Multi-agent cooperative energy production demand generation method and system

Abstract

The invention provides a multi-agent cooperative energy production demand generation method and system, and relates to the field of intelligent energy, and the method comprises the steps: collecting multi-modal data of a to-be-generated energy field, and carrying out the preprocessing; performing data analysis and information purification by utilizing a multi-modal information processing agent driven by a large model, taking an input energy production demand target as a cue word Prompt of the agent and taking the preprocessed multi-modal data as a memory bank of the agent, and generating energy production demand scheme items in a discrete token sequence form; automatically generating an energy production demand reference document based on the energy production demand scheme items through a large model-driven iterative dialogue between a session manager and an energy expert agent; according to the method, a generative multi-agent technology is introduced, the designed agents autonomously take actions according to information purification and answers of other agents, and energy production requirements are quickly and efficiently generated through own observation, reasoning and memory capabilities.

Description

Technical Field

[0001] The present disclosure relates to the field of smart energy, and in particular to a method and system for generating energy production demand through multi-agent collaboration. Background Art

[0002] Energy production demand generation is a demand model development process involving information such as policy documents, expert knowledge, and market dynamics. Traditional demand generation methods often rely on human experience, resulting in a lack of objectivity. In addition, when faced with multimodal information such as video, audio, and text, these traditional methods are inefficient in integrating and analyzing large amounts of heterogeneous data and cannot meet the requirements of intelligent energy structures. Moreover, the traditional generation process relies heavily on the professional level of energy experts, making it difficult to effectively guarantee the completeness and accuracy of the demand. These problems directly affect the efficiency of energy production demand generation, so there is an urgent need to introduce innovative methods.

[0003] With the rapid development of artificial intelligence, the emergence and improvement of multimodal large models have provided new opportunities to solve the above challenges. These large models have a rich knowledge base and can provide strong support for the generation of energy production demand. Nowadays, these large models have achieved remarkable results in a wide range of applications in various fields, especially in the generation of energy production demand, showing significant potential.

[0004] However, relying solely on large models to generate energy production demand still faces some difficulties: first, the natural language input standards in the energy field are high, and different inputs may lead to inconsistent results; second, large models rely on a large amount of historical data and long-term training, and they may not be able to adjust in time to adapt to new situations when data is scarce. Therefore, in emergencies or dynamic environments, large models often cannot accurately understand and respond. In addition, in specific energy adjustment tasks, using large models alone may result in answers that are not deep enough or professional enough.

[0005] Therefore, the existing energy production demand generation scheme has problems of information isolation and delayed response, resulting in unstable quality, poor adaptability and insufficient robustness. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a method and system for generating energy production demand through multi-agent collaboration, introduces generative multi-agent technology, designs agents to take autonomous actions based on information purification and the answers of other agents, and generates energy production demand quickly and efficiently through its own observation, reasoning and memory abilities.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] A multi-agent collaborative energy production demand generation method, comprising:

[0009] Collect multimodal data in the energy field to be generated and perform preprocessing;

[0010] Using a multimodal information processing agent driven by a large model, the input energy production demand target is used as the prompt of the agent, and the pre-processed multimodal data is used as the memory bank of the agent to perform data analysis and information purification, and generate energy production demand plan items in the form of discrete token sequences;

[0011] Based on the energy production demand plan items, the energy production demand reference document is automatically generated through iterative dialogue between the conversation manager and the energy expert agent driven by the big model.

[0012] According to some embodiments, the present disclosure adopts the following technical solutions:

[0013] A multi-agent collaborative energy production demand generation system, comprising:

[0014] The data collection module is configured to: collect multimodal data in the field of energy to be generated and perform preprocessing;

[0015] The information processing module is configured to: utilize a multimodal information processing agent driven by a large model, use the input energy production demand target as the prompt word of the agent, and the pre-processed multimodal data as the memory bank of the agent to perform data analysis and information purification, and generate energy production demand plan items in the form of discrete token sequences;

[0016] The document generation module is configured to automatically generate energy production demand reference documents based on energy production demand plan items through iterative dialogues between conversation managers and energy expert agents driven by a large model.

