Air conditioner and fresh air control method and system based on enthalpy value
By constructing an enthalpy-based air conditioning and fresh air control system, the operation strategy of the computer room air conditioning and fresh air system is dynamically adjusted, solving the problems of response lag and energy waste in traditional control schemes. This achieves efficient and stable computer room environment control and improves the energy efficiency and stability of equipment operation.
Patent Information
- Application Number
- CN202511306437.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-12
- Publication Date
- 2025-11-18
AI Technical Summary
Traditional air conditioning and fresh air system control solutions for computer rooms cannot predict load changes in advance, resulting in problems such as delayed response, energy waste, cooling and heating offsetting, and poor equipment coordination, making it difficult to meet the high-efficiency operation requirements of computer rooms with high heat density.
An enthalpy-based air conditioning and fresh air control method acquires equipment, environmental, and historical data to build a predictive model, generate forward-looking data and execution instructions, and achieves closed-loop optimization of perception-prediction-execution, dynamically adjusting the operation strategy of the air conditioning and fresh air system.
Increase the utilization rate of natural cold sources to over 85%, reduce PUE from 2.1 to 1.65, achieve an average daily energy saving rate of ≥15%, reduce the temperature gradient from ±3℃ to ±1℃, reduce the hot spot occurrence rate from 3 times per day to 0, improve equipment operation stability by 40%, and reduce operation and maintenance costs by 30%.
Smart Images

Figure CN120970005A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of computer room environmental control technology, specifically relating to a control method and system for air conditioning and fresh air based on enthalpy value. Background Technology
[0002] In the control of data center air conditioning and fresh air systems, traditional solutions mostly rely on static thresholds or simple feedback adjustment methods. This decision-making logic cannot predict future load changes in advance, leading to a series of problems.
[0003] First, there's the issue of response lag. Traditional solutions adjust equipment operation solely based on real-time data. If there's a sudden change in data center load, such as a dramatic increase in server cluster computing power, or a drastic change in the outdoor environment, like a sudden temperature rise, the system can take more than 30 minutes to return to a stable state. During this process, localized hotspots can easily appear, potentially leading to energy waste.
[0004] Secondly, there's the issue of model rigidity. Traditional solutions use fixed enthalpy difference thresholds or air supply parameters, failing to consider dynamic factors such as equipment heating patterns and seasonal changes. It's like adjusting the fresh air volume during the hottest summer months using strategies for spring and autumn; this would drastically increase the air conditioning load.
[0005] Another issue is poor coordination. The air conditioning and fresh air systems make decisions independently. For example, when the fresh air system introduces low-temperature fresh air, the air conditioning does not reduce its load in time, which leads to a situation of "cold and heat canceling each other out", ultimately resulting in an energy efficiency loss of 15%-20%.
[0006] While existing technologies have attempted to introduce simple algorithms like PID control, a prediction-based decision-making system has not been established, thus failing to achieve closed-loop optimization of "perception-prediction-execution." Consequently, unresolved issues remain, such as missing equipment heat dissipation data, coarse air supply regulation, limited switching conditions, and passive control strategies, making it difficult to meet the high-efficiency operation requirements of high-heat-density data centers. Summary of the Invention
[0007] The present invention aims to at least solve one of the technical problems existing in the prior art. To this end, the present invention proposes a control method and system for air conditioning and fresh air based on enthalpy value.
[0008] In a first aspect, this invention provides a method for controlling air conditioning and fresh air based on enthalpy. The method includes: acquiring equipment layer data, environmental layer data, and historical data from inside and outside a computer room; obtaining forward-looking data based on a constructed prediction model, the equipment layer data, environmental layer data, and historical data, wherein the forward-looking data includes predicted cooling load, predicted outdoor enthalpy, and load sensitivity coefficient correction parameters; generating benchmark judgment data based on the forward-looking data and collected first real-time data, wherein the first real-time data includes real-time equipment power, real-time equipment heat generation, real-time indoor and outdoor temperature, and real-time indoor and outdoor humidity in the computer room, and the benchmark judgment data includes corrected enthalpy difference and dynamic threshold; constructing mode instructions using the benchmark judgment data and collected second real-time data, wherein the second real-time data includes pollutant data, weather data, and equipment status data of the atmosphere outside the computer room; generating execution instructions based on the mode instructions and collected third real-time data, wherein the third real-time data includes temperature gradient, wind speed, and wind pressure, and the execution instructions include air conditioning adjustment instructions and fresh air adjustment instructions; acquiring real-time feedback data, and adjusting the parameters of the prediction model based on the real-time feedback data.
