Long-distance pipe network heat supply system automatic driving platform based on large model and knowledge graph

Abstract

The invention discloses an automatic driving platform for a long-distance pipe network heat supply system based on a large model and a knowledge graph, and the platform comprises an infrastructure layer which is used for connecting an actual heat source, a long-distance pipe network and a multi-stage pump station heat supply facility of the long-distance pipe network heat supply system into the platform; the data processing layer comprises a data acquisition unit and a multi-source data fusion unit; the digital twinborn layer is used for establishing a digital twinborn model of the long-distance pipe network heat supply system; the large language model-knowledge graph layer comprises a graph construction unit which is used for constructing each business knowledge graph; the system also comprises a large language model construction unit which is used for constructing and training large language models of all businesses and processing automatic driving related businesses of the long-distance pipe network heat supply system. The business application decision-making layer comprises a business application unit which is used for providing a business application module of the long-distance pipe network heat supply system; and the intelligent decision-making unit is used for carrying out simulation analysis according to the digital twin model and calling each business large model to output a corresponding business decision-making result.

Description

Technical Field

[0001] The present invention belongs to the technical field of long-distance pipeline network heating systems, and specifically relates to an automatic driving platform for long-distance pipeline network heating systems based on a large model and a knowledge graph. Background Art

[0002] With the increasing demand for heat from urban heat users year by year, the long-distance pipeline heating system is being used more and more widely during the heating season in various cities. Due to the long distance of the long-distance pipeline network, the large amount of operating data, the complex equipment, and the strong coupling of multi-stage pumping stations, the scheduling and operation of the long-distance pipeline network heating system has become more complicated, and the accuracy requirements of the scheduling technology level have become increasingly higher.

[0003] At present, large language models are being applied in various fields as a forward-looking technical means. The language model based on deep learning can learn and train from massive amounts of data, showing more powerful semantic understanding, generation capabilities and context perception capabilities, etc., providing unprecedented opportunities for the scheduling and operation of long-distance pipeline heating systems and autonomous driving. However, there is currently little technical basis for using large language models for the scheduling and operation of long-distance pipeline heating systems. Therefore, how to integrate large language models into long-distance pipeline heating systems, provide effective technical support for decision-making to ensure the safe and stable operation of long-distance pipeline heating systems, and realize autonomous driving of long-distance pipeline heating systems is an urgent problem that needs to be solved. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide an automatic driving platform for the long-distance pipeline heating system based on a large model and knowledge graph, which closely combines data, model calculation, expert experience and business knowledge, and utilizes the powerful natural language parsing ability, semantic understanding ability and knowledge reasoning ability provided by the large language model to carry out unified management of business and multi-source data, provide effective technical support for decision-making to ensure the safe and stable operation of the long-distance pipeline heating system, and realize automatic driving of the long-distance pipeline heating system.

[0005] In order to solve the above technical problems, the technical solution of the present invention is:

[0006] The present invention provides an autonomous driving platform for a long-distance pipeline heating system based on a large model and a knowledge graph, comprising:

[0007] The infrastructure layer includes a facility access unit, which is used to connect the actual heat source of the long-distance pipeline network heating system, the long-distance pipeline network, and the multi-stage pump station heating facilities to the platform;

[0008] The data processing layer includes a data acquisition unit for collecting operating data of the heating facilities in the infrastructure layer, outdoor weather data, and external dynamic data; and a multi-source data fusion unit for storing, pre-processing, and processing multi-type data from different sources using data fusion technology;

[0009] The digital twin layer includes a long-distance pipeline network heating system digital twin model establishment unit, which is used to map the actual long-distance pipeline network heating system into a virtual environment using mechanism modeling and data-driven methods to establish a digital twin model of the long-distance pipeline network heating system;

[0010] The large language model-knowledge graph layer includes a graph construction unit for extracting, processing, and fusing knowledge to construct knowledge graphs for each business. It also includes a large language model construction unit for building and training large language models for each business, handling autonomous driving-related businesses for long-distance pipeline heating systems.

[0011] The business application decision-making layer includes a business application unit, which is used to provide business application modules for the long-distance pipeline network heating system, including a long-distance heating load forecasting module, a long-distance heating network monitoring module, a long-distance heating operation protection module, a long-distance heating operation strategy decision module, and a long-distance heating working condition rehearsal module; it also includes an intelligent decision-making unit, which is used to perform simulation analysis based on the digital twin model of the long-distance pipeline network heating system and call each business large model to output the corresponding business decision results.

[0012] Furthermore, the infrastructure layer also includes a computing power center and a high-performance network; the computing power center is used to utilize high-performance computing resources to meet model training and computing needs; the high-performance network is used to utilize high-speed technology to improve data transmission efficiency and reduce communication delays.

[0013] Furthermore, the data types in the data collection unit include structured data tables, time series data, and unstructured data;

[0014] The operating data includes the steam volume and heat production of the heat source, the temperature of the hot water or steam output by the heat source, fuel consumption and pollutant emissions, the temperature, flow rate and pressure of the long-distance pipeline network, the pressure, flow rate and operating power of the relay pump station and pressure isolation station in the multi-stage pump station, the thermal data of the urban primary network between the pressure isolation station and the thermal power station, and the control parameters of each electric regulating valve and circulating pump;

[0015] The external dynamic data includes user demand response data, heat transaction data, extreme cold weather data and abnormal fault data;

[0016] The data fusion technology is used to store, preprocess and process multi-type data from different sources, including: establishing distributed column databases, large-scale parallel databases, relational databases, and graph databases for data storage at different scales; performing missing value processing, outlier processing, format conversion and standardization on multi-type data from different sources; and performing data distribution perception, data value activation, data value assessment, and cross-modal feature fusion on multi-type data from different sources.

