Multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system and method

Abstract

The present application belongs to the technical field of industrial waste heat recovery and intelligent operation and maintenance, and particularly relates to a multi-source industrial waste heat collaborative gradient utilization and intelligent operation and maintenance management system and method, which is applied to a forging production workshop deployed with an AI intelligent calculation center, multiple AI cabinets and power distribution cabinets are arranged in the AI intelligent calculation center, and the system comprises a multi-source waste heat gradient utilization subsystem, a phase change heat storage unit, an AI intelligent operation and maintenance subsystem, at least one mobile robot and a dynamic scheduling controller; the integrated intelligent operation and maintenance system integrating AI predictive maintenance, natural language interaction and mobile robots is adopted, so as to solve the problems of low energy utilization rate, poor collaboration, low operation and maintenance efficiency and insufficient scattered heat coverage in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of industrial waste heat recovery and intelligent operation and maintenance technology, specifically involving a multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system and method, which is particularly suitable for comprehensive industrial parks or intelligent manufacturing production workshops that simultaneously have forging process waste heat, AI intelligent computing center cooling exhaust and mobile robot collaborative operation. Background Technology

[0002] With the rapid development of artificial intelligence and high-performance computing, the power density of AI computing centers (especially GPU server clusters) continues to rise. The hot air discharged from their cooling channels typically reaches temperatures of 40-55°C, which is typical low-grade waste heat. Meanwhile, traditional heavy industries such as forging and metallurgy generate large amounts of high-grade waste heat during production. For example, the flue gas temperature in forging furnaces can reach 500-900°C, and the radiant heat temperature during the cooling process of forgings can reach 150-500°C. Directly releasing this waste heat not only causes enormous energy waste but also exacerbates environmental thermal pollution.

[0003] Currently, technologies for recovering single types of waste heat are relatively mature. However, existing technologies suffer from several drawbacks: energy level matching is often crude, resulting in low recovery efficiency. Waste heat of different grades is frequently mixed or used for a single purpose, failing to be finely matched according to the principle of "temperature matching and tiered utilization." Multi-source collaboration is lacking, resulting in isolated system operation and a lack of unified energy scheduling and information exchange. Maintenance relies heavily on manual labor, with low levels of intelligence. Traditional waste heat recovery systems rely on fixed threshold alarms for equipment status monitoring, making fault prediction and root cause analysis impossible. Industrial robots are widely used in production tasks such as handling, welding, and painting, but they lack energy and information collaboration with waste heat recovery and energy scheduling systems. The mobility and autonomy of robots are not utilized for energy transmission or equipment maintenance.

[0004] In summary, there is an urgent need for an integrated intelligent operation and maintenance system that can utilize multi-grade industrial waste heat, especially high-temperature waste heat from forging, in a tiered and coordinated manner, and integrate AI predictive maintenance, natural language interaction, and mobile robot execution to solve the problems of low energy utilization, poor coordination, low operation and maintenance efficiency, and insufficient coverage of decentralized heat use in existing technologies. Summary of the Invention

[0005] The purpose of this invention is to solve the problems existing in the prior art by proposing a multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system and method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system, applied to a forging production workshop equipped with an AI intelligent computing center. The forging production workshop is equipped with a Manufacturing Execution System (MES) for outputting forging production plans. The AI ​​intelligent computing center is equipped with multiple AI cabinets and power distribution cabinets. The system is characterized by including a multi-source waste heat cascade utilization subsystem, a phase change heat storage unit, an AI intelligent operation and maintenance subsystem, and at least one mobile robot. The mobile robot includes a mobile robot controller, which wirelessly communicates with the AI ​​intelligent operation and maintenance subsystem to receive and execute scheduling instructions. The multi-source waste heat cascade utilization subsystem includes a high-temperature recovery unit, a medium-temperature recovery unit, and a low-temperature recovery unit. The high-temperature recovery unit includes a flue gas heat exchanger for recovering 500-900°C flue gas discharged from the forging heating furnace. The medium-temperature recovery unit includes a radiant heat collection hood and an air preheater. The radiant heat collection hood is used to recover 150-500°C radiant heat generated during the forging cooling process. The heat source-side inlet of the air preheater is connected to the outlet of the radiant heat collection hood, and the air-side outlet of the air preheater is connected to the combustion air inlet of the forging heating furnace. The low-temperature recovery unit includes an air-water heat exchanger, a plate heat exchanger, and a high-temperature heat pump for recovering 40-55°C hot air discharged from the AI ​​cabinet cooling channel and 40-70°C cooling water from the forging equipment. The multi-source waste heat cascade utilization subsystem also includes a fixed pipeline network, which consists of heating pipes connecting the plate heat exchanger outlet of the low-temperature stage recovery unit to each fixed heat-consuming terminal. The fixed heat-consuming terminals include heating terminals, domestic hot water terminals, low-temperature drying sections of spray drying lines, and industrial cleaning tanks. The mobile robot is used to provide mobile heating in decentralized heat demand scenarios that cannot be covered by the fixed pipeline network, through its onboard phase change heat storage module. The hot air discharged from the AI ​​cabinet cooling channel heats the circulating water through the air-water heat exchanger. The heated circulating water mixes with the cooling water of the forging equipment in the plate heat exchanger to form medium-temperature water at 50~60℃. A diversion valve is installed on the outlet pipe of the plate heat exchanger. The phase change thermal energy storage unit includes a first thermal energy storage module and a second thermal energy storage module. The first thermal energy storage module is connected to the flue gas heat exchanger via a heat transfer oil pipeline. The heat transfer oil pipeline is provided with a bypass outlet, which is connected to the organic Rankine cycle generator set via a first electric regulating valve. The second thermal energy storage module is connected to the plate heat exchanger via a water pipeline. The water pipeline is provided with a second electric regulating valve for controlling the charge and discharge heat flow rate. The AI-powered intelligent operation and maintenance subsystem includes a data acquisition module, a predictive alarm module, and a natural language interaction module. The data acquisition module includes: a temperature sensor and a flow sensor installed on the flue gas heat exchanger, the air-water heat exchanger, the plate heat exchanger, the high-temperature heat pump, and the AI ​​cabinet; and a vibration sensor A and a current sensor installed on the compressor of the high-temperature heat pump and the fan of the air-water heat exchanger. The predictive warning module includes an LSTM inference unit deployed on an edge computing gateway, which is communicatively connected to the data acquisition module. The weight parameters of the LSTM inference unit are obtained through offline training using historical fault samples. The LSTM inference unit receives timing data from the vibration sensor and the current sensor, and outputs the predicted remaining service life of the compressor of the high-temperature heat pump and the fan of the air-water heat exchanger or the probability of failure in the next 48 hours. The LSTM inference unit also receives time-series data from the temperature sensor and the flow sensor, and outputs the predicted remaining service life or the predicted heat exchange efficiency decay of the flue gas heat exchanger and the plate heat exchanger. The natural language interaction module includes a natural language processing unit, a voice input unit, a voice synthesis unit, an intent parsing unit, and a security verification unit deployed on a local server; the natural language interaction module also includes a system database, which is communicatively connected to the data acquisition module and is used to store time-series data and equipment ledger information collected by the temperature sensor, flow sensor, vibration sensor, and current sensor. The multi-source waste heat cascade utilization subsystem also includes a programmable logic controller (PLC), which is communicatively connected to the diversion valve, the high-temperature heat pump, the first electric regulating valve, the second electric regulating valve, and the mobile robot controller. After the safety verification unit passes the verification, it sends control commands to the target device through the PLC.

[0007] Furthermore, the first heat storage module is internally equipped with a first heat exchange coil, a first intermediate circulation pump, and a heat transfer oil pipeline; the inlet of the first heat exchange coil is connected to the heat transfer oil outlet of the flue gas heat exchanger of the high-temperature stage recovery unit through the first intermediate circulation pump and the heat transfer oil pipeline, and the outlet of the first heat exchange coil is connected to the heat transfer oil inlet of the flue gas heat exchanger through the heat transfer oil pipeline, forming a heat transfer oil circulation loop; a bypass outlet is provided on the pipeline between the outlet of the first heat exchange coil and the heat transfer oil inlet of the flue gas heat exchanger, and the bypass outlet is connected to the organic Rankine cycle generator set; The second heat storage module is equipped with a second heat exchange coil, a second intermediate circulation pump and water pipes; the inlet of the second heat exchange coil is connected to the hot water outlet of the plate heat exchanger of the low temperature stage recovery unit through the second intermediate circulation pump and water pipes, and the outlet of the second heat exchange coil is connected to the return water port of the plate heat exchanger.

[0008] Furthermore, the diversion valve is divided into two outlets, including a first outlet and a second outlet. All first outlets are connected to heating terminals or domestic hot water terminals, and the second outlet is connected to the evaporator side of the high-temperature heat pump as a heat source.

