Air conditioner fault diagnosis method and system based on cooperation of large models in multiple intelligent fields

Abstract

The invention discloses an air conditioner fault diagnosis method and system based on multi-intelligent-field large model cooperation, and the method comprises the steps: inputting the equipment information and operation parameters of an air conditioner in a multi-mode form of voice, text or image, and converting the operation state information into text representation in a unified manner through a multi-mode conversion model; extracting the information of the text through the diagnosis agent and predicting the type of a potential fault; and finally, obtaining a corresponding fault reason and a disposal suggestion by utilizing the disposal agent. According to the method, callable professional tools in the field of air conditioner system fault diagnosis are introduced, and professional knowledge in the field of air conditioner fault diagnosis is provided for the disposal agent through retrieval enhancement, so that the problem that a general large model is insufficient in professionality in application in the field of air conditioner systems is solved, and accurate and reliable air conditioner system fault diagnosis is realized.

Description

Technical Field

[0001] The present invention relates to a technology in the field of building air conditioning, specifically an air conditioning fault diagnosis method and system based on the collaboration of multi-intelligent field large models. Background Art

[0002] As one of the key areas for implementing energy conservation and emission reduction, the efficient and reliable operation of building air conditioning is of great significance for the construction industry to achieve the dual carbon goals. Various faults are inevitable in the long-term operation of air conditioners, resulting in reduced system operating efficiency, increased energy consumption, and even equipment damage. With the development of artificial intelligence, various advanced algorithms are applied to air conditioning systems for fault diagnosis to achieve intelligent operation and maintenance. The big model is a general artificial intelligence with large-scale parameters. It has attracted widespread attention due to its powerful reasoning ability and is considered to be the leading force in promoting the next generation of industrial innovation. However, the air-conditioning system involves complex engineering principles and specific failure modes. The general big model cannot fully understand and respond to professional problems, making it difficult to achieve accurate diagnosis of the air-conditioning system. Summary of the invention

[0003] In view of the problems that current large models have low accuracy and are difficult to adapt to complex and changeable operating environments and fault scenarios in the professional field of air-conditioning systems, the present invention proposes an air-conditioning fault diagnosis method and system based on the collaboration of large models in multiple intelligent fields, which includes a diagnostic intelligent agent for fault identification and a disposal intelligent agent for fault handling with the large model as the core and basis, introduces callable professional tools in the field of air-conditioning system fault diagnosis for the diagnostic intelligent agent, and provides professional knowledge in the field of air-conditioning fault diagnosis for the disposal intelligent agent through retrieval enhancement generation, so as to overcome the problem of insufficient professionalism of general large models in the application of air-conditioning systems, and realize accurate and reliable air-conditioning system fault diagnosis.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to an air-conditioning fault diagnosis method based on the collaboration of a large model in multiple intelligent fields. The equipment information and operating parameters of the air-conditioning are input in the multimodal form of voice, text or image. The operating status information is uniformly converted into a text representation using a multimodal conversion model. The text information is then extracted and potential fault categories are predicted using a diagnostic agent. Finally, the corresponding fault cause and disposal suggestion are obtained using a disposal agent.

[0006] The method specifically includes:

[0007] Step 1: Input the equipment information and operating parameters of the air conditioner: Input the current status information of the air conditioner in multiple modes such as voice, text or image.

[0008] For example: use the microphone to say "the evaporation pressure is too low after the air conditioner is started", enter text on the keyboard to describe "the indoor room temperature cannot reach the set value", or upload a photo of "the air conditioner appearance is deformed and damaged, and liquid is leaking".

[0009] Step 2: Use the multimodal conversion model to convert the input into a unified text representation: When the input is speech, it is transcribed into text describing the air conditioning system based on the speech recognition model. The speech recognition model is a decoder-encoder model based on a deep neural network, which is used to extract the spectral information of the audio for encoding and predicting the corresponding text. When the input is an image, it is converted into the corresponding text description of the air conditioning system using the image recognition model. The image recognition model is a language-text model based on a deep self-attention mechanism, which is used to extract image information for encoding and predict the corresponding text. When the input is text, the multimodal conversion model does not process it and outputs it directly.