[0009] In some embodiments of the present invention, the device layer data includes the cabinet's real-time power and the device's runtime.
[0010] In some embodiments of the present invention, the environmental layer data includes indoor temperature, indoor humidity, outdoor temperature, outdoor humidity, and rainfall.
[0011] In some embodiments of the present invention, the historical data includes device-level data and environmental-level data from the past two years, which are stored in a time-series database.
[0012] In some embodiments of the present invention, the steps of constructing the prediction model include: constructing multiple samples based on the sliding window method and the historical data, each sample including input features at the first 24 time steps, a label at the 25th time step, and simulation error; constructing a training set and a validation set from the multiple samples, training the prediction model using the training set, and validating it using the validation set; and constructing a test set from the multiple samples, and testing the validated model using the test set.
[0013] In some embodiments of the present invention, a mode instruction is constructed using the benchmark judgment data and the collected second real-time data. The second real-time data includes pollutant data, weather data, and equipment status data of the atmosphere outside the computer room, specifically including:
[0014] When the corrected enthalpy difference is less than the negative dynamic threshold and the PM2.5 concentration is less than or equal to the preset pollution threshold, the first mode command and first mode parameters are generated.
[0015] When the corrected enthalpy difference is less than the negative dynamic threshold, the PM2.5 concentration is greater than the preset pollution threshold, and the real-time rainfall is less than or equal to the preset rainfall threshold, the second mode command and second mode parameters are generated.
[0016] When the real-time rainfall exceeds the preset rainfall threshold, a third-mode command and third-mode parameters are generated.
[0017] When the correction enthalpy difference is greater than or equal to a negative dynamic threshold, a fourth mode instruction and fourth mode parameters are generated.
[0018] In some embodiments of the present invention, the first mode parameters include a first fresh air load ratio, a first air conditioning load reduction ratio, and a first protection level.
[0019] In some embodiments of the present invention, the second mode parameters include a second fresh air load ratio, a second air conditioning load reduction ratio, and a second protection level.
[0020] In some embodiments of the present invention, the third mode parameters include a third fresh air load ratio, a third air conditioning load reduction ratio, and a third protection level.
[0021] In some embodiments of the present invention, the fourth mode parameters include a fourth fresh air load ratio, a fourth air conditioning load reduction ratio, and a fourth protection level.
[0022] In some embodiments of the present invention, the first fresh air load ratio is 0.7, the first air conditioning load reduction ratio is 0.3, the first protection level is primary, and the resistance is ≤50Pa.
[0023] In some embodiments of the present invention, the second fresh air load ratio is 0.5; the second air conditioning load reduction ratio is 0.5; the second protection level is high efficiency, with resistance ≤200Pa and filtration efficiency ≥99.97%.
[0024] In some embodiments of the present invention, the third fresh air load ratio is 0.1 (maintaining only the minimum air exchange rate of 2 times / h); the third air conditioning load reduction ratio is 0.9 (operating at 90% full load); and the third air valve has an IP54 protection rating and is splash-proof.
[0025] In some embodiments of the present invention, the fourth fresh air load ratio is 0.2, the fourth air conditioning load reduction ratio is 0.8, and the fourth protection level is primary.
[0026] In some embodiments of the present invention, a first mode instruction is converted into a first execution instruction based on a first mode parameter and a third real-time data; a second mode instruction is converted into a second execution instruction based on a second mode parameter and a third real-time data; a third mode instruction is converted into a third execution instruction based on a third mode parameter and a third real-time data; and a fourth mode instruction is converted into a fourth execution instruction based on a fourth mode parameter and a third real-time data.
[0027] In some embodiments of the present invention, based on a first fresh air load ratio, a reference wind speed in the computer room, and a temperature gradient, the first mode command is transformed into a first execution command that includes controlling the air supply speed. Based on a first protection level, the first mode command is transformed into a first execution command that includes the opening degree of the air valve. Based on a first air conditioning load reduction ratio, the first mode command is transformed into a first execution command that includes controlling the frequency of the air conditioning compressor.