[0017] According to some embodiments, the present disclosure adopts the following technical solutions:

[0018] A computer program product includes a computer program, which, when executed by a processor, implements the method for generating energy production demand through multi-agent collaboration.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for generating energy production demand through multi-agent collaboration is implemented.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the method for generating energy production demand through multi-agent collaboration.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. The present invention realizes the integration and conversion of information from different sources by comprehensively applying multimodal data processing technology, forms a unified processing standard, thereby enhancing the system's ability to understand complex information and making the generation process of energy production demand more accurate. The traditional energy production demand generation process often encounters data island problems, and it is difficult to obtain comprehensive information support; the present invention constructs a multimodal information processing intelligent agent to effectively convert data in various forms such as text, images, and numbers into an operational token sequence, captures more contextual information and implicit relationships, and lays the foundation for subsequent reasoning and decision-making; in addition, the present invention uses energy big data to bring together information such as energy feature types, historical collection data, new demand samples, policy documents, company regulations, expert knowledge and market dynamics. Through in-depth mining and analysis of these extensive data, the model can understand the complexity and diversity required for energy production in a larger information dimension. This comprehensive data collection enables the demand generation system to have stronger adaptability and can quickly adjust demand generation strategies according to real-time changing market dynamics and policy requirements; combining these innovations, the present invention not only improves the scientificity and practicality of demand generation, ensuring that it can better meet actual energy production needs, but also provides more extensive information support for energy production demand reference documents through the combination of comprehensiveness and real-time data, ensuring that the generated demand content not only meets the current technical level, but also adapts to future market development trends.

[0025] 2. The present invention introduces a multi-agent collaboration mechanism to achieve a flexible division of labor for different roles in the process of information acquisition and demand generation, and gives full play to the unique advantages of each agent. Different agents have the ability of independent innovation, and can not only process information independently, but also work together to achieve in-depth discussion and comprehensive analysis. In this way, not only can demands be responded to quickly, but also high-quality demand content can be effectively formed in simulated expert discussion scenarios. Especially during interaction, large models and agents with memory capabilities can perform effective reasoning and analysis, continuously improve and correct generated demands, and provide real-time feedback support for decision-making. This multi-agent collaboration mechanism improves the overall system efficiency, ensures that the generated production demands are more practical, and also demonstrates the broad potential for the application of agent technology in the energy field. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.

[0027] Figure 1 This is a flow chart of the method of Example 1.

[0028] Figure 2 A model framework diagram for generating a reference document for energy production requirements of multi-agent collaboration in Example 1.

[0029] Figure 3 This is the energy expert agent analysis flow chart of Example 1. DETAILED DESCRIPTION

[0030] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0031] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0032] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0033] In order to solve the problems of information isolation and delayed response in existing technologies, the academic community has proposed a multi-agent collaboration framework driven by a large model. This method can not only make full use of the excellent language comprehension ability of the large model, but also effectively overcome the limitations of a single large model in specific application scenarios through the distributed decision-making mechanism of multiple agents; the interaction and collaboration of agents can simulate the behavior and collective wisdom of real society, and provide a new perspective for solving complex problems; in addition, distributing decisions to multiple agents makes it easier to understand and track the decision logic of the system, thereby improving the transparency and credibility of the system; in the energy field, the transparency and reliability of decisions are closely related to the quality and safety of decisions; the decentralized decision-making model of multiple agents not only enhances the system's ability to adapt to new situations, but also ensures the accuracy of decisions in emergency situations.

[0034] In view of the above analysis, the present invention proposes a method and system for generating energy production demand with multi-agent collaboration. By building a multi-agent framework, integrating "multimodal information processing agents" and "energy expert agents", diversified data and large-scale energy data are effectively converged to form a dynamic information sharing network; multimodal information processing agents are used to process and analyze various data forms such as text, images, and numerical values ​​to ensure the comprehensiveness and accuracy of information. On this basis, the energy expert agent has the ability to observe, reason and make decisions through comprehensive analysis of the information obtained to form accurate energy production demand; the agents work together to optimize the demand generation process with the help of mutual feedback and tool calls, and finally output high-quality production demand reference documents; the application of this method can not only greatly improve the efficiency and flexibility of energy demand generation, but also promote the intelligence and refinement of resource allocation, and achieve rapid response to policy adjustments and market changes.

