Power transaction auxiliary decision-making method based on MCP

Abstract

The invention belongs to the technical field of power transaction management, and particularly relates to an MCP-based power transaction aid decision-making method, which comprises the following steps of: receiving and standardizing a user request through an MCP client layer; the MCP arrangement gateway analyzes the request, and dispatches and activates a corresponding MCP server agent group; the agent group obtains power operation data, historical transaction data, prediction model data and policy support information by calling the service API group, and completes data processing and task execution; the arrangement gateway integrates, verifies and formats the processing results of the intelligent agents; and finally, feeding back a unified response result to the user. According to the invention, standardized interaction and context sharing among multi-source data, heterogeneous models and business functions are realized through the MCP protocol, a modularized and extensible agent collaborative decision framework is constructed, and the efficiency, accuracy and policy compliance of power transaction decision are effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power transaction management, and particularly relates to a power transaction auxiliary decision-making method based on MCP. BACKGROUND

[0002] In the field of energy today, as the core link of guaranteeing the stable operation of the power system and optimizing the allocation of resources, the process of power transaction is increasingly complex. The power market covers various transaction mechanisms such as day-ahead market, real-time market, bilateral contract and rolling matching, which span different time scales from long-term planning to second-level response. At the same time, a large amount of multi-source heterogeneous data such as power, price, weather and new energy power is continuously generated, and the update frequency is very high, which requires market participants to make rapid and accurate real-time judgments and decisions in the dynamically changing market environment.

[0003] However, the existing power transaction auxiliary decision-making technology cannot effectively cope with the above challenges. Traditional systems mostly rely on single or mutually isolated prediction and optimization models, and lack a unified collaborative framework. This results in the inability to share context information between different models, and the disconnection between data processing, market prediction, risk assessment and strategy optimization, forming a serious "information island". In addition, the system has low integration and transformation efficiency for complex and high-frequency data, and the level of intelligence is limited, so that the traders still need to invest a lot of time in manual data arrangement, model comparison and information synthesis in the decision-making process, not only delaying the decision-making process, but also significantly increasing the operational risk due to the limitations of human experience.

[0004] In recent years, although large language models and multi-agent systems have shown potential in complex task processing, how to adapt them to the professional scene of power transaction and solve the fundamental problem of efficient collaboration and context sharing between multiple models has not been effectively broken through. In this background, a new protocol that can provide standardized interaction and context sharing capabilities for different models, the Model Context Protocol (MCP), provides a new technical path for building the next generation of intelligent power transaction auxiliary decision-making system. SUMMARY

[0005] The purpose of the present application is to overcome the deficiencies in the prior art, and to provide a power transaction auxiliary decision-making method based on MCP, which realizes the standardized integration and intelligent scheduling of multi-source data, prediction models and business functions through the MCP protocol, and automates and intelligentizes the traditional complex decision-making process that relies on manual work, significantly shortening the decision-making time.

[0006] The technical scheme adopted by the present application to solve its technical problems is: a power transaction auxiliary decision-making method based on MCP is provided, comprising the following steps:

[0007] S1: User requests input and standardized conversion - Receive user-initiated power transaction assistance decision demand through the MCP client layer, which converts the demand into a request in the MCP standard format, which includes transaction type, time range, and core demand elements;

[0008] S2: Request scheduling and agent activation - The MCP orchestration gateway receives the MCP standard format request, parses the demand type of the request, automatically activates the corresponding MCP server agent group based on the demand type, and routes the MCP standard format request to the MCP server agent group, synchronizing task parameters;

[0009] S3: Agent data acquisition and processing - The MCP server agent group receives the scheduling instructions from the MCP orchestration gateway, acquires target data and policy support information by calling the business API group, the target data includes power operation data, historical transaction data, and prediction model data, the policy support information comes from the policy document library, the MCP server agent group completes the specified processing task based on the target data and policy support information, and generates the processing result;

[0010] S4: Result integration and formatting - The MCP orchestration gateway receives the processing result from the MCP server agent group, performs format conversion, logical verification, and information aggregation on the processing result according to the output requirements of the user's original demand, and generates a unified response result;

[0011] S5: Result feedback to user - The MCP orchestration gateway transmits the unified response result to the MCP client layer, which converts the unified response result into a form readable by the user and feeds it back to the user;

[0012] Wherein, the MCP client layer, MCP orchestration gateway, MCP server agent group, business API group, and policy document library realize data interaction and context sharing through MCP protocol, and the policy document library is used to store power transaction field policy files, providing policy retrieval and compliance verification support for the MCP server agent group.