[0017] Furthermore, the mechanism modeling and data-driven approach is used to map the actual long-distance pipeline network heating system into a virtual environment to establish a digital twin model of the long-distance pipeline network heating system, including:

[0018] The long-distance pipeline heating system's heat source, long-distance pipeline network, and multi-stage pumping station physical entities, as well as the structural relationships and heat flow relationships between these physical entities, are integrated into a long-distance pipeline heating system mechanism model. Relevant parameters of the long-distance pipeline heating system are input to verify the dynamic characteristics of the mechanism model. The actual operating results are compared with the simulation results. After iteratively optimizing the parameters of the mechanism model, a high-precision mechanism model of the long-distance pipeline heating system is obtained.

[0019] Obtain the initial data set from the data processing layer and use machine learning algorithms to train objects related to the actual operation of the long-distance pipeline network heating system or unknown mechanisms to obtain a data-driven model. After forming a hybrid drive with the long-distance pipeline network heating system mechanism model, a digital twin model of the long-distance pipeline network heating system is established;

[0020] Obtain a new long-distance pipeline network heating system operation data set from the data processing layer and divide it into a training set and a test set;

[0021] The training set is input into the established digital twin model of the long-distance pipeline network heating system for training, and the model is fine-tuned using an intelligent optimization algorithm;

[0022] The fine-tuned digital twin model is evaluated using the test set. If the model evaluation results do not meet expectations, further fine-tuning is performed; otherwise, the new long-distance pipeline network heating system operation dataset is added to the initial dataset, and the model is iteratively trained and updated again to establish the final long-distance pipeline network heating system digital twin model.

[0023] Furthermore, the construction of each business knowledge graph after knowledge extraction, processing, and fusion includes:

[0024] For each business of the long-distance pipeline heating system, including long-distance heating load forecasting, long-distance heating network monitoring, long-distance heating operation protection, long-distance heating operation strategy decision-making, and long-distance heating operating condition rehearsal, we obtain data sources related to each business and manually annotate the data related to each business, including entity annotation, relationship annotation, and business type annotation.

[0025] The manually labeled business-related data is used as a training sample set, and a machine learning algorithm or natural language processing technology is used to train an intelligent labeling model. The trained model is then used to automatically label the unlabeled business data to obtain the labeled business data.

[0026] After the annotated business data is subjected to entity recognition and relationship extraction, the entity data and the relationship between entities are obtained. Duplicate entities and relationships are removed, similar information is merged, and entity disambiguation, entity alignment, and knowledge merging are performed. Finally, the entities and relationships are integrated and output in triple format.

[0027] The labeled business data, entity data and relationship data are stored in a graph data structure to obtain various business knowledge graphs, including the long-distance heating load forecast business knowledge graph, the long-distance heating network monitoring business knowledge graph, the long-distance heating operation protection business knowledge graph, the long-distance heating operation strategy decision-making business knowledge graph, and the long-distance heating working condition rehearsal business knowledge graph.

[0028] Furthermore, the construction and training of large language models for various businesses to handle autonomous driving related businesses of the long-distance pipeline heating system include:

[0029] Acquire multiple types of data from different sources in the actual system from the data processing layer; obtain system virtual operation data and future operation status data based on the digital twin model of the long-distance pipeline network heating system; obtain business-related entities and relationships based on the constructed business knowledge graphs, and jointly build the business knowledge corpus;

[0030] Through the business knowledge corpus, the large language model is trained and constructed in various business application directions. The business knowledge corpus is trained using deep learning algorithms to obtain the large language model of each business, which serves as the intelligent brain of each business. According to the operating status of the long-distance pipeline heating system and matching the knowledge graphs of each business, the long-distance pipeline heating system automatic driving related business is processed, and the long-distance heating load forecast results, long-distance heating network monitoring results, long-distance heating operation protection results, long-distance heating optimal operation strategy, and long-distance heating operation condition rehearsal results are output.

[0031] Furthermore, the deep learning algorithm is used to train each business knowledge corpus to obtain a large language model for each business, including:

[0032] Each business knowledge corpus is used as a pre-training sample set for the large language model of each business, and is divided into a training set and a test set for each business;

[0033] Select a deep learning algorithm that matches each business, use the training set data as input and the business results as output, conduct learning and pre-training, and build a universal language model for each business with general knowledge capabilities;

[0034] Evaluate the large language model common to each business using the test set data. If the evaluation does not meet expectations, continue pre-training and parameter fine-tuning until the large language model common to each business meets expectations.

[0035] Based on the specific application scenarios of each business, we use pre-set business-specific data fine-tuning samples, reinforcement learning algorithms based on human feedback, and knowledge graphs of each business. At the same time, we combine the knowledge and experience of experts from each business to fine-tune and optimize the model parameters of the common large language model of each business, improve the accuracy of the large language model of each business and the model's ability to understand natural language, and output the large language model of each business.

