Integrated cold station intelligent control system based on multi-modal data fusion

Abstract

The invention discloses an integrated cold station intelligent control system based on multi-modal data fusion, and relates to the technical field of data fusion, an intelligent decision and optimization module adopts reinforcement learning and model prediction control based on fused data to realize dynamic optimization of start and stop of a water chilling unit, chilled water temperature setting and water pump frequency adjustment; the fault detection and adaptive optimization module evaluates the data fusion accuracy by calculating an environment entropy value and cold station dynamic response delay time, and dynamically adjusts a control strategy based on a prediction deviation rate and the data fusion accuracy; the closed-loop feedback module analyzes and adjusts control parameters through operation data, intelligent control models between different cold stations are optimized through transfer learning, intelligent deployment of a new cold station is accelerated, and the generalization ability of the models is improved. More accurate, more energy-saving and more stable intelligent control of the cold station is realized, and the operation efficiency and reliability of the cold station system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion, and particularly relates to an integrated chiller plant intelligent control system based on multi-modal data fusion. Background Art

[0002] The intelligent control of an integrated chiller plant based on multi-modal data fusion refers to, during the intelligent control process of the chiller plant system, using various data from different sources and of different types (i.e., multi-modal data), and performing optimization analysis through data fusion methods to improve the operating efficiency of the chiller plant, reduce energy consumption, and optimize equipment scheduling. For example, when the chiller plant is operating, it involves various data such as temperature, humidity, pressure, power load, and equipment status. The intelligent control system fuses these data to establish accurate energy consumption prediction and optimization scheduling models, thereby achieving intelligent regulation.

[0003] Taking the central air-conditioning chiller plant of a large commercial complex as an example, its intelligent control system fuses sensor data (such as temperature and humidity sensors, flow meters), historical operation data, weather forecasts, energy prices, etc., and performs real-time analysis through artificial intelligence algorithms. Some systems use deep learning or reinforcement learning models to optimize the on / off of chillers, the adjustment of cooling water pumps, and the setting of chilled water temperatures. For example, Alibaba's data center once used multi-modal data fusion and AI optimization technologies to reduce the power usage effectiveness (PUE) of the cooling system to below 1.3, effectively saving energy and reducing consumption.

[0004] The existing technologies have the following deficiencies: Different data modalities (such as images, temperature, pressure, energy consumption, etc.) often have different acquisition frequencies, units, noise characteristics, and spatio-temporal distributions. For example, weather data is usually updated periodically (such as once an hour), while sensor data may flow in at a high frequency in real time (such as multiple times per second). If the fusion method is not accurate enough, it will lead to data distortion. In addition, data distortion may cause the chiller to misjudge the working conditions, resulting in control decisions that are too early or too late, increasing energy consumption and even damaging equipment. Summary of the Invention

[0005] The purpose of the present invention is to provide an integrated chiller plant intelligent control system based on multi-modal data fusion to solve the deficiencies in the background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: An integrated chiller plant intelligent control system based on multi-modal data fusion, including a data acquisition module, a data preprocessing and fusion module, an intelligent decision-making and optimization module, a fault detection and adaptive optimization module, and a closed-loop feedback module; The data acquisition module is used to collect the operation data of the central air-conditioning chiller plant, external environment data, and historical operation data; The data preprocessing and fusion module performs time series alignment and outlier cleaning on different modality data, and extracts and fuses multi-modal features through a deep neural network; The intelligent decision-making and optimization module, based on the fused data, uses reinforcement learning and model predictive control to dynamically optimize the start / stop of the chiller, the setting of the chilled water temperature, and the adjustment of the pump frequency, improving the energy efficiency of the chilled water station; The fault detection and adaptive optimization module monitors the status of the chilled water station equipment, calculates the environmental entropy value and the dynamic response lag time of the chilled water station to evaluate the data fusion accuracy, and dynamically adjusts the control strategy based on the prediction deviation rate and the data fusion accuracy; The closed-loop feedback module adjusts the control parameters according to the operation data of the chilled water station, and optimizes the intelligent control model between different chilled water stations through transfer learning.

[0007] Preferably, the data acquisition module performs real-time monitoring on the core equipment of the central air-conditioning chilled water station through a distributed sensor network, and collects the temperature, pressure, flow rate, and energy consumption data of the chiller, cooling tower, and pump.

