Model identification method and system based on multi-agent

Abstract

The present invention provides a model identification method and system based on multi-agents, the method comprising: when the algorithm model for executing a target task needs to be updated, the training data processed by the data preprocessing agent is input into multiple algorithm model learning agents, and the models to be screened for executing the target task are output by several algorithm model learning agents among the multiple algorithm model learning agents; each model to be screened is input into a model decision agent, and the best algorithm model for executing the target task is output by the model decision agent. Compared with traditional physical mechanism modeling, this method improves the modeling speed and model accuracy; in addition, this method adopts an agent combined with a data-driven approach, and does not require the staff to preprocess the collected HVAC system data, nor does it require the staff to manually build an artificial intelligence model, which reduces the professional threshold of engineering personnel and improves the automation and intelligence level of the system.

Description

Technical Field

[0001] This invention relates to the field of building energy conservation technology, and in particular to a model identification method and system based on multi-agent intelligence. Background Technology

[0002] Energy consumption is related to global climate change. With economic development, building energy consumption accounts for an increasingly larger proportion of total energy consumption. In my country, building energy consumption accounts for about 23% of social energy consumption, and related carbon emissions account for about 22% of total social carbon emissions.

[0003] In building energy consumption, HVAC systems are the largest energy consumers in large public buildings, accounting for 50%-60% of the total electricity consumption. Therefore, research on energy conservation and emission reduction in HVAC systems is of great significance for energy conservation and emission reduction.

[0004] The foundation of HVAC system optimization lies in establishing accurate mathematical models of the equipment and systems; only accurate mathematical models can guarantee the accuracy of optimization results. Existing industry modeling methods for HVAC equipment and systems can be mainly divided into physics-based mechanism modeling methods and data-driven model identification methods. Physics-based mechanism modeling methods offer strong interpretability and broad coverage of operating conditions; however, due to simplifications during model construction, certain errors are unavoidable. Data-driven model identification is an emerging field in control science. It refers to control theories and methods that do not include mathematical model information of the controlled process, but only utilize online or offline monitoring data of the controlled system and knowledge gained through data learning. Under certain assumptions, it achieves convergence, stability guarantees, and robustness conclusions. However, this data-driven model identification method requires a high level of skill from personnel; general engineers lacking HVAC and control expertise often find it difficult to perform this work.

[0005] It is evident that existing data modeling methods for HVAC systems and equipment either fail to accurately map the characteristics of the equipment and systems, or have high professional barriers, requiring intervention from professionals, and lack sufficient levels of intelligence and automation. Summary of the Invention

[0006] This invention provides a model identification method and system based on multi-agent technology to solve the problems of insufficient modeling accuracy caused by model simplification in the prior art, or the application defects that require professional engineers to perform. It can improve the modeling accuracy while reducing the professional technical threshold, thereby reducing labor costs.

[0007] This invention provides a model identification method based on multiple agents, comprising the following steps.

[0008] When the algorithm model for performing the target task needs to be updated, the training data processed by the data preprocessing agent is input into multiple algorithm model learning agents to obtain the candidate models for performing the target task output by several of the multiple algorithm model learning agents. Each of the models to be screened is input into the model decision agent to obtain the optimal algorithm model output by the model decision agent for performing the target task; The data preprocessing agent is used to collect real-time data from the site and preprocess the real-time data to obtain the training data. Each of the algorithm model learning agents is used to determine whether to select a training model for performing the target task. If selected, the training model is trained based on the training data to obtain a model to be selected.

[0009] According to the multi-agent model identification method provided by the present invention, before the step of inputting the training data processed by the data preprocessing agent into multiple algorithm model learning agents to obtain the candidate models for performing the target task output by several of the multiple algorithm model learning agents when the algorithm model for performing the target task needs to be updated, the method further includes: The training status information after the instruction forwarding agent makes a decision is sent to the data preprocessing agent and the multiple algorithm model learning agents respectively; the training status information is used to indicate that the algorithm model needs to be updated. The instruction forwarding agent makes decisions based on at least one of the following methods: In response to a user-inputted model update command, a decision is made as to whether the algorithm model needs to be updated; In response to periodically updated information, a decision is made as to whether the algorithm model needs to be updated.