[0036] Furthermore, the deep learning algorithm is embedded in the large language model of each business, and the deep learning algorithm is a multimodal algorithm that recognizes, understands and generates multiple types of data from different sources in the system, including image data, text data, voice data, and video data;

[0037] The human feedback-based reinforcement learning algorithm is based on a pre-trained large language model common to each business. By fine-tuning the reward model and the optimization algorithm, it iteratively outputs a natural language generation model that is more consistent with the operation of the long-distance pipeline network heating system.

[0038] Furthermore, the large language model-knowledge graph layer also includes a large language model-knowledge graph fusion unit, which is used to deeply fuse the constructed knowledge graph and the large language model, extract entities and relationships between entities from each business data through the large language model, fuse the entities according to the similarity between the entities, integrate the fused entities and the relationships between entities, and output them in triple format to form each business knowledge graph, and use each business knowledge graph to assist each business large language model in understanding the context, use the association relationship in the knowledge graph as a constraint condition and combine the information of the knowledge graph to guide the output results of each business large language model, and train and adjust the large language model by continuously correcting and improving the knowledge graph.

[0039] Furthermore, the long-distance pipeline heating system automatic driving platform also includes an information analysis layer, which is used to automatically parse and structure the received real-time business signals, and match business needs from various business knowledge graphs and large language models.

[0040] The beneficial effects of the present invention are:

[0041] The present invention constructs an automatic driving platform for the long-distance pipeline network heating system, which includes an infrastructure layer, a data processing layer, a digital twin layer, a large language model-knowledge graph layer, and a business application decision layer. It mainly combines digital twin technology, large language model technology, and knowledge graph technology to realize the simulation analysis of the actual operation of the long-distance pipeline network heating system through the digital twin model, and the intelligent analysis, decision-making, and operation management of the long-distance heating load forecasting business, the long-distance heating network monitoring business, the long-distance heating operation protection business, the long-distance heating operation strategy decision-making business, and the long-distance heating working condition preview business through the large language model and knowledge graph, so that the dispatching personnel can scientifically It can scientifically and rationally carry out dispatching and operation management. For example, after the weather changes, it can automatically adjust the heat source load, execute equipment control parameter adjustments, and synchronize the operation of the first station, relay pump station, and pressure isolation station of the long-distance pipeline network. It can improve the comprehensive utilization of internal knowledge of the long-distance pipeline network heating system, closely combine data, model calculations, expert experience and business knowledge, and use the powerful natural language parsing ability, semantic understanding ability and knowledge reasoning ability provided by the large language model to carry out unified management of business and multi-source data, provide effective technical support for decision-making to ensure the safe and stable operation of the long-distance pipeline network heating system, and realize automatic driving of the long-distance pipeline network heating system.

[0042] Other features and advantages will be described in the following description, and in part will become apparent from the description, or understood by practicing the invention. The purpose and other advantages of the invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a schematic diagram of the architecture of an autonomous driving platform for a long-distance pipeline heating system based on a large model and knowledge graph in the present invention;

[0046] Figure 2 Establish a knowledge map flow chart of each business of the long-distance pipeline network heating system for the present invention;

[0047] Figure 3 A large language model flow chart of each business of the long-distance pipeline network heating system is established for the present invention. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0049] Example 1

[0050] like Figure 1 As shown, this embodiment 1 provides an automatic driving platform for a long-distance pipeline network heating system based on a large model and a knowledge graph, including:

[0051] The infrastructure layer includes a facility access unit, which is used to connect the actual heat source of the long-distance pipeline network heating system, the long-distance pipeline network, and the multi-stage pump station heating facilities to the platform;

[0052] The data processing layer includes a data acquisition unit for collecting operating data of the heating facilities in the infrastructure layer, outdoor weather data, and external dynamic data; and a multi-source data fusion unit for storing, pre-processing, and processing multi-type data from different sources using data fusion technology;

[0053] The digital twin layer includes a long-distance pipeline network heating system digital twin model establishment unit, which is used to map the actual long-distance pipeline network heating system into a virtual environment using mechanism modeling and data-driven methods to establish a digital twin model of the long-distance pipeline network heating system;

[0054] The large language model-knowledge graph layer includes a graph construction unit for extracting, processing, and fusing knowledge to construct knowledge graphs for each business. It also includes a large language model construction unit for building and training large language models for each business, handling autonomous driving-related businesses for long-distance pipeline heating systems.

[0055] The business application decision-making layer includes a business application unit, which is used to provide business application modules for the long-distance pipeline network heating system, including a long-distance heating load forecasting module, a long-distance heating network monitoring module, a long-distance heating operation protection module, a long-distance heating operation strategy decision module, and a long-distance heating working condition rehearsal module; it also includes an intelligent decision-making unit, which is used to perform simulation analysis based on the digital twin model of the long-distance pipeline network heating system and call each business large model to output the corresponding business decision results.

[0056] It should be noted that the long-distance pipeline heating system autonomous driving platform also includes:

[0057] The emergency response layer is used to automatically generate emergency response plans, assessment plans, and implementation plans for the long-distance pipeline network heating system under emergency conditions including sudden weather changes, emergency operating conditions, and changes in heat sources;

[0058] The safe operation assessment unit is used to set corresponding security algorithms for each business decision result output by the business application decision layer to ensure the safe execution of each business decision.

[0059] In this embodiment, the infrastructure layer also includes a computing power center and a high-performance network; the computing power center is used to utilize high-performance computing resources to meet model training and computing needs; the high-performance network is used to utilize high-speed technology to improve data transmission efficiency and reduce communication delays.

