Automatic tuning and energy-saving method and system for ammonia refrigeration compressor and storage medium

Abstract

The invention discloses an ammonia refrigeration compressor automatic tuning and energy saving method and system and a storage medium, and the method comprises the steps that through a sensor, operation parameters and equipment states of an ammonia refrigeration system are collected in real time; splicing to form a system state feature vector at a preset time interval; in the domain knowledge base, domain knowledge which is matched with the current operation state of the ammonia refrigeration system and can enable the production efficiency and the energy efficiency level of the quick-freezing production line to be optimal is retrieved, and multiple sets of candidate control parameter sets are generated; and sending the candidate control parameters of each group of candidate control parameter sets into the digital twin in parallel for dynamic simulation, and determining an optimal control parameter set based on a dynamic simulation result, so that the ammonia refrigeration compressor realizes automatic tuning and energy saving after being adjusted based on the optimal control parameters of the optimal control parameter set. And dynamic optimization and energy-saving control of operation parameters of the ammonia refrigeration compressor can be realized by utilizing a large language model (LLM) and a multi-modal data fusion technology.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence technology and energy-saving control technology of refrigeration systems, and in particular to an automatic optimization and energy-saving method and system for ammonia refrigeration compressors, and a storage medium. Background Technology

[0002] Driven by dual-carbon goals, high operating costs, and the extensive management model of traditional systems, the need to refine and intelligently transform traditional ammonia refrigeration systems using modern information technology to achieve significant energy savings, reduced consumption, and improved safety is an inevitable trend. Ammonia refrigeration systems are widely used in cold chain logistics, chemical industries, and other fields due to their environmental friendliness and high efficiency; however, their energy consumption has long constrained the industry's development. Summary of the Invention

[0003] In view of this, this application provides an automatic optimization and energy-saving method and system for ammonia refrigeration compressors, as well as a storage medium. After acquiring data from relevant equipment using Internet of Things (IoT) technology, it combines digital twin technology to replicate digital scenarios, thereby creating a large language model structure for the application scenario. By collecting energy and capacity output results under different process parameter inputs in real time, and based on the data learning analysis optimization stage and the model accuracy correction and improvement stage, it realizes a paradigm leap from "experience-driven" to "data-mechanism dual-driven" for ammonia refrigeration compressors, which can increase economic benefits, ensure product quality, enhance competitiveness, and achieve the comprehensive goals of energy conservation and emission reduction.

[0004] According to one aspect of this application, an automatic optimization and energy-saving method for an ammonia refrigeration compressor is provided, applied to a quick-freezing production line. The quick-freezing production line uses an ammonia refrigeration system for cooling, the ammonia refrigeration system including an ammonia refrigeration compressor and a motor for supplying power to the ammonia refrigeration compressor. The method includes: Sensors deployed on the quick-freezing production line are used to collect the operating parameters and equipment status of the ammonia refrigeration system in real time. The operating parameters include at least one of the following: suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor. The equipment status includes at least one of the following: vibration spectrum of the ammonia refrigeration compressor, lubricating oil status of the ammonia refrigeration compressor, and electrical characteristics of the motor. At preset intervals, based on the collected operating parameters and equipment status, a system state feature vector representing the current operating status of the ammonia refrigeration system is constructed. In the domain knowledge base, the system state feature vector is used to retrieve domain knowledge that is compatible with the current operating state of the ammonia refrigeration system and can optimize the production efficiency and energy efficiency of the quick-freezing production line. Based on the retrieved domain knowledge, multiple sets of candidate control parameters are generated, including candidate control parameters for controlling the ammonia refrigeration compressor. Candidate control parameters from each set of candidate control parameters are fed into a digital twin in parallel for dynamic simulation. The optimal control parameter set is determined based on the dynamic simulation results, so that the ammonia refrigeration compressor can achieve automatic optimization and energy saving after adjusting the optimal control parameters based on the optimal control parameter set. The digital twin performs real-time dynamic digital modeling of the quick-freezing production line based on the real-time collected operating parameters and equipment status.

[0005] According to another aspect of this application, an automatic optimization and energy-saving system for an ammonia refrigeration compressor is provided, applied to a quick-freezing production line. The quick-freezing production line uses an ammonia refrigeration system for cooling. The ammonia refrigeration system includes an ammonia refrigeration compressor and a motor for supplying power to the ammonia refrigeration compressor. The system includes: The multi-source data acquisition module is used to collect the operating parameters and equipment status of the ammonia refrigeration system in real time through sensors deployed on the quick-freezing production line. The operating parameters include at least one of the following: suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor. The equipment status includes at least one of the following: vibration spectrum of the ammonia refrigeration compressor, lubricating oil status of the ammonia refrigeration compressor, and electrical characteristics of the motor. The system state feature vector splicing module is used to splice together a system state feature vector representing the current operating state of the ammonia refrigeration system based on the collected operating parameters and equipment status at preset intervals. The candidate control parameter set filtering module is used to retrieve domain knowledge from the domain knowledge base that is compatible with the current operating state of the ammonia refrigeration system and can optimize the production efficiency and energy efficiency of the quick-freezing production line, respectively. Based on the retrieved domain knowledge, multiple sets of candidate control parameters are generated, including candidate control parameters for controlling the ammonia refrigeration compressor. The automatic optimization and energy-saving module is used to send the candidate control parameters of each set of candidate control parameters into the digital twin in parallel for dynamic simulation. Based on the dynamic simulation results, the optimal control parameter set is determined so that the ammonia refrigeration compressor can achieve automatic optimization and energy saving after adjusting the optimal control parameters based on the optimal control parameter set. The digital twin performs real-time dynamic digital modeling of the quick-freezing production line based on the real-time collected operating parameters and equipment status.

[0006] According to another aspect of this application, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described automatic optimization and energy-saving method for ammonia refrigeration compressors.

[0007] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described automatic optimization and energy-saving method for ammonia refrigeration compressors.