[0010] Step 3: Construct a multi-agent system based on a large model in the air conditioning field: The agent is a system that can autonomously perceive the environment and make decisions. Based on the large model, a diagnostic agent and a disposal agent are constructed to form a multi-agent system, and the large model is combined with professional tools and professional knowledge to form a large model in the air conditioning field. Specifically, it refers to introducing callable professional tools in the field of air conditioning system fault diagnosis for the diagnostic agent to obtain diagnostic results, and providing professional knowledge in the field of air conditioning fault diagnosis for the disposal agent through retrieval enhancement generation to provide maintenance suggestions.

[0011] The large model refers to: a large-scale pre-trained language model that uses a self-attention deep neural network architecture with a parameter count of up to tens of billions. The input and output of the large model are both natural language text sequences. For example, the operating parameters of the current air-conditioning system are input and whether the air-conditioning system is operating normally is consulted, and the specific fault type of the current air-conditioning system is output. In the training phase, the large model self-supervises and learns a large amount of text data such as air-conditioning system literature, patents, and operation and maintenance manuals to learn the probability distribution of text sequences in the air-conditioning field. In the reasoning phase, given an input text sequence, such as a problem of diagnosing the fault status of an air-conditioning system, the large model generates a corresponding text sequence response based on the probability distribution, such as potential faults and maintenance suggestions for the air-conditioning system.

[0012] The large model in the field of air conditioning refers to: a large model that integrates knowledge in the field of air conditioning. Domain knowledge refers to professional tools and expertise acquired by humans through long-term practice in air conditioning system fault diagnosis. The large model that integrates professional tools in the field of air conditioning does not directly judge the fault category, but extracts physical parameters used for fault diagnosis, and then predicts the fault category by calling professional tools in the field of fault diagnosis. The large model that integrates professional knowledge in the field of air conditioning does not directly respond to fault repair suggestions, but generates professional knowledge related to retrieval and handling of faults through retrieval enhancement, and then generates fault repair suggestions based on fault categories and professional knowledge.

[0013] Step 4: Fault diagnosis of air conditioning system is realized through multi-agent collaboration: The constructed diagnostic agent extracts the parameters required for fault diagnosis based on the input air conditioning equipment information and operating parameter text content, and inputs them into the professional tool in the field of fault diagnosis to predict the type of air conditioning system fault. The treatment agent retrieves the professional knowledge in the field of fault diagnosis and outputs the corresponding cause and treatment process based on the predicted type of air conditioning system fault, so as to obtain comprehensive fault diagnosis results and maintenance suggestions.

[0014] The parameters required for fault diagnosis are: internal physical parameters and external physical parameters that can reflect changes in the state of the air-conditioning system or characterize faults, wherein the internal physical parameters include the power, current and voltage of the main components of the air-conditioning system and their inlet and outlet temperatures, pressures and flows, and the external physical parameters include ambient temperature, humidity, wind speed and radiation intensity.

[0015] The professional tool in the field of fault diagnosis refers to a deep fully connected neural network that outputs the fault type of the air-conditioning system according to the operating parameters of the air-conditioning system. The network diagnosis process is as follows: in: is the fault category predicted by the professional tool, (T, P, …, I) is the air-conditioning system operating parameters extracted by the large model from the input text, such as evaporation temperature, evaporation pressure, current, etc., C is the total number of fault categories, c is a specific type of fault, such as refrigerant leakage, and f(·) is the forward propagation process of the deep fully connected neural network.

[0016] The deep fully connected neural network uses the system parameter data of historical normal operation and fault operation to train the professional tools in the field of fault diagnosis to reduce the cross entropy loss between the predicted fault category and the actual fault category: L = - Where: y is the unique hot encoding of the correct fault category. The neural network parameters such as weight w and bias b are continuously adjusted by gradient descent method to reduce the cross entropy loss until it becomes stable, specifically: and Among them: η is the learning rate, which is used to control the step size of parameter update. The trained professional tool extracts the operating data in the input text based on the large model to predict the fault type of the air-conditioning system in real time online.

[0017] The professional knowledge in the field of fault diagnosis includes: texts on the operating principles of air-conditioning systems, common faults, diagnostic methods and maintenance suggestions. The professional knowledge can come from papers and patents in the field of air-conditioning systems, technical manuals, operating instructions, standards and regulations, and operation and maintenance experience and opinions summarized by experts. The professional knowledge base is constructed in the following way: first, the collected text library is cleaned to remove redundant and erroneous knowledge, then the text library is divided into blocks according to a fixed number of words to improve the efficiency of retrieval enhancement generation, and finally, the divided text fragments are mapped to the vector space through the embedding model and stored in the vector database for retrieval enhancement generation.