[0060] In practical applications, computing centers include high-performance hardware, which provides computational support for large language models. These centers are primarily composed of AI chips and AI servers. The long-distance pipeline heating system is complex, with large data volumes and computational workloads. Large language models for each service are built on multimodal data, requiring significant training data volumes. Therefore, high-performance hardware computing power is crucial for effectively training large language models with larger parameters, higher accuracy, and higher capabilities.

[0061] In this embodiment, the data types in the data collection unit include structured data tables, time series data, and unstructured data;

[0062] The operating data includes the steam volume and heat production of the heat source, the temperature of the hot water or steam output by the heat source, fuel consumption and pollutant emissions, the temperature, flow rate and pressure of the long-distance pipeline network, the pressure, flow rate and operating power of the relay pump station and pressure isolation station in the multi-stage pump station, the thermal data of the urban primary network between the pressure isolation station and the thermal power station, and the control parameters of each electric regulating valve and circulating pump;

[0063] The external dynamic data includes user demand response data, heat transaction data, extreme cold weather data and abnormal fault data;

[0064] The data fusion technology is used to store, preprocess and process multi-type data from different sources, including: establishing distributed column databases, large-scale parallel databases, relational databases, and graph databases for data storage at different scales; performing missing value processing, outlier processing, format conversion and standardization on multi-type data from different sources; and performing data distribution perception, data value activation, data value assessment, and cross-modal feature fusion on multi-type data from different sources.

[0065] In this embodiment, the mechanism modeling and data-driven approach is used to map the actual long-distance pipeline network heating system into a virtual environment to establish a digital twin model of the long-distance pipeline network heating system, including:

[0066] The long-distance pipeline heating system's heat source, long-distance pipeline network, and multi-stage pumping station physical entities, as well as the structural relationships and heat flow relationships between these physical entities, are integrated into a long-distance pipeline heating system mechanism model. Relevant parameters of the long-distance pipeline heating system are input to verify the dynamic characteristics of the mechanism model. The actual operating results are compared with the simulation results. After iteratively optimizing the parameters of the mechanism model, a high-precision mechanism model of the long-distance pipeline heating system is obtained.

[0067] Obtain the initial data set from the data processing layer and use machine learning algorithms to train objects related to the actual operation of the long-distance pipeline network heating system or unknown mechanisms to obtain a data-driven model. After forming a hybrid drive with the long-distance pipeline network heating system mechanism model, a digital twin model of the long-distance pipeline network heating system is established;

[0068] Obtain a new long-distance pipeline network heating system operation data set from the data processing layer and divide it into a training set and a test set;

[0069] The training set is input into the established digital twin model of the long-distance pipeline network heating system for training, and the model is fine-tuned using an intelligent optimization algorithm;

[0070] The fine-tuned digital twin model is evaluated using the test set. If the model evaluation results do not meet expectations, further fine-tuning is performed; otherwise, the new long-distance pipeline network heating system operation dataset is added to the initial dataset, and the model is iteratively trained and updated again to establish the final long-distance pipeline network heating system digital twin model.

[0071] like Figure 2 As shown, in this embodiment, the construction of each business knowledge graph after knowledge extraction, processing, and fusion includes:

[0072] For each business of the long-distance pipeline heating system, including long-distance heating load forecasting, long-distance heating network monitoring, long-distance heating operation protection, long-distance heating operation strategy decision-making, and long-distance heating operating condition rehearsal, we obtain data sources related to each business and manually annotate the data related to each business, including entity annotation, relationship annotation, and business type annotation.

[0073] The manually labeled business-related data is used as a training sample set, and a machine learning algorithm or natural language processing technology is used to train an intelligent labeling model. The trained model is then used to automatically label the unlabeled business data to obtain the labeled business data.

[0074] After the annotated business data is subjected to entity recognition and relationship extraction, the entity data and the relationship between entities are obtained. Duplicate entities and relationships are removed, similar information is merged, and entity disambiguation, entity alignment, and knowledge merging are performed. Finally, the entities and relationships are integrated and output in triple format.

[0075] The labeled business data, entity data and relationship data are stored in a graph data structure to obtain various business knowledge graphs, including the long-distance heating load forecast business knowledge graph, the long-distance heating network monitoring business knowledge graph, the long-distance heating operation protection business knowledge graph, the long-distance heating operation strategy decision-making business knowledge graph, and the long-distance heating working condition rehearsal business knowledge graph.

[0076] In practical applications, storing annotated business data, entity data, and relationship data in a graph data structure to build a specific business knowledge graph is an important step in achieving intelligent management of long-distance pipeline heating systems. The following is the knowledge graph construction process for each specific business scenario:

[0077] 1. Long-distance heating load forecasting business knowledge graph

[0078] Entities: heat sources, heat users, pumping stations, heat exchange stations;

[0079] Attributes: historical load data, real-time temperature, pressure, flow, etc.;

[0080] Relationships: supply paths from heat sources to users, impact of weather on load, mutual impact between users, etc.

[0081] Application scenario: Use historical data and current status to predict future heat demand.

[0082] 2. Knowledge Graph for Long-Distance Heating Network Monitoring

[0083] Entities: sensors, pipe segments, valves, pumping stations, etc.

[0084] Attributes: temperature, pressure, flow, equipment operating status, etc.;

[0085] Relationship: connection relationship between devices, coupling relationship of the entire network, etc.;

[0086] Application scenario: Real-time monitoring of the operation status of the entire network.