[0035] Example 1

[0036] In one embodiment of the present disclosure, a method for generating energy production demand by multi-agent collaboration is provided, comprising:

[0037] Step S1: Collect multimodal data in the energy field to be generated and perform preprocessing.

[0038] Furthermore, the multimodal data includes measurement indicator data, policy documents, company regulations, meteorological data, holiday data and various social information in the form of text, audio and video;

[0039] The preprocessing includes data cleaning, data deletion, data completion and normalization.

[0040] Step S2: Utilize a multimodal information processing agent driven by a large model, use the input energy production demand target as the agent's prompt and the preprocessed multimodal data as the agent's memory bank to perform data analysis and information purification, and generate energy production demand plan entries in the form of discrete token sequences.

[0041] Furthermore, the multimodal information processing agent specifically processes as follows:

[0042] Transcribe non-text data in the form of audio and video into text data and integrate the transcribed multimodal data;

[0043] Reasoning and analyzing the integrated multimodal data to extract key information related to energy production;

[0044] Based on the extracted key information, a large language model is used to generate energy production demand plan items in the form of discrete token sequences through real-time communication through a dynamic interaction mechanism.

[0045] Step S3: Based on the energy production demand plan items, the energy production demand reference document is automatically generated through iterative dialogue between the conversation manager and the energy expert agent driven by the big model.

[0046] Furthermore, the conversation manager, driven by the big model, is responsible for providing necessary background information and prompts based on the energy production demand plan items, coordinating the conversation process to guide the conversation towards a clear goal direction, using the state tracking algorithm to manage the conversation state, and defining the state transition probability of the conversation round:

[0047]

[0048] Among them, P(s t |s t-1 ,a t ) means that given the previous state s t-1 and the action taken t Under the circumstances, the current state s is reached t conditional probability to effectively coordinate the dialogue process and meet the needs.

[0049] Furthermore, the energy expert agent obtains the latest news related to the energy field through Internet search engines, uses a large language model to deeply understand the integrated multimodal data and the latest news, provides more detailed information for the dialogue topic, promotes dialogue generation through prompt words during iterative dialogues, and automatically generates in-depth dialogue texts and analysis opinions based on energy production demand plan items.

[0050] Furthermore, in the iterative dialogue process, a main loop iteration mechanism is set to dynamically adjust the generated content according to the deviation coefficient of the dialogue, which is expressed by the formula:

[0051] Adjustment = α Deviation

[0052] Among them, Adjustment is a quantitative representation of the adjustment made by the agent based on the deviation coefficient Deviation. The deviation coefficient Deviation is the gap between the current generated response and the actual demand. α is the adjustment coefficient, which is used to control the amplitude of content adjustment.

[0053] As an embodiment, the present invention discloses a method for generating energy production demand by multi-agent collaboration, introduces generative multi-agent technology, designs agents to take autonomous actions based on information purification and the answers of other agents, and quickly and efficiently generates energy production demand through its own observation, reasoning and memory capabilities. The method is implemented based on the energy big data of a power company of the State Grid Corporation of China. The data collection link includes input and output information connected to the peripheral system. The specific implementation process is as follows:

[0054] A. Extract electricity consumption characteristics, historical data and new electricity demand type sample data through the Energy Big Data Center, collect policy documents, expert knowledge, market dynamics and other information in the energy field, and summarize the data;

[0055] The electricity consumption information collection system collects more than 300,000 electricity consumption samples from July 2021 to June 2023 in a certain city. The corresponding users obtain metering index data, imported policy documents, company regulations, methods, etc. from the electricity consumption information collection system. In addition, meteorological data is obtained from the meteorological system, combined with holiday data and various social information crawled from the Internet for collection and aggregation. The initial selection of some feature data examples is shown in Table 1:

[0056] Table 1 Characteristic data table

[0057]

[0058]

[0059]

[0060] The data in the above table are multimodal data, which can be summarized as measurement indicator data, policy documents, company regulations, meteorological data, holiday data and various social information in the energy field to be generated, especially social conditions and policy documents, which may be audio and video data.

[0061] B. Perform data preprocessing on the aggregated multimodal data, including data cleaning, data deletion, data completion and normalization.

[0062] C. Determine energy production demand targets and module statements.