[0018] The retrieval enhancement generation means that when consulting the big model for maintenance suggestions based on the fault diagnosis results, first search for information related to the problem in the established professional knowledge base in the field of air conditioning system fault diagnosis, and provide the retrieved information and the question to the big model to supplement the context information and background knowledge, and then generate the corresponding answer according to the needs. The retrieval method calculates the cosine distance between the input question text and the text in the professional knowledge base Among them: q is the vector representation of the input question, p is the vector representation of the text in the professional knowledge base, and finally the content in the professional knowledge base with a small cosine distance to handle this type of fault is provided to the large model to enhance the quality of answering maintenance suggestions.

[0019] The multi-intelligent air-conditioning field big model collaboration refers to: the diagnostic agent that can call professional tools in the field of fault diagnosis and the disposal agent that can retrieve professional knowledge in the field of fault diagnosis collaborate to implement the fault diagnosis task of the air-conditioning system. First, the diagnostic agent extracts the parameters required by the professional tool based on the input air-conditioning system operation information. The prompt word requires the big model not to make a diagnosis directly but to extract the parameter name and parameter value in the input text, input the extracted parameters into the trained professional tool to predict the fault category, and input the fault category into the disposal agent to obtain the corresponding maintenance suggestion. The prompt word gives the fault type of the current system and requires the big model to give the corresponding fault cause and how to deal with the fault based on the retrieved knowledge. The whole process of fault diagnosis from discovery to resolution is realized through the collaboration of multiple agents.

[0020] The diagnostic results refer to: diagnosis of potential fault types of the air-conditioning system, including identification of sudden faults and gradual faults, wherein: sudden faults refer to faults that occur instantly and cause strong interference to the system, which are easy to diagnose; gradual faults refer to faults caused by gradual deterioration of the air-conditioning system during use, such as refrigerant leakage, reduced evaporator water flow, reduced condenser water flow, etc., which require professional tools for diagnosis.

[0021] The maintenance suggestions are to systematically analyze the causes of specific air conditioning failures and provide solutions.

[0022] For example, one of the reasons for the evaporator water flow reduction failure is the evaporator being dirty and clogged. The evaporator should be cleaned in time to ensure the smooth flow of the chilled water circuit. In addition, it also includes the performance degradation of the chilled water pump, sensor and control system failure, etc. By retrieving professional knowledge and guiding the operation and maintenance personnel to thoroughly repair the fault, the air-conditioning system can resume normal operation.

[0023] The present invention relates to an air-conditioning fault diagnosis system of a domain knowledge-guided large model for implementing the above-mentioned method, comprising: a multimodal input module, a central processing module, a storage module and an output module, wherein: the multimodal input module receives multimodal data from different sensors, devices or inputs, converts the multimodal data into unified text information based on a multimodal conversion model and then outputs it to the large model, the central processing module converts various forms of input information into natural language using a multimodal deep learning model, generates diagnostic results and maintenance suggestions through a multi-intelligent air-conditioning domain large model and calls professional knowledge and professional tools in the storage module, and the output module presents the diagnostic results and maintenance suggestions.

[0024] The storage module includes local storage and cloud storage. The local storage uses a solid-state hard disk with high read and write speeds, which is used to store: 1) system operating environment, software and programs; 2) professional tools and professional knowledge; 3) training data and model parameters. Cloud storage uses a cloud server to store a large number of intermediate records during the use process, such as conversation content, usage time, etc. in a MySQL database.

[0025] The output module is implemented using but not limited to a display screen, and is used to display a web page consisting of a fault text description, an icon of a fault process parameter change trend, a fault statistical report, etc. The web page is developed based on the Vue framework and CSS, HTML and JS. Important fault alarms are notified to operation and maintenance personnel via email, mobile phone text messages, etc. Technical Effects