[0028] In some embodiments of the present invention, based on the second fresh air load ratio, the reference wind speed of the computer room air, the temperature gradient, and the actual resistance of the filter, the second mode command is transformed into a second execution command that includes controlling the air supply speed. Based on the actual resistance of the filter, the second mode command is transformed into a second execution command that includes controlling the opening degree of the air valve; based on the second air conditioning load reduction ratio, the second mode command is transformed into a second execution command that includes controlling the frequency of the air conditioning compressor.
[0029] In some embodiments of the present invention, based on the third fresh air load ratio, the reference wind speed of the computer room air, and the indoor-outdoor pressure difference, the third mode command is transformed into a third execution command that includes controlling the air supply speed. Based on rainfall, the third mode command is transformed into a third execution command that includes controlling the opening degree of the air valve. Based on the third air conditioning load reduction ratio, the third mode command is transformed into a third execution command that includes controlling the frequency of the air conditioning compressor.
[0030] In some embodiments of the present invention, based on the fourth fresh air load ratio and the reference wind speed of the computer room air, the fourth mode command is transformed into a fourth execution command that includes controlling the air supply speed. Based on the protection level, the fourth mode command is transformed into a fourth execution command that includes controlling the opening degree of the air valve. Based on the fourth air conditioning load reduction ratio, the fourth mode command is transformed into a fourth execution command that includes controlling the frequency of the air conditioning compressor.
[0031] In some embodiments of the present invention, obtaining real-time feedback data and adjusting the parameters of the prediction model based on the real-time feedback data specifically includes:
[0032] Calculate the prediction deviation based on the actual and predicted cooling loads; adjust the hidden layer weights of the prediction model based on the prediction deviation.
[0033] The correction deviation is calculated based on the actual enthalpy difference and the corrected enthalpy difference, and the correction coefficient is recalibrated based on the correction deviation.
[0034] The execution deviation is calculated based on the actual temperature gradient obtained, and the drag coefficient of the flow field model is corrected based on the execution deviation.
[0035] In a second aspect, the present invention provides a control system for computer room air conditioning and fresh air based on enthalpy. The system includes a sensing layer, a decision-making layer, and an execution layer.
[0036] This system includes a temperature sensor, humidity sensor, power sensor, PM2.5 sensor, rain gauge, wind speed sensor, wind pressure sensor, and data acquisition module. The temperature sensor, humidity sensor, power sensor, PM2.5 sensor, rain gauge, wind speed sensor, and wind pressure sensor are all connected to the data acquisition module.
[0037] The decision-making layer includes an edge controller, an industrial server, and a time-series database server. The edge controller is connected to the industrial server via gigabit Ethernet, and the database server is connected to the industrial server via fiber optic cable.
[0038] This decision-making layer includes an edge controller with PLC functionality, an industrial server for deploying LSTM models, and a time-series database server. The edge controller and the industrial server are connected via gigabit Ethernet.
[0039] The execution layer includes an air conditioning inverter driver, an air conditioning compressor, a fresh air inverter driver, a damper actuator, a damper, a coarse filter, a medium-efficiency filter, and a high-efficiency filter. The air conditioning inverter driver is directly connected to the air conditioning compressor via a data line, and the fresh air inverter driver is directly connected to the damper actuator via a data line. The damper actuator is divided into 3 groups, which are connected to the dampers at the inlet of the coarse filter, medium-efficiency filter, and high-efficiency filter respectively via mechanical linkage or control lines to achieve individual control.
[0040] The data acquisition module is connected to the edge controller. The air conditioning inverter driver, fresh air inverter driver, and air valve actuator are all connected to the relay output module of the edge controller through control cables.
[0041] The system acquires equipment layer data, environmental layer data, and historical data from inside and outside the computer room. Based on the constructed prediction model, the equipment layer data, environmental layer data, and historical data, it obtains forward-looking data. The forward-looking data includes steps such as predicting cooling load, predicting outdoor enthalpy, and correcting load sensitivity coefficient parameters, which are implemented at the decision layer.
