In an embodiment of the present application, the second evaluation parameter of the new solution is calculated according to the same rule as the first evaluation parameter, and is compared with the first evaluation parameter. The solution with better evaluation parameters is retained as the update result. For example, the first evaluation parameter of the original solution is 200, and the second evaluation parameter of the new solution is 220. 220 is better, and the new solution is retained as the update result.

[0122] c4. Repeat the adjustment, calculation, and retention steps, retaining the updated results after each execution.

[0123] Among them, adjustment refers to changing the regional power distribution ratio according to the speed vector, calculation refers to calculating the evaluation parameters of the new plan, retention refers to keeping the better plan, and the update result is the better plan retained each time. Repeated execution refers to the process of adjustment, calculation, and retention multiple times. The generated result of this step is the cumulative update result.

[0124] In an embodiment of the present application, the steps of adjusting the regional electricity distribution ratio, calculating the evaluation parameters of the new scheme, and retaining the better scheme are repeated multiple times, and the updated results are retained each time. For example, after the first adjustment, the scheme with evaluation parameter 492 is retained, and the scheme is adjusted for the second time to obtain a new scheme with parameter 510 and retained. Multiple updated results are obtained after multiple repetitions.

[0125] c5. From all updated results, select the contract electricity splitting scheme with the best evaluation parameters as the global optimal solution.

[0126] Among them, the updated result is the better solution retained after each adjustment. The optimal evaluation parameter means that the evaluation parameter best meets the preset quality standards among all results. The global optimal solution is the best contract electricity splitting solution selected from all updated results. The optimal solution is the generated result of this step.

[0127] In an embodiment of the present application, the evaluation parameters of all cumulative update results are compared, and the contract electricity splitting scheme with the best parameters is selected as the global optimal solution. For example, the evaluation parameters of the cumulative results are 492, 495, and 500, among which 500 is the best, and the corresponding scheme is the global optimal solution.

[0128] This application provides the following specific example: a contract electricity split plan in the search set is 30% in area A, 50% in area B, and 20% in area C, with a total electricity of 1000 units. Step c1 maps it into particles with a position vector of [0.3, 0.5, 0.2] and a velocity vector of [-0.02, 0.04, -0.02]. Step c2 adjusts the position vector and acceleration vector to 0.28 in area A, 0.54 in area B, and 0.18 in area C, corresponding to electricity of 280, 540, and 180 units. Step c3 adjusts the position vector and acceleration vector to 0.28 in area A, 0.54 in area B, and 0.18 in area C, corresponding to electricity of 280, 540, and 180 units. The second evaluation parameter is calculated as 280×60%+540×50%+180×30%=168+270+54=492, which is better than the original parameter of 200, so this result is retained. Step C4 is repeated, with the new speed vector [-0.01, 0.03, -0.02], the new power units 270, 570, and 160, and the parameters 270×60%+570×50%+160×30%=162+285+48=495. After further adjustment, the solution with parameter 505 is obtained. Step C5 selects the solution with parameter 505 from the updated results as the global optimal solution.

[0129] By executing c1 to c5, the embodiment of the present application maps the plan into particles and continuously adjusts and optimizes it, repeatedly screening better plans to find the global optimal solution, so that the contract electricity splitting plan can better match the market trend, thereby improving the rationality of the plan and its adaptability to market changes.

[0130] Figure 2 A schematic diagram of the structure of a power trading strategy generation system based on Monte Carlo simulation and deep learning provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes: The acquisition module 21 is used to obtain historical price data of multi-regional power markets, decompose the historical price data to extract trend components, periodic components and random components, and construct a multidimensional feature set based on the trend components, periodic components and random components.

[0131] The mapping module 22 is used to use the blockchain network to receive unstructured data from multi-regional electricity markets, perform hash verification and smart contract packaging on the unstructured data to obtain structured data packets, and map the structured data packets into multi-dimensional semantic feature vectors based on a preset text vector mapping table.

[0132] The association module 23 is used to use a multimodal model based on deep learning to associate the multidimensional feature data in the multidimensional feature set with the multidimensional semantic feature vector, and output price trend prediction probability distribution data associated with the confidence level.

[0133] The generation module 24 is used to generate multiple contract power splitting schemes for multi-regional power markets based on Monte Carlo simulation, and to search for multiple contract power splitting schemes based on price trend prediction probability distribution data using a particle swarm optimization algorithm to obtain a global optimal solution, and to generate an optimal power trading strategy based on the global optimal solution.