[0008] Preferably, the data preprocessing and fusion module performs time series alignment on different modality data, where: for high-frequency data, the sampling frequency is reduced by using a moving window average; for low-frequency data, missing data is filled by using spline interpolation or linear interpolation; for non-equidistant data, prediction and filling are performed by using Kalman filtering or long short-term memory network.

[0009] Preferably, the intelligent decision-making and optimization module optimizes the start / stop strategy of the chiller based on reinforcement learning and model predictive control, where: Q-learning or deep Q-network is used to learn the optimal unit combination under different load conditions; The setting of the chilled water temperature is optimized through MPC to maximize the refrigeration COP and meet the indoor temperature requirements; the pump frequency conversion is adjusted by combining PID control and reinforcement learning to optimize the water flow rate and pump power consumption, improving the energy efficiency of the chilled water station.

[0010] Preferably, the environmental entropy value and the dynamic response lag time of the chilled water station are calculated to evaluate the data fusion accuracy, specifically: The method for obtaining the environmental entropy value is: let the external environmental variables be X = {Q, S, W}, where: Q is the outdoor temperature, S is the solar radiation intensity, and W is the wind speed; The observed values at the past n moments are binned, and the occurrence probability of the data in each interval i is calculated : ; the information entropy H(X) is calculated for each variable X, and the expression is: ; where: m is the number of bins, and the weighted total entropy value is calculated by integrating the entropy values of each environmental variable, that is, the environmental entropy value EE is calculated, and the expression is: ; where: is the weight of the environmental variable.

[0011] Preferably, the method for obtaining the cold station dynamic response lag time is as follows: Set the chilled water temperature set value adjusted by the cold station control system or the pump frequency and then monitor the response of the chilled water temperature and the change in energy consumption ; Record the moment t0 when the control signal changes. Let the steady state moment ts be the time point when the system reaches the set target value, and define the error threshold ϵ: ; Calculate the system response curve , and the expression is: ; Calculate the cold station dynamic response lag time CDRL. Let the time when the error is less than the threshold for the first time be ts, then the CDRL is calculated as: .

[0012] Preferably, normalize the environmental entropy value and the cold station dynamic response lag time so that they are both within [0, 1]. After weighted averaging the normalized environmental entropy value and the cold station dynamic response lag time, obtain the data fusion accuracy index, and the expression is: ; where is the data fusion accuracy index, and α, β are weights; FUI > 0.7: The data fusion uncertainty is high, and it is necessary to adjust the data weights or optimize the control algorithm; FUI < 0.3: The data fusion is stable and can be directly used for intelligent control optimization decision-making.

[0013] Preferably, dynamically adjust the control strategy based on the prediction deviation rate and the data fusion accuracy: Calculate the prediction deviation rate PDR. Let the predicted cooling load at a certain moment t and the actual cooling load ; is the intelligent prediction load based on historical data and environmental variables; is the true cooling load measured by the cold station sensor, and the expression is: ; If PDR < 5%: The prediction is accurate and the data fusion is reliable; 5% ≤ PDR < 15%: There is an error in the prediction, and it is necessary to optimize the data fusion; PDR ≥ 15%: The prediction error is large, there is an abnormality in the data fusion or a deviation in the model parameters, and it is necessary to adjust the control strategy; Use the prediction deviation rate and the data fusion accuracy index as the input items of fuzzy logic; Define the fuzzy set and establish the fuzzy rules under different control conditions; Take the unit start-stop adjustment, the chilled water temperature set adjustment, and the pump frequency adjustment as the output items to dynamically optimize the control parameters.

[0014] In the above technical solution, the technical effects and advantages provided by the present invention are as follows: 1. Through the real-time monitoring of core equipment such as chillers, cooling towers, and pumps, and by combining external environmental data and historical operation data, the present invention realizes the precise control of the cold station system. Compared with the prior art, the present invention uses a deep neural network for multi-modal data fusion, optimizes the temporal alignment, outlier cleaning, and feature extraction of different data modalities, and avoids decision-making errors caused by problems such as inconsistent data acquisition frequencies and noise interference. At the same time, through the collaborative optimization of reinforcement learning and model predictive control for the chiller start-stop strategy, chilled water temperature setting, and pump frequency adjustment, the cold station system can maintain a high-energy-efficiency and low-energy-consumption operating state under different load conditions, improving the overall stability and intelligent level of the refrigeration system.