[0042] Based on the prospective data and the collected first real-time data, a benchmark judgment data is generated. The first real-time data includes the real-time equipment power, real-time equipment heat generation, real-time indoor and outdoor temperature, and real-time indoor and outdoor humidity in the computer room. The benchmark judgment data includes steps for correcting enthalpy difference and dynamic threshold, which are implemented in the PLC core module of the edge controller.
[0043] The step of constructing a mode instruction using the benchmark judgment data and the collected second real-time data, where the second real-time data includes pollutant data, weather data, and equipment status data of the atmosphere outside the computer room, is executed in the logic control module of the edge controller.
[0044] Execution instructions are generated based on the mode instructions and the collected third real-time data, including temperature gradient, wind speed and wind pressure. The execution instructions, including air conditioning adjustment instructions and fresh air adjustment instructions, are executed in the execution control module of the edge controller.
[0045] The system acquires real-time feedback data and adjusts the parameters of the prediction model based on this data, running the system on an edge controller and an industrial server.
[0046] Compared to traditional solutions, this invention improves the utilization rate of natural cooling sources to over 85% through dynamic enthalpy difference correction, reduces PUE from 2.1 to 1.65, and achieves an average daily energy saving rate of ≥15%. Through precise air supply control, this invention reduces the temperature gradient from ±3℃ to ±1℃, lowers the hotspot occurrence rate from 3 times per day to 0, and improves equipment operational stability by 40%. Furthermore, through a multi-dimensional switching mechanism, this invention reduces false alarms during heavy rain by 75%, reduces PM2.5 intrusion during highly polluted weather by 80%, and lowers maintenance costs by 30%. Attached Figure Description
[0047] Figure 1 These are flowcharts illustrating steps 100-500 provided in some embodiments.
[0048] Figure 2 This is a flowchart illustrating step 100 provided in some embodiments.
[0049] Figure 3 This is a flowchart illustrating step 300 provided in some embodiments.
[0050] Figure 4 This is a flowchart illustrating step 400 provided in some embodiments.
[0051] Figure 5 This is a flowchart illustrating step 500 provided in some embodiments.
[0052] Figure 6 These are control system architecture diagrams for data center air conditioning and fresh air systems based on enthalpy values, provided in some embodiments. Detailed Implementation
[0053] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0054] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.
[0055] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0056] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0057] It should also be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such uses of these terms can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described.
[0058] Figure 1 A schematic flowchart of a control method for computer room air conditioning and fresh air based on enthalpy value is shown in one embodiment, which may include the following steps 100 to 500.
[0059] In step 100 provided in some embodiments, equipment layer data, environmental layer data and historical data inside and outside the computer room are obtained, and prospective data is obtained based on the constructed prediction model, the equipment layer data, environmental layer data and historical data. The prospective data includes predicted cooling load, predicted outdoor enthalpy value and load sensitivity coefficient correction parameters.
[0060] The prediction model is based on long-term and short-term time series models and includes a multi-parameter perception layer, a time series feature extraction layer, and a prediction decision layer.
[0061] The multi-parameter sensing layer collects and standardizes data on equipment, environment, and time-series correlation in the computer room, providing a unified input for subsequent feature extraction. Input data includes equipment status parameters (e.g., real-time cabinet power, equipment load rate); environmental status parameters (e.g., indoor and outdoor temperature and humidity, outdoor enthalpy, PM2.5 concentration); and time-series correlation parameters (e.g., average load of the previous hour, load deviation for the same period within the week). The time-series feature extraction layer includes short-term, medium-term, and long-term feature layers. The short-term feature layer extracts short-term equipment load fluctuation characteristics, the medium-term feature layer extracts medium-term equipment load cycle characteristics, and the long-term feature layer extracts long-term trend characteristics. The prediction layer maps the extracted features to predictions for the next 6 hours, providing target-oriented output as look-ahead data for subsequent control.