[0009] Furthermore, the second outlet is also connected to the condenser side of the high-temperature heat pump to output hot water or hot air at 75~85°C, which is then connected to the low-temperature drying section of the spray drying line or the industrial cleaning tank via a pipeline.

[0010] Furthermore, the predictive warning module also includes a root cause analysis submodule, which includes a causal graph knowledge base stored in the edge computing gateway. The causal graph knowledge base contains pre-stored causal association rules between fault modes and sensor features.

[0011] Furthermore, it also includes a dynamic scheduling controller, which includes a reinforcement learning inference module deployed on an industrial control computer. This module employs a near-end policy optimization algorithm, and its state space vector includes the load prediction value of the AI ​​intelligent computing center for the next N hours, the running time of the forging heating furnace in the forging production plan, and the forging production plan obtained from the Manufacturing Execution System (MES). The real-time power demand of at least one of the following heat-using terminals is also included: heating terminal, domestic hot water terminal, low-temperature drying terminal of the spray drying line, industrial cleaning tank, high-temperature curing terminal of the spray drying line, and temporary office space in the factory area. Public warehouse, outdoor equipment antifreeze heating belt, mobile spray painting trolley; real-time electricity price signal; the current position of the mobile robot, the current heat load status and remaining power of the phase change thermal storage module of the mobile robot; its motion space vector includes the opening degree of the diversion valve controlling the distribution of medium-temperature water; the start and stop and power of the high-temperature heat pump; the opening degree of the heat transfer oil bypass outlet valve connected to the organic Rankine cycle generator set; the opening degree of the water circulation loop valve controlling the charging and discharging of the second thermal storage module; issuing charging instructions, discharging instructions and moving to the designated position instructions to the mobile robot; its reward function R is calculated by the following formula: Where R is the reward function, in yuan. The electricity cost consumed by the system, in yuan; The gas cost consumed by the system is expressed in yuan. Equipment operation and maintenance costs, in yuan. This is the device switching penalty coefficient, expressed in yuan. Let be the dimensionless action value of the i-th device at time t. Let be the action value of the i-th device at time t-1, which is dimensionless; n is the total number of controllable devices, which is dimensionless. The energy consumption cost and task revenue of the mobile robot are listed in yuan.

[0012] Furthermore, the mobile robot also includes a sensor module, which includes any one of an infrared thermal imager, a vibration sensor B, and a gas sensor; the controller of the mobile robot uploads the data collected by the sensor module to the data acquisition module in real time.

[0013] Furthermore, the natural language interaction module also supports robot interaction commands. Its workflow includes: the voice input unit receiving voice commands input by the user, converting them into a digital audio stream and transmitting it to the natural language processing unit; the natural language processing unit performing speech recognition and semantic understanding on the digital audio stream, generating text tags containing actions, mobile robot identifiers, and task parameters, and transmitting them to the intent parsing unit; the intent parsing unit parsing the text tags into structured commands and transmitting them to the security verification unit; the security verification unit verifying whether the task parameters in the structured commands are within the preset safe operating range of the robot; if the verification passes, the security verification unit sends the structured commands to the mobile robot controller via a wireless network; if the verification fails, the security verification unit generates error message information and transmits it to the speech synthesis unit; the speech synthesis unit synthesizes the execution result or error message information into speech and broadcasts it through a speaker.

[0014] A method for the coordinated cascade utilization and intelligent operation and maintenance of multi-source industrial waste heat based on the multi-source industrial waste heat coordinated cascade utilization and intelligent operation and maintenance management system described in any one of the above includes the following steps: Step A: Collect temperature and flow data of forging heating furnace flue gas, forging radiant heat, AI cabinet exhaust and forging equipment cooling water through the data acquisition module; Step B: Pass 500~900℃ flue gas into the flue gas heat exchanger to generate 300~600℃ hot air, which is supplied to the high-temperature curing end of the spray drying line or the organic Rankine cycle generator set. Step C: After the radiant heat at 150~500℃ is collected by the radiant heat collection hood, it is used to heat the combustion air through the air preheater or directly as medium-temperature hot air. Step D: The 40-55℃ hot air discharged from the AI ​​cabinet heats the circulating water through the air-water heat exchanger, raising the temperature of the circulating water to 45-55℃. Then, the circulating water is mixed with the 40-70℃ cooling water from the forging equipment in a plate heat exchanger to form 50-60℃ medium-temperature water. A portion of this medium-temperature water is directly supplied to the heating terminal or domestic hot water terminal through the first outlet via a diversion valve, while the other portion is used as the evaporator heat source for the high-temperature heat pump through the second outlet. After being raised to 75-85℃ by the high-temperature heat pump, it is supplied to the low-temperature drying section of the spray drying line or the industrial cleaning tank. The opening degree of the diversion valve is determined by heat balance calculation based on the target temperature of the mixed medium-temperature water and the heat demand of each outlet. Step E: Process the data from the temperature sensor, flow sensor, vibration sensor, and current sensor using the LSTM inference unit to predict the remaining service life of the flue gas heat exchanger, high-temperature heat pump compressor, and fan; when the predicted remaining service life is lower than a set threshold, generate alarm information for the root cause of the fault through the root cause analysis submodule; the predicted remaining service life is obtained by linear extrapolation or threshold crossing calculation from the health index sequence output by the LSTM inference unit. Step F: Receive query or control commands through the natural language interaction module, and after parsing by the intent parsing unit and verification by the security verification unit: If it is a query command, the intent parsing unit generates a structured query statement to query the time-series data or equipment ledger information in the system database, and broadcasts the query results through the speech synthesis unit; If it is a control command, after the safety verification unit passes the verification, it sends control parameters to the target device through the PLC. The control parameters include the opening degree of the diversion valve, the output power of the high-temperature heat pump, the opening degree of the first electric regulating valve and the opening degree of the second electric regulating valve, as well as one of the following: a heat charging command, a heat dissipation command or a movement command sent to the mobile robot. Step G: When the dynamic scheduling controller determines that there is a decentralized heating demand and the fixed pipeline network cannot cover it, it issues a heating charge command to the mobile robot in an idle state; the mobile robot autonomously navigates to the hot water outlet of the plate heat exchanger of the low temperature stage recovery unit, and completes the heating charge of the phase change heat storage module after docking. Step H: After the heat charging is completed, the mobile robot carrying the phase change thermal storage module autonomously navigates to the target heat-using terminal, docks, and releases heat; after the heat release is completed, the mobile robot returns to the standby point or performs the next heat charging task. Step 1: When the predictive alarm module outputs a device fault warning, the AI ​​intelligent operation and maintenance subsystem dispatches the nearest mobile robot to the target device, collects infrared images and vibration data, and sends them back to the root cause analysis submodule for fault confirmation. Step J: The maintenance personnel issue task instructions to the mobile robot via natural language. After parsing, the natural language interaction module generates the mobile robot's motion path and task sequence, which are then sent to the mobile robot controller for execution after security verification.

[0015] Further, in step D, when the forging equipment is intermittent or the AI ​​cabinet load is low, resulting in heat generation exceeding heat consumption, the excess heat is stored in the first or second heat storage module; when heat generation is less than heat consumption, heat is released from the phase change heat storage module to supplement it; the charging and discharging process of the phase change heat storage module is optimized with the goal of minimizing the available energy loss rate, and the available energy loss rate during charging is calculated by the following formula: , The available energy loss rate during heat release is calculated using the following formula: , In the formula: and These are the available energy loss rates for heat charging and heat dissipation, respectively, in kW; For heating power, These are the heat release power values, all positive, in kW. This refers to the operating temperature of the phase change thermal energy storage module, expressed in Kelvin (K). The ambient temperature is expressed in Kelvin (K). The thermal efficiency of the phase change thermal storage module is dimensionless.