[0008] By means of the above technical solution, the present application provides an automatic optimization and energy-saving method and system for ammonia refrigeration compressors, as well as a storage medium, which utilizes Large Language Model (LLM) and multimodal data fusion technology to achieve dynamic optimization and energy-saving control of the operating parameters of ammonia refrigeration compressors.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A schematic flowchart of an automatic optimization and energy-saving method for an ammonia refrigeration compressor provided in an embodiment of this application is shown. Figure 2 This paper illustrates a schematic diagram of the process architecture of an automatic optimization and energy-saving method for an ammonia refrigeration compressor provided in an embodiment of this application. Figure 3 A flowchart illustrating another automatic optimization and energy-saving method for ammonia refrigeration compressor provided in an embodiment of this application is shown. Figure 4 This illustration shows a schematic diagram of a domain knowledge base construction method provided in an embodiment of this application; Figure 5 This illustration shows a schematic diagram of a large language model update process in a domain knowledge base provided in an embodiment of this application; Figure 6 A schematic diagram of an automatic optimization and energy-saving system for an ammonia refrigeration compressor provided in an embodiment of this application is shown. Detailed Implementation

[0011] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0012] This embodiment provides an automatic optimization and energy-saving method for an ammonia refrigeration compressor, such as... Figure 1 As shown, this method is applied to a quick-freezing production line, which uses an ammonia refrigeration system for cooling. The ammonia refrigeration system includes an ammonia refrigeration compressor and a motor for supplying power to the ammonia refrigeration compressor. The method includes: Step 101: Real-time acquisition of operating parameters and equipment status of the ammonia refrigeration system by sensors deployed on the quick-freezing production line. The operating parameters include at least one of the following: suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor. The equipment status includes at least one of the following: vibration spectrum of the ammonia refrigeration compressor, lubricating oil status of the ammonia refrigeration compressor, and electrical characteristics of the motor. Step 102: At preset intervals, based on the collected operating parameters and equipment status, a system state feature vector representing the current operating status of the ammonia refrigeration system is spliced ​​together. Step 103: In the domain knowledge base, the system state feature vector is used to retrieve the domain knowledge that is compatible with the current operating state of the ammonia refrigeration system and can optimize the production efficiency and energy efficiency of the quick-freezing production line. Based on the retrieved domain knowledge, multiple sets of candidate control parameters are generated, including candidate control parameters for controlling the ammonia refrigeration compressor. Step 104: The candidate control parameters of each set of candidate control parameters are fed into the digital twin in parallel for dynamic simulation. The optimal control parameter set is determined based on the dynamic simulation results so that the ammonia refrigeration compressor can achieve automatic optimization and energy saving after adjusting the optimal control parameters based on the optimal control parameter set. The digital twin performs real-time dynamic digital modeling of the quick-freezing production line based on the real-time collected operating parameters and equipment status.

[0013] Currently, the following bottlenecks exist for ammonia refrigeration systems: First, the control strategy is rigid. Traditional PID control relies on preset rules and struggles to adapt to dynamic cooling load fluctuations, resulting in lag in compression ratio and condensing temperature regulation. Second, there is insufficient coordination among multiple devices. The complex coupling between screw compressors, condensers, evaporators, and other devices prevents existing control methods from achieving global optimization. Third, data utilization is inefficient. Existing systems only collect single parameters (such as pressure and temperature) and lack fusion analysis of multimodal data such as vibration spectrum and lubricating oil status, leading to insufficient accuracy in fault warning and energy efficiency optimization. Fourth, AI applications are limited. Traditional machine learning models (such as LSTM) struggle to analyze the thermodynamic mechanisms of refrigeration systems, while large language models lack domain expertise, resulting in weak strategy generalization capabilities.

[0014] In the above embodiments of this application, the process architecture is as follows: Figure 2As shown, it includes four layers: execution and feedback layer, decision and optimization layer, digital twin and model layer, and perception layer. Specifically, various sensors can be deployed in the quick-freezing production line, such as pressure transmitters, temperature sensors, vibration acceleration sensors, oil condition sensors, and power quality analyzers. These sensors can be installed in key locations such as suction and exhaust pipes, heat exchange nodes, compressor bearing housings (drive and non-drive end bearing housings of ammonia refrigeration compressors), lubrication circuits, and motor distribution cabinets.

[0015] Next, the operating parameters of the ammonia refrigeration system are collected in real time, including the suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor; at the same time, the equipment status is collected, such as the vibration spectrum of the ammonia refrigeration compressor, the condition of the lubricating oil, and the electrical characteristics of the motor.

[0016] The collected data (operating parameters and device status) can be transmitted and processed through IoT gateways, serial servers, 4G DTUs and other devices to ensure stable data transmission and preliminary processing.

[0017] Next, the collected data is used to construct a digital twin of the ammonia refrigeration system on the quick-freezing production line, which includes digital models of components such as the compressor (ammonia refrigeration compressor), mechanism model, condenser, and evaporator.

[0018] The accuracy of the digital twin is continuously optimized through data-driven correction. At the same time, the large language model is enhanced by leveraging domain knowledge. By combining domain knowledge bases (such as refrigeration thermodynamics, compressor performance graphs, system operating boundaries, etc.) with the large language model and reinforcement learning strategies, the model's ability to understand and predict system operation is improved.

[0019] At preset time intervals, the collected operating parameters and equipment status information are concatenated to form a system status feature vector. This feature vector can comprehensively characterize the operating status of the ammonia refrigeration system at a given moment.

[0020] At the decision and optimization layer, the system state is first assessed. Using a large language model decision engine, the domain knowledge that optimizes the production and energy efficiency of the quick-freezing production line is identified in the domain knowledge base based on the system state feature vector.

[0021] Multiple sets of candidate control parameters are generated based on the identified domain knowledge, with each set of candidate control parameters corresponding to specific candidate control parameters.

[0022] Preliminary validation of multiple candidate control parameter sets was conducted using a digital twin sandbox. Then, combining multi-objective trade-offs and optimal decision-making methods, a more suitable candidate strategy was selected as the final optimal control parameter set. Specifically: Candidate control parameters from each set of candidate control parameters are fed in parallel into a digital twin for dynamic simulation. Based on real-time acquired operating parameters and equipment status, the digital twin performs real-time dynamic digital modeling of the quick-freezing production line, simulating the operation of the system (ammonia refrigeration system) and the quick-freezing production line under different strategies (one strategy corresponds to one set of candidate control parameters). Based on the dynamic simulation results, the impact of each candidate control parameter set on the production and energy efficiency of the quick-freezing production line is evaluated, and the optimal control parameter set is ultimately determined.

[0023] The determined optimal control parameter set is sent to the field actuators to automatically optimize and save energy for the ammonia refrigeration compressor.

