Internet of Things edge control method based on central air conditioner water chilling unit

Abstract

The invention belongs to the technical field of central air conditioner control, and particularly relates to an internet-of-things edge control method based on a central air conditioner water chilling unit, which comprises the following steps of: acquiring multi-dimensional operation parameters in real time; the parameters are classified according to hospital operating rooms, wards and public areas and then input into a cold load prediction model, and future cold load values and the like of all the areas are predicted through a long-short term memory (LSTM) network. Through deep fusion of Internet of Things edge calculation, a deep learning model and a dynamic optimization algorithm, the key problems of insufficient control precision, large energy efficiency fluctuation, cooperative control lagging and the like of a traditional central air conditioning system in a complex scene are solved, and the purposes of deeply fusing real-time data features, dynamically optimizing a control strategy and improving the control efficiency are achieved. And a multivariable coupling relationship can be effectively processed, so that the effect of high-precision and high-response temperature control requirements in a medical scene can be met.

Description

Technical Field

[0001] The present invention belongs to the technical field of central air-conditioning control, and in particular relates to an Internet of Things edge control method based on a central air-conditioning chiller. Background Art

[0002] Traditional central air conditioning control systems typically consist of a basic architecture consisting of distributed sensors, actuators, edge computing gateways, and cloud servers. The sensor network collects chiller operating parameters, the actuators receive control commands and adjust equipment status, the edge gateway handles local data processing and preliminary decision-making, and the cloud server provides long-term data storage and remote monitoring. This architecture works well in typical scenarios, but it exhibits significant drawbacks in specialized environments like hospitals, where extremely high precision temperature and humidity control is required. In existing technologies, cloud servers often generate control instructions based on empirical parameters. When sensors are affected by environmental interference, causing data drift, the collaborative decision-making between the edge gateway and the server will produce systematic deviations. This deviation directly leads to a significant drop in the coefficient of refrigeration (COP value). Measured data show that the COP value of traditional methods in emergency cooling scenarios in operating rooms can fluctuate by up to ±30%, which in severe cases can lead to the risk of exceeding temperature and humidity standards. The reason for this is that the hospital cooling system involves the coupling of multiple factors such as chillers, cooling towers, transmission pipelines, and terminal environments. There are strong nonlinear correlations between variables: changes in evaporator temperature will trigger compressor frequency adjustments, which in turn affects the heat exchange efficiency of the condenser, while the delayed response of the cooling tower fan will in turn restrict the dynamic balance of the entire system. This "multi-variable, strongly coupled, nonlinear, and large inertia" system characteristics make it difficult for traditional linear control models to accurately establish the correlation matrix between factors. Although previous studies have attempted to predict cooling loads using LSTM neural networks, they have not addressed the key issue of coordinated control of cooling towers and chillers. Patent application CN119353765A proposes a hierarchical control strategy, but its path priority calculation still relies on static weight allocation and fails to achieve optimal resource allocation in a dynamic environment. Furthermore, traditional multivariate regression models suffer from matrix ill-conditioning when processing high-dimensional data, which can easily lead to distortion of regression coefficients. This is particularly prominent in the construction of cooling efficiency models. Summary of the Invention

[0003] (1) Technical issues to be resolved In order to overcome the shortcomings of the above technical problems, the present invention provides an IoT edge control method based on a central air-conditioning chiller, which can deeply integrate real-time data features, dynamically optimize control strategies, and effectively handle multi-variable coupling relationships to meet the high-precision and high-response temperature control requirements in medical scenarios.

[0004] (2) Technical solution The present invention is implemented through the following technical solution: The present invention proposes an IoT edge control method based on a central air-conditioning chiller, comprising the following steps: S1: Real-time collection of multi-dimensional operating parameters, including: Chiller parameters: evaporator temperature , condenser temperature , compressor frequency , Chilled water flow ; Cooling tower parameters: cooling water inlet temperature , cooling water outlet temperature , fan speed , cooling water flow rate ; Pipeline parameters: Pipeline length , External insulation parameters , Sealed insulation parameters at the inlet and outlet ; Environmental parameters: indoor and outdoor temperature and humidity and their maximum / minimum limits (indoor temperature , indoor humidity , outdoor temperature , outdoor humidity , and its maximum limit 、 and minimum limit 、 ); Auxiliary parameters: crowd density and logistics dynamic data ; S2: The above parameters are classified into hospital operating rooms, wards, and public areas and then input into the cooling load prediction model. The long short-term memory network (LSTM) is used to predict the future cooling load value of each area. ; S3: Based on cooling load prediction value , building a cooling efficiency model through IoT edge computing , the multivariate nonlinear regression method is used to fit the following relationship: , Right now ;and The nonlinear equation is converted into a multiple linear regression equation by substitution method: ;Right now