[0063] Build as Figure 2 The generative model shown is used to generate energy production demand reference documents, and the generative model includes an input module, a multi-agent module and a tool library module.

[0064] The input module is responsible for receiving energy production demand targets and multimodal data. The input energy production demand targets are questions from users about energy needs, such as "Please write a plan for the city's peak summer electricity production based on this year's relevant policies and market environment."

[0065] The multi-agent module is composed of multiple agents (multimodal information processing agents and energy expert agents) and has the ability to collaborate to complete tasks, including information acquisition, in-depth discussions, and the generation of requirement lists.

[0066] The tool library module contains multiple independent functions, giving each agent the ability to innovate independently, thereby improving the effectiveness of the collective in complex information environments.

[0067] D. In the multi-agent framework, we first build a multimodal information processing agent, which incorporates multimodal data processing capabilities and can convert multimodal data into discrete token sequences. Then, we use a large model to understand and reason about these token sequences, and store the relevant knowledge in the agent's memory, so that it has the ability to conduct question-answering dialogues.

[0068] In step D, a multimodal information processing agent is constructed to simulate personnel who organize information such as relevant policies, regulations, documents, and news. Driven by a large model, it can obtain and process text information, video, and audio, and has thinking and memory functions. It can reason and analyze relevant data and generate preliminary information required for the energy production demand reference document, that is, energy production demand plan items in the form of discrete token sequences. The specific process is as follows:

[0069] D1. By building an efficient text data import and processing module, it aims to load and parse text data in a structured manner.

[0070] This module can access users' electricity consumption data, energy consumption calculation data and other related multimodal information, and perform preliminary cleaning, normalization and preliminary analysis on the original data, thus providing a solid foundation for subsequent analysis.

[0071] D2. Multimodal information processing agents use deep learning technology to achieve efficient automatic transcription capabilities.

[0072] For audio signals, the decoding-encoder structure based on the recurrent neural network (RNN) divides the continuous audio signal into small segments or frames, and then extracts features for each frame to capture the key speech information in the audio signal; subsequently, these features are input into the RNN decoding-encoder model; a fixed-length context vector containing the core information of the input audio frame is generated, which is processed by the decoder to gradually generate the corresponding text sequence.

[0073] When processing video content, the multimodal information processing agent not only relies on the decoder and encoder of the audio signal, but also parses the visual information of the video through deep learning technology. Specifically, the agent uses convolutional neural networks (CNN) to extract key visual features from video frames. These features capture information such as objects, scenes, and actions in the video. Subsequently, these visual features are input into a sequence-to-sequence (Seq2Seq) model together with the text information transcribed from the audio. The model can convert mixed audio-visual input sequences into coherent text outputs. The mathematical expression of this process is similar to that of audio signal processing, that is, the conditional probability of generating an output sequence y when a given input sequence (including visual and audio features) is calculated, and the formula is expressed as:

[0074]

[0075] Among them, P(y|x) represents the conditional probability of generating the output sequence y given the input sequence x, where the output sequence with the highest probability is taken as the text transcription result, y t is the tth element of the output sequence, and T is the total time steps of the output.

[0076] This process aims to eliminate the barriers of non-text information and ensure the smooth progress of subsequent analysis; by inputting a sequence of non-text information about electricity consumption (such as meeting minutes, expert interviews, etc.), and calculating the probability of maximizing the output sequence to obtain the best conversion into analyzable text transcription results, this step ensures that information from various sources is included in the analysis scope and achieves comprehensiveness of information.

[0077] D3. In the process of integrating electricity-related text data and structured information, the multimodal information processing agent constructs a systematic dataset to support subsequent data analysis and reasoning; the input includes transcribed text data, such as meeting minutes and expert interviews, and structured information, such as electricity consumption pattern analysis and real-time electricity prices, and the output is an optimized and weighted dataset that can improve the efficiency and relevance of data aggregation and provide a solid data foundation for subsequent analysis, prediction and decision-making.