[0015] 2. The present invention conducts intelligent regulation by combining the prediction deviation rate. When the uncertainty of data fusion is relatively high, the system automatically adjusts the data weights and dynamically optimizes the control parameters based on fuzzy logic control to reduce unnecessary starts and stops of the chiller, optimize the chilled water temperature setting, and precisely adjust the pump frequency, thereby reducing the energy consumption fluctuation of the cold station system and improving the long-term operation efficiency. At the same time, transfer learning is used to optimize the intelligent control model between different cold stations, enabling new cold stations to quickly adapt to different load demands and environmental conditions, reducing the commissioning cost and accelerating the intelligent deployment. The present invention not only improves the adaptive ability of the cold station intelligent control but also realizes more precise and energy-saving operation management, ensuring the long-term stable and efficient operation of the cold station. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] For the embodiments, please refer to Figure 1As shown in the figure, an integrated chiller intelligent control system based on multi-modal data fusion in this embodiment includes a data acquisition module, a data preprocessing and fusion module, an intelligent decision-making and optimization module, a fault detection and adaptive optimization module, and a closed-loop feedback module; The data acquisition module is used to collect the operation data of the central air-conditioning chiller station, external environment data, and historical operation data; The data preprocessing and fusion module performs time series alignment and outlier cleaning on different modal data, and extracts and fuses multi-modal features through a deep neural network; The intelligent decision-making and optimization module, based on the fused data, uses reinforcement learning and model predictive control to dynamically optimize the start and stop of the chiller, the setting of the chilled water temperature, and the adjustment of the pump frequency to improve the energy efficiency of the chiller station; The fault detection and adaptive optimization module monitors the status of the chiller station equipment, calculates the environmental entropy value and the dynamic response lag time of the chiller station to evaluate the data fusion accuracy, and dynamically adjusts the control strategy based on the prediction deviation rate and the data fusion accuracy; The closed-loop feedback module adjusts the control parameters according to the operation data of the chiller station, and optimizes the intelligent control model between different chiller stations through transfer learning.

[0020] The data acquisition module is used to obtain the operation data, external environment data, and historical operation data of the central air-conditioning chiller station, providing basic data support for intelligent control. The following describes the specific implementation of the data acquisition module in detail.

[0021] The data acquisition module monitors the core equipment (chiller, cooling water pump, chilled water pump, cooling tower) of the central air-conditioning chiller station in real time through a distributed sensor network, and collects the following key data: Chiller data: evaporation temperature, condensation temperature, chilled water inlet and outlet temperature, condensation pressure, cooling water flow, unit power consumption (kW), compressor frequency, etc.; Cooling tower data: fan speed, cooling water temperature, ambient humidity, cooling water flow, wind speed, etc.; Pump data: the operating status, flow, head, power consumption, frequency converter frequency of the chilled water pump and the cooling water pump; Sensor types: temperature and humidity sensors, pressure sensors, ultrasonic flow meters, electric energy meters, smart meters, etc. The data is transmitted to the central control system through Modbus, BACnet, MQTT protocols and stored in a time series database (such as InfluxDB, TSDB) to support real-time analysis and historical backtracking.

[0022] To optimize the cold station control strategy, the data acquisition module also obtains external environmental data through the Internet of Things interface (IoT API) or a local weather station, including: Weather information: outdoor temperature, humidity, wind speed, solar radiation, precipitation, etc. The data source can be the API of the China Meteorological Administration or commercial weather services (such as OpenWeather, WeatherAPI); Building load: Obtain air-conditioning load data from the building automation system (BAS), including the current air-conditioning area load, occupancy rate, personnel density, and historical load curve; Energy price: Obtain real-time electricity prices and peak-valley electricity price strategies from the electricity market API to optimize the cold station energy consumption strategy.

[0023] To achieve intelligent control, the system stores and analyzes past operation data, including: Equipment start-stop records: Record the start and stop times and operation durations of chillers, pumps, and cooling towers; Historical energy consumption data: Analyze the energy consumption under different operating conditions and predict future energy consumption trends in combination with the AI model; Abnormal fault data: Store historical alarm information, such as compressor overload, condenser fouling, and flow anomalies, for the fault detection model to learn and optimize.