[0005] in, are all the parameters from steps a to e, is the error term, is the regression coefficient, and One-to-one correspondence, the regression coefficient is dynamically adjusted through historical data learning; S4: Based on the cooling efficiency model, dynamically optimize the operation strategies of the chillers and cooling towers, adjust the regional distribution paths and priorities of cooling water and chilled water, and prioritize the temperature and humidity control in the operating room area. The cooling water distribution path optimization includes: According to the pipe resistance coefficient R Dynamically adjust the cooling water flow rate based on the water pump variable frequency characteristics ,satisfy: ; =1.2m / s is the velocity reference coefficient; is the rated speed of the fan.

[0006] S5: Collect parameters and actual cooling load data at historical time points as training samples, and periodically iterate and update the cooling load prediction model and cooling efficiency model.

[0007] Preferably, a long short-term memory network (LSTM) is used, which takes the chiller parameters, cooling tower parameters, environmental parameters and auxiliary parameters at past time points as input and the actual cooling load data at future time points as output; The hidden state update and prediction of the long short-term memory network LSTM are achieved by the following formula: Forget Gate: ; Input Gate: ; Candidate memory cells: ; Memory unit update: ; Output gate: ; Current hidden status: ; Cooling load forecast output: ; in, is the hidden state of the previous moment, and its dimension is , initial state ; Input parameter for the current moment, its dimension is , including all 18 parameters defined above; 、 are the dimensions of the hidden state. For example, if the hidden layer has 128 neurons, then 、 Both are 128; that is, the number of units in each hidden layer; 、 、 、 is the weight matrix, and its dimensions are ; tanh is the hyperbolic tangent function; 、 、 、 is the bias term, and its dimension is ; is the Sigmoid function, is element-wise multiplication.

[0008] Preferably, the weight adjustment of each variable in the multivariate linear regression equation includes: Chiller parameter weights Based on the temperature difference between the evaporator and the condenser Dynamic correction to meet , is the proportional coefficient, that is And so on; Cooling tower parameter weights According to the correlation optimization between fan speed and cooling water flow, , is the regulating factor, And so on; Pipeline parameter weights Degradation coefficient of thermal insulation performance with external application Periodic updates to meet , is the decay period, t is the running time, And so on; Preferably, in step S4, dynamically optimizing the operation strategy of the chiller and the cooling tower includes: Adjust the chiller water supply temperature setpoint based on cooling load forecast , satisfying the following formula to improve energy efficiency: ; in, is the design reference temperature; is the temperature adjustment range; Design the cooling capacity for the chiller.

[0009] Preferably, the method further includes: monitoring hidden faults of chillers, cooling towers and pipelines in real time through edge gateways, and using a Bayesian conditional probability network for diagnosis, the formula of which is: ; FA collection of fault types, including insufficient cooling water flow, evaporator scaling, and refrigerant leakage; E For evidence collection, including 、 、 、 Abnormal fluctuation data.

[0010] Preferably, the cooling load prediction model supports a time-of-use electricity price strategy and optimizes the operating cost through the following formula: ;in, Time-of-use electricity price, is the energy consumption in period t, Equipment maintenance costs.

[0011] Preferably, the method further comprises: displaying the operation status of the cooling station (including the evaporator temperature) in real time through the mobile terminal and the web terminal. , condenser temperature , actual energy efficiency ,in , total power consumption), alarm records, and energy-saving reports, supporting remote control and troubleshooting by operation and maintenance personnel.