[0026] The present invention predicts the fault category by calling professional tools in the field of fault diagnosis based on the air-conditioning operation information, and gives maintenance suggestions by retrieving professional knowledge in the field of fault diagnosis through retrieval enhancement generation. Compared with the prior art, the diagnostic accuracy of the normal operation of the air-conditioning system and three typical faults of the present invention is improved from 30.8% to 95.8%. At the same time, by introducing professional knowledge in the field of fault diagnosis, the generated maintenance suggestions more comprehensively cover the fault characteristics of each sub-equipment of the air-conditioning system, and avoid erroneous and redundant responses, greatly improving the quality of generated maintenance suggestions, which helps to promote the practical application of large models and other artificial intelligence technologies in the field of air-conditioning system fault diagnosis, and realize the safe and energy-saving operation of the air-conditioning system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic diagram of the fault diagnosis system of the present invention;

[0028] Figure 2 It is a flow chart of the present invention;

[0029] Figure 3 A schematic diagram of the principle of a diagnostic agent that calls professional tools;

[0030] Figure 4 Schematic diagram of the processing agent principle for retrieving professional knowledge;

[0031] Figure 5 This is the air conditioning failure experiment system diagram;

[0032] Figure 6 This is a comparison chart of fault diagnosis results;

[0033] Figure 7 Comparison chart for fault repair suggestions. DETAILED DESCRIPTION

[0034] like Figure 1 As shown, this embodiment relates to an air conditioning diagnosis system guided by a domain knowledge large model, including: a multimodal input module, a central processing module, a storage module and an output module.

[0035] The multimodal input module includes: an acquisition unit and a conversion unit, wherein: the acquisition unit uses a keyboard to input text information, uses a microphone to input voice information, and inputs image information based on camera shooting; the conversion unit uses a multimodal conversion model to convert image information and voice information into text information, the multimodal conversion model includes a speech recognition model and an image recognition model, the speech recognition model converts voice information into text information, and the image recognition model converts image information into text information, thereby converting the multimodal input into a unified text form.

[0036] The speech recognition model is a deep neural network based on an encoder-decoder architecture, wherein: the encoder extracts feature information from the spectrogram converted from speech data, and the decoder generates corresponding text based on the feature information, thereby converting speech information into text information.

[0037] The speech recognition deep neural network is implemented using, but not limited to, the technology described in "Robust Speech Recognition via Large-Scale Weak Supervision" by Radford et al.

[0038] The image recognition model is a deep neural network based on an encoder-decoder architecture, wherein: a visual model serves as an encoder and a pre-trained language model serves as a decoder, the encoder extracts image features, and the decoder generates corresponding image descriptions, thereby converting image information into text information.

[0039] The image recognition deep neural network is implemented using, but not limited to, the technology described in "The Illustrated Image Captioning using transformers" by Kumar Ankur.

[0040] like Figure 2 As shown, the present embodiment provides a method for identifying air conditioning system faults and generating maintenance suggestions based on the above system, including:

[0041] Step 1, extract physical parameters from text information: The input of the large model in the diagnostic large model includes three parts: text information, instructions for extracting physical parameters, and the return format of physical parameters. The text information is the input air conditioning system equipment parameters and operation information. The instructions for obtaining physical parameters require the model not to directly determine the type of fault but to extract physical parameters from the text. The return format of physical parameters is a Json string, specifically "{parameter name 1: parameter value 1, parameter name 2: parameter value 2, ..., parameter name n: parameter value n}".

[0042] The large model is a neural network model based on a deep self-attention mechanism, which generates responses to questions based on natural language text, namely, questions about fault diagnosis and maintenance suggestions for air-conditioning systems. Specifically, self-supervised pre-training is performed in a large amount of unlabeled text. The pre-training method is to predict the next word in the text based on the previous paragraph or sentence in the text, and then supervised training is performed on the basis of the pre-trained model by manually annotating questions and answers. Then, the supervised model is continuously fine-tuned based on reinforcement learning based on human feedback, so that it can better understand the meaning of human instructions, and finally realize the automatic generation of corresponding responses from the input text.

[0043] The neural network model based on the deep self-attention mechanism is implemented using but not limited to the technology recorded by Ouyang et al. in "Training Language Models to Follow Instructions with Human Feedback".

[0044] Step 2, call the professional tool in the field of air conditioning fault diagnosis established by the present invention to obtain the diagnosis result: first, use Python's json library to convert the physical parameters from a string into a dictionary, obtain the corresponding parameter values ​​in the dictionary and convert them into tensors through Python's Pytorch library, and then input them into the established professional tool in the field of air conditioning fault diagnosis. Finally, the output of the professional tool is the probability of each type of fault, and the fault category with the highest probability is taken as the diagnosis result.