[0013] Further, the MCP client layer includes at least one of a chat client, an API client, a file client, and a UI client;

[0014] The chat client receives user text conversations, performs legality and security verification on the text conversations, extracts transaction type, time range, and core demand elements through semantic analysis, and converts the text conversations into MCP standard format requests;

[0015] The API client receives RESTful API structured requests sent by external systems, directly reads the parameters in the structured requests, and generates MCP standard format requests.

[0016] The file client supports batch data import in CSV / Excel format, receives historical load data and electricity price data uploaded by users, and encapsulates the batch data into an MCP standard format request.

[0017] The UI client is a mobile or large-screen client that supports real-time visual interaction, receives user requests initiated through the visual interface, and converts them into MCP standard format requests.

[0018] Furthermore, the MCP orchestration gateway also performs the following operations in step S2:

[0019] Security and Access Control: User roles are used for identity authentication. These roles include power generation company users, electricity sales company users, and system administrators. Data access permissions and function usage permissions are defined based on these user roles to prevent unauthorized access.

[0020] Compliance audit and log tracking: Record the receiving time, scheduling path, agent invocation status and result feedback time of the MCP standard format request to form a traceable audit log;

[0021] Fault tolerance and rollback mechanism: If the target agent in the MCP server agent group or the target API in the business API group fails, the historical version of the agent or the backup API will be automatically called and step S3 will be re-executed.

[0022] Furthermore, the MCP server intelligent agent group includes a data integration intelligent agent. The processing procedure of the data integration intelligent agent in step S3 is as follows: the data integration intelligent agent calls the real-time operation data API and historical transaction data API in the business API group through the MCP protocol to obtain power operation data and historical transaction data within a specified time range and for a specified transaction type; integrates the power operation data and historical transaction data according to preset rules, including associating day-ahead market data and real-time market data by time dimension and summarizing bilateral contract data and rolling matching data by transaction type; generates structured data results from the integrated data and returns them to the MCP orchestration gateway through the MCP protocol.

[0023] Furthermore, the MCP server agent group includes a prediction and adaptation agent, and the processing procedure of the prediction and adaptation agent in step S3 is as follows:

[0024] The predictive adaptation agent receives scenario parameters synchronized by the MCP orchestration gateway via the MCP protocol. The scenario parameters include trading instruments and time periods.

[0025] Call the scene data API and prediction model API in the business API group to obtain the scene feature data and candidate prediction model list corresponding to the scene parameters;

[0026] Based on the scene feature data, the models in the candidate prediction model list are evaluated for suitability, the optimal prediction model that is suitable for the current scene is selected, a model recommendation list and the basis for suitability are generated, and the results are returned to the MCP orchestration gateway through the MCP protocol.

[0027] Furthermore, the MCP server agent group includes a decision support agent, and the processing procedure of the decision support agent in step S3 is as follows:

[0028] The decision support agent cascades the data integration agent and the prediction adaptation agent through the MCP protocol to obtain the integrated structured data and the prediction results of the optimal prediction model.

[0029] Based on the policy provisions provided by the policy document library, power trading decision suggestions are generated according to preset decision rules. The decision suggestions include power allocation schemes, bidding range references, and contract signing suggestions.

[0030] The decision recommendations are verified for policy compliance, the corresponding policy basis clauses are marked, and the results are returned to the MCP orchestration gateway via the MCP protocol.

[0031] Furthermore, the business API group includes data service API, model service API, and decision service API;

[0032] The data service API connects to the power data pool through the MCP protocol, providing structured data query services and unstructured data parsing services. The structured data query service includes rolling matching transaction volume queries within a specified time range, and the unstructured data parsing service includes extraction of key clauses from bilateral contract texts.

[0033] The model service API connects to the dynamic model adaptation and scheduling module through the MCP protocol, providing prediction model invocation service and model parameter adjustment service. The prediction model invocation service includes invocation of short-term load forecasting model and electricity price forecasting model.

[0034] The decision service API links with the MCP server intelligent agent group through the MCP protocol to output integrated power trading decision suggestions, which include data support, model prediction results and policy basis.

[0035] Furthermore, the policy document repository is built on Elasticsearch, and the construction and retrieval process of the policy document repository includes:

[0036] Document entry: Standardize policy documents in the field of power trading, including word segmentation to extract core words, converting text semantics into word vector data, tagging by transaction type, applicable scenario, and release time, and establishing an association index between policy clauses. The core words include "rolling matching", "price priority", and "new energy grid connection".