[0078] Specifically, the agent integrates data from different sources by building a similarity graph. In this graph, nodes represent data points in the data set, and edges between nodes represent similarities between data points, which can be determined based on indicators such as power consumption patterns, time series characteristics, or electricity price fluctuations. In this way, the similarity graph helps the agent identify and integrate data from different sources and achieve aggregation of multimodal data. This step generates a highly comprehensive data set to support subsequent data analysis and reasoning. A graph-based clustering algorithm is used to identify and integrate data from different sources by building a similarity graph, which is expressed as:

[0079]

[0080] Modularity (Q) represents the network modularity, which is used to quantify or judge the quality of complex network community division. ij is the similarity matrix, k i and k j are the degrees of nodes i and j respectively, m is the total number of edges, δ(c i ,c j ) is the Kroneckerdelta function, which determines the affiliation of node c; the higher the value of network modularity, the tighter the connection within the community, and the sparser the connection between communities, which means the better the effect of community division.

[0081] By determining the community division through the above calculations, the intelligent agent can identify groups with similar energy usage characteristics in the dataset. These groups may represent specific energy usage behavior patterns or market trends. Such divisions help the intelligent agent to more accurately analyze and predict energy demand, thereby providing more targeted data support for the formulation of energy production demand plans.

[0082] D4. Create in-memory vector storage to vectorize the acquired electricity-related text and multimodal information for fast retrieval and processing.

[0083] D5. The multimodal information processing agent uses its built-in thinking and memory functions to reason and analyze the acquired electricity consumption data and extract key information related to energy production. For example, based on the historical electricity consumption pattern and current electricity price, it determines the peak period of electricity consumption. Based on the Bayesian reasoning method, it is expressed by the following formula:

[0084]

[0085] Where H represents the hypothesis of "high electricity demand", E is the observed evidence (such as real-time electricity price changes), P(H|E) represents the posterior probability of hypothesis H given the evidence E, P(E|H) is the likelihood of the evidence under hypothesis H, P(H) is the prior probability of the hypothesis, and P(E) is the marginal probability of the evidence. By calculating the posterior probability, the intelligent agent can provide more reasonable energy usage recommendations, thereby supporting decision-making and optimization of energy production.

[0086] D6. Based on the analysis results, the agent generates preliminary energy production demand plan entries (such as electricity demand forecasts for a specific time period) to provide a basis for subsequent decision-making, which can effectively meet energy demand and optimize power distribution. In this step, the agent converts the key information obtained from the analysis into discrete token sequences, which represent the required knowledge for the preliminary energy production demand plan; by converting complex energy data and demand into standardized token form, the agent can more effectively process and compare different energy demand plans, thereby providing more accurate and structured data support for decision-making.

[0087] Specifically, in the process of generating demand plan items, a dynamic interaction mechanism is established between users and intelligent agents, through which real-time communication is carried out to ask questions or request further information to optimize the quality and relevance of demand items. This interaction can improve the quality of demand generation, make it more in line with actual operational needs, and ensure that the generation of token sequences can reflect the latest energy market dynamics and policy changes.

[0088] D7. After receiving user feedback, the agent will perform targeted iterative optimization to enrich and improve the knowledge required for the energy production demand plan. For example, after using the initially generated energy production demand plan entries, user feedback pointed out that the current plan failed to fully consider the impact of the new environmental protection regulations that will be implemented on energy use. The agent will iteratively optimize its required knowledge, add the specific content of the new regulations and analyze their potential impact on energy demand, so as to better reflect these changes in future energy production demand plans.

[0089] The intelligent agent continuously adjusts and optimizes the plan by simulating the process of natural selection and genetic variation. This process is similar to biological evolution, in which the best plan is retained and used to generate a new generation of plans, thereby gradually improving the overall performance of the plan. In this process, the dynamic adjustment of the token sequence ensures that the energy production demand plan is knowledge-rich and can flexibly respond to changes in the external environment.

[0090] The multimodal information processing agent can act as an assistant and can conduct in-depth discussions and Q&A exchanges on energy information content; at the same time, the agent also has information analysis capabilities and can independently aggregate and summarize multimodal data. In addition, to ensure the reliability of the agent, this embodiment imposes corresponding constraints on it so that it can only make decisions based on the collected multimodal information sources, and prohibits it from fabricating and making up answers to unknown information, thereby effectively reducing the risk of the agent generating hallucinations.