[0045] like Figure 3As shown, the diagnostic agent calls the professional tool to first establish a fault diagnosis professional tool and encapsulate it as a callable function. The fault diagnosis professional tool of this embodiment is a deep fully connected neural network model, which is established using Python's third-party library Pytorch. The input is the operating parameters of the air-conditioning system, and the output is the fault category. The fault diagnosis model can be trained through normal data and fault data in the historical operating data of the air-conditioning system. When asking questions, the diagnostic agent uses prompts to require the large model not to directly answer questions about fault diagnosis but to return the operating parameters in the user's question in the format. The obtained operating parameters must be matched with the input required by the professional tool. When all the required parameters are met, they are input into the professional tool to obtain the probability of various types of faults in the current state, and then the fault category corresponding to the maximum probability is obtained through Pytorch's torch.argmax() function as the diagnosis result.

[0046] Step 3: Figure 4 As shown, the professional knowledge in the field of fault diagnosis is retrieved to obtain maintenance suggestions, specifically including: building a professional knowledge base by collecting the operation and maintenance experience of experts in the field of air-conditioning system fault diagnosis, and using Langchain's CharacterTextSplitter() function to split the collected text into text blocks according to a fixed number of words. The purpose is to introduce relevant professional knowledge into a large language model without introducing the entire document. After only the most relevant part is introduced, the text block is converted into a vector and stored in the vector database Chroma of the storage module. This embodiment uses OpenAI's text embedding model text-embedding-3-small to map the text block into a vector. When replying to maintenance suggestions for a certain type of fault, the professional knowledge most relevant to the problem in the knowledge vector library is calculated based on the cosine distance, and the retrieved professional knowledge is input into the large model as part of the prompt word to assist in generating maintenance suggestions.

[0047] The storage module of this embodiment is a 512GB solid-state hard disk, which is used to store fault diagnosis expertise and professional tools, specifically including a model parameter file of a fault diagnosis professional tool and a professional knowledge vector library file. In addition, the module also stores programs and software related to the present invention and operating systems and drivers necessary for computer operation.

[0048] The output module of this embodiment is a 27-inch LCD display, which is used to display diagnostic results and maintenance suggestions. Specifically, when the diagnostic agent outputs the diagnostic results, the output module displays the fault type, and then the disposal agent generates maintenance suggestions based on the fault type and displays the cause and disposal method of the fault.

[0049] exist Figure 5The fault model experiment was carried out in the real air-conditioning system shown in the figure. The experimental air-conditioning system is a chiller, including a semi-enclosed screw compressor, a plate evaporator, a shell and tube condenser, an electronic expansion valve and some auxiliary equipment. The refrigerant used in the chiller is R22, and the rated power and rated cooling capacity are 90kW and 350kW respectively. The experimental system is used to simulate the normal state (N) and some fault states under different working conditions. The fault state includes three common types of faults: reduced condenser water flow (RC), reduced evaporating water flow (RE) and refrigerant leakage (RL). The measured physical parameters are shown in the following table, which contains a total of 9 characteristic parameters: Serial number symbol Variable Name 1 <![CDATA[T suc ]]> Suction temperature 2 <![CDATA[T dis ]]> Exhaust temperature 3 <![CDATA[P suc ]]> Inspiratory pressure 4 <![CDATA[T dis ]]> Exhaust pressure 5 I Current 6 <![CDATA[T wci ]]> Condenser inlet water temperature 7 <![CDATA[T wco ]]> Condenser outlet water temperature 8 <![CDATA[T wei ]]> Evaporator inlet water temperature 9 <![CDATA[T weo ]]> Evaporator outlet water temperature

[0050] A total of 654 data samples were obtained in the fault simulation experiment, and the data were divided into a training data set and a test data set. The number of data set samples is shown in the following table: Fault category Training Dataset Test Dataset N 162 30 RC 132 30 RE 111 30 RL 129 30 total 534 120

[0051] The training data set is used to train a professional fault diagnosis tool, which is a deep learning model with a six-layer fully connected neural network. In addition to the input layer, it contains 5 hidden layers and 1 output layer. The number of neuron nodes in each layer of the neural network is 9, 16, 32, 64, 32, 16 and 4 respectively, and ReLU is used as the activation function. The number of iterations in the training parameters is 1000 times, the learning rate is 0.001, the loss function is the cross entropy loss, and the optimizer is the adaptive moment estimation. The test data set is used to verify the fault diagnosis performance of the method of the present invention.