[0024] All data is preprocessed (such as data cleaning, time series alignment) through an edge computing gateway and stored in a big data platform (such as Hadoop, Kafka) to ensure data integrity and availability, providing support for subsequent data fusion, intelligent optimization, and fault prediction.

[0025] The data preprocessing and fusion module performs time series alignment and outlier cleaning on different modality data, and extracts and fuses multi-modal features through a deep neural network.

[0026] Since the sampling frequencies of different modality data (operation data, environmental data, historical data) involved in the central air-conditioning cold station are inconsistent. For example: Sensor data (such as flow, temperature, pressure) is sampled at the second level; Weather data (such as temperature, humidity, solar radiation) is updated hourly; Historical operation data may be stored at the day level or minute level. To ensure the time series consistency of data fusion, the present invention adopts a time interpolation and downsampling strategy: For high-frequency data (second level), use a sliding window average to reduce the sampling frequency and align it with low-frequency data; For low-frequency data (hourly), use spline interpolation or linear interpolation to fill in missing time points to form a unified time series; For non-equidistant data (such as data missing during abnormal equipment shutdown), use Kalman filtering or long short-term memory network (LSTM) prediction to fill in.

[0027] Cold station data may be affected by sensor drift, communication failures, or sudden environmental changes, resulting in outliers. The present invention adopts a triple outlier detection strategy to clean data anomalies: Based on statistical methods (3σ principle), detect extreme outliers of parameters such as temperature and flow rate, and perform mean filling; based on time series anomaly detection (such as ARIMA or LSTM-AE autoencoder), detect the changing trends of flow rate and temperature, identify short-term data mutations, and perform smoothing processing; based on physical models (such as thermodynamic equations), calculate the deviation between the sensor readings and the theoretical parameters of the equipment, and eliminate data points that do not conform to physical laws.

[0028] The present invention uses a deep neural network (DNN) to extract features from multi-modal data and fuse them to form a unified representation for input into an intelligent control decision-making system. The specific steps are as follows: Use a one-dimensional convolutional neural network (1D-CNN) to extract time series patterns, such as the changing trend of chilled water temperature; use a long short-term memory network (LSTM) to model long-term time dependencies and predict future energy consumption changes; use an autoencoder to reduce dimensionality, remove redundant information, and improve computational efficiency. Use a self-attention mechanism to enhance the interaction between different modal data and improve the model's adaptability to environmental changes; use a weighted fusion method to dynamically adjust weights according to the importance of data modalities and improve the accuracy of data fusion.

[0029] The intelligent decision-making and optimization module, based on the fused data, uses reinforcement learning and model predictive control to dynamically optimize the start / stop of the chiller, the setting of the chilled water temperature, and the adjustment of the pump frequency to improve the energy efficiency of the chilled water station.

[0030] The start / stop control of the chiller is optimized through reinforcement learning (RL), with the goal of minimizing the total energy consumption and maintaining the chilled water station load demand : ; where: is the energy consumption (kW) of the i-th chiller; is the control time step (h); N is the total number of chillers, and T is the total control time; is the current cooling load demand of the building (kW), satisfying the constraint: ; is the cooling capacity of the i-th unit, and the optimal unit combination {ON, OFF} is selected through a reinforcement learning algorithm to avoid unnecessary start / stop losses and improve energy efficiency.

[0031] Chilled water temperature setting affects the refrigeration COP (coefficient of performance): ; through the MPC optimization objective, maximize the COP while ensuring indoor comfort: ; where, is the chilled water outlet temperature (°C); is the chilled water temperature setting range, such as 6 - 10 °C; is the compressor input power (kW), varying with changes; is the refrigeration capacity of the unit (kW), affected by the chilled water temperature. The future load is predicted by MPC, and is adjusted in advance to reduce energy consumption.

[0032] The pump frequency affects the system water flow rate , and variable frequency control (VFD) is adopted to optimize energy consumption: ; where: is the pump power (kW); k is the pump characteristic coefficient, determined by the pump characteristic curve; is the system water flow rate (L / s), meeting the cooling capacity demand of the chilled water station. The optimization objective is: ; and the flow rate demand constraint is satisfied: ; where is determined by the chilled water station load demand.