[0012] Preferably, the method is implemented by a control system, the control system comprising: The data acquisition module is used to collect multi-dimensional operating parameters. The data acquisition module is implemented by the following devices: temperature sensors (accuracy: ±0.1°C) are installed on the evaporator, condenser, cooling tower inlet and outlet, and indoor and outdoor environments; Flow meters (ultrasonic type, accuracy: ±1%), installed in chilled water and cooling water pipes; Frequency converter (Modbus protocol), used to collect compressor frequency and fan speed; Infrared sensors and logistics RFID tags are used to obtain crowd density and logistics dynamic data; The edge computing module is deployed on the IoT edge gateway and performs the following functions: Run cooling load forecasting models; Construct a cooling efficiency model and dynamically update the regression coefficients through a multiple linear regression equation; Generate optimization strategies, including chiller supply water temperature settings, cooling water flow rates, and zone allocation priorities; Policy execution module, and dynamic control through the following devices: Electric regulating valve (accuracy: ±2%), regulating the flow of chilled water and cooling water; The frequency converter controls the compressor frequency and fan speed; Priority controller, which distributes cooling capacity by hospital operating room, ward, and public area classification, with operating room temperature and humidity control response time ≤ 5 seconds; The communication module uses the Modbus-RTU protocol to realize real-time data transmission connection between the data acquisition module, edge computing module and policy execution module. The sampling period is 1 minute and the data packet loss rate is ≤ 0.1%.

[0013] (3) Beneficial effects The present invention uses an edge computing gateway to collect multi-dimensional parameters such as chillers, cooling towers, pipelines, environment, and human / logistics flows in real time, and combines it with an LSTM neural network to predict cooling loads, avoiding the delay problem of traditional cloud-based decision-making. High-frequency data processing at the edge (sampling period of 1 minute) makes the temperature and humidity control response time in key areas such as operating rooms ≤5 seconds, significantly improving dynamic response capabilities. In addition, the present invention adopts a multivariate nonlinear regression model and converts it into a linear equation through the substitution method, solving the matrix ill-conditioning problem of traditional models under high-dimensional data. The regression coefficient is dynamically adjusted based on historical data (such as the dynamic correction weight of the temperature difference between the evaporator and the condenser), effectively capturing the strong nonlinear correlation between variables, and improving the COP value stability by more than 30%. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings: Figure 1 Schematic diagram of the control method of the present invention.

[0015] Figure 2 This is a functional block diagram of the control system of the present invention. DETAILED DESCRIPTION

[0016] The present invention is achieved through the following technical solutions: Reference Figure 1 As shown, the present invention proposes an IoT edge control method based on a central air-conditioning chiller, comprising the following steps: S1: Real-time collection of multi-dimensional operating parameters, including: Chiller parameters: evaporator temperature , condenser temperature , compressor frequency , Chilled water flow ; Cooling tower parameters: cooling water inlet temperature , cooling water outlet temperature , fan speed , cooling water flow rate ; Pipeline parameters: Pipeline length , External insulation parameters , Sealed insulation parameters at the inlet and outlet ; Environmental parameters: indoor and outdoor temperature and humidity and their maximum / minimum limits (indoor temperature , indoor humidity , outdoor temperature , outdoor humidity , and its maximum limit 、 and minimum limit 、 ); Auxiliary parameters: crowd density and logistics dynamic data ; S2: The above parameters are classified into hospital operating rooms, wards, and public areas and then input into the cooling load prediction model. The long short-term memory network (LSTM) is used to predict the future cooling load value of each area. ; S3: Based on cooling load prediction value , building a cooling efficiency model through IoT edge computing , the multivariate nonlinear regression method is used to fit the following relationship: , Right now ;and The nonlinear equation is converted into a multiple linear regression equation by substitution method: ;Right now

[0017] in, are all the parameters from steps a to e, is the error term, is the regression coefficient, and One-to-one correspondence, the regression coefficient is dynamically adjusted through historical data learning; S4: Based on the cooling efficiency model, dynamically optimize the operation strategies of the chillers and cooling towers, adjust the regional distribution paths and priorities of cooling water and chilled water, and prioritize the temperature and humidity control in the operating room area. The cooling water distribution path optimization includes: According to the pipe resistance coefficient R Dynamically adjust the cooling water flow rate based on the water pump variable frequency characteristics ,satisfy: ; =1.2m / s is the velocity reference coefficient; is the rated speed of the fan.