[0010] According to the multi-agent model identification method provided by the present invention, before the training state information after the instruction forwarding agent makes a decision is respectively sent to the data preprocessing agent and the multiple algorithm model learning agents, the method further includes: The training configuration agent obtains the training configuration information input by the user; the training configuration information includes the periodic update information; the training configuration agent is used to determine whether the algorithm model needs to be updated based on the periodic update information, and if so, sends the periodic update information to the instruction forwarding agent; The periodic update information of the training configuration agent is sent to the instruction forwarding agent.

[0011] According to the multi-agent model identification method provided by the present invention, before the step of inputting the training data processed by the data preprocessing agent into multiple algorithm model learning agents to obtain the candidate models for performing the target task output by several of the multiple algorithm model learning agents when the algorithm model for performing the target task needs to be updated, the method further includes: The training configuration agent obtains training configuration information input by the user; the training configuration information includes model selection information, which is used by the algorithm model learning agent to determine whether to select the training model for itself to perform the target task; The model selection information is sent to the multiple algorithm model learning agents.

[0012] According to the multi-agent-based model identification method provided by the present invention, the method further includes: The model decision agent is further configured to determine the ranking results of each of the models to be screened. After inputting each of the models to be screened into the model decision agent and obtaining the optimal algorithm model output by the model decision agent for performing the target task, the method further includes: The ranking results of the model decision-making agent are input into the plurality of algorithm model learning agents; each of the algorithm model learning agents is also used to learn again based on the ranking results. This invention also provides a multi-agent model identification system, comprising a data preprocessing agent, multiple algorithm model learning agents, and a model decision-making agent; wherein, The data preprocessing agent is used to collect real-time data from the field, preprocess the real-time data to obtain the training data, and input the processed training data into multiple algorithm model learning agents when the algorithm model for executing the target task needs to be updated. Any of the multiple algorithm model learning agents is used to determine whether to select the training model for itself to perform the target task. If selected, the training model is trained based on the training data to obtain the model to be screened. A model decision agent is used to select the best algorithm model from the models to be screened.

[0013] The multi-agent-based model identification system provided in this application further includes an instruction forwarding agent; The instruction forwarding agent is used to send its own post-decision training state information to the data preprocessing agent and the multiple algorithm model learning agents respectively; the training state information is used to indicate that the algorithm model needs to be updated. The instruction forwarding agent makes decisions based on at least one of the following methods: In response to a user-inputted model update command, a decision is made as to whether the algorithm model needs to be updated; In response to periodically updated information, a decision is made as to whether the algorithm model needs to be updated.

[0014] According to the multi-agent model identification system provided in this application, the system further includes training and configuring agents; The training configuration agent is used to obtain the training configuration information input by the user, and based on the periodic update information contained in the training configuration information, determine whether the algorithm model needs to be updated. If so, the periodic update information is sent to the instruction forwarding agent.

[0015] According to the multi-agent model identification system provided in this application, the training configuration agent is used to obtain training configuration information input by the user and send the model selection information in the training configuration information to the multiple algorithm model learning agents. The model selection information is used by the algorithm model learning agent to determine whether to select a training model for performing the target task.

[0016] According to the multi-agent model identification system provided in this application, the model decision agent is further configured to determine the ranking results of each of the models to be screened, and input the ranking results into the plurality of algorithm model learning agents; any of the algorithm model learning agents is further configured to learn again based on the ranking results.