[0037] Policy retrieval: The MCP server intelligent agent group calls the retrieval interface of the policy document library through the business API group, matches the target policy clauses based on full-text search or semantic vector search, and returns structured results containing original text fragments, core clause summaries and related policy recommendations;

[0038] Cross-version comparison: The policy document library stores policy versions from different periods, supports difference analysis of different versions of the same policy, generates policy change reports, and indicates the potential impact of policy changes on power trading strategies.

[0039] Furthermore, the format of the unified response result mentioned in step S4 includes at least one of the following: a visual chart, a structured table, and a natural language report;

[0040] The visualization charts include an electricity price forecast trend chart and a bar chart of electricity consumption execution.

[0041] The structured tables include a detailed table of bilateral contract electricity volume execution and a table of policy compliance verification results;

[0042] The natural language reports include transaction decision analysis reports and risk assessment reports, which contain data sources, model basis, and policy references.

[0043] Furthermore, the MCP server intelligent agent group also includes a user profile intelligent agent, which performs the following operations in step S3:

[0044] The user profiling agent obtains users' historical transaction behavior data and risk preference data through the business API group;

[0045] User profiles are generated based on the historical transaction behavior data and risk preference data. The user profiles include transaction habit tags and risk tolerance level tags.

[0046] The user profile is synchronized to the decision support agent, which adjusts the content and form of decision suggestions based on the user profile to provide personalized decision support.

[0047] The present application has the following beneficial effects:

[0048] (1) The present application realizes the standardized integration and intelligent scheduling of multi-source data, prediction models and business functions through the MCP protocol, automates and intelligenizes the traditional complex decision-making process relying on manual work, and significantly shortens the decision-making time.

[0049] (2) The present application improves the prediction accuracy of core indicators through a prediction adaptation mechanism, and performs real-time, automatic compliance verification and association based on a structured policy library, so that the decision-making suggestions have accuracy, policy compliance and explainability.

[0050] (3) The present application adopts a loose coupling architecture based on MCP, so that the system can quickly adapt to changes in market rules, transaction varieties and data sources through modular addition and deletion (such as agents and APIs), reducing the complexity and cost of system evolution and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 is a flowchart of a power transaction auxiliary decision-making method based on MCP of the present application;

[0052] Figure 2 is a flowchart of a data integration agent of a power transaction auxiliary decision-making method based on MCP of the present application;

[0053] Figure 3 is a flowchart of a prediction adaptation agent of a power transaction auxiliary decision-making method based on MCP of the present application;

[0054] Figure 4 is a flowchart of a decision support agent of a power transaction auxiliary decision-making method based on MCP of the present application;

[0055] Figure 5 is a structural diagram of a power transaction auxiliary decision-making system based on MCP of the present application. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0057] Embodiment 1:

[0058] A power transaction auxiliary decision-making method based on MCP, as shown in Figure 1 includes the following steps:

[0059] S1: User requests input with standardized translation - User-initiated power trading assistance decision demand is received through the MCP client layer, which translates the demand into a request in the MCP standard format, including transaction type, time range, and core demand elements.

[0060] Scenario: A power trading company trader inputs in the system chat interface: "Please help me analyze the purchase strategy for tomorrow's (October 27, 2023) intraday market in the province, considering the high wind power forecast and the newly introduced peak price policy."

[0061] Implementation process:

[0062] Request reception: The user's natural language request is input into the system through the chat client (belonging to the MCP client layer).

[0063] Security and compliance screening: The chat client's built-in filter first performs a basic security scan (such as preventing injection attacks) and a preliminary compliance judgment (such as screening sensitive words) on the request text.

[0064] Semantic analysis and element extraction: The client calls the integrated lightweight semantic understanding module (which can be based on a small language model or a rule engine) to deeply analyze the user's request.

[0065] Intention recognition: The core intention of the user is identified as "formulating an intraday market purchase strategy."

[0066] Key entity and parameter extraction:

[0067] Time: 2023-10-27 (tomorrow).

[0068] Transaction variety: Intra-provincial intraday market.

[0069] Decision goal: Purchase strategy.

[0070] Constraint condition: High wind power forecast, peak price policy.

[0071] Context completion: If the request is missing necessary information (such as specific regions), the client can prompt the user to supplement it, or complete it based on user history and default settings.