[0087] 3. Knowledge map of long-distance heating operation protection business

[0088] Entity: key equipment (such as boilers, pumps), control systems, key nodes of the pipeline network, etc.;

[0089] Attributes: equipment parameter settings, operation logs, alarm information, etc.;

[0090] Relationships: dependencies between devices, control logic, failure mode analysis, etc.

[0091] Application scenarios: Ensure the stable operation of the system and quickly locate risk points and fault types.

[0092] 4. Long-distance heating operation strategy decision-making business knowledge map

[0093] Entity: multi-stage pumping station, water pumps, valves, units, etc.;

[0094] Attributes: cost-benefit analysis, carbon emissions, energy consumption, etc.

[0095] Relationships: impacts between pumping stations, impact of pump operating frequency on flow, impact of valve opening on flow, impact of multiple pump combinations on flow, impact of external factors on strategy selection, etc.

[0096] Application scenario: Develop optimized operation strategies to reduce operating costs while meeting service requirements.

[0097] 5. Long-distance heating operation condition preview business knowledge map

[0098] Entity: simulation scenarios, emergency plans, operating procedures, etc.;

[0099] Attributes: expected results, execution time, resource consumption, etc.

[0100] Relationship: the correspondence between the plan and the actual event, the substitution or complementary relationship between different plans, etc.;

[0101] Application scenario: Test different operation plans in a virtual environment and evaluate their feasibility and effectiveness.

[0102] In this embodiment, the construction and training of large language models for various services to handle services related to autonomous driving in a long-distance pipeline heating system include:

[0103] Acquire multiple types of data from different sources in the actual system from the data processing layer; obtain system virtual operation data and future operation status data based on the digital twin model of the long-distance pipeline network heating system; obtain business-related entities and relationships based on the constructed business knowledge graphs, and jointly build the business knowledge corpus;

[0104] Through the business knowledge corpus, the large language model is trained and constructed in various business application directions. The business knowledge corpus is trained using deep learning algorithms to obtain the large language model of each business, which serves as the intelligent brain of each business. According to the operating status of the long-distance pipeline heating system and matching the knowledge graphs of each business, the long-distance pipeline heating system automatic driving related business is processed, and the long-distance heating load forecast results, long-distance heating network monitoring results, long-distance heating operation protection results, long-distance heating optimal operation strategy, and long-distance heating operation condition rehearsal results are output.

[0105] In practical applications, long-distance heat load forecasting leverages historical and real-time load data, weather conditions, holiday conditions, and user behavior to predict future heat load demand for long-distance pipeline networks. This is achieved by pre-training and learning a large language model on general data and fine-tuning it with a small amount of specific data. This model is then used to generate load forecasts, providing a basis for subsequent operational decisions.

[0106] The long-distance heating network monitoring service can dynamically simulate and calculate real-time water pressure diagrams based on the digital twin model of the long-distance pipeline network heating system, and graphically monitor and analyze the operating data. By comparing the digital twin model simulation data, the efficient operating range of multi-stage pump station pumps is comprehensively considered, and the operating parameters are monitored in real time. Data thresholds and change gradient standards are established, and alarms are generated when the range of the judgment value is exceeded. The constructed long-distance heating network monitoring large language model can be used to monitor and evaluate the operating status of the long-distance heating network in real time, and the monitoring results can be converted into a natural language representation of the long-distance pipeline network heating system network operation monitoring that can be understood by heating dispatchers. Then, combined with the graphical plug-in, a visual display result is formed to assist heating dispatchers in quickly understanding the operating status of the entire system network.

[0107] The long-distance heating operation monitoring service can sense the actual operating conditions based on the digital twin model of the long-distance pipeline network heating system, predict the subsequent changes in the operating data of the conditions, and issue timely warnings for risks and faults. It also has a built-in monitoring rule engine to conduct real-time analysis and matching of the operation conditions of the entire network, issue warnings for risks and fault conditions, and recommend treatment solutions. In actual operation, sensor data is mainly used to identify risks and fault types, but the small amount of data and the difficulty of installing sensors to collect some data are a difficulty. In addition, the risk and fault diagnosis are complex, and it is difficult to diagnose multiple risks and fault types at the same time. Therefore, the constructed long-distance heating operation monitoring large language model can be used to solve this problem. Small sample fine-tuning learning is used to identify and diagnose risks and faults. At the same time, a comprehensive operation monitoring report is automatically generated based on the identification and diagnosis results and the system operation status. The knowledge learning of different types of risks and fault diagnoses is integrated into one model to achieve simultaneous diagnosis of multiple types of risks and faults.

[0108] The long-distance heating operation strategy decision-making business is to analyze the pump operation characteristics of multi-stage pumping stations under different flow rates, optimize the water pressure distribution of the entire long-distance heating pipeline network based on the digital twin model, and issue the operation strategy of the multi-stage pumping station based on the different electricity prices and pump efficiencies of each pump room. In addition, based on the digital twin model, combined with the actual operating conditions of the units in the heat source and the production plan required by the long-distance pipeline network, the target flow rate is adjusted. The constructed long-distance heating operation strategy decision-making large language model can provide heating dispatchers with the optimal operation strategy for the pumps of the multi-stage pumping stations in the long-distance pipeline network, provide intelligent decision-making support based on the operating status and heat load demand value, and automatically generate an operation decision plan report. It can also simulate the business operations of various equipment based on voice, text and other commands, and implement dynamic feedback from the system.