[0017] The multi-agent model identification method and system provided by this invention, when the algorithm model for performing the target task needs to be updated, inputs the training data processed by the data preprocessing agent into multiple algorithm model learning agents to obtain the candidate models for performing the target task output by several algorithm model learning agents; and inputs each candidate model into the model decision agent to obtain the optimal algorithm model for performing the target task output by the model decision agent. This model identification method does not require modeling based on actual physical mechanisms. Instead, it utilizes data generated by the actual HVAC system and combines it with an artificial intelligence learning model to extract potential knowledge. This not only incorporates known physical data but also identifies unknown model parameters, improving modeling speed and accuracy compared to traditional physical mechanism modeling. Furthermore, compared to existing data-driven model identification methods, this application employs an intelligent agent combined with a data-driven approach. This eliminates the need for personnel to preprocess the collected HVAC system data or build artificial intelligence models. The entire system can train multiple algorithm models based on the collected system data and evaluate and select the best algorithm model. This requires minimal human intervention, lowers the professional threshold for engineers, reduces labor costs, and improves the automation and intelligence level of model identification. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the multi-agent-based model identification method provided by the present invention.

[0020] Figure 2 This is a schematic diagram of the system structure of the multi-agent model identification system provided by the present invention.

[0021] Figure 3 This is a schematic diagram of the operating status of the multi-agent-based model identification system provided by the present invention.

[0022] Figure 4 This is a schematic diagram of the user operation process of the multi-agent-based model identification system provided by the present invention.

[0023] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] The following is combined with Figures 1-5 Specific embodiments of the present invention are described below.

[0026] Figure 1 This is one of the flowcharts illustrating the multi-agent-based model identification method provided by this invention, such as... Figure 1 As shown, this method is applied to HVAC systems, primarily focusing on collecting data from these systems to extract patterns and regularities, thereby establishing mathematical models of the HVAC system. These models can be models of the entire HVAC system or models of any subsystem within it. For example, it could be a model of the relationship between the outlet temperature of the cooling water from the chiller unit and the inlet and outdoor temperatures, or a model for predicting the load on air conditioning ducts. The aforementioned multi-agent model identification method can be executed collaboratively by multiple agents, which are manifested in, for example... Figure 2 In the multi-agent model identification system shown, the method specifically includes the following steps.

[0027] Step 101: If the algorithm model for executing the target task needs to be updated, the training data processed by the data preprocessing agent 201 is input into multiple algorithm model learning agents 202 to obtain the models to be selected for executing the target task output by several algorithm model learning agents 202.

[0028] The data preprocessing agent 201 is used to collect real-time data from the site and perform data preprocessing on the real-time data to obtain the training data.

[0029] Each of the algorithm model learning agents is used to determine whether to select a training model for performing the target task. If selected, the training model is trained based on the training data to obtain a model to be selected.

[0030] It should be noted that an intelligent agent (also known as an artificial intelligence agent) is a proxy system composed of a large model as the core controller, possessing memory, planning skills, and tool usage. In this application, it refers to an intelligent entity capable of perceiving the environment, making decisions, and executing actions. A multi-agent system is a distributed system composed of several intelligent agents. Each agent is an autonomous entity that continuously observes the environment, communicates with neighboring agents, and makes decisions based on objectives. Agents collaborate and communicate with each other to complete the target task. Due to its distributed, self-organizing, and modular characteristics, multi-agent systems have been widely researched and applied in fields such as industrial control, transportation, and unmanned aerial vehicles. The target task refers to a specific model prediction task or other required task, such as the task of predicting air conditioning load based on real-time perceived data. This task requires an air conditioning load model, and the parameters in the model may need to be constantly updated to adapt to the characteristics of different seasons. Model identification in this application refers to the identification and selection of which algorithm (including the specific parameters in the algorithm) to choose for the air conditioning load model in order to obtain the optimal algorithm model. This application uses different algorithms to run simultaneously, selects the preferred algorithm and parameters, and uses the preferred algorithm and parameters to construct the aforementioned air conditioning load model.

[0031] exist Figure 2 In this multi-agent system, there are three agents: a data preprocessing agent 201, an algorithm model learning agent 202, and a model decision-making agent 203. The algorithm model learning agent 203 includes different types of specific algorithm model agents, each corresponding to a different parameter identification algorithm, such as an XGBoost model learning agent, a multinomial model learning agent, and a random forest model learning agent.