[0024] In addition, the operating performance of the quick-freezing production line can be monitored in real time through performance deviation monitoring and data annotation. The actual operating data can be fed back to the large language model for online fine-tuning and reinforcement learning, continuously optimizing subsequent decision-making and optimization processes, forming a closed-loop optimization system, and continuously improving the operating efficiency and energy efficiency of the ammonia refrigeration system of the quick-freezing production line.

[0025] Therefore, by following the above steps, intelligent optimization and control of the ammonia refrigeration system in the quick-freezing production line can be achieved, thereby improving production efficiency, reducing energy consumption, and ensuring the stable operation of the system.

[0026] Optionally, in step 104, the candidate control parameters of each set of candidate control parameters are fed into the digital twin in parallel for dynamic simulation, and the optimal control parameter set is determined based on the dynamic simulation results, including: Step 1041: For any set of candidate control parameters, predict the performance index values ​​of the quick-freezing production line under the candidate control parameters using a digital twin. The performance index includes at least one of energy efficiency ratio, exhaust temperature, vibration intensity, and degree of approaching the safety boundary. Each performance index has a corresponding index weight. Step 1042: Based on the predicted index values ​​of each performance index and the corresponding index weights of the performance index, calculate the comprehensive performance score corresponding to the candidate control parameter set. Step 1043: Among the comprehensive performance scores corresponding to each group of candidate control parameter sets, select the candidate control parameter set with the highest comprehensive performance score as the final optimal control parameter set.

[0027] In the embodiments described above, after generating multiple sets of candidate control parameters (each set containing adjustable operational variables, i.e., candidate control parameters) based on the system state feature vector used for state assessment and the large language model combined with embedded domain knowledge, a digital twin can be used for sandbox verification. Specifically, each generated set of candidate control parameters is synchronously input into the constructed digital twin for rapid simulation. The digital twin predicts the future performance indicators of the quick-freezing production line under the given set of candidate control parameters, including coefficient of performance (COP), exhaust temperature, vibration intensity, and the degree to which it approaches the safety boundary.

[0028] By combining the multi-objective prediction results provided by the digital twin, a weighted scoring method is used to select the set of control parameters with the best overall performance.

[0029] Furthermore, the performance metrics are shown in Table 1: Table 1

[0030] The core of the weighted scoring method is to use weights to reflect the priority of business objectives. Combining the core requirements of quick-freezing production lines of "safety first, energy saving second, and stability as an auxiliary", the weight assignments can be as follows: energy efficiency ratio (COP) 0.35, exhaust temperature 0.30, vibration intensity 0.20, and degree of approaching the safety boundary 0.15.

[0031] Specifically, the digital twin is a virtual mirror of the physical ammonia refrigeration system. It predicts the future performance of candidate control parameter sets through real-time data calibration and mechanistic modeling. The specific prediction methods for four indicators are as follows: 1. Coefficient of Performance (COP) Prediction: Input: Candidate control parameters (such as compressor suction pressure setpoint, motor frequency, condensate flow rate).

[0032] Mechanism model: Combining the thermodynamic properties of ammonia refrigerant (such as enthalpy and entropy), calculations are performed using formulas: , Where m is the refrigerant circulation rate (calculated from suction pressure / temperature), h is the enthalpy (from the ammonia thermodynamic property table), and P... 电机 Calculated from current / frequency.

[0033] 2. Exhaust temperature prediction: Input: Candidate control parameters (such as inhalation pressure, exhaust pressure, and inhalation temperature).

[0034] Mechanism model: Based on the formula for adiabatic compression process, combined with overheating correction: , Where k is the adiabatic index of ammonia (for example, it can be 1.31), ΔT 过热 This is the intake overheat temperature (corrected from sensor data).

[0035] 3. Vibration intensity prediction: Input: Candidate control parameters (such as compressor load, bearing clearance setting).

[0036] Model: Combining mechanical dynamics models (such as bearing wear-vibration transfer function) and historical vibration data, predict the effective value (RMS) of vibration.

[0037] 4. Prediction of the degree to which the object approaches the safety boundary (safety margin): A weighted average of exhaust temperature margin and vibration intensity margin (each with a weight of 0.5) can be used: , In particular, different indicators have different dimensions (e.g., COP is dimensionless, exhaust temperature is in °C), so the original predicted values ​​need to be converted into a standardized score of 0-100 (the higher the "positive indicator", the higher the score; the lower the "negative indicator", the higher the score). Standardized formula for positive metrics (COP, safety margin): , Standardized formulas for negative indicators (exhaust temperature, vibration intensity): , Next, the comprehensive score of each candidate control parameter set is obtained by summing the "weight × standardized score", and the highest score satisfies the safety constraints.

[0038] Therefore, the above process ensures both the scientific nature of decision-making and takes into account business priorities, making it a core tool for the "intelligent optimization" of the ammonia refrigeration system in the quick-freezing production line.

[0039] Optionally, in step 104, after determining the optimal control parameter set based on the dynamic simulation results, the method further includes: Step 105: The optimal control parameter set is sent to the PLC or DCS drive actuator corresponding to the ammonia refrigeration compressor through an industrial communication protocol, wherein the industrial communication protocol includes OPCU and Modbus.

[0040] In the above embodiments of this application, the optimal control parameters can be sent to the PLC or DCS on site to drive the actuators for control parameter adjustment via industrial communication protocols such as OPCUA or Modbus.

[0041] Optionally, in step 104, before feeding the candidate control parameters of each set of candidate control parameters into the digital twin for dynamic simulation in parallel, the method further includes: Step 106: Based on the physical structure, equipment model parameters and refrigerant thermophysical properties of the ammonia refrigeration compressor, construct a digital model of the ammonia refrigeration compressor; Step 107: Based on the laws of conservation of mass, energy, and momentum, establish a digital model of the core component, and characterize the mechanism model of the core component through heat transfer coefficient and efficiency coefficient. The core component includes at least one of condenser, evaporator, and throttling device. Step 108: After calibrating the heat transfer coefficient and efficiency coefficient in the mechanism model using the historical operating parameters and historical equipment status of the quick-freezing production line, a digital twin for the quick-freezing production line is constructed based on the digital models of the ammonia refrigeration compressor and core components, as well as the calibrated mechanism model.