[0016] Compared with existing technologies, the core advantages of this invention are: This invention discloses a multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system and method. By integrating AI predictive maintenance, natural language interaction and mobile robot execution into an integrated intelligent operation and maintenance system, it solves the problems of low energy utilization, poor coordination, low operation and maintenance efficiency and insufficient coverage of decentralized heat use in the prior art. Attached Figure Description

[0017] Figure 1 This is a framework diagram of the system in this invention; Figure 2 This is a structural diagram of the multi-source waste heat cascade utilization subsystem in this invention; Figure 3 This is a flowchart illustrating the framework structure of the AI-powered intelligent operation and maintenance subsystem in this invention. Figure 4 This is a flowchart illustrating the framework structure of the phase change thermal energy storage unit in this invention. Figure 5 This is a flowchart of step AD in the method of this invention; Figure 6 This is a flowchart of step EI in the method of this invention; Figure 7 This is a flowchart of the instruction process for the natural language interaction module in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1, please refer to the accompanying drawings in the instruction manual. Figures 1-7 As shown, a multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system is applied to a forging production workshop equipped with an AI intelligent computing center. The forging production workshop is equipped with a Manufacturing Execution System (MES) for outputting forging production plans. The AI ​​intelligent computing center is equipped with multiple AI cabinets and power distribution cabinets, including a multi-source waste heat cascade utilization subsystem, a phase change heat storage unit, an AI intelligent operation and maintenance subsystem, and at least one mobile robot. The mobile robot includes a mobile robot controller, which wirelessly communicates with the AI ​​intelligent operation and maintenance subsystem to receive and execute scheduling instructions. The multi-source waste heat cascade utilization subsystem includes a high-temperature recovery unit, a medium-temperature recovery unit, and a low-temperature recovery unit. The high-temperature recovery unit includes a flue gas heat exchanger for recovering 500-900°C flue gas discharged from the forging heating furnace. The medium-temperature recovery unit includes a radiant heat collection hood and an air preheater. The radiant heat collection hood is used to recover 150-500°C radiant heat generated during the forging cooling process. The heat source-side inlet of the air preheater is connected to the outlet of the radiant heat collection hood, and the air-side outlet of the air preheater is connected to the combustion air inlet of the forging heating furnace. The low-temperature recovery unit includes an air-water heat exchanger, a plate heat exchanger, and a high-temperature heat pump for recovering 40-55°C hot air discharged from the AI ​​cabinet cooling channel and 40-70°C cooling water from the forging equipment. The multi-source waste heat cascade utilization subsystem also includes a fixed pipeline network, which consists of heating pipes connecting the plate heat exchanger outlet of the low-temperature stage recovery unit to each fixed heat-consuming terminal. The fixed heat-consuming terminals include heating terminals, domestic hot water terminals, low-temperature drying sections of spray drying lines, and industrial cleaning tanks. The mobile robot is used to provide mobile heating in decentralized heat demand scenarios that cannot be covered by the fixed pipeline network, through its onboard phase change heat storage module. The hot air discharged from the AI ​​cabinet cooling channel heats the circulating water through the air-water heat exchanger. The heated circulating water mixes with the cooling water of the forging equipment in the plate heat exchanger to form medium-temperature water at 50-60°C. A diversion valve is installed on the outlet pipe of the plate heat exchanger. The diversion valve has a first outlet and a second outlet. The first outlet is connected to the heating terminal or the domestic hot water terminal, and the second outlet is connected to the evaporator side of the high-temperature heat pump, which is used to distribute the medium-temperature water to different heat-using branches. The phase change thermal energy storage unit includes a first thermal energy storage module and a second thermal energy storage module. The first thermal energy storage module is connected to the flue gas heat exchanger via a heat transfer oil pipeline. The heat transfer oil pipeline is provided with a bypass outlet, which is connected to the organic Rankine cycle generator set via a first electric regulating valve. The second thermal energy storage module is connected to the plate heat exchanger via a water pipeline. The water pipeline is provided with a second electric regulating valve for controlling the charge and discharge heat flow rate. The AI-powered intelligent operation and maintenance subsystem includes a data acquisition module, a predictive alarm module, and a natural language interaction module. The data acquisition module includes: temperature and flow sensors installed on the flue gas heat exchanger, the air-water heat exchanger, the plate heat exchanger, the high-temperature heat pump, and the AI ​​cabinet; and vibration sensor A and current sensor installed on the compressor of the high-temperature heat pump and the fan of the air-water heat exchanger. In this embodiment, the data acquisition module is used to collect real-time sensor data such as temperature, flow, vibration, current, infrared image, and gas concentration distributed on the flue gas heat exchanger, air-water heat exchanger, plate heat exchanger, high-temperature heat pump, AI cabinet, and mobile robot, as shown in Table 1 below: Sensor Measurement Object Details. Table 1. List of Sensor Measurement Objects

[0020] The data acquisition module primarily provides the time-series data foundation for the LSTM predictive alarm module, offers real-time status feedback to the dynamic scheduling controller, and provides data sources for natural language queries. It forms the data foundation for achieving intelligent operation and maintenance.

[0021] The predictive warning module includes an LSTM inference unit deployed on an edge computing gateway, which is communicatively connected to the data acquisition module. The weight parameters of the LSTM inference unit are obtained through offline training using historical fault samples. The LSTM inference unit receives timing data from the vibration sensor and the current sensor, and outputs the predicted remaining service life of the compressor of the high-temperature heat pump and the fan of the air-water heat exchanger or the probability of failure in the next 48 hours. The LSTM inference unit also receives time-series data from the temperature sensor and the flow sensor, and outputs predicted remaining useful life (RUL) or heat exchange efficiency decay values ​​for the flue gas heat exchanger and the plate heat exchanger. In this embodiment, an LSTM (Long Short-Term Memory) time-series model is used to process historical and real-time sensor data to predict the RUL or failure probability in the next 48 hours for key equipment (such as high-temperature heat pump compressors, fans in air-to-water heat exchangers, flue gas heat exchangers, and plate heat exchangers). Simultaneously, root cause analysis is performed using a causal graph knowledge base. When the failure probability exceeds a failure probability trigger threshold (e.g., 0.7), the fault source equipment is automatically located, and a text alarm containing the equipment number, fault type, and maintenance suggestions is generated. When the predicted RUL is lower than a preset RUL warning threshold (e.g., 72 hours), a maintenance reminder is issued in advance. This transforms traditional "post-event alarms" into "pre-event predictions," avoiding unplanned downtime. Root cause analysis significantly shortens fault diagnosis time, guides precise maintenance, and improves system reliability.

[0022] The natural language interaction module includes a natural language processing unit, a voice input unit, a voice synthesis unit, an intent parsing unit, and a security verification unit deployed on a local server; the natural language interaction module also includes a system database, which is communicatively connected to the data acquisition module and is used to store time-series data and equipment ledger information collected by the temperature sensor, flow sensor, vibration sensor, and current sensor. The multi-source waste heat cascade utilization subsystem also includes a programmable logic controller (PLC), which is communicatively connected to the diversion valve, the high-temperature heat pump, the first electric regulating valve, the second electric regulating valve, and the mobile robot controller. After the safety verification unit passes the verification, it sends control commands to the target device through the PLC.

[0023] The intent parsing unit parses the text tag into a structured instruction and transmits it to the security verification unit; the structured instruction includes the target device identifier and the target parameter value; in this embodiment, the natural language interaction module integrates a Large Language Model (LLM), supporting maintenance personnel to input natural language instructions via voice or text. This includes: Query commands such as "What is the current waste heat recovery efficiency?" and "What is the remaining service life of the No. 2 high-temperature heat pump?" - The system automatically interprets the intent, queries the database, and returns the results. Control commands, such as "Set the high-temperature heat pump output power to 75 kW" and "Open the diversion valve to 80%", first perform a safety check (comparing the target parameters with the equipment's safety limits, for example, the valve's safety range is 5%~95%, and the high-temperature heat pump's safety range is 0%~100% of its rated power). After the check passes, the commands are executed via the PLC. Additionally, maintenance personnel can also control the allocation ratio of high-temperature preheating power generation and low-temperature thermal storage separately via voice commands such as "Increase the opening of the heat transfer oil bypass outlet valve". This mechanism significantly lowers the operational threshold, improves maintenance efficiency, and achieves "conversational" industrial control, while also incorporating a built-in safety mechanism to prevent misoperation.

[0024] The safety verification unit pre-stores the safe operating parameter ranges for each target device. These ranges include: the lower and upper limits of the diversion valve opening, the lower and upper limits of the high-temperature heat pump output power, and the lower and upper limits of the fan speed. Upon receiving a control command, the safety verification unit compares the target parameter values ​​with the corresponding safe operating parameter ranges for each target device. Only when the target parameter value falls within the safe operating parameter range will the command be executed via the PLC to the target device; otherwise, execution is refused and a warning message indicating that the safe range has been exceeded is returned. In the actual system of this embodiment, the safety verification unit pre-stores the corresponding upper and lower limits based on the device's nameplate parameters, manufacturer-recommended parameters, and on-site debugging results. For example, if the allowable power adjustment range for a certain model of high-temperature heat pump is 10% to 100% (the upper limit of 10% is to ensure lubricating oil circulation), the safety verification unit will use 10KW as the lower limit and 90% as the lower limit. kW is used as the upper limit; if the fan is frequency converter controlled, its lower limit is usually set to 30% of the rated speed (to avoid the critical speed range), and the upper limit is the rated speed (or 110% for short periods); the opening of the diversion valve usually retains a 5% margin (lower limit 5%, upper limit 95%) to avoid the valve seat from jamming.