[0032] Specifically, when a specific target task needs to be performed, such as predicting the current air conditioning load, the air conditioning load model needs to be activated. This requires updating the model parameters, necessitating model identification. The entire multi-agent system can determine whether a specific target task needs to be performed based on a preset operating cycle. For example, if the system is set to perform air conditioning load prediction every 5 days, model identification can be performed when the preset operating cycle is reached to update the model parameters in the air conditioning load model. The entire multi-agent system can also activate model identification based on user input commands, such as receiving a model update command from the user.

[0033] In the model identification process, the data preprocessing agent 201 first receives the training data, which can be real-time data or historical data. The data preprocessing agent 201 preprocesses the training data, including: checking the validity of the training data, retrieving relevant training data based on training status information, and cleaning and judging the data volume to avoid inaccurate model parameters due to insufficient effective training data, which would lead to low accuracy of the model performing the target task (e.g., air conditioning load prediction model). Specifically, data cleaning can be achieved by using a K-means algorithm combined with the 3σ rule to mark outliers in the training data, thus completing data cleaning and improving model quality.

[0034] The data preprocessing agent 201 preprocesses the data to be trained to obtain the training data, and then sends this training data to multiple algorithm model learning agents 202, such as... Figure 2 As shown, the multiple algorithm model learning agents 202 can be XGBoost model learning agents, MLP (Multilayer Perceptron) model learning agents, multinomial model learning agents, random forest model learning agents, support vector regression model learning agents, etc. These algorithm model learning agents are pre-configured with corresponding algorithm models. Each agent inputs the aforementioned training data into its local algorithm model for training, resulting in a trained algorithm model, which is the model to be selected for performing the target task (e.g., predicting air conditioning load). Specifically, each algorithm model learning agent automatically completes model training and optimization with the goal of obtaining the highest coefficient of determination R². It iterative optimization of model parameters is achieved by calling methods such as GridSearchCV to obtain the optimal model configuration, thereby improving the model's regression accuracy.

[0035] Optionally, in this application, all algorithm model learning agents 202 can be enabled simultaneously, or only some of them can be enabled. Before training the algorithm model, all algorithm model learning agents 202 can determine which agents are selected based on the model selection information. If selected, the selected algorithm learning agents train the corresponding algorithm model based on the above training data to obtain the model to be selected.

[0036] Step 102: Input each of the models to be screened into the model decision agent 203 to obtain the optimal algorithm model output by the model decision agent 203 for performing the target task.

[0037] Specifically, after completing the local model training task, the selected algorithm model agent 202 obtains the corresponding trained algorithm model (i.e., the model to be trained). These models are then sent to the model decision agent, which runs them on the test dataset, obtains the results, ranks and evaluates them, and determines the best algorithm model. More specifically, the model decision agent 203 uses the coefficient of determination R² and root mean square error (RMSE) as evaluation metrics. Based on the performance of each trained algorithm model on the test dataset, it determines the best algorithm model for this model identification, outputs the best model information, and marks the best algorithm model in the database for automatic use during subsequent user downloads and simulations.

[0038] In the above embodiment, when the algorithm model for performing the target task needs to be updated, the training data processed by the data preprocessing agent is input into multiple algorithm model learning agents to obtain the candidate models for performing the target task output by several algorithm model learning agents; each candidate model is input into the model decision agent to obtain the optimal algorithm model for performing the target task output by the model decision agent. This model identification method does not require modeling based on actual physical mechanisms. Instead, it utilizes data generated by the actual HVAC system and combines it with an artificial intelligence learning model to extract potential knowledge. This not only incorporates known physical data but also identifies unknown model parameters, improving modeling speed and accuracy compared to traditional physical mechanism modeling. Furthermore, compared to existing data-driven model identification methods, this application employs an intelligent agent combined with a data-driven approach. This eliminates the need for personnel to preprocess the collected HVAC system data or build artificial intelligence models. The entire system can train multiple algorithm models based on the collected system data and evaluate and select the best algorithm model. This requires minimal human intervention, lowers the professional threshold for engineers, reduces labor costs, and improves the automation and intelligence level of model identification.