[0042] In the above embodiments of this application, a hybrid model combining mechanism and data-driven approaches can be constructed based on the physical structure of the ammonia refrigeration system, equipment model parameters, and refrigerant thermophysical properties, wherein: Mechanism Model Section: Based on the laws of conservation of mass, energy, and momentum, mathematical models of core components such as compressors, condensers, evaporators, and throttling devices are established, which can accurately simulate the steady-state and dynamic characteristics of the system.

[0043] Data-driven calibration: Using historical operating data, key parameters such as heat transfer coefficient and efficiency coefficient in the mechanism model are calibrated to ensure that the error between the output of the digital twin and the real system is less than a preset threshold, thereby forming a high-fidelity virtual system mapping.

[0044] Furthermore, as Figure 3 As shown, multi-source data acquisition is first performed, namely, collecting a set of operating parameters (including pressure, temperature, and power) and a set of equipment states (including vibration, hydraulic, and electrical parameters). Then, data fusion and preprocessing are performed to form a system state feature vector for the ammonia refrigeration system. This vector is input into the constructed high-fidelity digital twin hybrid model. Training is considered complete when the training error is less than a preset threshold. For the high-fidelity digital twin construction process, a mechanistic model can be built based on the conservation of mass, energy, and momentum. Then, operating data is used to calibrate the model parameters, such as the heat transfer coefficient and efficiency coefficient.

[0045] Optionally, in step 103, before retrieving domain knowledge from the domain knowledge base that is compatible with the current operating state of the ammonia refrigeration system and can optimize the production efficiency and energy efficiency of the quick-freezing production line using the system state feature vector, the method further includes: Step 109: Obtain domain knowledge in text form, wherein the domain knowledge includes at least one of the following: refrigeration thermodynamics principles, compressor performance curve graphs, system safe operation boundaries, equipment maintenance manuals, historical fault case databases, and expert operation experience rules; Step 110: Using natural language processing technology, the domain knowledge is segmented and vectorized to obtain domain knowledge feature vectors, which are stored in the domain knowledge feature vector library. Based on the domain knowledge feature vector library and the large language model running on the domain knowledge feature vector library, a domain knowledge library is constructed. The constructed domain knowledge library matches the input system state feature vector with the corresponding domain knowledge feature vector through the large language model.

[0046] In the above embodiments of this application, structured and unstructured domain expertise can be digitized, including, for example, principles of refrigeration thermodynamics, compressor performance curves, system safety operating boundaries, equipment maintenance manuals, historical fault case libraries, expert operating experience rules, etc.

[0047] Using natural language processing techniques, the aforementioned knowledge text is segmented, embedded, and stored in a vector database to form a domain knowledge base that can be quickly retrieved and invoked by large language models.

[0048] Next, a general-purpose large language model with a suitable parameter size can be selected as the base model, and the base model can be trained using a domain-specific dataset consisting of historical running data, expert operation sequences, and simulation scenarios generated by digital twins, employing instruction fine-tuning and human feedback-based reinforcement learning (RLHF) strategies.

[0049] The training objective is to enable the model to not only understand natural language instructions, but also to deeply understand the operating logic, constraints, and optimization objectives of the ammonia refrigeration system, thereby transforming it from a general dialogue model into an "ammonia refrigeration system expert".

[0050] In practical use, the generated real-time system state feature vector, along with the user-defined optimization objectives (such as optimizing the production efficiency and energy efficiency of the quick-freezing production line), is input into the domain knowledge-enhanced large language model. The large language model first performs a deep interpretation of the current system state, for example: "The current condensing temperature is too high, and the compressor vibration energy is rising at the 2nd harmonic, indicating that there may be a slight misalignment, and the energy efficiency ratio is lower than the historical level for the same period." Specifically, such as Figure 4As shown, the first step is to construct and vectorize a domain knowledge base. Domain-specific knowledge can be derived from refrigeration thermodynamics principles, compressor performance curves, safe operating boundaries, maintenance manuals, fault case libraries, and expert experience rules. This knowledge is then digitized and vectorized to construct a vectorized domain knowledge base. Specifically, domain-adaptive fine-tuning can be performed on the large language model using instruction fine-tuning and RLHF. Data sources also include historical operation and management data and digital twin simulation data. A general-purpose language model is used, for example, the Qwen-7B base model.

[0051] Optionally, in step 104, after the ammonia refrigeration compressor performs automatic energy-saving adjustment based on the optimal control parameter set, the method further includes: Step 111: Monitor the energy efficiency ratio of the quick-freezing production line under the adjustment of optimal control parameters and compare it with the energy efficiency ratio simulated by the digital twin. Step 112: When the deviation exceeds the set adaptive threshold, the optimal control parameters corresponding to the optimal control parameter set are stored as a set of feedback samples in the experience playback pool. Step 113: When the number of feedback samples in the experience replay pool reaches a preset number, retrain the large language model based on the feedback samples in the experience replay pool.

[0052] In the above embodiments of this application, the actual effect of operation under the new control strategy (optimal control parameter set) can be continuously monitored, the energy efficiency ratio can be calculated and compared with the energy efficiency ratio predicted by the digital twin. When the deviation exceeds the set adaptive threshold, the system automatically labels the complete data sequence of "state-strategy-actual result" and stores it as an important feedback sample in the experience playback pool.

[0053] Next, the model update process is initiated periodically or when a certain number of feedback samples are accumulated. With the goal of improving the model's decision-making efficiency and ensuring more stable operation, reinforcement learning algorithms are used to fine-tune the large language model online.

[0054] This process enables large language models to learn from both successes and failures, continuously refine their decision-making logic, and gradually adapt to slowly changing factors such as equipment performance degradation and environmental changes, thus achieving self-evolution from "good" to "excellent".

[0055] Specifically, such as Figure 5As shown, the system (ammonia refrigeration system) is first evaluated for its state and its optimization objectives are analyzed. This includes inputting real-time system state feature vectors and user-defined optimization objectives, such as optimizing the production and energy efficiency of the quick-freezing production line. Next, multi-strategy generation and digital twin generation are performed. Multi-strategy generation is performed using a large language model, and candidate control parameter sets are output. After verification through a digital twin sandbox, performance indicators are quickly simulated and predicted. Then, multi-objective trade-offs and optimal decision-making are performed. Through multi-objective trade-offs and optimal decision-making, the optimal control parameter set is output, and control commands are issued to the physical ammonia refrigeration system.