[0072] Generate MCP standard format request: Encapsulate the structured information parsed into a JSON format standard request that conforms to the MCP protocol definition. For example:

[0073] "request_id":"REQ_20231026_142335_001",

[0074] "client_type":"chat",

[0075] "user_id":"trader_wang",

[0076] "user_role":"Electricity Sales Company",

[0077] "timestamp":"2023-10-26T14:23:35Z",

[0078] "intent":"strategy_analysis",

[0079] "parameters":{

[0080] "market_type":"day_ahead",

[0081] "region":"within the province",

[0082] "target_date": "2023-10-27",

[0083] "focus": "purchase_strategy",

[0084] "constraints": ["high_wind_forecast", "new_peak_price_policy"]},

[0085] "response_preference": {

[0086] "format": ["nl_report", "chart"],

[0087] "detail_level": "detailed"}

[0088] Request Sending: The encapsulated MCP standard request is sent to the MCP orchestration gateway through an internal secure channel (such as HTTPS).

[0089] Alternative Implementation: If a user sends a structured request (e.g., {"action": "get_strategy", "date": "2023-10-27"}) via a RESTful API interface, the API client will directly map and supplement it into a complete MCP request. File clients, on the other hand, will include the uploaded CSV data file (e.g., historical load curves) as additional data source information in the request parameters.

[0090] S2: Request scheduling and agent activation - The MCP orchestration gateway receives the MCP standard format request, parses the request demand type, automatically activates the corresponding MCP server agent group based on the demand type, and routes the MCP standard format request to the MCP server agent group, synchronizing task parameters;

[0091] Implementation process:

[0092] Request reception and parsing: The request reception module of the MCP orchestration gateway receives the MCP request from the chat client.

[0093] Deep analysis and task decomposition: The parsing engine of the gateway performs more detailed analysis on the request. It is identified that the request is a composite task involving data query (getting historical load, electricity price, wind power prediction), policy retrieval (querying peak electricity price policy details), prediction analysis (evaluating the impact of wind power uncertainty on the market), strategy optimization (generating electricity purchase strategy), etc.

[0094] Identity authentication and permission verification: The security module of the gateway verifies user_id and user_role, confirming that the user "trader_wang" as the "power seller" role has the right to access day-ahead market data, prediction models and strategy analysis functions.

[0095] Agent routing planning: Based on the task decomposition results, the scheduler of the gateway plans the execution link. It is decided to activate the following agents in sequence or in parallel:

[0096] Data integration agent: Get historical data and real-time prediction data.

[0097] Policy query agent: Retrieve the peak electricity price policy.

[0098] Prediction adaptation agent: Adapt appropriate electricity price prediction models for high wind power penetration scenarios.

[0099] Risk assessment agent: Quantify the risk brought by wind power prediction uncertainty.

[0100] Decision support agent (as coordinator): Integrate the above information to generate strategy recommendations.

[0101] Generate and distribute scheduling instructions: The gateway generates scheduling instructions containing original request parameters and sub-task descriptions, and sends them to the above agents through the MCP protocol. At the same time, the log module of the gateway starts to record the complete life cycle events of the request.

[0102] Fault-tolerant mechanism: When planning the route, the gateway checks the health status of each target agent. If the optimization solving agent (a more complex strategy generation method) is temporarily unavailable, the scheduler will automatically downgrade the strategy generation task to the decision support agent (based on rules and heuristic methods), ensuring uninterrupted service.

[0103] S3: Agent data acquisition and processing - The MCP server agent group receives scheduling instructions from the MCP orchestration gateway, acquires target data and policy support information by calling the business API group, the target data includes power operation data, historical transaction data, prediction model data, the policy support information comes from the policy document library, the MCP server agent group completes the specified processing task based on the target data and policy support information, generates processing results;

[0104] As shown in Figure 2 , Figure 3 and Figure 4 , each agent receives a scheduling instruction and independently or cooperatively completes the work by calling the business API group.

[0105] Data integration agent processing:

[0106] API call: Call real-time operation data API (get the latest grid state), historical transaction data API (get the day-ahead market clearing price and transaction volume of the same period in the past month), and new energy prediction data API (get the next day's provincial wind power and photovoltaic power prediction curve) in the business API group through the MCP protocol.