[0052] like Figure 6 As shown in the figure, in order to verify the effectiveness of introducing professional tools in the field of fault diagnosis into the diagnostic agent, 120 samples in the test data are used to test the diagnostic method proposed by the present invention, and the performance is compared when the professional tools are not called, that is, the input is the equipment information of the air-conditioning system, the actual value of the operating parameters, the mean and standard deviation under normal conditions, and the candidate fault type options, and the large model directly determines the fault type based on the input. Compared with directly outputting the fault category by the large model, the diagnostic accuracy of the present invention for the normal operation of the air-conditioning system and three typical faults is improved from 30.8% to 95.8%, indicating that the introduction of professional tools improves the accuracy of the large model in air-conditioning system diagnosis.

[0053] like Figure 7As shown, in order to verify the effectiveness of introducing professional knowledge in the field of fault diagnosis into the handling agent, the handling agent is input with the diagnosis results obtained by the diagnosis agent and asked for maintenance suggestions. At the same time, the performance without introducing professional knowledge is compared, that is, the input is the fault type, and the large model is required to give the cause of the fault and the method of handling the fault according to the fault type. Compared with the case without introducing professional knowledge, the maintenance suggestions generated by the present invention are more complete. For example, for the condenser water flow reduction fault, the sensor and control system failures are considered, and the generation of erroneous content is also avoided. For example, when professional knowledge is not introduced, the leakage location of the refrigerant leakage is determined by observing the smell, but many refrigerants are colorless and odorless. In addition, the overall output is also very concise and avoids redundancy, which shows that the introduction of professional knowledge improves the quality of the large model in generating maintenance suggestions for air-conditioning systems.

[0054] Compared with the prior art, the technical effects of the present invention include: First, through the collaboration of multiple intelligent air-conditioning field large models, the corresponding maintenance suggestions can be output after diagnosing the fault type of the air-conditioning system, solving the problem that the traditional method can only obtain the fault type. Second, by using the diagnostic agent to call professional tools to diagnose air-conditioning system faults, the diagnostic accuracy rate is increased from 30.8% to 95.8%, improving the accuracy of the large model in air-conditioning system diagnosis. Third, by embedding professional knowledge into the large model through the disposal agent to generate fault diagnosis and maintenance suggestions, based on the retrieval enhancement generation to retrieve the professional knowledge in the air-conditioning field to generate a more comprehensive, accurate and concise response, the quality of the large model in the generation of air-conditioning system maintenance suggestions.

[0055] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. An air conditioning fault diagnosis method based on multi-intelligence domain large model collaboration, characterized in that: The equipment information and operating parameters of the air conditioner are input in the form of voice, text or image. The operating status information is uniformly converted into text representation using a multimodal conversion model. The diagnostic agent then extracts this information and predicts potential fault categories. Finally, the treatment agent is used to obtain the corresponding fault cause and treatment suggestions. The diagnostic results refer to: diagnosing the potential fault types of the air conditioning system, including the identification of sudden faults and gradual faults, wherein: sudden faults refer to faults that occur instantly and cause strong interference to the system; gradual faults refer to faults that gradually deteriorate during the use of the air conditioning system; The maintenance suggestions are to systematically analyze the causes of specific air conditioning failures and provide solutions.

2. The air conditioning fault diagnosis method based on multi-intelligence domain large model collaboration according to claim 1 is characterized in that: include: Step 1: Input the equipment information and operating parameters of the air conditioner: input the current status information of the air conditioner in multiple modes such as voice, text or image; Step 2: Use the multimodal conversion model to convert the input into a unified text representation: When the input is speech, it is transcribed into text describing the air conditioning system based on the speech recognition model. The speech recognition model is a decoder-encoder model based on a deep neural network, which is used to extract the spectral information of the audio, encode it, and predict the corresponding text; Step 3: Construct a multi-agent system based on a large model in the air-conditioning field: The agent is a system that can autonomously perceive the environment and make decisions. Based on the large model, a diagnostic agent and a disposal agent are constructed to form a multi-agent system, and the large model is combined with professional tools and professional knowledge to form a large model in the air-conditioning field. Specifically, it refers to introducing callable professional tools in the field of air-conditioning system fault diagnosis for the diagnostic agent to obtain diagnostic results, and providing professional knowledge in the field of air-conditioning fault diagnosis for the disposal agent through retrieval enhancement generation to provide maintenance suggestions; Step 4: Realize air-conditioning system fault diagnosis through multi-agent collaboration: The constructed diagnostic agent extracts the parameters required for fault diagnosis based on the input air-conditioning equipment information and operating parameter text content, and inputs them into the professional tools in the field of fault diagnosis to predict the air-conditioning system fault type; the handling agent retrieves the professional knowledge in the field of fault diagnosis based on the predicted air-conditioning system fault type and outputs the corresponding cause and handling process to obtain comprehensive fault diagnosis results and maintenance suggestions.