[0025] Example 2: In the case of Example 1 above, the phase change temperature of the first thermal storage module is 200~300℃, and it is internally equipped with a first heat exchange coil, a first intermediate circulation pump, and a heat transfer oil pipeline. The inlet of the first heat exchange coil is connected to the heat transfer oil outlet of the flue gas heat exchanger of the high-temperature stage recovery unit via the first intermediate circulation pump and the heat transfer oil pipeline. The outlet of the first heat exchange coil is connected to the heat transfer oil inlet of the flue gas heat exchanger via the heat transfer oil pipeline, forming a heat transfer oil circulation loop. A bypass outlet is provided on the pipeline between the outlet of the first heat exchange coil and the heat transfer oil inlet of the flue gas heat exchanger, and this bypass outlet is connected to the organic Rankine cycle generator set. The phase change temperature of the second heat storage module is 50~60℃, and it is equipped with a second heat exchange coil, a second intermediate circulation pump and water pipes inside. The inlet of the second heat exchange coil is connected to the hot water outlet of the plate heat exchanger of the low temperature stage recovery unit through the second intermediate circulation pump and water pipes, and the outlet of the second heat exchange coil is connected to the return water port of the plate heat exchanger.

[0026] Furthermore, the air-water heat exchanger heats the circulating water to 45-55°C. This circulating water mixes with the cooling water from the forging equipment (40-70°C) in a plate heat exchanger to form medium-temperature water at 50-60°C. This medium-temperature water is divided into two paths by a diversion valve, including a first outlet and a second outlet. The first outlet is connected to a heating terminal or a domestic hot water terminal, and the second outlet is connected to the evaporator side of the high-temperature heat pump as a heat source. In this embodiment, the high-temperature heat pump is a vapor compression heat pump unit, mainly composed of an evaporator, a compressor, a condenser, and an expansion valve connected sequentially through refrigerant pipelines to form a closed loop. Its working principle is as follows: Liquid refrigerant absorbs heat from a low-temperature heat source (i.e., medium-temperature water at 50-60°C in the second outlet) in the evaporator and evaporates into a low-pressure gaseous refrigerant; the low-pressure gaseous refrigerant is compressed by the compressor into a high-temperature, high-pressure gaseous refrigerant; the high-temperature, high-pressure gaseous refrigerant enters the condenser, releases heat to the heat-using terminal, and condenses into a liquid refrigerant; the liquid refrigerant returns to the evaporator after being throttled and depressurized by the expansion valve, completing the cycle. The heat source side inlet of the evaporator is connected to the second outlet, and the heat source side outlet of the evaporator is connected to the return water pipe; the user side inlet of the condenser is connected to the return water pipe of the heat-using terminal, and the user side outlet of the condenser is connected to the low-temperature drying section of the spray drying line or the industrial cleaning tank, outputting hot water or hot air at 75-85°C. In this embodiment, the first heat storage module uses molten salt phase change material (phase change temperature 200~300℃) and is connected to a high-temperature flue gas heat exchanger via a heat transfer oil circulation loop. When there is excess waste heat from forging, the heat is stored; when forging stops or the demand for high-temperature drying increases, the heat is released to supplement the high-temperature hot air. The second heat storage module uses paraffin-based phase change material (phase change temperature 50~60℃) and is connected to a plate heat exchanger via a water circulation loop to store waste heat from medium-temperature water. When the AI ​​load is low or the heating demand decreases at night, excess heat is stored in this module; it is released the next morning when the heating demand increases.

[0027] Example 3, based on Example 1, further includes a root cause analysis submodule. This submodule includes a causal graph knowledge base stored in the edge computing gateway. In this solution, the causal graph knowledge base is a structured knowledge storage system that predefines the causal logical relationships between different fault modes and their observable sensor characteristics in the form of a directed graph (nodes represent events or variables, edges represent causal relationships). Furthermore, in this solution, the causal graph knowledge base is an offline-built, online-queryable rule base stored in the edge computing gateway. Its core content consists of multiple causal association rules in the form of "if...then...", as follows: The causal graph knowledge base pre-stores causal association rules between fault modes and sensor characteristics: Rule 1: The correlation between bearing wear failure in high-temperature heat pump compressors and the increase in the second harmonic component of the vibration sensor and the rise in the effective value of the current sensor; Rule 2: The correlation between impeller imbalance faults in air-to-water heat exchanger fans and the increase in the fundamental frequency component of vibration sensors; Rule 3: The correlation between ash accumulation and blockage in flue gas heat exchangers and decreased heat exchange efficiency of temperature sensors and increased differential pressure of flow sensors; The correlation rules between scaling failure of plate heat exchangers and decreased heat exchange efficiency of temperature sensors and increased differential pressure of flow sensors; When the failure probability output by the LSTM inference unit exceeds the threshold, the root cause analysis submodule uses the abnormal sensor features output by the LSTM inference unit as evidence and inputs them into the causal graph knowledge base for reverse reasoning. It then matches the most likely fault source device and its fault type, and generates a text alarm message containing the device number, fault type, and maintenance suggestions.

[0028] In this embodiment, the cause-effect graph knowledge base is the core component of the "root cause analysis submodule in the predictive alarm module," which connects the two stages of LSTM time-series prediction and fault location decision-making. The LSTM inference unit outputs a "failure probability" and a set of "abnormal sensor characteristics" (e.g., the vibration second harmonic component increases from 0.2 mm / s to 1.5 mm / s, and the current RMS value increases from 50 A to 65 A). At this point, the LSTM inference unit can only tell you "the device may have a problem," but it cannot tell you "what the specific problem is."

[0029] Cause-effect graph knowledge base: Receives the above "abnormal sensor characteristics" as evidence and performs reverse reasoning (searching for the most likely cause from the observed phenomenon). For example, if the "precondition" of rule 1 in the knowledge base is met (the second harmonic component increases and the flow rate rises), then the "conclusion" points to "bearing wear".

[0030] Correlation: The causal graph knowledge base maps the feature space of abnormal sensors to the fault mode space, realizing a leap from "anomaly detection" to "fault diagnosis". Without the causal graph knowledge base, the output of the LSTM inference unit is just a bunch of numbers, and maintenance personnel still need to analyze it manually; with the causal graph knowledge base, the system can automatically give specific fault types and maintenance suggestions. For example, in Example 1, the LSTMT inference unit predicts that the failure probability of the high-temperature heat pump compressor reaches 85%, and at the same time detects the following abnormal sensor features: vibration sensor 2nd harmonic component = 2.3 mm / s (normal <0.5 mm / s); current sensor effective value = 72A (normal 45~55 A). The root cause analysis submodule inputs these features into the causal graph knowledge base, matches rule 1 (bearing wear association rule), and inversely infers that: "the most likely fault source device is the high-temperature heat pump compressor, and the fault type is bearing wear". An alarm message was then generated: "Equipment No.: HP-03, Fault Type: Bearing Wear, Maintenance Recommendation: Replace the bearing within 48 hours, Estimated downtime: 2 hours, Spare parts inventory is sufficient." Maintenance personnel did not need to manually analyze the data and could directly arrange maintenance according to the recommendation, avoiding sudden downtime.

[0031] Example 4 further includes a dynamic scheduling controller, which includes a reinforcement learning inference module deployed on an industrial control computer. The reinforcement learning inference module employs a near-end policy optimization algorithm, and its state space vector includes the load prediction value of the AI ​​intelligent computing center for the next N hours, the running period of the forging heating furnace in the forging production plan, and the forging production plan obtained from the Manufacturing Execution System (MES). The real-time power demand of at least one of the following heat-using terminals is also included: heating terminal, domestic hot water terminal, low-temperature drying terminal of the spray drying line, industrial cleaning tank, high-temperature curing terminal of the spray drying line, and temporary office space in the factory area. Public warehouse, outdoor equipment antifreeze heating belt, mobile spray painting trolley; real-time electricity price signal; the current position of the mobile robot, the current heat load status and remaining power of the phase change thermal storage module of the mobile robot; its motion space vector includes the opening degree of the diversion valve controlling the distribution of medium-temperature water; the start and stop and power of the high-temperature heat pump; the opening degree of the heat transfer oil bypass outlet valve connected to the organic Rankine cycle generator set; the opening degree of the water circulation loop valve controlling the charging and discharging of the second thermal storage module; issuing charging instructions, discharging instructions and moving to the designated position instructions to the mobile robot; its reward function R is calculated by the following formula: Where R is the reward function, in yuan. The electricity cost consumed by the system, in yuan; The gas cost consumed by the system is expressed in yuan. Equipment operation and maintenance costs, in yuan. This is the device switching penalty coefficient, expressed in yuan. Let be the dimensionless action value of the i-th device at time t. Let be the action value of the i-th device at time t-1, which is dimensionless; n is the total number of controllable devices, which is dimensionless. The energy consumption cost and task revenue of the mobile robot are listed in yuan.