[0107] Data processing: Align the obtained multi-source data according to the time axis (for example, unify to 15-minute granularity), and perform data quality verification (such as filling small missing values and marking abnormal values). Finally, integrate into a multi-dimensional structured data set containing timestamp, load prediction, wind power prediction, photovoltaic prediction, historical electricity price, etc.

[0108] Result return: Package the integrated data set into an MCP format response and return it to the orchestration gateway, and note that the data is ready.

[0109] Policy query agent processing:

[0110] Query construction: According to the "peak electricity price policy" constraint, construct a semantic query vector.

[0111] In-database retrieval: By invoking the policy retrieval API of the business API group, a query is initiated to the policy document library. The policy document library is built based on Elasticsearch, and its internal NLP processing pipeline performs word segmentation and vectorization on the original policy text. The retrieval uses a "semantic matching + keyword weighting" approach to accurately locate the latest published "Notice on the Trial Seasonal Peak Electricity Price Mechanism" and its specific provisions.

[0112] Information extraction and summary: After the agent obtains the policy text segment, it uses text summarization technology to extract the core content: "Tomorrow (October 27) evening peak 18:00-20:00 will implement peak electricity price, which will be 20% higher than the flat segment price."

[0113] Result return: The policy summary, original text link and impact range (time period, amplitude) are returned to the gateway.

[0114] Cooperative processing of prediction adaptation and risk assessment agents:

[0115] Scenario adaptation: The prediction adaptation agent receives the "high wind power prediction" scenario parameter. It calls the scenario data API to obtain the statistical characteristics of the current wind power prediction (such as predicted power mean, volatility), and then calls the prediction model API to query the model repository. Each model in the model repository has metadata tags, such as applicable scenarios: [high new energy, day-ahead market], good at volatility: high. The agent recommends using "hybrid electricity price prediction model based on attention mechanism and Monte Carlo simulation (Model-A)" through matching.

[0116] Model execution and uncertainty quantification: The prediction adaptation agent calls Model-A, inputs the data set provided by the data integration agent, and runs the prediction. Model-A not only outputs point prediction (expected electricity price curve), but also outputs probability distribution (such as quantile prediction) to represent uncertainty.

[0117] Risk assessment: The risk assessment agent receives the probability prediction results output by Model-A, combines historical volatility and market liquidity indicators to calculate the expected loss value (VaR) and conditional risk value (CVaR) of each time period's electricity purchase tomorrow, and gives "low, medium, high" risk level labels. For example, it determines that the risk level of the evening peak period is "high" due to the policy superimposed wind power fluctuations.

[0118] S4: Result integration and formatting - the MCP orchestration gateway receives the processing results of the MCP server agent group, performs format conversion, logical verification and information aggregation on the processing results according to the output requirements of the user's original demand, and generates a unified response result;

[0119] Result collection: The integration module of MCP orchestration gateway continuously listens and collects the returned results from data integration, policy query, decision support (with prediction and risk assessment cascaded), etc.

[0120] Logical association and verification:

[0121] The policy suggestion generated by the decision support agent (such as "suggest more electricity purchase during the low electricity price period in the afternoon, and be cautious during the peak period in the evening") is associated and verified with the policy provisions returned by the policy query agent, to confirm that the policy meets the policy requirements.

[0122] The electricity allocation suggestion in the policy is superimposed and analyzed with the risk level given by the risk assessment agent, and a warning mark is added to the suggestion in the high-risk period.

[0123] Check whether the time stamp and caliber of all data are consistent.

[0124] Format synthesis and rendering: according to the [“nl_report”, “chart”] specified in the response_preference in the user request:

[0125] Generate natural language report: use the report generation template to organize structured data, policy suggestions, policy basis, risk prompts, etc. into a coherent and professional analysis report text.

[0126] Generate visual charts: call chart service to generate two core charts: (a) 24-hour electricity price prediction curve (including confidence interval) tomorrow, and superimpose the peak price period; (b) recommended electricity purchase distribution bar chart.

[0127] Structured data attachment: prepare the cleaned data and model prediction details table in CSV format as an attachment.

[0128] Generate unified response package: encapsulate the above text report, chart (embedded in URL or base64 encoding), data attachment link, confidence score of this analysis, version information of models and data used, into a final data package that meets the MCP response format.

[0129] S5: result feedback to user - the MCP orchestration gateway transmits the unified response result to the MCP client layer, which converts the unified response result into a form readable by the user and feeds back to the user;

[0130] Response transmission: orchestration gateway sends unified response package back to the chat client that originally initiated the request.

[0131] Client rendering: chat client parses response package.