[0039] In one embodiment, before step 101, the method further includes: sending the training state information after the instruction forwarding agent 204 makes a decision to the data preprocessing agent 201 and the plurality of algorithm model learning agents 203 respectively; the training state information is used to indicate that the algorithm model needs to be updated; wherein, the instruction forwarding agent 204 makes the decision based on at least one of the following methods.

[0040] (1) In response to the user's input of a model update command, decide whether the algorithm model needs to be updated.

[0041] (2) In response to periodically updated information, decide whether the algorithm model needs to be updated.

[0042] Combination Figure 2 The schematic diagram of the multi-agent system shown is used for illustration. The multi-agent system also includes an instruction forwarding agent 204, which can decide whether the algorithm model used to perform the target task needs to be updated based on the received instructions; the decision-making methods include...

[0043] (1) Determine whether the algorithm model needs to be updated based on the model update command input by the user in real time.

[0044] (2) The user sets up periodic update information in the training configuration agent 205. The training configuration agent 205 transmits the above periodic update information to the instruction forwarding agent 204. The instruction forwarding agent 204 determines whether the algorithm model needs to be updated based on the periodic update information.

[0045] Instruction forwarding agent 204, responds to the model update instructions input by the user in real time (in... Figure 2 In the process, the user's running instructions include model update instructions, or the periodic update information transmitted by the training configuration agent 205 is used to make judgments and decisions, and the decided training status information is sent to the data preprocessing agent 201 and the various algorithm model learning agents 202. The training status information is used to indicate that the algorithm models used to perform the target task need to be updated. The training status information also includes data range indication information for the data to be trained, such as specifying that the range of data to be trained used in this model update is real-time collected data, or whether historical data is included. Based on this, the data preprocessing agent 201 can select suitable data to be trained.

[0046] In the above embodiments, the instruction forwarding agent 204 receives model update instructions input by the user, or receives periodic update information transmitted by the training configuration agent 205, and decides whether the algorithm model needs to be updated. This not only enables the completion of urgent model identification work based on the user's real-time update instructions, but also enables the automatic updating of the model periodically based on the user's customized periodic update information. This improves the system's automation level while ensuring flexible use by the user.

[0047] In one embodiment, before the above-mentioned training state information after the instruction forwarding agent makes a decision is sent to the data preprocessing agent and the multiple algorithm model learning agents, it further includes:

[0048] The training configuration agent 205 obtains the training configuration information input by the user; the training configuration information includes the periodic update information; the training configuration agent is used to determine whether the algorithm model needs to be updated based on the periodic update information, and if so, sends the periodic update information to the instruction forwarding agent; the periodic update information of the training configuration agent 205 is sent to the instruction forwarding agent 204.

[0049] Specifically, such as Figure 2 As shown, the multi-agent system also includes: training and configuration agent 205.

[0050] The training configuration agent 205 is able to receive and transmit the training configuration information input by the user before the multi-agent system runs. The contents of the training configuration information are shown in Table 1 below.

[0051] Table 1

[0052] The above embodiments train and configure the agent to receive training configuration information input by the user, providing the necessary data foundation for the system to perform model identification work regularly in the future.

[0053] In one embodiment, before step 101 above, the method further includes: obtaining training configuration information input by the user through a training configuration agent; the training configuration information includes model selection information, which is used by the algorithm model learning agent to determine whether to select a training model for performing the target task; and sending the model selection information to the plurality of algorithm model learning agents.

[0054] Specifically, as shown in Table 1 above, the training configuration information also includes model selection information, which indicates which models are selected. The training configuration agent sends the model selection information to multiple algorithm model learning agents 202 to trigger several of the algorithm model learning agents 202 to start the model training task.