[0109] The long-distance heating supply condition rehearsal service uses a digital twin model to rehearse different operational strategies. This allows users to visualize the hydraulic conditions caused by each strategy, verify the rationality of the strategy, and provide an effective operational basis for actual production. The constructed large language model for long-distance heating condition rehearsal provides heating dispatchers with the operating conditions after executing each strategy and automatically generates rehearsal reports for each strategy.

[0110] It should be noted that the large language models for various business operations in the long-distance pipeline heating system utilize the Transformer architecture during pre-training. This architecture is capable of solving a variety of natural language processing problems and can be trained using unlabeled corpus samples. Data or device information collected by various sensors in the long-distance pipeline heating system can be directly input into the Transformer model for training, saving significant annotation labor. The Transformer model consists of an encoder and decoder, and is primarily optimized using the decoder and a masked self-attention mechanism.

[0111] like Figure 3 As shown, in this embodiment, the deep learning algorithm is used to train each business knowledge corpus to obtain each business language model, including:

[0112] Each business knowledge corpus is used as a pre-training sample set for the large language model of each business, and is divided into a training set and a test set for each business;

[0113] Select a deep learning algorithm that matches each business, use the training set data as input and the business results as output, conduct learning and pre-training, and build a universal language model for each business with general knowledge capabilities;

[0114] Evaluate the large language model common to each business using the test set data. If the evaluation does not meet expectations, continue pre-training and parameter fine-tuning until the large language model common to each business meets expectations.

[0115] Based on the specific application scenarios of each business, we use pre-set business-specific data fine-tuning samples, reinforcement learning algorithms based on human feedback, and knowledge graphs of each business. At the same time, we combine the knowledge and experience of experts from each business to fine-tune and optimize the model parameters of the common large language model of each business, improve the accuracy of the large language model of each business and the model's ability to understand natural language, and output the large language model of each business.

[0116] In this embodiment, the deep learning algorithm is embedded in the large language model of each business, and the deep learning algorithm is a multimodal algorithm that recognizes, understands and generates multiple types of data from different sources of the system. The multiple types of data include image data, text data, voice data, and video data;

[0117] The human feedback-based reinforcement learning algorithm is based on a pre-trained large language model common to each business. By fine-tuning the reward model and the optimization algorithm, it iteratively outputs a natural language generation model that is more consistent with the operation of the long-distance pipeline network heating system.

[0118] In this embodiment, the large language model-knowledge graph layer also includes a large language model-knowledge graph fusion unit, which is used to deeply fuse the constructed knowledge graph and the large language model, extract entities and relationships between entities from each business data through the large language model, fuse the entities according to the similarity between the entities, integrate the fused entities and the relationships between entities, and output them in triple format to form each business knowledge graph, and use each business knowledge graph to assist each business large language model in understanding the context, use the association relationship in the knowledge graph as a constraint condition and combine the information of the knowledge graph to guide the output results of each business large language model, and train and adjust the large language model by continuously correcting and improving the knowledge graph.

[0119] It should be noted that deep integration of knowledge graphs and large language models (LLMs) can fully leverage the advantages of both: knowledge graphs provide structured domain knowledge and relationships between entities, while large language models excel at natural language understanding and generation tasks. The following are the steps to achieve this process:

[0120] 1. Use large language models to generate knowledge graphs:

[0121] 1) Entity extraction and relationship extraction

[0122] Use large language models to extract entities and relationships between entities from unstructured or semi-structured text data;

[0123] Pre-trained large language model: Choose a suitable pre-trained large language model, such as BERT, RoBERTa, etc.

[0124] Fine-tuning: Fine-tune the large language model on labeled data in a specific domain to improve its performance in that domain;

[0125] Named Entity Recognition (NER): Train the model to identify key entities in text, such as device names and fault types;

[0126] Relationship extraction: The model is further trained to identify and classify the relationships between these entities, such as "located in", "connected to", "affects", etc.

[0127] 2) Entity Fusion

[0128] Entities are merged based on their similarity to reduce redundancy and enhance consistency;

[0129] Similarity calculation: Use word vectors (such as Word2Vec, GloVe) or sentence embeddings (such as Sentence-BERT) to calculate the semantic similarity between entities;

[0130] Rule engine: Sets certain rules to determine whether two entities should be merged, such as based on string matching, context information, etc.

[0131] Manual review: For uncertain situations, manual review can be used to decide whether to merge.

[0132] 3) Knowledge graph construction

[0133] Output the fused entities and inter-entity relationships in triple format and integrate them into the knowledge graph;

[0134] Triple format: Each entity and its relationship are represented in the form of (subject, relationship, object);

[0135] Graph database import: import these triples into the graph database and create nodes (entities) and edges (relationships);

[0136] Attribute addition: Add additional attributes to each node and edge, such as timestamp, confidence, etc.

[0137] 2. Using knowledge graphs to optimize large language models:

[0138] 1) Enhance comprehension

[0139] Context enrichment: By querying the knowledge graph, richer background information is provided for the input text, helping large language models better understand the context;

[0140] Constrained reasoning: Use known relationships in the knowledge graph as constraints to guide the model to generate more reasonable and accurate answers.

[0141] 2) Improve generation quality

[0142] Knowledge injection: During the generation process, information from the knowledge graph is combined to ensure that the generated content is consistent with the facts and avoid incorrect or misleading answers;

[0143] Template filling: Use the structured information provided by the knowledge graph to fill in pre-defined templates and generate high-quality text content.