[0132] Display natural language report part directly in the conversation window.

[0133] Render chart, show or provide click to view large chart in conversation interface.

[0134] Provide "download detailed data" button, link to data attachment.

[0135] User interaction complete: traders clearly and intuitively obtain the required strategy analysis report, visual chart and supporting data in the chat interface, which can be used to make more efficient decisions. The entire process from asking questions to obtaining a comprehensive report usually takes a few seconds to a minute, far exceeding the efficiency of manual processing.

[0136] As an embodiment, the implementation of the MCP protocol includes: the MCP protocol is the "common language" of internal communication of the system. The implementation points include:

[0137] Message structure: JSON Schema is used to strictly define the format of requests and responses, including header (metadata), payload (data load), context (session context), etc.

[0138] Context management: Each request-response session has a unique ID. When passing intermediate results between agents, intermediate data can be temporarily stored in the context cache of the gateway for subsequent agents to retrieve as needed, avoiding repeated calculations and data transmission.

[0139] Capability declaration: Each agent and API declares the types of tasks it can handle and the required parameter formats when starting, and the gateway establishes a dynamic routing table accordingly.

[0140] As an embodiment, the construction of the policy document library includes:

[0141] Data collection: Through web crawlers, official interface subscriptions, etc., continuously collect policy documents (PDF, Word, HTML) published by national, provincial, and municipal energy bureaus and power trading centers.

[0142] Structured analysis: Use OCR and document parsing technology to convert unstructured text into plain text. Use named entity recognition (NER) models to extract key entities from the file, such as issuing authority, effective date, involved market, and key clauses.

[0143] Deep indexing:

[0144] Vectorization: Use Sentence-BERT and other models to encode each paragraph into a semantic vector and store it in the dense_vector field of Elasticsearch.

[0145] Knowledge graph construction: Establish the association between clauses, such as "Rule A" X refers to "Method B" Y, forming a policy knowledge network that can be inferred.

[0146] Retrieval optimization: When searching, the user query is also vectorized, and semantic retrieval is performed by calculating the cosine similarity, while combining traditional keyword retrieval (BM25) for hybrid sorting to ensure recall and precision.

[0147] As an embodiment, the working principle of the user portrait agent includes: the agent implements offline and online analysis:

[0148] Offline analysis: Regularly (such as daily) analyze the user's historical transaction data, pricing behavior, and risk preference questionnaire results. Use clustering and classification algorithms to label users, such as style: aggressive, risk preference: moderate, and market: intraday rolling matching.

[0149] Online application: When the decision support agent generates a strategy, it queries the user's portrait. For "aggressive" users, the strategy may include higher-risk but potentially higher-yield arbitrage opportunity prompts; for "conservative" users, the strategy will emphasize risk avoidance and stable guaranteed schemes. At the same time, the language style and detail level of the report can be adjusted according to user habits.

[0150] Embodiment 2:

[0151] The application also provides a power transaction auxiliary decision-making system based on MCP, as shown in Figure 5 , comprising:

[0152] MCP client layer: The system's interaction entrance with users, responsible for receiving and converting user requests.

[0153] MCP orchestration gateway: The scheduling and control center of the system, responsible for request analysis, agent routing, and result integration.

[0154] MCP server agent group: Composed of multiple specialized agents, it is the core computing and reasoning engine of the system.

[0155] Business API group: A set of interfaces that provide standardized data, models, and decision-making services.

[0156] Policy document library: A knowledge base that stores and manages power transaction-related policy documents.

[0157] The layers communicate through a unified MCP protocol, ensuring standardized, efficient, and traceable data flow and instruction flow.

[0158] Through the implementation of the above embodiments, the present application has the following significant beneficial effects compared to traditional methods:

[0159] Decision efficiency leap: The manual process that originally required traders to spend hours reviewing data, analyzing policies, and invoking multiple tools is compressed into minutes of automatic completion, releasing manpower.

[0160] Decision accuracy improvement: Sharing context through MCP protocol enables the electricity price prediction model to fully consider the uncertainty of wind power prediction, and the risk assessment model to closely integrate policy changes, ensuring the scientificity and accuracy of the final strategy recommendation.

[0161] System adaptability enhancement: When new trading varieties (such as climbing products) or new policies are introduced into the market, only the corresponding new agent (such as a climbing product analysis agent) or policy library needs to be developed and accessed through the MCP protocol to quickly expand system functions without the need to restructure the overall architecture.