[0062] Specifically, the multi-parameter perception layer collects 15-dimensional feature vectors covering three key parameter categories: equipment status parameters (5-dimensional): real-time rack power (P1-P3, taking 3 representative racks), cumulative equipment runtime (tsum), and equipment load rate (Preal / Prated, where Prated is the rated power); environmental status parameters (7-dimensional): indoor temperature (Tin), indoor humidity (Hin), outdoor temperature (Tout), outdoor humidity (Hout), outdoor enthalpy (hout), rainfall (R), and PM2.5 concentration (Cpm); and time-series correlation parameters (3-dimensional): average load of the previous hour (Qavg, 1h), cumulative load of the day (Qsum, day), and load deviation of the same period within the week (Qdev, week). The input layer normalizes the parameters to the [0,1] interval (xnorm=(x−xmin) / (xmax−xmin)) to avoid dimensional differences affecting model learning.
[0063] Specifically, the time-series feature extraction layer employs a 3-layer LSTM network. The short-term feature layer consists of 128 neurons, extracting load fluctuation features over the past 1-3 hours (such as sudden increases in computing power), with the activation function being Tanh (for handling gradient stability of time-series data). The medium-term feature layer consists of 64 neurons, extracting load cycle features within one day (such as weekday / weekend patterns), with the activation function also being Tanh. The long-term feature layer consists of 32 neurons, extracting trend features at the weekly / monthly scale (such as the impact of seasonal changes), with the activation function also being Tanh. Each layer includes a dropout layer (dropout=0.2) to prevent overfitting.
[0064] Specifically, the prediction decision layer maps the extracted features to predictions for the next 6 hours, providing target guidance for subsequent control. The prediction results are fed into the dynamic enthalpy difference correction model as "look-through data," driving subsequent decisions. The prediction results are as follows: Predicted cooling load Qpred(t) (t=1−6h): using a linear activation function (because the load is a continuous value), the output range is [0,Qmax] (Qmax is the historical maximum load), reflecting the heat that needs to be removed from the server room in the future; Predicted outdoor enthalpy hout,pred(t): using a linear activation function, the output range is [0,hmax] (hmax is the historical maximum enthalpy), predicting the thermal and humidity state of the outdoor air to guide fresh air utilization. The output layer and the hidden layer are connected by a weight matrix, with the initial weight values using a He normal distribution to adapt to the gradient characteristics of the Tanh activation function. Load sensitivity coefficient correction command (α'): dynamically adjusted according to the peak value of Qpred(t) (e.g., α'=0.3 when predicting high load, α'=0.1 when predicting low load), used to enhance the sensitivity of subsequent enthalpy difference correction to load.
[0065] In some embodiments, step 100 includes equipment layer data such as real-time cabinet power and equipment runtime. Real-time cabinet power (P1−Pn, in kW) is acquired by a current / voltage sensor at the cabinet's inlet (sampling frequency 1Hz, accuracy ±0.5%). Equipment runtime (t1−tn, in hours) is recorded by a PLC timer.
[0066] In some embodiments, step 100 includes environmental layer data such as indoor temperature, indoor humidity, outdoor temperature, outdoor humidity, and rainfall. Indoor temperature (Tin, in °C) and humidity (Hin, in %) are collected using distributed cold / hot aisle sensors (one every 5 meters, accuracy ±0.3 °C). Outdoor temperature and humidity (Tout, Hout) and rainfall (R, in mm / h) are collected using outdoor weather stations.
[0067] In some embodiments, step 100 includes historical data, which includes device-level data and environment-level data from the past two years, and is stored in a time-series database (InfluxDB).
[0068] like Figure 2 Some embodiments provide step 100, which includes the step of constructing a prediction model. Specifically, this includes:
[0069] Obtain historical data from the computer room (step 101); construct multiple samples based on the sliding window method and the historical data, each sample including input features from the first 24 time steps, a label from the 25th time step, and simulation error (step 102); construct a training set and a validation set from the multiple samples, train the prediction model using the training set, and validate it using the validation set (step 103); and construct a test set from the multiple samples, and test the validated model using the test set (step 104).
[0070] A specific set of steps for building a predictive model includes:
[0071] Steps 101 and 102:
[0072] Time-series data (1 record every 15 minutes, totaling 35,040 records) from 2021 to 2022 were collected from a data center. This data covers input features such as equipment power and environmental parameters, as well as the actual cooling load and outdoor enthalpy at the corresponding time points (used as labels). A sliding window method was used (window size = 24, i.e., 1 day of data). Each sample contains "input features from the previous 24 time points + the label from the 25th time point," generating a total of 34,800 samples. ±5% Gaussian noise was added to simulate sensor error, improving model robustness.