[0018] S5: Collect parameters and actual cooling load data at historical time points as training samples, and periodically iterate and update the cooling load prediction model and cooling efficiency model.

[0019] Among them, the long short-term memory network LSTM is used, which takes the chiller parameters, cooling tower parameters, environmental parameters and auxiliary parameters at the past time point as input and the actual cooling load data at the future time point as output; The hidden state update and prediction of the long short-term memory network LSTM are achieved by the following formula: Forget Gate: ; Input Gate: ; Candidate memory cells: ; Memory unit update: ; Output gate: ; Current hidden status: ; Cooling load forecast output: ; in, is the hidden state of the previous moment, and its dimension is , initial state ; Input parameter for the current moment, its dimension is , including all 18 parameters defined above; 、 are the dimensions of the hidden state. For example, if the hidden layer has 128 neurons, then 、 Both are 128; that is, the number of units in each hidden layer; 、 、 、 is the weight matrix, and its dimensions are ; tanh is the hyperbolic tangent function; 、 、 、 is the bias term, and its dimension is ; is the Sigmoid function, is element-wise multiplication.

[0020] The weight adjustment of each variable in the multivariate linear regression equation includes: Chiller parameter weights Based on the temperature difference between the evaporator and the condenser Dynamic correction to meet , is the proportional coefficient, that is And so on; Cooling tower parameter weights According to the correlation optimization between fan speed and cooling water flow, , is the regulating factor, And so on; Pipeline parameter weights Degradation coefficient of thermal insulation performance with external application Periodic updates to meet , is the decay period, t is the running time, And so on; In step S4, the dynamic optimization of the operation strategy of the chiller and the cooling tower includes: Adjust the chiller water supply temperature setpoint based on cooling load forecast , satisfying the following formula to improve energy efficiency: ; in, is the design reference temperature; is the temperature adjustment range; Design the cooling capacity for the chiller.

[0021] This also includes: real-time monitoring of hidden faults in chillers, cooling towers, and pipelines through edge gateways, and diagnosis using a Bayesian conditional probability network, the formula of which is: ; F A collection of fault types, including insufficient cooling water flow, evaporator scaling, and refrigerant leakage; E For evidence collection, including 、 、 、 Abnormal fluctuation data.

[0022] The cooling load prediction model supports the time-of-use electricity price strategy and optimizes the operating cost through the following formula: ;in, Time-of-use electricity price, is the energy consumption in period t, Equipment maintenance costs.

[0023] The method further includes: displaying the operation status of the cooling station (including the evaporator temperature) in real time through the mobile terminal and the web terminal. , condenser temperature , actual energy efficiency ,in , total power consumption), alarm records, and energy-saving reports, supporting remote control and troubleshooting by operation and maintenance personnel.

[0024] Wherein, the method is implemented by a control system, the control system (refer to Figure 2 shown) include: The data acquisition module is used to collect multi-dimensional operating parameters. The data acquisition module is implemented by the following devices: temperature sensors (accuracy: ±0.1°C) are installed on the evaporator, condenser, cooling tower inlet and outlet, and indoor and outdoor environments; Flow meters (ultrasonic type, accuracy: ±1%), installed in chilled water and cooling water pipes; Frequency converter (Modbus protocol), used to collect compressor frequency and fan speed; Infrared sensors and logistics RFID tags are used to obtain crowd density and logistics dynamic data; The edge computing module is deployed on the IoT edge gateway and performs the following functions: Run cooling load forecasting models; Construct a cooling efficiency model and dynamically update the regression coefficients through a multiple linear regression equation; Generate optimization strategies, including chiller supply water temperature settings, cooling water flow rates, and zone allocation priorities; Policy execution module, and dynamic control through the following devices: Electric regulating valve (accuracy: ±2%), regulating the flow of chilled water and cooling water; The frequency converter controls the compressor frequency and fan speed; Priority controller, which distributes cooling capacity by hospital operating room, ward, and public area classification, with operating room temperature and humidity control response time ≤ 5 seconds; The communication module uses the Modbus-RTU protocol to realize real-time data transmission connection between the data acquisition module, edge computing module and policy execution module. The sampling period is 1 minute and the data packet loss rate is ≤ 0.1%.