[0162] User experience optimization: Natural language interaction reduces the use threshold; integrated report presentation avoids the cumbersome switching between multiple software interfaces; personalized recommendation makes decision support more targeted.

[0163] Compliance and audit improvement: Full-process log recording and context traceability make every decision recommendation traceable, facilitating internal review and responding to external regulatory audits.

[0164] The present application provides a specific implementation scheme of a highly intelligent, modular and scalable power trading auxiliary decision-making method based on MCP protocol. This scheme deeply integrates advanced model context protocol, multi-agent collaborative computing technology and professional knowledge in the field of power trading, and builds an integrated platform that can "understand" user needs, "dispatch" professional models, "integrate" multi-source information and "generate" comprehensive decisions. It effectively solves the core pain points of data silos, model fragmentation, poor real-time performance and low intelligence level in the current power trading auxiliary decision-making field, and provides a powerful intelligent tool for power market participants to cope with the increasingly complex and volatile market environment, which has important practical value and broad application prospects.

[0165] Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for part of the technical features, and any modification, equivalent substitution, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A power trading auxiliary decision-making method based on MCP, characterized in that, Includes the following steps: S1: User Request Input and Standardization Transformation – The MCP client layer receives user-initiated power trading auxiliary decision-making requests and transforms these requests into requests conforming to the MCP standard format. The MCP standard format request includes the transaction type, time range, and core requirement elements. S2: Request Scheduling and Agent Activation - The MCP orchestration gateway receives the MCP standard format request, parses the request's demand type, automatically activates the corresponding MCP server agent group based on the demand type, routes the MCP standard format request to the MCP server agent group, and synchronizes task parameters. S3: Intelligent Agent Data Acquisition and Processing — The MCP server intelligent agent group receives the scheduling instructions from the MCP orchestration gateway, and obtains target data and policy support information by calling the business API group. The target data includes power operation data, historical transaction data, and prediction model data. The policy support information comes from the policy document library. The MCP server intelligent agent group completes the specified processing tasks based on the target data and policy support information and generates processing results. S4: Result Integration and Formatting - The MCP orchestration gateway receives the processing results from the MCP server intelligent agent group, performs format conversion, logical verification, and information aggregation on the processing results according to the output requirements of the user's original needs, and generates a unified response result. S5: Result feedback to user - The MCP orchestration gateway transmits the unified response result to the MCP client layer, and the MCP client layer converts the unified response result into a user-readable form and feeds it back to the user; The MCP client layer, MCP orchestration gateway, MCP server intelligent agent group, business API group, and policy document library achieve data interaction and context sharing through the MCP protocol. The policy document library is used to store policy documents in the field of power trading and provides policy retrieval and compliance verification support for the MCP server intelligent agent group.

2. The power trading auxiliary decision-making method based on MCP according to claim 1, characterized in that, The MCP client layer includes at least one of a chat client, an API client, a file client, and a UI client; The chat client receives user text conversations, verifies the legality and security of the text conversations, extracts transaction type, time range, and core requirement elements through semantic parsing, and converts the text conversations into MCP standard format requests. The API client receives RESTful API structured requests sent by external systems, directly reads the parameters in the structured requests, and generates MCP standard format requests. The file client supports batch data import in CSV / Excel format, receives historical load data and electricity price data uploaded by users, and encapsulates the batch data into an MCP standard format request. The UI client is a mobile or large-screen client that supports real-time visual interaction, receives user requests initiated through the visual interface, and converts them into MCP standard format requests.

3. The power trading auxiliary decision-making method based on MCP according to claim 1 or 2, characterized in that, The MCP orchestration gateway also performs the following operations in step S2: Security and Access Control: User roles are used for identity authentication. These roles include power generation company users, electricity sales company users, and system administrators. Data access permissions and function usage permissions are defined based on these user roles to prevent unauthorized access. Compliance audit and log tracking: Record the receiving time, scheduling path, agent invocation status and result feedback time of the MCP standard format request to form a traceable audit log; Fault tolerance and rollback mechanism: If the target agent in the MCP server agent group or the target API in the business API group fails, the historical version of the agent or the backup API will be automatically called and step S3 will be re-executed.