[0056] In particular, it can also update the large language model in real time, realize closed-loop feedback and model self-evolution, monitor and evaluate performance deviation, evaluate actual results and pre-tests, and if the deviation is greater than the adaptive threshold, the data is calibrated and put into the experience replay pool for online fine-tuning of the model based on reinforcement learning, thus completing the update of the large language model.

[0057] Optionally, in step 101, the operating parameters and equipment status of the ammonia refrigeration system are collected in real time, including: Step 1011: Vibration acceleration sensors deployed on the drive end and non-drive end bearing seats of the ammonia refrigeration compressor are used to collect the effective value of vibration velocity, acceleration envelope value, rotor dynamic balance state and bearing wear state of the ammonia refrigeration compressor in real time, so as to quantify the vibration spectrum of the ammonia refrigeration compressor. Step 1012: By deploying an oil condition sensor in the lubrication circuit corresponding to the ammonia refrigeration compressor, the trace moisture content and particulate contamination of the lubricating oil are collected in real time to quantify the lubricating oil condition of the ammonia refrigeration compressor. Step 1013: By using a power quality analyzer deployed in the main motor distribution cabinet corresponding to the ammonia refrigeration compressor, the current, voltage, power, power factor, total harmonic distortion rate and specific harmonic content of the motor are collected in real time to quantify the electrical characteristics of the motor. Step 1014: The suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor are measured in real time using pressure transmitters and temperature sensors. The collected operating parameters and equipment status are pre-processed at the edge by a multi-protocol industrial IoT gateway deployed in the computer room, stored locally in a time series database, and uploaded to the control center of a preset private cloud platform via an industrial fiber optic ring network.

[0058] In the above embodiments of this application, a sensor network deployed at key nodes of the ammonia refrigeration system is used to collect operating parameter sets and equipment status sets in real time. The operating parameter sets include core thermodynamic parameters of the system such as compressor suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, evaporation temperature, motor operating current, and real-time power. The equipment status sets include multimodal data such as vibration spectrum data, lubricating oil status data, motor electrical characteristic data, and data fusion and preprocessing.

[0059] Currently, traditional control methods cannot effectively respond to rapid and nonlinear changes in operating conditions such as cooling load and ambient temperature, causing compressors to operate at suboptimal operating points for extended periods, resulting in low energy efficiency. Ammonia refrigeration systems are complex systems with multiple coupled devices. Existing control strategies are mostly local optimizations, lacking global collaborative optimization of devices such as compressors, condensers, evaporators, and throttling devices, making it difficult to maximize system energy efficiency. Traditional AI models lack an understanding of the physical mechanisms of thermodynamics and fluid mechanics in the refrigeration field, while general-purpose models lack domain-specific knowledge, leading to physically infeasible or poorly generalizable control strategies. Furthermore, current methods fail to fully utilize multimodal data such as vibration, current harmonics, and oil state generated during system operation for deep fusion analysis with traditional temperature and pressure data to provide early warnings of energy efficiency degradation trends and potential equipment failures.

[0060] By applying the technical solution of this embodiment, through system-level reasoning of a large language model and real-time simulation of digital twins, collaborative control of multiple devices in an ammonia refrigeration system is achieved. This enables rapid response to changes in operating conditions, leaping from static, local optimization to dynamic, global optimization, significantly improving the overall energy efficiency of the system. Embedding domain knowledge into the large language model ensures that the control strategy conforms to physical laws and can learn from data, solving the problems of physical infeasibility in traditional AI model decision-making and the lack of professional knowledge in general large language models, thus enhancing the reliability and generalization ability of the strategy. Systematically utilizing multimodal data such as vibration, oil, and current not only optimizes energy efficiency but also enables early fault warning, achieving a balance between energy saving and safe maintenance, and improving overall operational efficiency. Through closed-loop feedback and reinforcement learning mechanisms, the system can continuously learn from operational results, automatically correct model biases, adapt to equipment aging and environmental changes, and possess the self-evolutionary ability to maintain optimal performance over the long term.

[0061] In one specific embodiment, this can be applied, for example, to a central refrigeration room that provides a cooling source for the entire plant's quick-freezing production line, spiral tower, and high- and low-temperature cold storage. The central refrigeration room includes four 315kW screw-type ammonia refrigeration compressors operating in a "3-in-1-out" mode with automatic adjustment capabilities; three plate heat exchanger condensers equipped with a frequency converter-controlled cooling tower system; multiple evaporator groups with different temperature zones; and throttling using an electronic expansion valve cluster. The existing control system uses a PLC + host computer monitoring system, providing basic data acquisition and automatic control functions.

[0062] At this point, a comprehensive intelligent sensing upgrade will be carried out based on the existing control system: Vibration monitoring involves installing industrial-grade IEPE accelerometers in the drive and non-drive bearing housings of each compressor. These sensors, with a sampling frequency of 25.6kHz, are used to collect high-frequency vibration signals. The vibration velocity effective value (VRMS) and acceleration envelope value are calculated in real time via an edge computing gateway to monitor rotor dynamic balance and bearing wear status.

[0063] Oil quality monitoring involves installing an online oil quality sensor in the lubrication circuit of each compressor to monitor the trace moisture content (accuracy ±5ppm) and particulate contamination of the lubricating oil in real time, and outputting the data to the data acquisition system via a 4-20mA signal.

[0064] Electrical monitoring involves installing a three-phase power quality analyzer in the compressor's main motor distribution cabinet to continuously collect current, voltage, power, and power factor data. The analyzer focuses on the total harmonic distortion rate and specific harmonic content of the current to assess the electrical health of the motor.

[0065] Thermodynamic parameters are supplemented by high-precision pressure transmitters (accuracy 0.1%FS) and PT100 temperature sensors (accuracy ±0.1℃) at key nodes of the system to accurately measure parameters such as intake pressure, exhaust pressure / temperature, condensation temperature, evaporation temperature, subcooling and superheat.

[0066] Data is aggregated, and all sensor data undergoes edge preprocessing through a multi-protocol industrial IoT gateway deployed in the data center. The gateway has data caching and protocol conversion functions, stores data locally using a time-series database, and uploads it to the enterprise's private cloud platform control center via an industrial fiber optic ring network.