[0025] This invention solves key problems of traditional central air-conditioning systems in complex scenarios, such as insufficient control accuracy, large energy efficiency fluctuations, and collaborative control lag, through the deep integration of IoT edge computing, deep learning models, and dynamic optimization algorithms. This invention uses an edge computing gateway to collect multi-dimensional parameters such as chillers, cooling towers, pipelines, environment, and personnel / logistics flows in real time. Combined with an LSTM neural network, it predicts cooling load, avoiding the latency issues of traditional cloud-based decision-making. High-frequency data processing at the edge (with a sampling period of 1 minute) reduces the response time for temperature and humidity control in key areas such as operating rooms to ≤5 seconds, significantly improving dynamic response capabilities. This invention uses a multivariate nonlinear regression model and converts it into a linear equation through variable substitution, solving the matrix ill-conditioning problem of traditional models under high-dimensional data. The regression coefficients are dynamically adjusted based on historical data (such as the dynamic correction weight of the temperature difference between the evaporator and condenser), effectively capturing the strong nonlinear correlation between variables and improving the COP value stability by more than 30%. The present invention also proposes a correlation formula between fan speed and cooling water flow rate , dynamically adjust the cooling water distribution path, reduce the impact of system inertia on heat exchange efficiency, and achieve optimal energy efficiency distribution; The present invention also proposes energy efficiency optimization and cost saving, dynamically adjusting the water supply temperature setting value of the chiller ( ), combined with the time-of-use electricity price strategy to optimize energy consumption (objective function ), the comprehensive energy saving rate can reach 15%~20%; The present invention also proposes the effect of fault diagnosis and system reliability enhancement, introducing the Bayesian conditional probability network ( ), real-time monitoring of hidden faults such as insufficient cooling water flow and evaporator scaling, with diagnostic accuracy increased to 95%, reducing the risk of temperature and humidity out of control; The present invention also proposes high-precision temperature and humidity control. Through regional priority control (operating room > ward > public area) and dynamic optimization of the duct resistance coefficient, it ensures that the temperature and humidity fluctuation range in key areas is ≤±0.5°C, meeting the stringent requirements of scenarios such as hospital operating rooms. The control method and control system of the present invention have wide adaptability in various scenarios, namely, they support places such as hospitals, data centers, and laboratories that are sensitive to temperature and humidity. They are particularly suitable for complex systems with multi-variable coupling (such as cooling towers, pipe networks, and terminal environment linkage). At the same time, their operation and maintenance are intelligent, namely, the mobile terminal and the web terminal can display the operating status in real time (such as actual energy efficiency). ) and alarm records, reducing manual inspection costs by more than 30%. In addition, it is also compatible and scalable, that is, it adopts Modbus-RTU protocol and standardized hardware (such as ultrasonic flowmeter, inverter), seamlessly connects to the existing central air-conditioning system, and supports subsequent algorithm iteration and function expansion.

[0026] Example 1: High-precision temperature and humidity control in hospital operating rooms: Scenario: The operating room of a tertiary hospital needs to maintain a temperature of 22±0.5℃, a humidity of 50±3%, and a response time of ≤5 seconds. Parameter collection: Chiller parameters: evaporator temperature , condenser temperature , compressor frequency , Chilled water flow ; Cooling tower parameters: cooling water inlet temperature , cooling water outlet temperature , fan speed , cooling water flow rate =1.25m / s ; Pipeline parameters: Pipeline length , External insulation parameters , Sealed insulation parameters at the inlet and outlet ; Environmental parameters: operating room temperature , humidity in the operating room , maximum limit value 、 and minimum limit 、 ; Auxiliary parameters: crowd density and logistics dynamic data ; Specific steps: Data collection and preprocessing: Data is collected through a ±0.1°C accuracy temperature sensor and an ultrasonic flow meter (accuracy ±1%), and transmitted to an edge gateway using the Modbus-RTU protocol (sampling period 1 minute). Cooling load forecast: The LSTM model (hidden layer with 128 neurons) uses the past hour’s data as input to predict the cooling load for the next 30 minutes. , error ≤ 3%; Dynamic adjustment of cooling efficiency model: Multiple regression coefficient correction: evaporator temperature difference weight , fan speed weight ; Model output , cop stability ±7%; Dynamic optimization control: Adjust the chiller water supply temperature: ; Optimize cooling water flow rate: , reducing water pump energy consumption by 12%; Priority execution: The electric regulating valve (accuracy ±2%) prioritizes the distribution of chilled water to the operating room, and the temperature stabilizes to 22.1°C and the humidity to 50.5% within 5 seconds.