3. The air conditioning fault diagnosis method based on multi-intelligence domain large model collaboration according to claim 2 is characterized in that: The parameters required for fault diagnosis include: power, current and voltage of each main component and its inlet and outlet temperature, pressure, flow, ambient temperature, humidity, wind speed and radiation intensity.

4. The air conditioning fault diagnosis method based on multi-intelligence domain large model collaboration according to claim 1 or 2 is characterized in that: The professional tool in the field of fault diagnosis refers to a deep fully connected neural network that outputs the fault type of the air-conditioning system according to the operating parameters of the air-conditioning system. The network diagnosis process is as follows: in: is the fault category predicted by the professional tool, (T, P, …, I) is the air-conditioning system operating parameters extracted by the large model from the input text, C is the total number of fault categories, c is a specific type of fault, and f(·) is the forward propagation process of the deep fully connected neural network.

5. The air conditioning fault diagnosis method based on multi-intelligence domain large model collaboration according to claim 1 is characterized in that: The deep fully connected neural network uses the system parameter data of historical normal operation and fault operation to train the professional tools in the field of fault diagnosis to reduce the cross entropy loss between the predicted fault category and the actual fault category. Specifically: Where: y is the unique hot encoding of the correct fault category. The neural network parameters such as weight w and bias b are continuously adjusted by gradient descent method to reduce the cross entropy loss until it stabilizes. Specifically: and Among them: η is the learning rate, which is used to control the step size of parameter update. The trained professional tool extracts the operating data in the input text based on the large model to predict the fault type of the air-conditioning system in real time online.

6. The air conditioning fault diagnosis method based on multi-intelligence domain large model collaboration according to claim 1 is characterized in that: The professional knowledge in the field of fault diagnosis includes texts on the operating principles of air-conditioning systems, common faults, diagnostic methods and maintenance suggestions. The professional knowledge comes from papers and patents, technical manuals, operating instructions, standards and regulations, and operation and maintenance experience and opinions summarized by experts. The professional knowledge base is constructed in the following way: first, the collected text library is cleaned to remove redundant and erroneous knowledge, and then the text library is divided into blocks according to a fixed number of words to improve the efficiency of retrieval enhancement generation. Finally, the divided text fragments are mapped to the vector space through the embedding model and stored in the vector database for retrieval enhancement generation.

7. The air conditioning fault diagnosis method based on multi-intelligence domain large model collaboration according to claim 1 is characterized in that: The retrieval enhancement generation means that when consulting the big model for maintenance suggestions based on the fault diagnosis results, first search for information related to the problem in the established professional knowledge base in the field of air conditioning system fault diagnosis, and provide the retrieved information and the question to the big model to supplement the context information and background knowledge, and then generate the corresponding answer according to the needs. The retrieval method calculates the cosine distance between the input question text and the text in the professional knowledge base. Among them: q is the vector representation of the input question, p is the vector representation of the text in the professional knowledge base, and finally the content in the professional knowledge base with a small cosine distance to handle this type of fault is provided to the large model to enhance the quality of answering maintenance suggestions.

8. An air conditioning fault diagnosis system based on multi-intelligence domain large model collaboration for implementing any of the methods described in claims 1-7, characterized in that: include: Multimodal input module, central processing module, storage module and output module, wherein: the multimodal input module receives multimodal data from different sensors, devices or inputs, converts it into unified text information based on the multimodal conversion model and outputs it to the big model; the central processing module uses the multimodal deep learning model to convert various forms of input information into natural language, generates diagnostic results and maintenance suggestions through the multi-intelligent field big model and calls the professional knowledge and professional tools in the storage module; the output module presents the diagnostic results and maintenance suggestions.

Images