[0144] 3) Continuous learning and updating

[0145] Feedback loop: Establish a feedback mechanism so that users can continuously correct and improve the knowledge graph through interaction. At the same time, this new information can also be used to further train and adjust the large language model;

[0146] Incremental updates: As new data is continuously added, the knowledge graph is updated regularly and the large language model is adjusted accordingly to maintain its timeliness and accuracy.

[0147] The technology selection and tools for the large language model-knowledge graph layer are as follows:

[0148] Large language models: BERT, RoBERTa, GPT series, etc.;

[0149] Knowledge graph storage: Neo4j, JanusGraph, Amazon Neptune, etc.

[0150] Entity recognition and relation extraction: Spacy, Stanford NLP, Hugging Face Transformers, etc.

[0151] Similarity calculation: Word2Vec, Sentence-BERT, etc.

[0152] In this embodiment, the long-distance pipeline heating system automatic driving platform also includes an information analysis layer for automatically parsing and structured representation of received real-time business signals, and matching business needs from various business knowledge graphs and large language models.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the systems, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a portion of code, and the module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0154] In addition, the functional modules in each embodiment of the present invention can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part. If the function is implemented in the form of a software functional module and sold or used as an independent product, it 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 the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0155] With the above-described preferred embodiments of the present invention as a guide, and with reference to the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. An automatic driving platform for a long-distance pipeline heating system based on a large model and knowledge graph, characterized in that: include: The infrastructure layer includes a facility access unit, which is used to connect the actual heat source of the long-distance pipeline network heating system, the long-distance pipeline network, and the multi-stage pump station heating facilities to the platform; The data processing layer includes a data acquisition unit for collecting the operation data of the heating facilities in the infrastructure layer, outdoor weather data and external dynamic data; and also includes a multi-source data fusion unit for storing, pre-processing and processing multi-type data from different sources using data fusion technology; The digital twin layer includes a digital twin model building unit for the long-distance pipeline network heating system, which is used to map the actual long-distance pipeline network heating system into a virtual environment by using a mechanism modeling and data-driven approach, and to build a digital twin model for the long-distance pipeline network heating system; The large language model-knowledge graph layer includes a graph construction unit for extracting, processing, and fusing knowledge to construct knowledge graphs for each business; it also includes a large language model construction unit for constructing and training large language models for each business, and processing the long-distance pipeline network heating system autonomous driving related business; The business application decision-making layer includes a business application unit, which is used to provide business application modules for the long-distance pipeline heating system, including a long-distance heating load forecasting module, a long-distance heating full-network monitoring module, a long-distance heating operation protection module, a long-distance heating operation strategy decision module, and a long-distance heating operating condition rehearsal module; it also includes an intelligent decision-making unit, which is used to perform simulation analysis based on the digital twin model of the long-distance pipeline heating system and call each business large model to output the corresponding business decision results.

2. The automatic driving platform for the long-distance pipeline network heating system according to claim 1 is characterized in that: The infrastructure layer also includes a computing power center and a high-performance network; the computing power center is used to utilize high-performance computing resources to meet model training and computing needs; the high-performance network is used to utilize high-speed technology to improve data transmission efficiency and reduce communication delays.

3. The automatic driving platform for the long-distance pipeline network heating system according to claim 1 is characterized in that: The data types in the data collection unit include structured data tables, time series data, and unstructured data; The operation data include the steam volume, heat production, temperature of hot water or steam output by the heat source, fuel consumption and pollutant emissions, temperature, flow rate and pressure of the long-distance pipeline network, pressure, flow rate and operating power of the relay pump station and the pressure isolation station in the multi-stage pump station, thermal data of the city primary network between the pressure isolation station and the thermal power station, and control parameters of each electric regulating valve and circulating pump; The external dynamic data includes user demand response data, heat transaction data, extreme cold weather data and abnormal fault data; The use of data fusion technology to store, preprocess and process multiple types of data from different sources includes: establishing a distributed column database, a large-scale parallel database, a relational database, and a graph database to store data at different scales; Perform missing value processing, outlier processing, format conversion and standardization on multi-type data from different sources; perform data distribution perception, data value activation, data value assessment and cross-modal feature fusion on multi-type data from different sources.

4. The automatic driving platform for the long-distance pipeline network heating system according to claim 1 is characterized in that: The method of using mechanism modeling and data-driven to map the actual long-distance pipeline network heating system into a virtual environment and establish a digital twin model of the long-distance pipeline network heating system includes: The heat source of the long-distance pipeline network heating system, the long-distance pipeline network, the physical entities of the multi-stage pumping station, the structural relationship between the physical entities and the heat flow relationship are integrated into a long-distance pipeline network heating system mechanism model, and the relevant parameters of the long-distance pipeline network heating system are input to verify the dynamic characteristics of the mechanism model, compare the differences between the actual operation results and the simulation results, and iteratively optimize the parameters of the mechanism model to obtain a high-precision mechanism model of the long-distance pipeline network heating system. Obtain the initial data set from the data processing layer, and use machine learning algorithms to train the objects with unknown processes or mechanisms related to the actual operation of the long-distance pipeline network heating system to obtain a data-driven model. After forming a hybrid drive with the mechanism model of the long-distance pipeline network heating system, a digital twin model of the long-distance pipeline network heating system is established. Obtain a new long-distance pipeline network heating system operation data set from the data processing layer and divide it into a training set and a test set; The training set is input into the established digital twin model of the long-distance pipeline network heating system for training, and the model is fine-tuned using an intelligent optimization algorithm; The fine-tuned digital twin model is evaluated using the test set. If the model evaluation result does not meet expectations, further fine-tuning is performed; otherwise, the new long-distance pipeline heating system operation data set is added to the initial data set, and the model is iteratively trained and updated again to establish the final digital twin model of the long-distance pipeline heating system.