[0032] This embodiment, based on Embodiment 1 (forging and AI intelligent computing center and spray drying), further describes the implementation method of the dynamic scheduling controller: 1. Hardware Deployment A reinforcement learning inference module is deployed on an industrial control computer (ICC), employing the Proximal Policy Optimization (PPO) algorithm. The network structure is a two-layer fully connected neural network (128 neurons in each hidden layer). The control cycle is 15 minutes, meaning that an action is calculated every 15 minutes based on the current state and sent to each actuator. The actuators include: a diversion valve controlling the distribution of medium-temperature water, a compressor and frequency converter of the high-temperature heat pump, a bypass outlet valve for the heat transfer oil connected to the organic Rankine cycle generator set, a water circulation loop valve controlling the charging and discharging of the second heat storage module, and a mobile robot controller. Each valve actuator receives a 4-20mA control signal or Modbus command to adjust the opening degree; the high-temperature heat pump frequency converter receives a frequency setting signal to adjust the power; and the mobile robot controller receives wireless commands to perform charging, discharging, and movement tasks. Each actuator is equipped with safety limit protection (e.g., valve soft limit 5%~95%, heat pump power 0%~100%) to ensure that the actions do not exceed the equipment's safety range. 2. Definition of state-space vector (example values) At a certain decision time (t=14:00), the specific values ​​of each component of the state space vector are shown in Table 2 below, which provides an example of state space vector data at a certain decision time: Table 2 shows example data of the state space vector at a certain decision-making moment. AI Computing Center Load Forecast for the Next N=4 Hours [85%,82%,78%,75%] GPU utilization based on task scheduler prediction Heating furnace operation period in forging production plan Currently running From MES system Real-time power demand of each heat-consuming terminal High-temperature curing section: 50KW, low-temperature drying section: 150KW, heating: 30KW, hot water: 20KW Calculated in real time from the flow rate and temperature difference at each of the first and second outlets. Real-time electricity price signal 0.85 yuan / kWh Obtained from the power grid Current location of the mobile robot Coordinates (12.5m, 8.0m), standby point A Positioning system from robots Thermostatus of mobile robots The current temperature of the phase change thermal storage module is 52℃, and it has been charged to 60% capacity. Estimated by temperature sensor inside phase change thermal storage module Remaining battery power of the mobile robot 78% Battery Management System Data 3. Action space vector output (example) The reinforcement learning module outputs the following action values ​​based on the current state (all are normalized values, ranging from 0 to 1, and need to be converted into actual physical quantities), as shown in Table 3 below (example): Action Space Vector Output Table Table 3. Action Space Vector Output (Example) Diverter valve opening 0.65 65% Controlling the proportion of medium-temperature water entering the high-temperature heat pump High-temperature heat pump start-up and shutdown and power 0.70 70% (rated power) Rated power 90kW, actual power 63kW Heat transfer oil bypass outlet valve opening 0.20 20% Some of the high-temperature waste heat is fed into the Rankine cycle generator set. Second thermal storage module water circulation loop valve opening 0.35 35% Controlling the flow rate of medium-temperature water into the phase change thermal storage module Mobile robot instructions Heat charging command Send mobile robot controller The mobile robot goes to the plate heat exchanger interface to charge the heat. 4. Example of reward function calculation The data in this embodiment is based on the system configuration described in Embodiment 1. The rated power of the high-temperature heat pump is 90kW, and the current operating output is 70% of the rated power (operating value 0.70). Therefore, the actual input power = 90kW × 0.70 = 63kW, and the decision cycle is set to 15 minutes = 0.25 hours. Therefore, the power consumption of the high-temperature heat pump = 63kW × 0.25h = 15.75 kWh. The power consumption of other devices in this cycle is estimated as follows: Circulating water pump (air-water heat exchanger side): 3kW power, continuous operation → 0.75 kWh Cooling water circulation pump (forging equipment side): 4kW power, continuous operation → 1.0 kWh Flue gas heat exchanger induced draft fan: 5kW power, continuous operation → 1.25 kWh Radiant heat collection hood fan: 2kW power, continuous operation → 0.5 kWh Mobile robot charging (heating task performed this cycle): Charging power 2kW, operation time 0.25h → 0.5 kWh Phase change thermal energy storage unit circulating pump: 3kW power, 0.25h operation time → 0.75 kWh AI rack cooling fans (already included in the AI ​​system load, not repeated here) Organic Rankine cycle generator set (not used in this cycle, not included) Low-voltage electrical equipment such as industrial control computers, PLCs, and sensors: power 1kW → 0.25 kWh The total is: 0.75 + 1.0 + 1.25 + 0.5 + 0.5 + 0.75 + 0.25 = 5.0 kWh. The total power consumption is: 15.75 kWh (heat pump) + 5 kWh (other equipment) = 20.75 kWh. If the electricity price is taken as the typical value of the average electricity price during peak hours in China's industrial sector, which is 0.85 yuan / kWh, then the electricity cost for this period is 20.75 × 0.85 = 17.64 yuan.

[0033] Gas cost 43.75 yuan, source derivation (based on heat input of forging furnace): Assume the natural gas consumption of the forging heating furnace is 50 Nm³. 3 / h (typical value for small forging furnaces); Gas consumption per 0.25 hours this period = 50 × 0.25 = 12.5 Nm 3 ; Based on an industrial natural gas price of 3.5 yuan / Nm³ 3 The gas cost is calculated to be 12.5 × 3.5 = 43.75 yuan.

[0034] The equipment maintenance cost is 4.5 yuan, derived from the total equipment value and reasonable maintenance rate in Example 1: The total equipment value of the system is approximately 3 million yuan. High-temperature heat pump 500,000 RMB + heat exchangers (flue gas heat exchanger + air-to-water heat exchanger + plate heat exchanger) 600,000 RMB + fans / pumps (flue gas heat exchanger induced draft fan + radiant heat collection hood fan + air-to-water heat exchanger side circulating water pump + forging equipment side cooling water circulating pump + phase change heat storage unit circulating pump) 300,000 RMB + heat storage (first heat storage module + second heat storage module) 200,000 RMB + mobile robot 400,000 RMB + electrical control (industrial control computer + edge computing gateway + PLC programmable logic controller + local server + various instruments and sensors + power distribution cabinet) 200,000 RMB + piping 800,000 RMB; The annual operation and maintenance fee rate is calculated at 5% of the equipment value: 3 million × 5% = 150,000 yuan / year; Converted to each 15-minute period (35,040 15-minute periods per year): 150,000 ÷ 35,040 ≈ 4.28 yuan; The rounded amount is 4.50 yuan, which is a reasonable estimate.

[0035] Electricity cost C elec : 20.75 kWh × 0.85 yuan / kWh = 17.64 yuan.

[0036] Gas fee C gas The gas consumption of the forging heating furnace is equivalent to RMB 40.00 (natural gas consumption of 11.4 Nm³, unit price of RMB 3.5 / Nm³).

[0037] Equipment maintenance cost C om The depreciation and maintenance of the equipment are calculated as RMB 4.50.

[0038] Device switching penalty: Compare the action value with the previous time step (t-1): Flow divider valve: 0.65 → 0.60 (change of 0.05) Heat pump power: 0.70 → 0.68 (change of 0.02) Heat transfer oil bypass valve: 0.20 → 0.15 (change of 0.05) Hot water storage valve: 0.35 → 0.40 (change of 0.05) The summation is: 0.05 + 0.02 + 0.05 + 0.05 = 0.17 Let λ = 5 yuan per unit change, then the penalty term = 5 × 0.17 = 0.85 yuan.

[0039] Energy consumption cost and task revenue of mobile robots (Item C) robot The net contribution for this period is set at -3.00 yuan (i.e., a saving of 3 yuan). Substitute into the formula: , R = -[17.64 + 40.00 + 4.5 + 0.85 + (-3.00)] = -59.99 yuan.

[0040] The reward function R = -59.99 yuan.

[0041] This negative value indicates that the total cost for this period is 59.99 yuan. The goal of reinforcement learning is to maximize the reward function R (i.e., minimize the cost), so subsequent decisions will attempt to adjust actions to reduce this value.

[0042] In this embodiment, the dynamic scheduling controller is a core decision-making module in the system, deployed on an industrial control computer (ICC), and internally runs a reinforcement learning inference module (using a near-end policy optimization algorithm). Its function is to automatically decide how to adjust the control parameters of each energy flow device (diversion valve, high-temperature heat pump, phase change thermal storage unit) based on real-time changes in system status (such as AI load prediction, forging production plan, heat demand, electricity price, etc.) to minimize the total system operating cost (electricity cost + gas cost + maintenance cost + equipment switching penalty). It can be understood as an intelligent parameter tuner. Traditionally, heat allocation relies on manual experience or fixed rules (e.g., "prioritize heating in winter"), while the dynamic scheduling controller continuously optimizes through reinforcement learning, learning to make optimal decisions under different conditions (e.g., "use less high-temperature heat pump and more phase change thermal storage unit heat release when electricity prices are high").