[0134] Figure 2 The power trading strategy generation system based on Monte Carlo simulation and deep learning can be executed Figure 1 The implementation principles and technical effects of the Monte Carlo simulation and deep learning-based power trading strategy generation method described in the illustrated embodiment are not further elaborated. The specific manner in which each module and unit performs operations in the Monte Carlo simulation and deep learning-based power trading strategy generation system described in the above embodiment has been described in detail in the related embodiments and will not be further elaborated here.

[0135] In one possible design, Figure 2 The power trading strategy generation system based on Monte Carlo simulation and deep learning of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0136] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0137] The processing component 32 is used to execute the following process: obtaining historical price data of multi-regional electricity markets, decomposing the historical price data to extract trend components, periodic components and random components, and constructing a multidimensional feature set based on the trend components, periodic components and random components; using the blockchain network to receive unstructured data from multi-regional electricity markets, performing hash verification and smart contract packaging on the unstructured data to obtain a structured data packet, and mapping the structured data packet into a multi-dimensional semantic feature vector based on a preset text vector mapping table; using a multimodal model based on deep learning to associate the multi-dimensional feature data in the multi-dimensional feature set with the multi-dimensional semantic feature vector, and output price trend prediction probability distribution data associated with the confidence level; based on Monte Carlo simulation, generating multiple contract electricity splitting schemes for multi-regional electricity markets, based on the price trend prediction probability distribution data, using a particle swarm optimization algorithm to search for multiple contract electricity splitting schemes to obtain a global optimal solution, and based on the global optimal solution, generating an optimal electricity trading strategy.

[0138] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0139] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as random access memory (RAM), static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0140] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0141] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0142] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0143] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0144] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment provides a method for generating power trading strategies based on Monte Carlo simulation and deep learning.

[0145] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0146] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0147] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for generating power trading strategies based on Monte Carlo simulation and deep learning, characterized in that: include: Acquiring historical price data of multi-regional electricity markets, decomposing the historical price data to extract a trend component, a periodic component, and a random component, and constructing a multidimensional feature set based on the trend component, the periodic component, and the random component; Utilizing a blockchain network to receive unstructured data from the multi-regional electricity market, performing hash verification and smart contract packaging on the unstructured data to obtain a structured data packet, and mapping the structured data packet into a multi-dimensional semantic feature vector based on a preset text vector mapping table; Utilizing a multimodal model based on deep learning, the multidimensional feature data in the multidimensional feature set is associated with the multidimensional semantic feature vector, and price trend prediction probability distribution data associated with the confidence level is output; Based on Monte Carlo simulation, multiple contract electricity splitting schemes are generated for the multi-regional electricity market. Based on the price trend prediction probability distribution data, a particle swarm optimization algorithm is used to search for the multiple contract electricity splitting schemes to obtain a global optimal solution. Based on the global optimal solution, an optimal electricity trading strategy is generated.

2. The method for generating power trading strategies based on Monte Carlo simulation and deep learning according to claim 1, characterized in that: The method utilizes a multimodal model based on deep learning to associate the multidimensional feature data in the multidimensional feature set with the multidimensional semantic feature vector, and outputs price trend prediction probability distribution data associated with the confidence level, including: Inputting the multidimensional feature data in the multidimensional feature set into a first feature processing unit in a multimodal model to obtain a first feature sequence; Inputting the multi-dimensional semantic feature vector into a second feature processing unit in the multimodal model to obtain a second feature sequence; Combining the first feature sequence and the second feature sequence through a feature association unit in a multimodal model to form a fused feature sequence; Inputting the fused feature sequence into a multi-layer processing unit in a multimodal model, and obtaining an intermediate feature sequence after multiple feature conversions; Based on the generation unit in the multimodal model, the intermediate feature sequence is processed to generate initial price trend prediction probability distribution data; A confidence level corresponding to the price trend prediction probability distribution data is generated through a feature statistics unit in a multimodal model, and the price trend prediction probability distribution data is associated with the confidence level.

3. The method for generating power trading strategies based on Monte Carlo simulation and deep learning according to claim 2, characterized in that: Generating a confidence level corresponding to the price trend prediction probability distribution data includes: Dividing the price trend prediction probability distribution data into multiple price value intervals, and calculating the corresponding proportion of each price value interval; Calculate the difference between the proportions based on the preset measurement rules; Based on the difference degree values, confidence parameters for each corresponding price value interval are generated according to a preset mapping rule; All the confidence parameters are integrated according to a preset integration rule to obtain a confidence level corresponding to the price trend prediction probability distribution data.