5. The automatic driving platform for the long-distance pipeline network heating system according to claim 1 is characterized in that: The above-mentioned knowledge extraction, processing and fusion are used to construct the business knowledge graph, including: For the long-distance transmission pipeline network heating system services, including long-distance transmission heating load forecasting services, long-distance transmission heating network monitoring services, long-distance transmission heating operation protection services, long-distance transmission heating operation strategy decision-making services, and long-distance transmission heating working condition rehearsal services, obtain data sources related to each service, and manually label the data related to each service, including entity labeling, relationship labeling, and service type labeling; The manually annotated business-related data is used as a training sample set, and a machine learning algorithm or natural language processing technology is used to train an intelligent annotation model. The trained model is then used to automatically annotate unannotated business data to obtain annotated business data. After the annotated business data are respectively subjected to entity recognition and relationship extraction, entity data and relationships between entities are obtained, and duplicate entities and relationships are removed, similar information is merged, and entity disambiguation, entity alignment and knowledge merging are performed. Entities and relationships are integrated and output in triple format; The labeled business data, entity data and relationship data are stored in a graph data structure to obtain various business knowledge graphs, including the long-distance heating load forecasting business knowledge graph, the long-distance heating full network monitoring business knowledge graph, the long-distance heating operation protection business knowledge graph, the long-distance heating operation strategy decision-making business knowledge graph, and the long-distance heating operating condition rehearsal business knowledge graph.

6. The automatic driving platform for the long-distance pipeline network heating system according to claim 1 is characterized in that: The construction and training of large language models for various businesses to handle autonomous driving related businesses of long-distance pipeline heating systems include: Obtain multi-type data from different sources of the actual system from the data processing layer, obtain system virtual operation data and future operation status data based on the digital twin model of the long-distance pipeline network heating system, obtain entities and relationships related to each business based on the constructed business knowledge graphs, and jointly build the knowledge corpus of each business; Through the business knowledge corpus, the large language model is trained and constructed in the direction of various business applications. The business knowledge corpus is trained using deep learning algorithms to obtain the large language model of each business, which serves as the intelligent brain of each business. According to the operating status of the long-distance pipeline heating system and matching the business knowledge graphs, the long-distance pipeline heating system automatic driving related business is processed, and the long-distance heating load forecast results, long-distance heating full network monitoring results, long-distance heating operation protection results, long-distance heating optimal operation strategy, and long-distance heating operation conditions preview results are output.

7. The automatic driving platform for the long-distance pipeline network heating system according to claim 6 is characterized in that: The deep learning algorithm is used to train each business knowledge corpus to obtain each business language model, including: Each business knowledge corpus is used as a pre-training sample set for each business large language model, and is divided into a training set and a test set for each business; Select deep learning algorithms that match each business, use training set data as input and business results as output, conduct learning and pre-training, and build a universal large language model for each business with general knowledge capabilities; The constructed large language model common to each business is evaluated through the test set data. If the evaluation does not meet expectations, pre-training and parameter fine-tuning are continued until the large language model common to each business is output in accordance with expectations; According to the specific application scenarios of each business, we use pre-set business-specific data fine-tuning samples, reinforcement learning algorithms based on human feedback, and business knowledge graphs. At the same time, we combine the knowledge and experience of experts from each business to fine-tune and optimize the model parameters of the common big language model for each business, improve the accuracy of the big language model for each business and the model's ability to understand natural language, and output the big language model for each business.

8. The automatic driving platform for the long-distance pipeline network heating system according to claim 7 is characterized in that: The deep learning algorithm is embedded in the large language model of each business, and the deep learning algorithm is a multimodal algorithm that recognizes, understands and generates multi-type data from different sources of the system, including image data, text data, voice data, and video data; The reinforcement learning algorithm based on human feedback is based on a pre-trained large language model common to each business. Through fine-tuning of the reward model and the optimization algorithm, it iteratively outputs a natural language generation model that is more in line with the operation of the long-distance pipeline network heating system.

9. The automatic driving platform for the long-distance pipeline network heating system according to claim 1 is characterized in that: The large language model-knowledge graph layer also includes a large language model-knowledge graph fusion unit, which is used to deeply fuse the constructed knowledge graph and the large language model, extract entities and relationship between entities of each business data through the large language model, fuse the entities according to the similarity between the entities, integrate the fused entities and the relationship between the entities, and output them in triple format to form each business knowledge graph, and use each business knowledge graph to assist each business large language model to understand the context, use the association relationship in the knowledge graph as a constraint condition and combine the information of the knowledge graph to guide the output result of each business large language model, and train and adjust the large language model by continuously correcting and improving the knowledge graph.

10. The automatic driving platform for the long-distance pipeline network heating system according to claim 1, characterized in that: The long-distance pipeline heating system automatic driving platform also includes an information analysis layer, which is used to automatically parse and structure the received real-time business signals, and match business needs from various business knowledge graphs and large language models.

Images