[0091] E. To simulate the expert demand generation discussion scenario, an energy expert agent was designed to communicate through natural language dialogue. Before the agent dialogue begins, a manager driven by the big model is introduced to provide necessary background information and prompts. The energy expert agent uses the big model to observe, analyze and reason about information in order to form opinions and draw conclusions in the discussion. At the same time, the agent has the ability to autonomously call tools in the tool library to support its thinking and decision-making process. Within this framework, multi-agents can not only understand and process complex multimodal information, but also effectively generate high-quality demand content in an environment that simulates expert discussions.

[0092] The energy expert agent driven by the glm-4v large model has the ability to operate autonomously, such as Figure 3 As shown in the figure, through the iterative dialogue of the intelligent agent, a reference document for energy production demand is automatically generated; the generative intelligent agent iterative dialogue model designed in this embodiment enables the intelligent agent to autonomously generate prompt words and dialogue content according to the initial instructions. In the iterative process, the interaction of prompt words promotes the deepening of the dialogue, and then automatically forms a more detailed dialogue text and analysis opinions based on the demand items generated in step D, that is, the reference document for energy production demand. The specific process is as follows:

[0093] E1. To ensure the efficiency of agent interaction and promote effective coordination between agents at different stages, an efficient conversation manager mechanism is designed. This mechanism aims to provide necessary background information and prompts to guide the conversation towards a clear target direction, covering information such as electricity demand, energy production status and real-time policy environment; by extracting contextual information from multiple data sources and adaptively evaluating feedback and the evolution of conversation content, the agent can effectively predict the future state of the conversation.

[0094] Specifically, we design a conversation manager and establish an efficient conversation management mechanism to provide necessary background information and prompts for the conversation, so as to optimize the interaction quality of the agents and the collaboration between different agents. This mechanism is designed to ensure the fluency and goal clarity of the conversation, thereby improving the efficiency of communication. To this end, we can use a state tracking algorithm to manage the conversation state by defining the state transition probability of the conversation turn:

[0095]

[0096] Among them, P(s t |s t-1 ,a t ) means that given the previous state s t-1 and the action taken t Under the circumstances, the current state s is reached t conditional probability to effectively coordinate the dialogue process and meet the needs.

[0097] E2. Functional expansion of the intelligent agent, including designing a mechanism that integrates multiple processing capabilities, such as data analysis modules, real-time electricity price query tools, and electricity consumption pattern analysis tools, to enhance its information processing capabilities in complex tasks. This mechanism enables the intelligent agent to flexibly respond to diverse task scenarios and improve the adaptability of the system.

[0098] Specifically, by introducing a mechanism with diversified processing capabilities, the information processing ability of the agent in complex tasks is enhanced so that it can respond to diverse task scenarios more flexibly and improve the adaptability of the system. A learning algorithm based on a recurrent neural network (RNN) can be combined to process and generate dialogue content. Its basic form is:

[0099] h t =f(W h h t-1 +W x x t +b)

[0100] Among them, h t is the current hidden state, i.e. the “memory” at time t, x t is the current input, W h and W x is the weight matrix, b is the bias term, and is the activation function (such as tanh or ReLU).

[0101] E3. Design an energy expert agent driven by the glm-4v large model to complete the generation of energy production demand reference documents. In the agent dialogue, the energy expert agent generates opinion O at time t1. t1 , the energy expert agent generates opinion O at time t2 t2 In order to effectively integrate the dialogue evolution process over time, a dynamic knowledge base integrating the time dimension is formed, and the fusion information set M is constructed. t As shown in the formula:

[0102]

[0103] Among them, the view {O t1 ,O t2 The relationship between} represents different views or supplementary information on the same topic or issue over time. t The key insights and information from the dialogue process are accumulated over time, providing continuity and depth for the generation of the "Energy Production Demand Reference Document"; such information integration ensures that the document can be based on previously discussed content, making the final reference document more comprehensive and accurate, reflecting the accumulation of knowledge over time.

[0104] Develop energy expert agents and generate professional conversations related to energy production needs based on them. Through in-depth integration of domain-specific knowledge, ensure that the content generated by the conversations is both professional and practical.