[0043] The mobile robot includes: Mobile chassis for autonomous navigation within the factory area; The detachable phase change thermal energy storage module has a phase change temperature of 50~60℃ or 200~300℃. The module is equipped with a quick-connect interface for docking with the heat release interface of the flue gas heat exchanger in the high-temperature stage recovery unit or the plate heat exchanger in the low-temperature stage recovery unit. The "heat release interface" refers to the output port on the fixed flue gas heat exchanger or plate heat exchanger used to transfer heat to the mobile robot's thermal energy storage module. This interface includes a shut-off valve, a quick-connect connector, and a temperature sensor. Under normal conditions, the shut-off valve is closed to prevent heat loss. When the mobile robot docks with it via the quick-connect interface, the shut-off valve automatically opens, allowing high-temperature heat transfer oil or medium-temperature water from the heat exchanger to flow into the heat exchange coil of the thermal energy storage module, releasing heat to the phase change material and completing the charging process. After charging is complete, the shut-off valve closes, the mobile robot disengages, and carries the thermal energy storage module to the heat-using terminal.

[0044] The robot controller communicates wirelessly with the AI ​​intelligent operation and maintenance subsystem to receive heating / heating commands and navigation commands. The detachable phase change thermal storage module, in a heated state, is transported by the mobile robot to a heat-using terminal far from the fixed pipeline network, and releases heat through the quick-connect interface; the heat-using terminal includes: a temporary office cabin in the factory area, an anti-freeze heating belt for outdoor equipment, and a mobile spray painting cart.

[0045] Example 5, based on Example 1 above, further includes a sensor module, which comprises at least one of an infrared thermal imager, a vibration sensor, and a gas sensor. The infrared thermal imager is used to detect abnormal surface temperature distribution in the flue gas heat exchanger, plate heat exchanger, high-temperature heat pump casing, and pipe insulation layer, thereby identifying fault signs such as localized overheating, insulation layer damage, or heat exchanger blockage. The vibration sensor is used to detect the vibration spectrum characteristics of the high-temperature heat pump compressor and the fan of the air-water heat exchanger, including vibration amplitude, fundamental frequency component, and second harmonic component. The system is used to determine rotating machinery faults such as bearing wear and impeller imbalance; the gas sensor is used to detect flue gas leaks at the flue gas pipeline or flue gas heat exchanger interface of the forging heating furnace, as well as refrigerant leaks in the refrigerant pipeline of the high-temperature heat pump; the mobile robot controller uploads the data collected by the sensor module to the data acquisition module in real time, as supplementary input features of the LSTM inference unit, and as fault confirmation evidence of the root cause analysis submodule; through the mobile inspection of the mobile robot, the insufficient coverage of the fixed sensor can be made up for, and a comprehensive status perception of all equipment in the plant area can be achieved. The state space vector of the dynamic scheduling controller also includes the current position of each mobile robot, the current heat load state of the phase change thermal storage module, and the remaining power; the action space vector of the dynamic scheduling controller also includes issuing charging instructions, releasing instructions, and moving to designated positions to designated mobile robots; the reward function of the reinforcement learning inference module also includes the energy consumption cost item and the task completion reward item of the mobile robot.

[0046] Example 6: The natural language interaction module also supports robot interaction commands, issuing natural language commands such as "Go check [device name]", "Report the current phase change thermal storage module temperature", "Go to [location name] for charging", and "Transport the phase change thermal storage module to [hotspot]" to a designated mobile robot. Its workflow includes: the voice input unit receiving the user's voice command, converting it into a digital audio stream and transmitting it to the natural language processing unit; the natural language processing unit performing speech recognition and semantic understanding on the digital audio stream, generating text tags containing actions, target mobile robot identifiers, and task parameters, and transmitting them to the intent parsing unit; and the intent parsing unit parsing the text tags into... The structured instruction, including operation type, mobile robot ID, task type, and task parameters, is transmitted to the security verification unit. The security verification unit verifies whether the task parameters in the structured instruction are within a preset safe operating range for the robot, which includes: upper limit of movement speed, upper and lower limits of phase change thermal storage module temperature, and lower limit of battery power. If the verification passes, the security verification unit sends the structured instruction to the mobile robot controller via a wireless network. If the verification fails, the security verification unit generates an error message and transmits it to the speech synthesis unit. The speech synthesis unit synthesizes the execution result or error message into speech and broadcasts it through a speaker.

[0047] Example 7: A method for the coordinated cascade utilization and intelligent operation and maintenance of multi-source industrial waste heat based on any one of the above-described multi-source industrial waste heat coordinated cascade utilization and intelligent operation and maintenance management systems, comprising the following steps: Step A: Collect temperature and flow data of forging heating furnace flue gas, forging radiant heat, AI cabinet exhaust and forging equipment cooling water through the data acquisition module; Step B: Pass flue gas at 500~900℃ into the flue gas heat exchanger to generate hot air at 300~600℃, which is then supplied to the high-temperature drying section or the organic Rankine cycle generator set. Step C: After the radiant heat at 150~500℃ is collected by the radiant heat collection hood, it is used to heat the combustion air through the air preheater or directly as medium-temperature hot air. Step D: The 40-55℃ hot air discharged from the AI ​​cabinet is used to heat the circulating water through the air-water heat exchanger, raising the temperature of the circulating water to 45-55℃. Then, the circulating water is mixed with the 40-70℃ cooling water of the forging equipment in a plate heat exchanger to form 50-60℃ medium-temperature water. A portion of the medium-temperature water is directly supplied to the terminal or domestic hot water terminal through the first outlet through a diversion valve, while the other portion is used as the evaporator heat source of the high-temperature heat pump through the second outlet. The high-temperature heat pump then raises the temperature to 75-85℃ and sends it to the low-temperature drying section of the spray drying line or the industrial cleaning tank. The opening degree of the diversion valve is determined by heat balance calculation based on the target temperature of the mixed medium-temperature water and the heat demand of each branch. The method for determining the opening degree of the flow divider valve is as follows: Let the target temperature of the mixed medium-temperature water be T. mix,target (For example, at 55℃), the required flow rate of the first outlet (direct heating / hot water supply) is: The required flow rate of the second outlet (supplying the heat pump evaporator) is Total flow , According to the law of conservation of energy (heat balance equation): , Where T1 is the supply temperature of the first outlet (depending on the heating / hot water terminal requirements, e.g., 50°C), and T2 is the supply temperature of the second outlet (depending on the heat pump evaporator inlet requirements, e.g., 55°C). p The specific heat capacity of water can be reduced to a value less than 1.

[0048] Due to total flow The required flow rate at the first outlet is measured by the total flow sensor at the plate heat exchanger outlet. The required opening θ of the diversion valve is determined by the actual demand at the heating / hot water terminal (calculated using the terminal valve opening and return water temperature) and the required flow rate at the second outlet. Proportional relationship: ; Step E: Process the data from the temperature sensor, flow sensor, vibration sensor, and current sensor using the LSTM inference unit to predict the remaining service life of the flue gas heat exchanger, high-temperature heat pump compressor, and fan; when the predicted remaining service life is lower than a set threshold, generate alarm information for the root cause of the fault through the root cause analysis submodule; the predicted remaining service life is obtained by linear extrapolation or threshold crossing calculation from the health index sequence output by the LSTM inference unit. Step F: Receive query or control commands through the natural language interaction module, and then have them parsed by the intent parsing unit and verified by the security verification unit; If it is a query command, the intent parsing unit generates a structured query statement to query the time-series data or equipment ledger information in the system database, and broadcasts the query results through the speech synthesis unit; If it is a control command, after the safety verification unit passes the verification, it sends control parameters to the target device through the PLC. The control parameters include the opening degree of the diversion valve, the output power of the high-temperature heat pump, the opening degree of the first electric regulating valve and the opening degree of the second electric regulating valve, as well as one of the following: a heat charging command, a heat dissipation command or a movement command sent to the mobile robot. In step D, when the forging equipment is intermittent or the AI ​​cabinet load is low, resulting in heat generation exceeding heat consumption, the excess heat is stored in the first or second heat storage module; when the heat generation is less than the heat consumption, heat is released from the phase change heat storage module to supplement it; the charging and discharging process of the heat storage module is optimized with the goal of minimizing the available energy loss rate, and the available energy loss rate during charging is calculated by the following formula: , The available energy loss rate during heat release is calculated using the following formula: , In the formula: and These are the available energy loss rates for heat charging and heat dissipation, respectively, in kW; For heating power, These are the heat release power values, all positive, in kW. This refers to the operating temperature of the phase change thermal energy storage module, expressed in Kelvin (K). The ambient temperature is expressed in Kelvin (K). The thermal efficiency of the phase change thermal storage module is dimensionless.

[0049] Based on the above embodiments, this embodiment provides a specific example scenario to illustrate how to use this formula to guide the charging and discharging heat scheduling of a phase change thermal storage module in order to minimize available energy loss: For example, in the forging production workshop and AI computing center, during a certain period (e.g., 2:00 PM - 3:00 PM), heat generation exceeds heat consumption, resulting in a surplus of 200 kW of heat power. It is necessary to determine which phase change thermal storage module to store this surplus heat in. The system is configured with two phase change thermal storage modules: First thermal storage module (high temperature): Phase change temperature T store,high =250℃=523K, thermal storage efficiency η store,high =0.92; Second thermal storage module (low temperature): Phase change temperature T store,low =55℃=328K, thermal storage efficiency η store,low =0.88; Ambient temperature T amb =25℃=298K.