[0027] Data indicators: COP stability ±7% (compared to traditional ±30%); temperature and humidity control: 、 ; Response time is 4.8s; Energy consumption is reduced: Chilled water pump energy consumption is reduced by 15%, and overall energy efficiency is improved by 18%.

[0028] Comparative Example (Traditional Method) Scenario: Same as the hospital operating room in Example 1, using a cloud-based static control strategy: The parameters are the same as in Example 1: step: Cooling load prediction: Linear regression model, without considering the density of people, the cooling load prediction value is 230 kW (error +18%); Control instruction: Fixed cooling water flow rate 1.2 m / s， Water supply temperature 7°C (without dynamic optimization); Implementation effect: COP fluctuation ±28%, operating room temperature 22±1.8℃, humidity 50±6%; The response time is ≥20 seconds, and the evaporator scaling fault missed detection rate is 40%.

[0029] Comparison table of Example 1 and Comparative Example:

[0030] Through edge real-time computing, dynamic model optimization and multi-parameter coupling control, this invention demonstrates significant advantages in medical scenarios, especially in terms of control accuracy, energy efficiency and reliability, far surpassing traditional methods.

[0031] The present invention can also be applied to other scenarios besides hospitals, such as Example 2: Example 2: Data center room cooling optimization: Scenario: An IDC room needs to maintain a temperature of 25±1°C and a humidity of 40±5%, and operate continuously throughout the year. Parameter collection: Chiller parameters: evaporator temperature , condenser temperature , compressor frequency , Chilled water flow ; Cooling tower parameters: cooling water inlet temperature , cooling water outlet temperature , fan speed , cooling water flow rate =1.3m / s ; Environmental parameters: Indoor temperature of the computer room , Indoor humidity of the computer room , outdoor temperature ; Auxiliary parameters: server load rate and heat generated by IT equipment ; Specific steps: Cooling load forecast: LSTM model combined with server load to dynamically predict cooling load , error ≤ 5%; Time-of-use electricity price strategy optimization: During peak electricity price period (12:00-18:00), the compressor frequency is reduced to 45Hz, and increased to 55Hz during off-peak period, saving 22% of electricity costs; Compensation for pipe insulation degradation: Pipeline running time t=5000 h , thermal insulation degradation coefficient , the weight is 0.4×0.99=0.396; Cooling tower coordinated control: The correlation formula between fan speed and cooling water flow rate is: , condenser efficiency increased by 10%; Fault diagnosis: The Bayesian network detected abnormal cooling water flow (fluctuation >10%) and diagnosed the evaporator with a 92% probability of scaling, triggering an automatic cleaning procedure. Data indicators: The annual power usage effectiveness (PUE) value dropped from 1.6 to 1.45. Fault response time: reduced from 2 hours to 10 minutes; Annual electricity cost savings: approximately RMB 180,000 (compared to traditional methods); Temperature and humidity control: indoor temperature of the computer room , Indoor humidity of the computer room .

[0032] Example 2 illustrates that the present invention has significant advantages not only in medical scenarios, but also in data centers.

[0033] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

Claims

1. A method for edge control of the Internet of Things based on a central air-conditioning chiller, characterized by: Including the following step: S1: Real-time collection of multi-dimensional operating parameters, including: Chiller parameters: evaporator temperature, condenser temperature, compressor frequency, chilled water flow; Cooling tower parameters: cooling water inlet temperature, cooling water outlet temperature, fan speed, cooling water flow rate; Pipeline parameters: pipeline length, external insulation parameters, inlet and outlet sealing insulation parameters; Environmental parameters: indoor and outdoor temperature and humidity and their maximum / minimum limits; Auxiliary parameters: crowd density and logistics dynamic data; S2: The above parameters are classified into hospital operating rooms, wards, and public areas and then input into the cooling load prediction model. The long short-term memory network (LSTM) is used to predict the future cooling load value of each area. ; S3: Based on cooling load prediction value , building a cooling efficiency model through IoT edge computing , the multivariate nonlinear regression method is used to fit the following relationship: ; And transform the nonlinear equation into a multiple linear regression equation by substitution method: ； in, are all the parameters in steps a to e, is the error term, is the regression coefficient, which is dynamically adjusted through historical data learning; S4: Based on the cooling efficiency model, dynamically optimize the operation strategies of the chillers and cooling towers, adjust the regional distribution paths and priorities of cooling water and chilled water, with temperature and humidity control in the operating room area having the highest priority; S5: Collect parameters and actual cooling load data at historical time points as training samples, and periodically iterate and update the cooling load prediction model and cooling efficiency model.