[0033] The specific optimization control steps include: Step 1: Collect real-time chiller status, chilled water temperature, pump frequency, external environment data (temperature, humidity, etc.). Estimate the future cooling load and weather trend through the LSTM prediction model.

[0034] Step 2: Optimize the start / stop of the chiller by reinforcement learning: State space: ; Action space: The on / off strategy of the chiller A = {ON, OFF}; Reward function: ; where, is the start / stop loss weight of the unit. RL selects the optimal action through Q-learning or DQN to reduce the frequent start / stop of the unit and improve energy efficiency.

[0035] Step 3: Optimize the chilled water temperature by MPC: Predict the future 24h load curve, calculate the influence of different on the COP; Solve the nonlinear optimization problem and select the set value that maximizes the COP; Adjust the unit control parameters to ensure that the cooling capacity meets the demand.

[0036] Step 4: PID+ optimizes and adjusts the pump frequency: Calculate the instantaneous water flow rate demand of the chilled water station ; Adjust through PID control to make stable around ; Combine reinforcement learning for fine-tuning to reduce the pump power consumption.

[0037] The fault detection and adaptive optimization module monitors the status of the chilled water plant equipment, calculates the environmental entropy value and the chilled water plant dynamic response lag time to evaluate the data fusion accuracy, and dynamically adjusts the control strategy based on the prediction deviation rate and the data fusion accuracy.

[0038] The environmental entropy value is used to measure the complexity and change degree of the external environment, and has a great influence on the chilled water plant load demand. The EE calculation is based on external environmental variables (temperature, humidity, solar radiation, wind speed, etc.), and the uncertainty level is calculated through the information entropy theory.

[0039] Let the external environmental variables be X = {Q, S, W}, where: Q is the outdoor temperature (°C); S is the solar radiation intensity (W / m²), and W is the wind speed (m / s).

[0040] The observed values at the past n moments are binned (such as the histogram method), and the probability of data appearance in each interval i is calculated : ; Calculate the information entropy H(X) for each variable X, and the expression is: ; where: m is the number of bins (such as 10 intervals). Combine the entropy values of each environmental variable to calculate the weighted total entropy value, that is, calculate the environmental entropy value EE, and the expression is: ; where: is the weight of the environmental variable (such as temperature 0.4, humidity 0.3, solar radiation 0.2, wind speed 0.1), which can be determined through data training.

[0041] EE high (>0.8): The external environment changes violently, the chilled water plant load prediction is unstable, and the data fusion uncertainty is high; EE low (<0.3): The external environment is relatively stable, and the data fusion accuracy is relatively high.

[0042] The chilled water plant dynamic response lag time is used to quantify the time lag from the change of the control signal to the system reaching the steady state of the chilled water plant, and characterizes the adaptation degree of the data fusion to the physical characteristics of the chilled water plant.

[0043] Let the chilled water plant control system adjust the chilled water temperature set value or the pump frequency After that, monitor the response of the following parameters: the chilled water temperature response : After adjusting the set value, the actual chilled water outlet temperature changes with time; the energy consumption change : The response of the chiller power consumption to the control adjustment.

[0044] Record the control signal change time t0 and the system response data. Let the steady state time ts be the time point when the system reaches the set target value, and define the error threshold ϵ (such as 2%): ; Calculate the system response curve: ; Calculate the cold station dynamic response lag time CDRL. Let the time when the first error is less than the threshold be ts, then CDRL is calculated as follows: ; Among them, CDRL > 300s: The cold station responds sluggishly, and the accuracy of data fusion prediction decreases; CDRL < 100s: The cold station responds quickly to the control signal, and the reliability of data fusion is relatively high.

[0045] Normalize the environmental entropy value and the cold station dynamic response lag time so that they are both within [0, 1]. After weighted averaging the normalized environmental entropy value and the cold station dynamic response lag time, the data fusion accuracy index is obtained. The expression is as follows: ; Among them, is the data fusion accuracy index, α, β are weights (such as 0.6, 0.4), which can be optimized through historical data training. FUI > 0.7: The uncertainty of data fusion is high, and it is necessary to adjust the data weights or optimize the control algorithm; FUI < 0.3: The data fusion is stable and can be directly used for intelligent control optimization decisions.