[0050] Excess heat sources: 120kW from the high-temperature stage recovery unit (forging furnace flue gas) and 80kW from the low-temperature stage recovery unit (AI cabinet exhaust + forging equipment cooling water). However, during heat storage, the heat needs to be transferred through a flue gas heat exchanger or plate heat exchanger, and the actual heat charging power Q... charge Limited by heat exchange capacity and the heat absorption rate of the phase change thermal storage module, this example assumes that both the first and second thermal storage modules can accept a charging power of 200 kW (i.e., Q). charge =200kW), to compare the available energy loss rates of the two thermal storage methods. Calculate the available energy loss rate during charging: First thermal storage module (high temperature): , Substitute the values: , ; Second thermal storage module (low temperature): , Substitute the values: , , , , , .

[0051] See Table 4 below for results analysis and decision-making: Table 4 Results Analysis and Decision Table First thermal storage module (250℃) 6.88kW The losses are significant; high-grade heat storage involves large temperature differences, but the heat storage efficiency is slightly higher. Second thermal storage module (55℃) 2.20kW The loss is smaller because of the smaller temperature difference with the environment. Decision: Considering only the minimization of available energy loss rate, surplus heat should be stored in the second thermal storage module (low temperature), which has an available energy loss rate of only 32% of that of the high temperature module.

[0052] Example extended to exothermic scenarios: Suppose that at a certain moment, heat needs to be released from the first and second thermal storage modules, with a heat demand of 150kW (e.g., the low-temperature section of a spray drying line). Compare the available energy loss rates of the heat release from the two modules.

[0053] The formulas for exothermic and induction heat are the same in form:

[0054] in,

[0055] The first thermal storage module releases heat: E loss,dis,high =150×0.4302×0.08=150×0.034416=5.1624kw The second thermal storage module releases heat: E loss,dis,low =150×0.0915×0.12=150×0.01098=1.647kw When releasing heat, the second thermal storage module also experiences less energy loss. Therefore, in system scheduling, the second thermal storage module should be used first to meet the heating needs at medium and low temperatures, while the first thermal storage module should only be used in situations where high temperatures are required (such as solidification sections or ORC power generation).

[0056] This embodiment clearly demonstrates the engineering application of the available energy loss rate formula through specific numerical calculations, verifying its practicality in thermal energy storage scheduling optimization.

[0057] A method for the coordinated cascade utilization and intelligent operation and maintenance management system of multi-source industrial waste heat, comprising the following steps: Step G: When the dynamic scheduling controller determines that there is a decentralized heating demand and the fixed pipeline network cannot cover it, it issues a heating charge command to the mobile robot in an idle state; the mobile robot autonomously navigates to the hot water outlet of the plate heat exchanger of the low temperature stage recovery unit, and completes the heating charge of the phase change heat storage module after docking. Step H: After the heat charging is completed, the mobile robot carrying the phase change thermal storage module autonomously navigates to the target heat-using terminal, docks, and releases heat; after the heat release is completed, the mobile robot returns to the standby point or performs the next heat charging task. Step 1: When the predictive alarm module outputs a device fault warning, the AI ​​intelligent operation and maintenance subsystem dispatches the nearest mobile robot to the target device, collects infrared images and vibration data, and sends them back to the root cause analysis submodule for fault confirmation. Step J: The maintenance personnel issue task instructions to the mobile robot via natural language. After parsing, the natural language interaction module generates the mobile robot's motion path and task sequence, which are then sent to the mobile robot controller for execution after security verification.

[0058] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system, applied to a forging production workshop equipped with an AI computing center, wherein the forging production workshop is equipped with a Manufacturing Execution System (MES) for outputting forging production plans, and the AI ​​computing center is equipped with multiple AI cabinets and power distribution cabinets, characterized in that, It includes a multi-source waste heat cascade utilization subsystem, a phase change thermal storage unit, an AI intelligent operation and maintenance subsystem, and at least one mobile robot; the mobile robot includes a mobile robot controller, which wirelessly communicates with the AI ​​intelligent operation and maintenance subsystem to receive and execute scheduling instructions. The multi-source waste heat cascade utilization subsystem includes a high-temperature recovery unit, a medium-temperature recovery unit, and a low-temperature recovery unit. The high-temperature recovery unit includes a flue gas heat exchanger for recovering 500-900°C flue gas discharged from the forging heating furnace. The medium-temperature recovery unit includes a radiant heat collection hood and an air preheater. The radiant heat collection hood is used to recover 150-500°C radiant heat generated during the forging cooling process. The heat source-side inlet of the air preheater is connected to the outlet of the radiant heat collection hood, and the air-side outlet of the air preheater is connected to the combustion air inlet of the forging heating furnace. The low-temperature recovery unit includes an air-water heat exchanger, a plate heat exchanger, and a high-temperature heat pump for recovering 40-55°C hot air discharged from the AI ​​cabinet cooling channel and 40-70°C cooling water from the forging equipment. The multi-source waste heat cascade utilization subsystem also includes a fixed pipeline network, which consists of heating pipes connecting the plate heat exchanger outlet of the low-temperature stage recovery unit to each fixed heat-consuming terminal. The fixed heat-consuming terminals include heating terminals, domestic hot water terminals, low-temperature drying sections of spray drying lines, and industrial cleaning tanks. The mobile robot is used to provide mobile heating in decentralized heat demand scenarios that cannot be covered by the fixed pipeline network, through its onboard phase change heat storage module. The hot air discharged from the AI ​​cabinet cooling channel heats the circulating water through the air-water heat exchanger. The heated circulating water mixes with the cooling water of the forging equipment in the plate heat exchanger to form medium-temperature water at 50~60℃. A diversion valve is installed on the outlet pipe of the plate heat exchanger. The phase change thermal energy storage unit includes a first thermal energy storage module and a second thermal energy storage module. The first thermal energy storage module is connected to the flue gas heat exchanger via a heat transfer oil pipeline. The heat transfer oil pipeline is provided with a bypass outlet, which is connected to the organic Rankine cycle generator set via a first electric regulating valve. The second thermal energy storage module is connected to the plate heat exchanger via a water pipeline. The water pipeline is provided with a second electric regulating valve for controlling the charge and discharge heat flow rate. The AI-powered intelligent operation and maintenance subsystem includes a data acquisition module, a predictive alarm module, and a natural language interaction module. The data acquisition module includes: a temperature sensor and a flow sensor installed on the flue gas heat exchanger, the air-water heat exchanger, the plate heat exchanger, the high-temperature heat pump, and the AI ​​cabinet; and a vibration sensor A and a current sensor installed on the compressor of the high-temperature heat pump and the fan of the air-water heat exchanger. The predictive warning module includes an LSTM inference unit deployed on an edge computing gateway, which is communicatively connected to the data acquisition module. The weight parameters of the LSTM inference unit are obtained through offline training using historical fault samples. The LSTM inference unit receives timing data from the vibration sensor and the current sensor, and outputs the predicted remaining service life of the compressor of the high-temperature heat pump and the fan of the air-water heat exchanger or the probability of failure in the next 48 hours. The LSTM inference unit also receives time-series data from the temperature sensor and the flow sensor, and outputs the predicted remaining service life or the predicted heat exchange efficiency decay of the flue gas heat exchanger and the plate heat exchanger. The natural language interaction module includes a natural language processing unit, a voice input unit, a voice synthesis unit, an intent parsing unit, and a security verification unit deployed on a local server; the natural language interaction module also includes a system database, which is communicatively connected to the data acquisition module and is used to store time-series data and equipment ledger information collected by the temperature sensor, flow sensor, vibration sensor, and current sensor. The multi-source waste heat cascade utilization subsystem also includes a programmable logic controller (PLC), which is communicatively connected to the diversion valve, the high-temperature heat pump, the first electric regulating valve, the second electric regulating valve, and the mobile robot controller. After the safety verification unit passes the verification, it sends control commands to the target device through the PLC.

2. The multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system according to claim 1, characterized in that: The first heat storage module is internally equipped with a first heat exchange coil, a first intermediate circulation pump, and a heat transfer oil pipeline. The inlet of the first heat exchange coil is connected to the heat transfer oil outlet of the flue gas heat exchanger of the high-temperature stage recovery unit through the first intermediate circulation pump and the heat transfer oil pipeline. The outlet of the first heat exchange coil is connected to the heat transfer oil inlet of the flue gas heat exchanger through the heat transfer oil pipeline, forming a heat transfer oil circulation loop. A bypass outlet is provided on the pipeline between the outlet of the first heat exchange coil and the heat transfer oil inlet of the flue gas heat exchanger, and the bypass outlet is connected to the organic Rankine cycle generator set. The second heat storage module is equipped with a second heat exchange coil, a second intermediate circulation pump and water pipes; the inlet of the second heat exchange coil is connected to the hot water outlet of the plate heat exchanger of the low temperature stage recovery unit through the second intermediate circulation pump and water pipes, and the outlet of the second heat exchange coil is connected to the return water port of the plate heat exchanger.