2. The IoT edge control method based on a central air-conditioning chiller according to claim 1 is characterized in that: The long short-term memory network (LSTM) is used, with the chiller parameters, cooling tower parameters, environmental parameters, and auxiliary parameters at past time points as input, and the actual cooling load data at future time points as output; The hidden state update and prediction of the long short-term memory network LSTM are achieved by the following formula: Forget Gate: ; Input Gate: ; Candidate memory cells: ; Memory unit update: ; Output Gate: ; Current hidden status: ; Cooling load forecast output: ; in, is the hidden state at the previous moment; Enter parameters for the current moment; 、 、 、 is the weight matrix; tanh is the hyperbolic tangent function; 、 、 、 is the bias term; is the Sigmoid function, is element-wise multiplication.

3. The IoT edge control method based on a central air-conditioning chiller according to claim 1 is characterized in that: The weight adjustment of each variable in the multivariate linear regression equation includes: The chiller parameter weights are dynamically corrected based on the temperature difference between the evaporator and condenser; The cooling tower parameter weights are optimized based on the correlation between fan speed and cooling water flow; The pipeline parameter weights are updated periodically along with the external insulation performance degradation coefficient.

4. The IoT edge control method based on a central air-conditioning chiller according to claim 1 is characterized in that: In step S4, the dynamic optimization of the operation strategy of the chiller and the cooling tower includes: Adjust the chiller water supply temperature setpoint based on the cooling load forecast , satisfying the following formula to improve energy efficiency: ； in, is the design reference temperature; is the temperature adjustment range; Design the cooling capacity for the chiller.

5. The method for edge control of the Internet of Things based on a central air-conditioning chiller according to claim 1, characterized in that: Also includes: Hidden faults of chillers, cooling towers, and pipelines are monitored in real time through edge gateways, and diagnosed using a Bayesian conditional probability network.

6. The IoT edge control method based on a central air-conditioning chiller according to claim 1, characterized in that: The cooling load prediction model supports the time-of-use electricity price strategy and optimizes the operating cost through the following formula: ； in, Time-of-use electricity price, is the energy consumption in period t, Equipment maintenance costs.

7. The IoT edge control method based on a central air-conditioning chiller according to claim 1 is characterized in that: The method also includes: displaying the cold station operation status, alarm records and energy-saving reports in real time through mobile terminals and web terminals, supporting remote control and fault handling by operation and maintenance personnel.

8. The IoT edge control method based on a central air-conditioning chiller according to claim 1 is characterized in that: The method is implemented by a control system, which includes: The data acquisition module is used to collect multi-dimensional operating parameters. The data acquisition module is implemented by the following devices: temperature sensors are installed on the evaporator, condenser, cooling tower inlet and outlet, and indoor and outdoor environments; Flow meters are installed in chilled water and cooling water pipes; Frequency converter, used to collect compressor frequency and fan speed; Infrared sensors and logistics RFID tags are used to obtain crowd density and logistics dynamic data; The edge computing module is deployed on the IoT edge gateway and performs the following functions: Run cooling load forecasting models; Construct a cooling efficiency model and dynamically update the regression coefficients through a multiple linear regression equation; Generate optimization strategies, including chiller supply water temperature settings, cooling water flow rates, and zone allocation priorities; Policy execution module, and dynamic control through the following devices: Electric regulating valve to adjust the flow of chilled water and cooling water; The frequency converter controls the compressor frequency and fan speed; Priority controller, which distributes cooling capacity by hospital operating rooms, wards, and public areas; The communication module uses the Modbus-RTU protocol to realize real-time data transmission connection between the data acquisition module, edge computing module and policy execution module.

Images