[0073] Step 103:
[0074] 24,360 of these samples were used to form the training set, and the remaining samples were used to form the validation set.
[0075] The model was trained using the training set. The Adam optimizer was used (learning rate = 0.001, β1 = 0.9, β2 = 0.999). The loss function was...
[0076] Loss = w1MSE(Qpred,Qreal) + w2MSE(hout,pred;hout,real); where w1 = 0.7 (cooling load prediction has a higher weight), w2 = 0.3.
[0077] Training epochs: Initially set to 300 epochs, with a batch size of 64 per epoch. Early stopping (patience=10) is used to prevent overfitting. Every 5 epochs, the training set loss is calculated and the weights are updated via backpropagation. Training is stopped and the current model parameters are saved when the validation set loss shows no decrease for 10 consecutive epochs.
[0078] The validation set includes samples from different seasons (high load in summer, low load in winter), weather conditions (sunny days, rainy days), and equipment statuses (normal operation, partial shutdown) to ensure the comprehensiveness of the validation scenarios. The mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R^2) are used to evaluate the model's generalization ability.
[0079] Initial validation results: Cooling load prediction MAE=5.2kW, RMSE=6.8kW, R^2=0.89; Outdoor enthalpy prediction MAE=2.1kJ / kg, RMSE=2.8kJ / kg, R^2=0.91.
[0080] Optimization and adjustment: To address the issue of large deviation in winter load forecast (MAE=7.5kW), the winter sample weight was increased (from 1.0 to 1.2), and after retraining, the winter MAE decreased to 4.8kW.
[0081] Step 104:
[0082] A new dataset from January to June 2023 (17,520 records, independent of the training / validation sets) was used as the test set, covering extreme scenarios (such as high temperatures in June and a cold snap in March). Device and environmental parameters of the test set were input in real time, and the model output Qpred and hout,pred for the next 6 hours. The predicted values were compared with the actual values, and test metrics (consistent with validation metrics) were calculated. The causes of deviations were analyzed, such as load fluctuations caused by sudden equipment expansion (Qreal suddenly increasing by 20%) and enthalpy prediction deviations caused by extreme rainstorms.
[0083] Test results: Predicted cooling load MAE = 5.8kW (error rate 7.2%), RMSE = 7.5kW, R^2 = 0.88; Predicted outdoor enthalpy MAE = 2.3kJ / kg (error rate 5.1%), RMSE = 3.1kJ / kg, R^2 = 0.90, meeting the preset error rate requirement of ≤8%.
[0084] Model solidification: The tested model parameters (weights, biases) are written into the decision-making layer controller as the benchmark model for subsequent "forward-looking data" generation.
[0085] The logical structure, training, validation and testing of the prediction model draw on the ideas of "dynamic state perception-closed-loop update" and "multi-parameter collaborative decision-making-error feedback optimization". Through time-series feature extraction, multi-scenario verification and dynamic iteration, the output "forward data" (cold load, outdoor enthalpy prediction) is ensured to have high accuracy and strong robustness, providing a reliable forward basis for subsequent enthalpy difference correction and mode switching.
[0086] In step 100 of the embodiment, a future state model of the load and environment is constructed by mining the heating patterns of the equipment and the changing trends of the outdoor environment. This step provides an "expected target" for subsequent enthalpy difference correction and mode switching, avoiding energy waste or temperature fluctuations caused by lag in traditional control. Through this step, the optimization direction can be locked in 6 hours in advance, reducing the system's response time to load changes from 30 minutes to 5 minutes, and the load prediction error is ≤8% (better than the 15%-20% of traditional methods).
[0087] In a specific step 100, a data center server room starts prediction at 8:00 on a summer weekday.
[0088] Obtain the rack power (average 70kW, peak 90kW), outdoor temperature and humidity (average 32℃ / 60%), and indoor temperature (24℃) for the past two years from 8:00 to 14:00. The prediction model extracts features such as "load increases by 10% in the hour after 8:00" and "peak at 12:00," combined with the daily weather forecast (sunny, 34℃). Output forward-looking data: Qpred(9:00) = 77kW, Qpred(12:00) = 90kW; hout,pred(12:00) = 85kJ / kg; α' = 0.3 (due to predicted high load).