[0055] In the above embodiments, the training configuration agent 205 receives the model selection information configured by the user, providing data for subsequent model selection and improving the flexibility of system execution.

[0056] In one embodiment, the model decision agent 203 is further configured to determine the ranking results of each of the models to be screened; after step 102, the method further includes: inputting the ranking results of the model decision agent 203 into the plurality of algorithm model learning agents; any of the algorithm model learning agents is further configured to learn again based on the ranking results.

[0057] Specifically, the model decision agent 203 ranks and evaluates the algorithm models trained by each algorithm model learning agent, and determines the best algorithm model.

[0058] Furthermore, the model decision agent 203 can feed back model evaluation information to each algorithm model learning agent. This model evaluation information refers to the ranking information based on the coefficient of determination (R²) or root mean square error (RMSE). Specifically, the model decision agent 203 calculates the coefficient of determination (R²) or RMSE of each model's prediction results based on the training of each algorithm model on the test dataset, performing the aforementioned target task (e.g., predicting air conditioning load). It then ranks each algorithm model based on its R² or RMSE and feeds back the ranking results, including the R² and RMSE of each algorithm model, to each algorithm model learning agent. Upon receiving the current ranking of the model identification algorithms from the model decision agent 203, each algorithm model learning agent will change or broaden the search range of its algorithm model parameters, thereby promoting continuous learning and improvement.

[0059] In the above embodiments, the ranking results are fed back to each algorithm model learning agent by the model decision agent 203, which is beneficial for each algorithm model learning agent to improve its own strategy and further optimize the algorithm model used to perform the target task.

[0060] This application also provides a multi-agent-based model identification system based on the above-mentioned multi-agent-based model identification method, such as... Figure 2 As shown, the system includes: a data preprocessing agent 201, an algorithm model learning agent 202, and a model decision-making agent 203; wherein, The data preprocessing agent 201 is used to collect real-time data from the field, preprocess the real-time data to obtain the training data, and input the processed training data into multiple algorithm model learning agents when the algorithm model for executing the target task needs to be updated. Any of the multiple algorithm model learning agents 202 is used to determine whether to select the training model for itself to perform the target task. If selected, the training model is trained based on the training data to obtain the model to be screened. Model decision agent 203 is used to select the best algorithm model from the models to be screened.

[0061] In one embodiment, the system further includes an instruction forwarding agent; The instruction forwarding agent is used to send its own post-decision training state information to the data preprocessing agent and the multiple algorithm model learning agents respectively; the training state information is used to indicate that the algorithm model needs to be updated. The instruction forwarding agent makes decisions based on at least one of the following methods: In response to a user-inputted model update command, a decision is made as to whether the algorithm model needs to be updated; In response to periodically updated information, a decision is made as to whether the algorithm model needs to be updated.

[0062] In one embodiment, the system further includes training and configuring an agent; The training configuration agent is used to obtain the training configuration information input by the user, and based on the periodic update information contained in the training configuration information, determine whether the algorithm model needs to be updated. If so, the periodic update information is sent to the instruction forwarding agent.

[0063] In one embodiment, the training configuration agent is used to obtain training configuration information input by the user and send the model selection information in the training configuration information to the plurality of algorithm model learning agents; The model selection information is used by the algorithm model learning agent to determine whether to select a training model for performing the target task.

[0064] In one embodiment, the model decision agent is further configured to determine the ranking result of each of the models to be screened, and input the ranking result into the plurality of algorithm model learning agents; The learning agent of any of the aforementioned algorithm models is also used to learn again based on the ranking results.