[0067] For the construction of digital twins and domain-specific large language models: 1. Construction of a high-fidelity digital twin: The platform can be the ANSYS TwinBuilder digital twin platform, combined with Python scripts to develop custom components and build a hybrid model of the system.

[0068] In terms of establishing the mechanism model, a compressor model based on neural networks was established based on the three-dimensional performance map (efficiency, power, and flow rate changes with pressure ratio and guide vane opening) provided by the compressor manufacturer to achieve rapid and accurate calculations; a distributed parameter model was established using the finite volume method to establish condenser and evaporator models, simulating the heat exchange process between refrigerant and cooling water / air in detail, and taking into account actual factors such as fin efficiency and fouling coefficient; a pipe network model including local resistance coefficients was established to accurately calculate system pressure loss.

[0069] In terms of data-driven approach, historical data and operational data under different production plans are collected. Bayesian parameter estimation methods are used to invert and dynamically correct the overall heat transfer coefficient of the condenser, the overall heat transfer coefficient of the evaporator, and the compressor efficiency coefficient in the digital twin. After correction, the mean absolute error (MAE) of the key system parameters simulated by the digital twin is controlled within 2.5% compared to the actual values.

[0070] 2. Construction of a large language model enhanced with domain knowledge: Considering the enterprise's data security and real-time response requirements, the open-source model Qwen-7B with 7 billion parameters can be selected as the foundation and deployed on the enterprise's private cloud. A knowledge base is established by integrating the "Safety Specifications for Refrigeration Systems in the Food Industry," equipment technical manuals (including performance curves), nearly ten years of equipment maintenance records, temperature control requirements in the HACCP system, and expert rules for fault diagnosis summarized by senior engineers. Simultaneously, the BAAI / bge-large-zh model is used to convert the above text into 1024-dimensional vectors, perform vectorization processing, and store them in the Milvus vector database to achieve efficient similarity retrieval.

[0071] After completing the construction of the domain knowledge base, perform domain-adaptive fine-tuning.

[0072] The system extracts valid sequences of "system state - operation command - operation result" from historical systems. Simultaneously, it uses digital twins to generate double-set simulation data covering various production conditions, including quick-freezing line startup and batch warehousing. Training is performed using a combination of command fine-tuning and reinforcement learning. Energy efficiency, stability, and safety are comprehensively considered.

[0073] After training, the model can output professional and explainable decisions such as, "It is recommended to reduce the load of compressor No. 2 to 70% and increase the frequency of the cooling tower fan in group B to 45Hz. Reason: The current quick-freezing line task has ended, the load has decreased, this adjustment can avoid compressor surge, is expected to increase the system COP by about 6%, and ensure that the exhaust temperature is within a safe range."

[0074] For online operation and closed-loop optimization processes: Once the system is built, it enters an online automatic operation mode. Its core workflow is as follows: 1. Triggering and Perception: The optimization cycle is dynamically adjusted based on production characteristics: triggered every 10 minutes during peak production periods and every 30 minutes during nighttime or low-load periods. Emergency optimization is triggered immediately when the system detects the start-up of a quick-freezing line, large-scale product in / out of the warehouse, or abnormal key parameters. All real-time data is aggregated and processed through the enterprise's industrial internet platform to form a unified system status feature vector.

[0075] 2. Reasoning and Decision-Making (Typical Decision-Making Examples): Scenario description: During the peak summer production season, at 2:00 PM, three quick-freezing lines are operating simultaneously with an ambient temperature of 35°C and a system load rate of 95%. Status monitoring shows that the condensing temperature has risen to 34°C, and the total vibration of compressor No. 1 has increased by 15%.

[0076] After receiving the state vector and the instruction to "ensure optimal production and energy efficiency," the model searches the domain knowledge base and identifies two key points: "high ambient temperature leads to increased condensing pressure" and "increased vibration requires control." Three sets of candidate strategies (candidate control parameter sets) are generated: Strategy A: Ensure full production capacity. {Compressor No. 1 load: 100%, Cooling tower fan frequency: 100%}.

[0077] Strategy B: Intelligent Collaborative Optimization. {Compressor 1 load: 85%, Compressor 3 starts and loads to 40%, Cooling tower fan frequency: 85%}.

[0078] Strategy C: Conservative operation. {Compressor 1 load: 80%, Cooling tower fan frequency: 90%}.

[0079] The three sets of candidate control parameters for the vibration increase were fed in parallel into a digital twin for an 8-minute dynamic simulation for verification. The prediction results are as follows: Strategy A: Maximum cooling capacity, but lowest COP, and the predicted compressor vibration intensity will be close to the alarm value.

[0080] Strategy B: Cooling capacity meets demand, COP is highest, and operating parameters of all equipment are within the optimal range.

[0081] Strategy C: The safest option, but the cooling capacity may not be sufficient to meet the simultaneous operation requirements of three quick-freezing lines.

[0082] The large language model comprehensively weighs cooling capacity, energy efficiency, and equipment safety, ultimately selecting strategy B as the optimal solution. Control commands are sent to the field PLC system via the OPCUA protocol, and the system automatically starts compressor number 3 and executes the commands according to the parameters set in strategy B.

[0083] 3. Feedback and Evolution: Performance monitoring: In the next cycle after strategy B is implemented, the system calculates an actual average COP of 4.25, indicating that the cooling capacity fully meets production requirements.

[0084] Deviation Analysis and Learning: Case 1 (Model Enhancement): The digital twin predicted a COP of 4.32 for Strategy B under this condition, with a deviation of 1.6% between the actual and predicted values ​​(less than the set 3% threshold). This data was stored as a positive sample in the experience replay pool.

[0085] Case 2 (Model Correction): During a spring / autumn transition period, after the model made its decision, the actual COP was 4.9, but the predicted value was 5.1, a deviation of 4.1%. Analysis revealed that the cause was slight blockage of the cooling tower packing, leading to a decrease in actual heat exchange efficiency. The system marked this data as a high-value negative feedback sample.

[0086] Online fine-tuning: Every two weeks, the large language model can be fine-tuned online using accumulated feedback samples and a near-end strategy optimization algorithm. Through continuous learning, the model gradually masters more precise control strategies under different seasons and device conditions.