[0046] Dynamically adjust the control strategy based on the prediction deviation rate and the data fusion accuracy, which specifically includes: Calculate the prediction deviation rate PDR. PDR is used to measure the deviation between the predicted data and the actual measured data to evaluate the prediction accuracy of the model. The higher the PDR, the lower the accuracy of data fusion, and the control strategy needs to be adjusted. Let the predicted cooling load at a certain moment t be and the actual cooling load be ; is the intelligent predicted load (kW) based on historical data and environmental variables; is the actual measured cooling load (kW) measured by the cold station sensor. The expression is as follows: ; Among them: The larger the PDR, the greater the prediction error and the lower the data fusion accuracy. PDR < 5%: The prediction is accurate and the data fusion is reliable; 5% ≤ PDR < 15%: There are errors in the prediction, and it is necessary to optimize the data fusion; PDR ≥ 15%: The prediction error is large, there are abnormalities in data fusion or deviations in model parameters, and it is necessary to adjust the control strategy.

[0047] Take the prediction deviation rate and the data fusion accuracy index as the input items of fuzzy logic, and take the control strategy as the output item of fuzzy logic; the output control strategy adjustment includes: starting and stopping adjustment of the chiller : Adjust the number of operating units; adjustment of the chilled water temperature setting : Optimize the chilled water temperature setting value; variable frequency adjustment of the water pump : Adjust the operating frequency of the water pump.

[0048] Define fuzzy sets for input items (PDR, FUI): PDR = {small (S), medium (M), large (L)}; FUI = {low (L), medium (M), high (H)}.

[0049] For output items Define fuzzy sets: = {decrease (-1), remain unchanged (0), increase (+1)}; = {decrease (-1°C), remain unchanged (0°C), increase (+1°C)}; = {decrease (-5Hz), remain unchanged (0Hz), increase (+5Hz)}.

[0050] When PDR is small (prediction error is small): If the FUI is low (data fusion accuracy is high), it means that the system is running stably, keeping the current unit operating status, chilled water temperature setting and water pump frequency unchanged.

[0051] If in FUI (there is a certain uncertainty in data fusion), the chilled water temperature setting is appropriately fine-tuned to optimize the system energy efficiency, but the unit operating status and water pump frequency remain basically unchanged.

[0052] If the FUI is high (data fusion is unreliable), although the prediction error is small, due to the large uncertainty in data fusion, it is still necessary to optimize the chilled water temperature setting and make small adjustments to the water pump frequency to enhance system stability.

[0053] When PDR is medium (moderate prediction error): If the FUI is low, it means that the data fusion is more accurate and the system is more adaptable to environmental changes. Therefore, the unit operating status and chilled water temperature are kept unchanged, and only a small adjustment is made to the water pump frequency to optimize the water supply system.

[0054] If FUI is in, it means that there is uncertainty in data fusion, which may affect the operation of the cold station. At this time, it is necessary to slightly optimize the start and stop strategy of the chiller, and adjust the chilled water temperature setting and water pump frequency to reduce energy consumption.

[0055] If the FUI is high, it means that the data fusion reliability is low and the prediction results may have large deviations. Therefore, it is necessary to adjust the unit operating status, optimize the chilled water temperature setting, and significantly adjust the water pump frequency to reduce system response lag and improve energy efficiency.

[0056] When the PDR is large (the prediction error is large): If the FUI is low, although the prediction error is large, the data fusion is reliable, so the unit start and stop status, chilled water temperature and pump frequency are slightly optimized to reduce the impact of the prediction error.

[0057] If the data fusion uncertainty in the FUI is high, the system may be affected by environmental changes. Therefore, it is necessary to adjust the operating state of the unit, optimize the chilled water temperature, and adjust the pump frequency to enhance the robustness of the system.

[0058] If the FUI is high, it indicates that both the prediction error and the data fusion error are large, and the system stability is poor. At this time, it is necessary to significantly adjust the operating state of the chiller, optimize the chilled water temperature setting, and significantly adjust the pump frequency to restore the operating stability of the chilled water station and reduce energy consumption.