[0065] The above embodiments, by designing a multi-agent model identification system, are applied to the monitoring and model identification of HVAC systems. This eliminates the need for modeling based on actual physical mechanisms. Instead, it utilizes data actually generated by the HVAC system, combined with an artificial intelligence learning model to extract potential knowledge. This not only incorporates existing physical data but also identifies unknown model parameters, improving modeling speed and accuracy compared to traditional physical mechanism modeling. Furthermore, compared to existing data-driven model identification methods, this application employs an agent-based data-driven approach. It eliminates the need for preprocessing of collected HVAC system data or the construction of artificial intelligence models. The entire system can train multiple algorithm models based on the training data and evaluate and select the optimal algorithm model, requiring minimal human intervention, lowering the professional threshold for engineers, and reducing labor costs.

[0066] Figure 3The diagram illustrates the operational status of a multi-agent model identification system. This system operates continuously online, initiating a model identification task at regular intervals. The system primarily operates in three states.

[0067] 1) When the state is 0, the system does not perform model identification. Only the data preprocessing agent checks the validity of the data input in real time and saves the valid data to the database.

[0068] 2) Upon receiving the user's initial model identification instruction, the system will enter state 1. In this operation, the system's task is to use the historical device data provided by the user in the early stage and the data received during runtime as training data for this model identification, and then carry out the model identification work.

[0069] 3) When the system receives an immediate model update instruction from the user or when the system runs to the scheduled update condition, it enters state 2. The system's task in state 2 is also to perform model identification, but the difference from state 1 is that the training data range for this model identification is the user-specified data range.

[0070] When the system state is not 0, that is, when there is a model identification task, each agent autonomously completes its own sub-task while communicating with other agents to exchange effective information, thereby automatically cooperating to complete the system model identification work and selecting the best algorithm model in this model identification.

[0071] Figure 4 This is a user operation flowchart of a multi-agent model identification system. Users perform preliminary configuration work through the interactive interface, including inputting model selection information and periodically updating information; and configuring the field data interfaces of the HVAC system, including the sensor data interfaces of the HVAC system. The system performs model identification work according to these configuration information of the user.

[0072] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a multi-agent model identification method. This method includes: when the algorithm model for executing the target task needs updating, inputting the training data processed by the data preprocessing agent into multiple algorithm model learning agents to obtain candidate models for executing the target task output by several algorithm model learning agents; inputting each candidate model into the model decision agent to obtain the optimal algorithm model for executing the target task output by the model decision agent.

[0073] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the multi-agent-based model identification method provided by the above methods. The method includes: when the algorithm model for performing the target task needs to be updated, inputting the training data processed by the data preprocessing agent into multiple algorithm model learning agents to obtain the candidate models for performing the target task output by several algorithm model learning agents; inputting each candidate model into the model decision agent to obtain the optimal algorithm model for performing the target task output by the model decision agent.

[0075] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program is implemented to perform the multi-agent-based model identification method provided by the above methods. The method includes: when the algorithm model for performing the target task needs to be updated, inputting the training data processed by the data preprocessing agent into multiple algorithm model learning agents to obtain candidate models for performing the target task output by several of the multiple algorithm model learning agents; inputting each candidate model into the model decision agent to obtain the optimal algorithm model for performing the target task output by the model decision agent.

[0076] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0077] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-agent based model identification method, characterized in that: include: When the algorithm model for executing the target task needs to be updated, the training data processed by the data preprocessing agent is input into multiple algorithm model learning agents to obtain the models to be screened for executing the target task output by several algorithm model learning agents among the multiple algorithm model learning agents; Input each of the to-be-screened models into a model decision agent, and obtain the optimal algorithm model output by the model decision agent for executing the target task; The data preprocessing agent is used to collect on-site real-time data and perform data preprocessing on the on-site real-time data to obtain the training data; Any of the algorithm model learning agents is used to determine whether the model to be trained is selected for performing the target task. If selected, the model to be trained is trained based on the training data to obtain a model to be screened.