[0105] E4. Energy expert agents will integrate information from electricity consumption, market trends, and policy documents to provide more detailed information for the topic of the conversation, and enable the agent to autonomously generate prompts and dialogues based on the initial instructions, and mutually promote dialogue generation through prompts during iterative dialogues, and automatically generate in-depth dialogue texts and analytical opinions based on demand items. This process ensures that the generated content not only closely follows real-time needs, but also pays attention to the dynamic changes and policy orientation of the current energy market.

[0106] Guide the intelligent agent to deeply integrate thematic information and enrich the content of the dialogue. By accessing background information such as policy documents, news summaries and energy market data, ensure that the generated content is not only in line with actual needs but also has deep professional knowledge.

[0107] E5. By setting a main loop iteration mechanism, the agent can continuously adjust and optimize the generated content according to the deviation coefficient of the dialogue, thereby promoting continuous information updating and demand discovery; this deviation coefficient is used to quantify the difference between the generated content and expectations, thereby driving continuous information updating and demand discovery. This loop mechanism ensures that the agent can adapt to feedback information and flexibly adjust its strategy. The adjustment mechanism can be expressed as:

[0108] Adjustment = α Deviation

[0109] Among them, Adjustment is a quantitative representation of the agent's adjustment based on the deviation coefficient Deviation, which is part of the agent's self-optimization mechanism. Deviation refers to the gap between the current generated response and the actual demand, that is, the deviation coefficient. α is an adjustment coefficient used to control the amplitude of content adjustment. When the deviation is large, the agent will increase the adaptability of the output content according to the formulated adjustment strategy.

[0110] E6. In order to ensure that the intelligent agent can absorb new information in a timely manner and maintain the timeliness and accuracy of the memory bank, it is necessary to introduce a knowledge updating mechanism. This mechanism regularly updates the intelligent agent's built-in memory bank, including the latest energy policies, market dynamics and feedback information. This will significantly improve the intelligent agent's response quality and support capabilities in subsequent dialogues, ensure the timeliness and accuracy of its memory bank, and provide support for subsequent dialogues.

[0111] The energy expert agent can obtain the latest news related to the energy field through Internet search engines and form opinions based on this to provide rich background information. In addition, it can call on resources such as open source papers, policy documents and company regulations to quickly obtain relevant materials. The agent uses a large language model to deeply understand the content of the literature and master knowledge on topics such as energy adjustment, carbon emissions and carbon inclusiveness, so as to answer professional questions in a targeted manner. In information retrieval, the agent shows excellent problem analysis ability and sensitivity to data content, which enables it to answer questions efficiently and draw conclusions quickly. At the same time, when performing professional retrieval, it pays attention to user privacy and ensures that browsing history and IP addresses are not leaked to meet the need for confidentiality of energy data.

[0112] The method of this embodiment is compared with the existing technologies GPT-4o and Claude 3.5. As shown in Table 2, the user's question about energy demand is "Please write a plan for the city's peak summer electricity production based on this year's relevant policies and market environment." The energy production demand reference documents generated by the three methods are:

[0113] Table 2 Effect comparison

[0114]

[0115]

[0116] It can be seen that the solutions provided by GPT-4o and Claude 3.5 both emphasize the importance of relevant measures. However, this embodiment not only automatically identifies the year and geographical location, but also actively introduces the local policy background, which are not reflected in the first two models. In addition, this embodiment automatically expands the relevant information by deeply analyzing the current electricity consumption status of the city, and provides a more accurate and detailed energy production plan. This ability shows the advantages of this embodiment in understanding and processing complex information and generating highly targeted documents, so that the documents it generates are more outstanding in professionalism, accuracy and practicality. Therefore, the method of this embodiment performs better in generating high-quality energy production plan documents, and can provide stronger support for actual energy management and policy making.

[0117] Example 2

[0118] In one embodiment of the present disclosure, a multi-agent collaborative energy production demand generation system is provided, comprising:

[0119] The data collection module is configured to: collect multimodal data in the field of energy to be generated and perform preprocessing;

[0120] The information processing module is configured to: utilize a multimodal information processing agent driven by a large model, use the input energy production demand target as the prompt word of the agent, and the pre-processed multimodal data as the memory bank of the agent to perform data analysis and information purification, and generate energy production demand plan items in the form of discrete token sequences;

[0121] The document generation module is configured to automatically generate energy production demand reference documents based on energy production demand plan items through iterative dialogues between conversation managers and energy expert agents driven by a large model.