[0059] The closed-loop feedback module of the present invention optimizes the system operation efficiency by continuously monitoring the operating data of the chilled water station and dynamically adjusting the control parameters. First, the system collects key data such as the energy consumption of the chiller, the deviation of the chilled water temperature, the deviation of the pump frequency, the environmental entropy value, and the prediction deviation rate (PDR) in real time, and evaluates the execution effect of the control strategy through data analysis. When the actual operating state deviates from the optimization target, the system adjusts the control parameters based on the feedback data. For example, it optimizes the start-stop strategy of the chiller to reduce unnecessary start-stop losses, adjusts the chilled water temperature setting to improve the COP (Coefficient of Performance), or optimizes the pump frequency control to reduce energy consumption while maintaining the stability of the cooling demand.

[0060] In addition, in order to improve the generalization ability of intelligent control, the present invention uses transfer learning technology to enable the optimization experience of one chilled water station to be quickly applied to different chilled water stations. Specifically, first train the reinforcement learning model in the source chilled water station to form an optimized control strategy network, then collect the operating data in the target chilled water station, and calculate the feature similarity between the two. If the similarity is high, transfer the model parameters of the source chilled water station and perform fine-tuning to adapt to the operating characteristics of the target chilled water station. Through the Policy Distillation method, the new chilled water station can quickly optimize the control strategy with less data, thereby reducing the model training cost and accelerating the intelligent deployment.

[0061] Finally, the closed-loop feedback module combines adaptive optimization and transfer learning, continuously adjusts the control parameters during the long-term operation process, and optimizes the intelligent control model according to the actual operation situation to make it more accurate and efficient. With the accumulation of data from more chilled water stations, the system can form an intelligent optimization network across chilled water stations, enabling new chilled water stations to have optimization capabilities at the initial stage, improving the overall energy-saving level of the chilled water station system, reducing the commissioning time, and at the same time enhancing the adaptability to different environmental and load changes.

[0062] In this embodiment, first, the data acquisition module obtains the operation data, external environment data, and historical operation data of the central air-conditioning cooling station, providing basic information for intelligent control. Subsequently, the data preprocessing and fusion module performs time series alignment and outlier cleaning on different modality data, and uses a deep neural network for multi-modal feature extraction and fusion to ensure data consistency and reliability. The intelligent decision-making and optimization module, based on the fused data, adopts reinforcement learning and model predictive control to dynamically optimize the start / stop of the chiller, the setting of the chilled water temperature, and the adjustment of the pump frequency to improve the system energy efficiency. At the same time, the fault detection and adaptive optimization module monitors the status of the cooling station equipment, calculates the environmental entropy value and the dynamic response lag time of the cooling station to evaluate the accuracy of data fusion, and dynamically adjusts the control strategy based on the prediction deviation rate and the data fusion accuracy to enhance the system stability. Finally, the closed-loop feedback module adjusts the control parameters by analyzing the operation data of the cooling station, and uses transfer learning to optimize the intelligent control model between different cooling stations, improving the generalization ability and long-term optimization effect of the system, and realizing efficient, stable, and intelligent operation management of the cooling station.

[0063] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0064] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0065] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. An integrated cold station intelligent control system based on multimodal data fusion, characterized in that: It includes data acquisition module, data preprocessing and fusion module, intelligent decision-making and optimization module, fault detection and adaptive optimization module and closed-loop feedback module; Data collection module, used to collect the operation data of the central air-conditioning cooling station, external environment data and historical operation data; The data preprocessing and fusion module performs time series alignment and outlier cleaning on data of different modalities, and extracts and fuses multimodal features through deep neural networks; Intelligent decision-making and optimization module, based on fusion data, uses reinforcement learning and model predictive control to dynamically optimize the start and stop of chillers, chilled water temperature setting and pump frequency adjustment to improve the energy efficiency of the cooling station; The fault detection and adaptive optimization module monitors the status of cold station equipment, calculates the environmental entropy value and the dynamic response lag time of the cold station to evaluate the accuracy of data fusion, and dynamically adjusts the control strategy based on the predicted deviation rate and data fusion accuracy; The closed-loop feedback module adjusts control parameters according to the operation data of the cooling station and optimizes the intelligent control model between different cooling stations through transfer learning.