[0087] Furthermore, as Figure 1 In specific implementation of the method, this application provides an automatic optimization and energy-saving system for an ammonia refrigeration compressor, such as... Figure 6 As shown, this is applied to a quick-freezing production line, which uses an ammonia refrigeration system for cooling. The ammonia refrigeration system includes an ammonia refrigeration compressor and a motor for supplying power to the ammonia refrigeration compressor. The system includes: The multi-source data acquisition module 201 is used to collect the operating parameters and equipment status of the ammonia refrigeration system in real time through sensors deployed on the quick-freezing production line. The operating parameters include at least one of the following: suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor. The equipment status includes at least one of the following: vibration spectrum of the ammonia refrigeration compressor, lubricating oil status of the ammonia refrigeration compressor, and electrical characteristics of the motor. The system state feature vector splicing module 202 is used to splice together a system state feature vector representing the current operating state of the ammonia refrigeration system based on the collected operating parameters and equipment status at preset intervals. The candidate control parameter set filtering module 203 is used to retrieve, from the domain knowledge base, domain knowledge that is compatible with the current operating state of the ammonia refrigeration system and can respectively optimize the production efficiency and energy efficiency of the quick-freezing production line by using the system state feature vector. Based on the retrieved domain knowledge, multiple sets of candidate control parameters are generated, including candidate control parameters for controlling the ammonia refrigeration compressor. The automatic optimization and energy saving module 204 is used to send the candidate control parameters of each set of candidate control parameters into the digital twin in parallel for dynamic simulation. Based on the dynamic simulation results, the optimal control parameter set is determined so that the ammonia refrigeration compressor can achieve automatic optimization and energy saving after adjusting the optimal control parameters based on the optimal control parameter set. The digital twin performs real-time dynamic digital modeling of the quick-freezing production line based on the real-time collected operating parameters and equipment status.

[0088] It should be noted that other corresponding descriptions of the functional units involved in the automatic optimization and energy-saving system for an ammonia refrigeration compressor provided in this application embodiment can be found in the following references. Figures 1 to 5 The corresponding descriptions in the method will not be repeated here.

[0089] Based on the above, Figures 1 to 5 Accordingly, this application also provides a storage medium storing a computer program, which, when executed by a processor, implements the above-described method. Figures 1 to 5 The method for automatic optimization and energy saving of ammonia refrigeration compressors is shown.

[0090] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive) and includes several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the methods described in the various implementation scenarios of this application.

[0091] Based on the above, Figures 1 to 5 The method shown, and Figure 6 To achieve the above objectives, the virtual system embodiment shown in this application also provides a computer device, specifically a personal computer, server, network device, etc. This computer device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described objectives. Figures 1 to 5 The method for automatic optimization and energy saving of ammonia refrigeration compressors is shown.

[0092] Optionally, the computer device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB ports, card reader ports, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Bluetooth interfaces, Wi-Fi interfaces), etc.

[0093] Those skilled in the art will understand that the computer device structure provided in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0094] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages and stores the hardware and software resources of a computer device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software within the physical device.

[0095] Through the above description of the implementation methods, those skilled in the art can clearly understand that this application can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented using hardware to collect the operating parameters and equipment status of the ammonia refrigeration system in real time through sensors; at preset time intervals, a system state feature vector representing the current operating state of the ammonia refrigeration system is formed; in the domain knowledge base, the system state feature vector is used to retrieve domain knowledge that is compatible with the current operating state of the ammonia refrigeration system and can optimize the production efficiency and energy efficiency of the quick-freezing production line; based on the retrieved domain knowledge, multiple sets of candidate control parameters are generated, including candidate control parameters for controlling the ammonia refrigeration compressor; the candidate control parameters of each set of candidate control parameters are fed in parallel into a digital twin for dynamic simulation; based on the dynamic simulation results, the optimal control parameter set is determined, so that the ammonia refrigeration compressor can achieve automatic optimization and energy saving after adjusting based on the optimal control parameters of the optimal control parameter set. The large language model (LLM) and multimodal data fusion technology are used to realize dynamic optimization and energy-saving control of the operating parameters of the ammonia refrigeration compressor.

[0096] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the system of the embodiment scenario can be distributed throughout the system of the embodiment scenario as described, or they can be modified to reside in one or more systems different from this embodiment scenario. The modules of the above-described embodiment scenario can be combined into one module, or further divided into multiple sub-modules.

[0097] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any modifications that can be made by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for automatic optimization and energy saving of an ammonia refrigeration compressor, characterized in that, The method is applied to a quick-freezing production line, wherein the quick-freezing production line uses an ammonia refrigeration system for cooling, the ammonia refrigeration system including an ammonia refrigeration compressor and a motor for supplying power to the ammonia refrigeration compressor, the method comprising: Sensors deployed on the quick-freezing production line are used to collect the operating parameters and equipment status of the ammonia refrigeration system in real time. The operating parameters include at least one of the following: suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor. The equipment status includes at least one of the following: vibration spectrum of the ammonia refrigeration compressor, lubricating oil status of the ammonia refrigeration compressor, and electrical characteristics of the motor. At preset intervals, based on the collected operating parameters and equipment status, a system state feature vector representing the current operating status of the ammonia refrigeration system is constructed. In the domain knowledge base, the system state feature vector is used to retrieve domain knowledge that is compatible with the current operating state of the ammonia refrigeration system and can optimize the production efficiency and energy efficiency of the quick-freezing production line. Based on the retrieved domain knowledge, multiple sets of candidate control parameters are generated, including candidate control parameters for controlling the ammonia refrigeration compressor. Candidate control parameters from each set of candidate control parameters are fed into a digital twin in parallel for dynamic simulation. The optimal control parameter set is determined based on the dynamic simulation results, so that the ammonia refrigeration compressor can achieve automatic optimization and energy saving after adjusting the optimal control parameters based on the optimal control parameter set. The digital twin performs real-time dynamic digital modeling of the quick-freezing production line based on the real-time collected operating parameters and equipment status.