4. The power trading auxiliary decision-making method based on MCP according to claim 3, characterized in that, The MCP server intelligent agent group includes a data integration intelligent agent. The processing procedure of the data integration intelligent agent in step S3 is as follows: the data integration intelligent agent calls the real-time operation data API and historical transaction data API in the business API group through the MCP protocol to obtain power operation data and historical transaction data within a specified time range and for a specified transaction type; and integrates the power operation data and historical transaction data according to preset rules, including associating day-ahead market data and real-time market data by time dimension and summarizing bilateral contract data and rolling matching data by transaction type. The integrated data is used to generate structured data results, which are then returned to the MCP orchestration gateway via the MCP protocol.

5. The power trading auxiliary decision-making method based on MCP according to claim 4, characterized in that, The MCP server agent group includes a prediction and adaptation agent, and the processing procedure of the prediction and adaptation agent in step S3 is as follows: The predictive adaptation agent receives scenario parameters synchronized by the MCP orchestration gateway via the MCP protocol. The scenario parameters include trading instruments and time periods. Call the scene data API and prediction model API in the business API group to obtain the scene feature data and candidate prediction model list corresponding to the scene parameters; Based on the scene feature data, the models in the candidate prediction model list are evaluated for suitability, the optimal prediction model that is suitable for the current scene is selected, a model recommendation list and the basis for suitability are generated, and the results are returned to the MCP orchestration gateway through the MCP protocol.

6. The power trading auxiliary decision-making method based on MCP according to claim 5, characterized in that, The MCP server agent group includes a decision support agent, and the processing procedure of the decision support agent in step S3 is as follows: The decision support agent cascades the data integration agent and the prediction adaptation agent through the MCP protocol to obtain the integrated structured data and the prediction results of the optimal prediction model. Based on the policy provisions provided by the policy document library, power trading decision suggestions are generated according to preset decision rules. The decision suggestions include power allocation schemes, bidding range references, and contract signing suggestions. The decision recommendations are verified for policy compliance, the corresponding policy basis clauses are marked, and the results are returned to the MCP orchestration gateway via the MCP protocol.

7. The power trading auxiliary decision-making method based on MCP according to claim 6, characterized in that, The business API group includes data service API, model service API, and decision service API; The data service API connects to the power data pool through the MCP protocol, providing structured data query services and unstructured data parsing services. The structured data query service includes rolling matching transaction volume queries within a specified time range, and the unstructured data parsing service includes extraction of key clauses from bilateral contract texts. The model service API connects to the dynamic model adaptation and scheduling module through the MCP protocol, providing prediction model invocation service and model parameter adjustment service. The prediction model invocation service includes invocation of short-term load forecasting model and electricity price forecasting model. The decision service API links with the MCP server intelligent agent group through the MCP protocol to output integrated power trading decision suggestions, which include data support, model prediction results and policy basis.

8. The power trading auxiliary decision-making method based on MCP according to claim 7, characterized in that, The policy document repository is built on Elasticsearch, and the construction and retrieval process of the policy document repository includes: Document entry: Standardize policy documents in the field of power trading, including word segmentation to extract core words, converting text semantics into word vector data, tagging by transaction type, applicable scenario, and release time, and establishing an association index between policy clauses. The core words include "rolling matching", "price priority" and "new energy grid connection". Policy retrieval: The MCP server intelligent agent group calls the retrieval interface of the policy document library through the business API group, matches the target policy clauses based on full-text search or semantic vector search, and returns structured results containing original text fragments, core clause summaries and related policy recommendations; Cross-version comparison: The policy document library stores policy versions from different periods, supports difference analysis of different versions of the same policy, generates policy change reports, and indicates the potential impact of policy changes on power trading strategies.

9. The power trading auxiliary decision-making method based on MCP according to claim 8, characterized in that, The format of the unified response result mentioned in step S4 includes at least one of the following: visual charts, structured tables, and natural language reports; The visualization charts include an electricity price forecast trend chart and a bar chart of electricity consumption execution. The structured tables include a detailed table of bilateral contract electricity volume execution and a table of policy compliance verification results; The natural language reports include transaction decision analysis reports and risk assessment reports, which contain data sources, model basis, and policy references.

10. The power trading auxiliary decision-making method based on MCP according to claim 9, characterized in that, The MCP server intelligent agent group also includes a user profile intelligent agent, which performs the following operations in step S3: The user profiling agent obtains users' historical transaction behavior data and risk preference data through the business API group; User profiles are generated based on the historical transaction behavior data and risk preference data. The user profiles include transaction habit tags and risk tolerance level tags. The user profile is synchronized to the decision support agent, which adjusts the content and form of decision suggestions based on the user profile to provide personalized decision support.

Images