2. The multi-agent based model identification method according to claim 1, characterized in that: In the case where the algorithm model for executing the target task needs to be updated, the method further includes: before inputting the training data processed by the data preprocessing agent into multiple algorithm model learning agents, and obtaining the models to be screened output by several algorithm model learning agents among the multiple algorithm model learning agents for executing the target task, Sending the training status information after the instruction forwarding agent makes a decision to the data preprocessing agent and the multiple algorithm model learning agents respectively; the training status information is used to indicate that the algorithm model needs to be updated; The instruction forwarding agent makes decisions based on at least one of the following methods: In response to a model update instruction input by a user, deciding whether the algorithm model needs to be updated; In response to the periodic update information, it is decided whether the algorithm model needs to be updated.

3. The multi-agent based model identification method according to claim 2, characterized in that: Before sending the training state information after the instruction forwarding agent makes a decision to the data preprocessing agent and the plurality of algorithm model learning agents respectively, the method further comprises: The training configuration agent obtains the training configuration information input by the user; the training configuration information includes the periodic update information; the training configuration agent is used to determine whether the algorithm model needs to be updated based on the periodic update information, and if necessary, sends the periodic update information to the instruction forwarding agent; Sending periodic update information of the training configuration agent to the instruction forwarding agent.

4. The multi-agent based model identification method according to claim 1, characterized in that: In the case where the algorithm model for executing the target task needs to be updated, the method further includes: before inputting the training data processed by the data preprocessing agent into multiple algorithm model learning agents, and obtaining the models to be screened output by several algorithm model learning agents among the multiple algorithm model learning agents for executing the target task, The training configuration agent obtains the training configuration information input by the user; the training configuration information includes model selection information, and the model selection information is used for the algorithm model learning agent to determine whether to select the model to be trained for executing the target task; The model selection information is sent to the multiple algorithm model learning agents.

5. The multi-agent based model identification method according to claim 1, characterized in that: The model decision agent is also used to determine the ranking results of each of the models to be screened. After each of the models to be screened is input into the model decision agent to obtain the best algorithm model output by the model decision agent for performing the target task, the method further includes: The ranking results of the model decision agent are input into the several algorithm model learning agents; any of the algorithm model learning agents is also used to learn again based on the ranking results.

6. A multi-agent based model recognition system, characterized in that: The system includes: a data preprocessing agent, multiple algorithm model learning agents and a model decision agent; wherein, The data preprocessing agent is used to collect real-time data on site, perform data preprocessing on the real-time data on site to obtain the training data; when the algorithm model for executing the target task needs to be updated, the processed training data is input into multiple algorithm model learning agents; Any algorithm model learning agent among the multiple algorithm model learning agents is used to determine whether the model to be trained for executing the target task is selected, and if selected, train the model to be trained based on the training data to obtain a model to be screened; The model decision agent is used to select the best algorithm model from the models to be screened.

7. The multi-agent based model recognition system according to claim 6, characterized in that: It also includes instruction forwarding agents; The instruction forwarding agent is used to send the training status information after its own decision to the data preprocessing agent and the multiple algorithm model learning agents respectively; the training status information is used to indicate that the algorithm model needs to be updated; The instruction forwarding agent makes decisions based on at least one of the following methods: In response to a model update instruction input by a user, deciding whether the algorithm model needs to be updated; In response to the periodic update information, it is decided whether the algorithm model needs to be updated.

8. The multi-agent based model recognition system according to claim 6, characterized in that: The system also includes a training configuration agent; The training configuration agent is used to obtain the training configuration information input by the user, and based on the periodic update information contained in the training configuration information, determine whether the algorithm model needs to be updated, and if necessary, send the periodic update information to the instruction forwarding agent.

9. The multi-agent based model recognition system according to claim 6, characterized in that: The training configuration agent is used to obtain the training configuration information input by the user, and send the model selection information in the training configuration information to the multiple algorithm model learning agents; Among them, the model selection information is used for the algorithm model learning agent to determine whether to select the model to be trained for performing the target task.

10. The multi-agent based model recognition system according to claim 6, characterized in that: The model decision agent is further used to determine the ranking results of each of the to-be-screened models, and input the ranking results into the plurality of algorithm model learning agents; Any of the algorithm model learning agents is also used to learn again based on the ranking results.

Images