4. The method for generating power trading strategies based on Monte Carlo simulation and deep learning according to claim 1, characterized in that: The unstructured data is hashed and packaged with a smart contract to obtain a structured data packet, and the structured data packet is mapped into a multi-dimensional semantic feature vector based on a preset text vector mapping table, including: Processing the unstructured data according to preset encoding rules to generate a corresponding unique data code, and appending the data code to the unstructured data to complete hash verification; Based on preset grouping rules, the unstructured data that has completed hash verification is classified and packaged through smart contracts to form structured data packets; Based on a preset text vector mapping table, the content of each field in the structured data packet is converted into a corresponding vector value, and all the vector values ​​are integrated according to a preset arrangement rule to obtain a multi-dimensional semantic feature vector.

5. The method for generating power trading strategies based on Monte Carlo simulation and deep learning according to claim 4 is characterized in that: The method converts the content of each field in the structured data packet into a corresponding vector value based on a preset text vector mapping table, and integrates all the vector values ​​according to a preset arrangement rule to obtain a multi-dimensional semantic feature vector, including: According to the field type, each field in the structured data packet is divided into a text field, a numerical field and a classification field; Splitting the field content of the text field into multiple word units according to a preset word segmentation rule, and converting each word unit into a corresponding basic vector value according to the correspondence between the word units and vector values ​​in the preset text vector mapping table; Mapping the field content of the numerical field to a corresponding first-dimensional vector value based on an interval mapping rule in a preset text vector mapping table; Based on the category coding rules in the preset text vector mapping table, the field content of the classification field is converted into a corresponding second dimension vector value; All the basic vector values, the first dimensional vector values ​​and the second dimensional vector values ​​are integrated according to the field order and dimension arrangement rule in the preset arrangement rule to obtain a multi-dimensional semantic feature vector.

6. The method for generating power trading strategies based on Monte Carlo simulation and deep learning according to claim 1, characterized in that: The method of searching the plurality of contract electricity splitting schemes based on the price trend prediction probability distribution data using a particle swarm optimization algorithm to obtain a global optimal solution includes: Matching the regional electricity data in each of the contract electricity splitting schemes with the price trend prediction probability distribution data of the corresponding region to obtain correlation data, and calculating the correlation data according to a preset evaluation rule to obtain a first evaluation parameter for each of the contract electricity splitting schemes; Select all contract electricity splitting schemes whose first evaluation parameter meets the preset range as the search set; A particle swarm optimization algorithm is used to search based on the search set to obtain a global optimal solution.

7. The method for generating power trading strategies based on Monte Carlo simulation and deep learning according to claim 6, characterized in that: The particle swarm optimization algorithm is used to search based on the search set to obtain a global optimal solution, including: Map each of the contract electricity splitting schemes in the search set to a particle in the particle swarm, wherein the position vector of the particle represents the allocation ratio of the regional electricity data, and the velocity vector of the particle represents the adjustment direction and amplitude of the allocation ratio; Adjusting the allocation ratio of the regional electricity data according to a preset adjustment rule, and obtaining a new contract electricity splitting plan based on the adjusted allocation ratio of the regional electricity data; Calculating a second evaluation parameter of the new contract power splitting scheme, comparing the second evaluation parameter with the first evaluation parameter, and retaining the contract power splitting scheme with the better evaluation parameter as the update result; Repeat the adjustment, calculation and retention steps, and retain the updated results after each execution; From all the update results, the contract electricity splitting scheme with the best evaluation parameters is selected as the global optimal solution.

8. A power trading strategy generation system based on Monte Carlo simulation and deep learning, characterized by: include: an acquisition module, configured to acquire historical price data of multi-regional electricity markets, decompose the historical price data to extract a trend component, a periodic component, and a random component, and construct a multidimensional feature set based on the trend component, the periodic component, and the random component; a mapping module for receiving unstructured data from the multi-regional power market using a blockchain network, performing hash verification and smart contract packaging on the unstructured data to obtain a structured data packet, and mapping the structured data packet into a multi-dimensional semantic feature vector based on a preset text vector mapping table; an association module, configured to associate the multidimensional feature data in the multidimensional feature set with the multidimensional semantic feature vector using a multimodal model based on deep learning, and output price trend prediction probability distribution data associated with the confidence level; A generation module is used to generate multiple contract electricity splitting schemes for the multi-regional electricity market based on Monte Carlo simulation, and based on the price trend prediction probability distribution data, use a particle swarm optimization algorithm to search for the multiple contract electricity splitting schemes to obtain a global optimal solution, and generate an optimal electricity trading strategy based on the global optimal solution.

Images