2. The method according to claim 1, characterized in that, Candidate control parameters from each set of candidate control parameters are fed in parallel into a digital twin for dynamic simulation. Based on the dynamic simulation results, the optimal control parameter set is determined, including: For any set of candidate control parameters, the performance index values ​​of the quick-freezing production line under the candidate control parameters are predicted by a digital twin. The performance index includes at least one of energy efficiency ratio, exhaust temperature, vibration intensity and degree of approaching the safety boundary, and the performance index has a corresponding index weight. Based on the predicted index values ​​of each performance index and the corresponding index weights of the performance indexes, the comprehensive performance score corresponding to the candidate control parameter set is calculated. Among the comprehensive performance scores of each group of candidate control parameter sets, the candidate control parameter set with the highest comprehensive performance score is selected as the final optimal control parameter set.

3. The method according to claim 1, characterized in that, After determining the optimal control parameter set based on dynamic simulation results, the method further includes: The optimal control parameter set is sent to the PLC or DCS drive actuator corresponding to the ammonia refrigeration compressor through industrial communication protocols, including OPCU and Modbus.

4. The method according to claim 1, characterized in that, Before feeding the candidate control parameters of each set of candidate control parameters into the digital twin for dynamic simulation in parallel, the method further includes: A digital model of an ammonia refrigeration compressor is constructed based on its physical structure, equipment model parameters, and refrigerant thermophysical properties. Based on the laws of conservation of mass, energy, and momentum, a digital model of the core component is established, and the mechanism model of the core component is characterized by the heat transfer coefficient and efficiency coefficient. The core component includes at least one of a condenser, an evaporator, and a throttling device. By utilizing the historical operating parameters and equipment status of the quick-freezing production line, the heat transfer coefficient and efficiency coefficient in the mechanism model are calibrated. Based on the digital models of the ammonia refrigeration compressor and core components, as well as the calibrated mechanism model, a digital twin of the quick-freezing production line is constructed.

5. The method according to claim 2, characterized in that, Before retrieving domain knowledge from the domain knowledge base that is compatible with the current operating state of the ammonia refrigeration system and can optimize the production efficiency and energy efficiency of the quick-freezing production line, using system state feature vectors, the method further includes: Acquire domain knowledge in text form, wherein the domain knowledge includes at least one of the following: principles of refrigeration thermodynamics, compressor performance curves, system safe operating boundaries, equipment maintenance manuals, historical fault case databases, and expert operating experience rules; Using natural language processing techniques, domain knowledge is segmented and vectorized to obtain domain knowledge feature vectors, which are stored in a domain knowledge feature vector library. Based on the domain knowledge feature vector library and a large language model running on the domain knowledge feature vector library, a domain knowledge base is constructed. The constructed domain knowledge base matches the input system state feature vector with the corresponding domain knowledge feature vector through the large language model.

6. The method according to claim 5, characterized in that, After the ammonia refrigeration compressor performs automatic optimization and energy saving based on the optimal control parameter set, the method further includes: Monitor the energy efficiency ratio of the quick-freezing production line under the adjustment of optimal control parameters and compare it with the energy efficiency ratio simulated by digital twin; When the deviation exceeds the set adaptive threshold, the optimal control parameters corresponding to the optimal control parameter set are stored as a set of feedback samples in the experience playback pool. When the number of feedback samples in the experience replay pool reaches a preset number, the large language model is retrained based on the feedback samples in the experience replay pool.

7. The method according to any one of claims 1 to 6, characterized in that, Real-time acquisition of operating parameters and equipment status of the ammonia refrigeration system, including: Vibration acceleration sensors deployed on the drive and non-drive bearing housings of the ammonia refrigeration compressor are used to collect the effective value of vibration velocity, acceleration envelope value, rotor dynamic balance state and bearing wear state of the ammonia refrigeration compressor in real time, which is used to quantify the vibration spectrum of the ammonia refrigeration compressor. By deploying oil condition sensors in the lubrication circuit of the ammonia refrigeration compressor, the trace moisture content and particulate contamination of the lubricating oil are collected in real time to quantify the lubricating oil condition of the ammonia refrigeration compressor. By deploying a power quality analyzer in the main motor distribution cabinet corresponding to the ammonia refrigeration compressor, the motor's current, voltage, power, power factor, total harmonic distortion rate and specific harmonic content are collected in real time to quantify the motor's electrical characteristics. The suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor are measured in real time using pressure transmitters and temperature sensors. The collected operating parameters and equipment status are pre-processed at the edge by a multi-protocol industrial IoT gateway deployed on-site in the computer room, stored locally in a time series database, and uploaded to the control center of a pre-set private cloud platform via an industrial fiber optic ring network.

8. An automatic optimization and energy-saving system for an ammonia refrigeration compressor, characterized in that, This system is applied to a quick-freezing production line, which uses an ammonia refrigeration system for cooling. The ammonia refrigeration system includes an ammonia refrigeration compressor and a motor for supplying power to the ammonia refrigeration compressor. The multi-source data acquisition module is used to collect the operating parameters and equipment status of the ammonia refrigeration system in real time through sensors deployed on the quick-freezing production line. The operating parameters include at least one of the following: suction pressure, discharge pressure, suction temperature, discharge temperature, condensation temperature, and evaporation temperature of the ammonia refrigeration compressor. The equipment status includes at least one of the following: vibration spectrum of the ammonia refrigeration compressor, lubricating oil status of the ammonia refrigeration compressor, and electrical characteristics of the motor. The system state feature vector splicing module is used to splice together a system state feature vector representing the current operating state of the ammonia refrigeration system based on the collected operating parameters and equipment status at preset time intervals. The candidate control parameter set filtering module is used to retrieve domain knowledge from the domain knowledge base that is compatible with the current operating state of the ammonia refrigeration system and can optimize the production efficiency and energy efficiency of the quick-freezing production line, respectively. Based on the retrieved domain knowledge, multiple sets of candidate control parameters are generated, including candidate control parameters for controlling the ammonia refrigeration compressor. The automatic optimization and energy-saving module is used to send the candidate control parameters of each set of candidate control parameters into the digital twin in parallel for dynamic simulation. Based on the dynamic simulation results, the optimal control parameter set is determined so that the ammonia refrigeration compressor can achieve automatic optimization and energy saving after adjusting the optimal control parameters based on the optimal control parameter set. The digital twin performs real-time dynamic digital modeling of the quick-freezing production line based on the real-time collected operating parameters and equipment status.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the automatic optimization and energy-saving method for the ammonia refrigeration compressor as described in any one of claims 1 to 7.

10. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the automatic optimization and energy-saving method for the ammonia refrigeration compressor as described in any one of claims 1 to 7.

Images