[0122] Example 3

[0123] In one embodiment of the present disclosure, a computer program product is provided, including a computer program, which, when executed by a processor, implements the method for generating energy production demand through multi-agent collaboration.

[0124] Example 4

[0125] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method for generating energy production demand through multi-agent collaboration is implemented.

[0126] Example 5

[0127] In one embodiment of the present disclosure, an electronic device is provided, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the method for generating energy production demand through multi-agent collaboration.

[0128] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0130] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A method for generating energy production demand by multi-agent collaboration, characterized in that: include: Collect multimodal data in the energy field to be generated and perform preprocessing; Using a multimodal information processing agent driven by a large model, the input energy production demand target is used as the prompt of the agent, and the pre-processed multimodal data is used as the memory bank of the agent to perform data analysis and information purification, and generate energy production demand plan items in the form of discrete token sequences; Based on the energy production demand plan items, the energy production demand reference document is automatically generated through iterative dialogue between the conversation manager and the energy expert agent driven by the big model.

2. A method for generating energy production demand by multi-agent collaboration as claimed in claim 1, characterized in that: The multimodal data includes measurement indicator data, policy documents, company regulations, meteorological data, holiday data and various social information in the form of text, audio and video; The preprocessing includes data cleaning, data deletion, data completion and normalization.

3. A method for generating energy production demand by multi-agent collaboration as claimed in claim 1, characterized in that: The multimodal information processing agent specifically processes as follows: Transcribe non-text data in the form of audio and video into text data and integrate the transcribed multimodal data; Reasoning and analyzing the integrated multimodal data to extract key information related to energy production; Based on the extracted key information, a large language model is used to generate energy production demand plan items in the form of discrete token sequences through real-time communication through a dynamic interaction mechanism.

4. The method for generating energy production demand by multi-agent collaboration according to claim 1, characterized in that: The conversation manager, driven by the big model, is responsible for providing necessary background information and prompts based on the energy production demand plan entries, coordinating the conversation process to guide the conversation towards a clear goal, using the state tracking algorithm to manage the conversation state, and defining the state transition probability of the conversation round: Among them, P(s t |s t-1 ,a t ) means that given the previous state s t-1 and the action taken t Under the circumstances, the current state s is reached t conditional probability to effectively coordinate the dialogue process and meet the needs.

5. The method for generating energy production demand by multi-agent collaboration according to claim 1, characterized in that: The energy expert agent obtains the latest news related to the energy field through Internet search engines, uses a large language model to deeply understand the integrated multimodal data and the latest news, provides more detailed information for the dialogue topic, promotes dialogue generation through prompt words during iterative dialogues, and automatically generates in-depth dialogue texts and analysis opinions based on energy production demand plan items.

6. A method for generating energy production demand by multi-agent collaboration as claimed in claim 1, characterized in that: In the iterative dialogue process, a main loop iteration mechanism is set to dynamically adjust the generated content according to the deviation coefficient of the dialogue, which is expressed as follows: Adjustment = α Deviation Among them, Adjustment is a quantitative representation of the adjustment made by the agent based on the deviation coefficient Deviation. The deviation coefficient Deviation is the gap between the current generated response and the actual demand. α is the adjustment coefficient, which is used to control the amplitude of content adjustment.

7. A multi-agent collaborative energy production demand generation system, characterized in that: include: The data collection module is configured to: collect multimodal data in the energy field to be generated and perform preprocessing; The information processing module is configured to: utilize a multimodal information processing agent driven by a large model, use the input energy production demand target as the prompt word of the agent, and the pre-processed multimodal data as the memory bank of the agent to perform data analysis and information purification, and generate energy production demand plan items in the form of discrete token sequences; The document generation module is configured to automatically generate energy production demand reference documents based on energy production demand plan items through iterative dialogues between conversation managers and energy expert agents driven by a large model.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the method for generating energy production demand through multi-agent collaboration as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by the processor, a method for generating energy production demand through multi-agent collaboration as described in any one of claims 1-6 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement a multi-agent collaborative energy production demand generation method as described in any one of claims 1-6.

Images