3. The multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system according to claim 1, characterized in that: The air-water heat exchanger heats the circulating water to 45~55℃. This circulating water mixes with the cooling water of the forging equipment at 40~70℃ in the plate heat exchanger to form medium-temperature water at 50~60℃. This medium-temperature water is divided into two outlets by a diversion valve, including a first outlet and a second outlet. All first outlets are connected to heating terminals or domestic hot water terminals, and the second outlet is connected to the evaporator side of the high-temperature heat pump as a heat source.

4. The multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system according to claim 3, characterized in that: The second outlet is also connected to the condenser side of the high-temperature heat pump to output hot water or hot air at 75~85℃, which is then connected to the low-temperature drying section of the spray drying line or the industrial cleaning tank through a pipeline.

5. The multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system according to claim 1, characterized in that: The predictive warning module also includes a root cause analysis submodule, which includes a causal graph knowledge base stored in the edge computing gateway. The causal graph knowledge base contains pre-stored causal association rules between fault modes and sensor features.

6. A multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system according to any one of claims 1 to 5, characterized in that: It also includes a dynamic scheduling controller, which includes a reinforcement learning inference module deployed on an industrial control computer. This module employs a near-end policy optimization algorithm, and its state space vector includes the load prediction value of the AI ​​intelligent computing center for the next N hours, the running period of the forging heating furnace in the forging production plan, and the forging production plan obtained from the Manufacturing Execution System (MES). The real-time power demand of at least one of the following heat-using terminals is also included: heating terminal, domestic hot water terminal, low-temperature drying terminal of a spray drying line, industrial cleaning tank, high-temperature curing terminal of a spray drying line, and temporary office warehouse in the factory area. The system includes: outdoor equipment antifreeze heating belt; mobile spray painting trolley; real-time electricity price signal; the current position of the mobile robot, the current heat load status and remaining power of the phase change heat storage module of the mobile robot; its motion space vector includes the opening degree of the diversion valve controlling the distribution of medium-temperature water; the start-up and shutdown and power of the high-temperature heat pump; the opening degree of the heat transfer oil bypass outlet valve connected to the organic Rankine cycle generator set; the opening degree of the water circulation loop valve controlling the charging and discharging of the second heat storage module; issuing charging instructions, discharging instructions and instructions to move to the designated position to the mobile robot; and its reward function R is calculated by the following formula: Where R is the reward function, in yuan. The electricity cost consumed by the system, in yuan; The gas cost consumed by the system is expressed in yuan. Equipment operation and maintenance costs, in yuan. This is the device switching penalty coefficient, expressed in yuan. Let be the dimensionless action value of the i-th device at time t. Let be the action value of the i-th device at time t-1, which is dimensionless; n is the total number of controllable devices, which is dimensionless. The energy consumption cost and task revenue of the mobile robot are listed in yuan.

7. The multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system according to claim 6, characterized in that, The mobile robot also includes a sensor module, which includes any one of an infrared thermal imager, a vibration sensor B, and a gas sensor; the controller of the mobile robot uploads the data collected by the sensor module to the data acquisition module in real time.

8. The multi-source industrial waste heat collaborative cascade utilization and intelligent operation and maintenance management system according to claim 1, characterized in that: The natural language interaction module also supports robot interaction commands. Its workflow includes: the voice input unit receives voice commands input by the user, converts them into a digital audio stream, and transmits them to the natural language processing unit; the natural language processing unit performs speech recognition and semantic understanding on the digital audio stream, generates text tags containing actions, mobile robot identifiers, and task parameters, and transmits them to the intent parsing unit; the intent parsing unit parses the text tags into structured commands and transmits them to the security verification unit; the security verification unit verifies whether the task parameters in the structured commands are within the preset safe operating range of the robot. If the verification passes, the security verification unit sends the structured commands to the mobile robot controller via a wireless network; if the verification fails, the security verification unit generates error messages and transmits them to the speech synthesis unit; the speech synthesis unit synthesizes the execution result or error messages into speech and broadcasts it through a speaker.

9. A method for the coordinated cascade utilization and intelligent operation and maintenance of multi-source industrial waste heat based on the multi-source industrial waste heat coordinated cascade utilization and intelligent operation and maintenance management system according to any one of claims 2, 6, or 7, characterized in that, Includes the following steps: Step A: Collect temperature and flow data of forging heating furnace flue gas, forging radiant heat, AI cabinet exhaust and forging equipment cooling water through the data acquisition module; Step B: Pass 500~900℃ flue gas into the flue gas heat exchanger to generate 300~600℃ hot air, which is supplied to the high-temperature curing end of the spray drying line or the organic Rankine cycle generator set. Step C: After the radiant heat at 150~500℃ is collected by the radiant heat collection hood, it is used to heat the combustion air through the air preheater or directly as medium-temperature hot air. Step D: The 40-55℃ hot air discharged from the AI ​​cabinet heats the circulating water through the air-water heat exchanger, raising the temperature of the circulating water to 45-55℃. Then, the circulating water is mixed with the 40-70℃ cooling water from the forging equipment in a plate heat exchanger to form 50-60℃ medium-temperature water. A portion of this medium-temperature water is directly supplied to the heating terminal or domestic hot water terminal through the first outlet via a diversion valve, while the other portion is used as the evaporator heat source of the high-temperature heat pump through the second outlet. After being raised to 75-85℃ by the high-temperature heat pump, it is supplied to the low-temperature drying section of the spray drying line or the industrial cleaning tank. The opening degree of the diversion valve is determined by heat balance calculation based on the target temperature of the mixed medium-temperature water and the heat demand of each branch. Step E: Process the data from the temperature sensor, flow sensor, vibration sensor, and current sensor using the LSTM inference unit to predict the remaining service life of the flue gas heat exchanger, high-temperature heat pump compressor, and fan; when the predicted remaining service life is lower than a set threshold, generate alarm information for the root cause of the fault through the root cause analysis submodule; the predicted remaining service life is obtained by linear extrapolation or threshold crossing calculation from the health index sequence output by the LSTM inference unit. Step F: Receive query or control commands through the natural language interaction module, and after parsing by the intent parsing unit and verification by the security verification unit: If it is a query command, the intent parsing unit generates a structured query statement to query the time-series data or equipment ledger information in the system database, and broadcasts the query results through the speech synthesis unit; If it is a control command, after the safety verification unit passes the verification, it sends control parameters to the target device through the PLC. The control parameters include the opening degree of the diversion valve, the output power of the high-temperature heat pump, the opening degree of the first electric regulating valve and the opening degree of the second electric regulating valve, as well as one of the following: a heat charging command, a heat dissipation command or a movement command sent to the mobile robot. Step G: When the dynamic scheduling controller determines that there is a decentralized heating demand and the fixed pipeline network cannot cover it, it issues a heating charge command to the mobile robot in an idle state; the mobile robot autonomously navigates to the hot water outlet of the plate heat exchanger of the low temperature stage recovery unit, and completes the heating charge of the phase change heat storage module after docking. Step H: After the heat charging is completed, the mobile robot carrying the phase change thermal storage module autonomously navigates to the target heat-using terminal, docks, and releases heat; after the heat release is completed, the mobile robot returns to the standby point or performs the next heat charging task. Step 1: When the predictive alarm module outputs a device fault warning, the AI ​​intelligent operation and maintenance subsystem dispatches the nearest mobile robot to the target device, collects infrared images and vibration data, and sends them back to the root cause analysis submodule for fault confirmation. Step J: The maintenance personnel issue task instructions to the mobile robot via natural language. After parsing, the natural language interaction module generates the mobile robot's motion path and task sequence, which are then sent to the mobile robot controller for execution after security verification.

10. A method for the coordinated and cascaded utilization and intelligent operation and maintenance of multi-source industrial waste heat according to claim 9, characterized in that: In step D, when the forging equipment is intermittent or the AI ​​cabinet load is low, resulting in heat generation exceeding heat consumption, the excess heat is stored in the first or second heat storage module; when heat generation is less than heat consumption, heat is released from the phase change heat storage module to supplement it; the charging and discharging process of the phase change heat storage module is optimized with the goal of minimizing the available energy loss rate, and the available energy loss rate during charging is calculated by the following formula: ， The available energy loss rate during heat release is calculated using the following formula: ， In the formula: and These are the available energy loss rates for heat charging and heat dissipation, respectively, in kW; For heating power, These are the heat release power values, all positive, and the unit is kW. This refers to the operating temperature of the phase change thermal energy storage module, expressed in Kelvin (K). The ambient temperature is expressed in Kelvin (K). The thermal efficiency of the phase change thermal storage module is dimensionless.

Images