2. The integrated cold station intelligent control system based on multimodal data fusion according to claim 1 is characterized by: The data acquisition module monitors the core equipment of the central air-conditioning cold station in real time through a distributed sensor network, and collects temperature, pressure, flow and energy consumption data of the chiller, cooling tower and water pump.

3. The integrated cold station intelligent control system based on multimodal data fusion according to claim 2 is characterized in that: The data preprocessing and fusion module performs time series alignment on data of different modalities. For high-frequency data, the sliding window average is used to reduce the sampling frequency; for low-frequency data, spline interpolation or linear interpolation is used to fill in missing data; for non-equally spaced data, Kalman filtering or long short-term memory network is used for prediction and completion.

4. The integrated cold station intelligent control system based on multimodal data fusion according to claim 3 is characterized in that: The intelligent decision-making and optimization module optimizes the start-stop strategy of the chiller based on reinforcement learning and model predictive control, including: Use Q-learning or deep Q network to learn the optimal unit commitment under different load conditions; MPC is used to optimize the chilled water temperature setting to maximize the cooling COP and meet the indoor temperature requirements. PID control and reinforcement learning are used to adjust the water pump frequency to optimize the water flow and pump power consumption, thereby improving the energy efficiency of the cold station.

5. The integrated cold station intelligent control system based on multimodal data fusion according to claim 4 is characterized in that: The environmental entropy value and the dynamic response lag time of the cold station are calculated to evaluate the accuracy of data fusion, specifically: The method for obtaining the environmental entropy value is as follows: let the external environmental variables be X={Q,S,W}, where: Q is the outdoor temperature, S is the solar radiation intensity, and W is the wind speed; Bin the observations of the past n moments and calculate the probability of data occurrence in each interval i : ; Calculate the information entropy H(X) for each variable X, the expression is: ; Where: m is the number of bins, the entropy values ​​of each environmental variable are integrated to calculate the weighted total entropy value, that is, the environmental entropy value EE is calculated, and the expression is: ;in: is the weight of the environment variable.

6. The integrated cold station intelligent control system based on multimodal data fusion according to claim 5 is characterized in that: The method for obtaining the dynamic response lag time of the cooling station is as follows: the cooling station control system adjusts the chilled water temperature setting value Or pump frequency After that, monitor the chilled water temperature response and energy consumption changes ; Record the time t0 when the control signal changes, set the steady-state time ts as the time point when the system reaches the set target value, and define the error threshold ϵ: ; Calculate the system response curve , the expression is: ; Calculate the dynamic response delay time CDRL of the cold station. Assume that the time when the first error is less than the threshold is ts, then the CDRL is calculated as: .

7. The integrated cold station intelligent control system based on multimodal data fusion according to claim 6 is characterized by: The environmental entropy value and the dynamic response lag time of the cold station are normalized so that they are both between [0,1]. The normalized environmental entropy value and the dynamic response lag time of the cold station are calculated by weighted average to obtain the data fusion accuracy index, which is expressed as: ;in, is the data fusion accuracy index, α and β are weights; FUI > 0.7: Data fusion uncertainty is high, and it is necessary to adjust the data weight or optimize the control algorithm; FUI < 0.3: Data fusion is stable and can be directly used for intelligent control optimization decision-making.

8. The integrated cold station intelligent control system based on multimodal data fusion according to claim 7 is characterized in that: Dynamically adjust the control strategy based on the prediction deviation rate and data fusion accuracy: Calculate the prediction deviation rate PDR, assuming the predicted cooling load at a certain time t and actual cooling load ; Intelligent load forecasting based on historical data and environmental variables; is the actual cooling load measured by the cooling station sensor, and its expression is: ; If PDR < 5%: the prediction is accurate and data fusion is reliable; 5% ≤ PDR < 15%: there is an error in the prediction and data fusion needs to be optimized; PDR ≥ 15%: the prediction error is large, there is data fusion anomaly or model parameter deviation, and the control strategy needs to be adjusted; The prediction deviation rate and data fusion accuracy index are used as input items of fuzzy logic; Define fuzzy sets and establish fuzzy rules under different control conditions; The control parameters are dynamically optimized with unit start and stop adjustment, chilled water temperature setting adjustment, and water pump frequency adjustment as output items.

Images