[0089] In step 200 provided in some embodiments, benchmark judgment data is generated based on the prospective data and the collected first real-time data. The first real-time data includes the real-time equipment power, real-time equipment heat generation, real-time indoor and outdoor temperature, and real-time indoor and outdoor humidity in the computer room. The benchmark judgment data includes the corrected enthalpy difference and dynamic threshold.
[0090] This step incorporates the real-time power and heat generation of the equipment into the enthalpy difference calculation, eliminating the bias of traditional enthalpy difference calculations that only reflect air conditions and ignore equipment heat generation. This step provides an accurate benchmark for prioritizing the coordination of air conditioning and fresh air (e.g., the smaller the corrected enthalpy difference, the higher the utilization value of fresh air). Through this step, the enthalpy difference calculation error is reduced from ±15% to ±5%, and the fresh air utilization rate is increased by 20%, avoiding the misjudgment of "making decisions based on low-load enthalpy differences when the equipment is under high load".
[0091] The real-time heat generation of the devices is calculated based on the real-time power of the devices in the computer room. The real-time heat generation calculation is: Qdevice=∑(Preal,i·ηi), where Preal,i is the real-time power of the device, which is collected in real time by a power sensor (once every 5 minutes), and ηi is the device heat generation efficiency (0.9 for servers and 0.85 for switches, calibrated experimentally).
[0092] The real-time indoor and outdoor temperature and humidity (Tin,real, Hin,real, Tout,real, Hout,real) are collected by temperature and humidity sensors (once every 2 minutes) and used to calculate the original enthalpy value hin,real = 1.005Tin,real + (2501 + 1.84Tin,real)din, where din is the humidity content.
[0093] The corrected enthalpy difference is calculated by comparing the outdoor enthalpy with the original enthalpy. Specifically, the corrected enthalpy difference is calculated as follows: ∆h' = ∆h + kQdevice, where ∆h = hout,real - hin,real (original enthalpy difference), and k = 0.02 kg / kJ (calibrated through a full-load test: when Qdevice = 100 kW, ∆h' is 2 kJ / kg higher than ∆h).
[0094] The dynamic threshold is calculated using the real-time calorific value and the maximum real-time calorific value of the device. The dynamic threshold is calculated as follows: ɛ(t) = ɛ0(1 + α'Qdevice / Qdevice,max), where ɛ0 = 2 kJ / kg (baseline threshold, referencing ASHRAE standards).
[0095] Among them, the corrected enthalpy difference reflects the comprehensive heat and humidity difference, which combines the outdoor air enthalpy with the heat generated by the equipment. The larger the negative value, the stronger the cooling potential of the fresh air. The dynamic threshold is the critical value for judging the start and stop of the fresh air supply. The higher the load, the larger the threshold. By setting the threshold, it is possible to avoid excessive introduction of fresh air, which could cause fluctuations in the air conditioning load.
[0096] In another specific step 200, the real-time data at 12:00 in the computer room is as follows: Equipment data: Preal,i total = 90kW, ηi average 0.9, Qdevice = 90 × 0.9 = 81kW. Environmental data: hout,real = 85kJ / kg, hin,real = 50kJ / kg, ∆h = 35kJ / kg. Processing calculation: ∆h' = 35 + 0.02 × 81 = 36.62kJ / kg; ɛ(t) = 2 × (1 + 0.3 × 81 / 100) = 2 × 1.243 = 2.486kJ / kg. Output baseline data: ∆h' = 36.62kJ / kg (positive value, indicating that the outdoor enthalpy is higher than the indoor + equipment heat generation, and the utilization value of fresh air is low), ɛ(t) = 2.486kJ / kg.
[0097] In step 300 of some embodiments, a mode instruction is constructed using the benchmark judgment data and the collected second real-time data. The second real-time data includes pollutant data, weather data, and equipment status data of the atmosphere outside the computer room. This step balances "energy efficiency priority" and "safety priority" in complex scenarios (such as high pollution or heavy rain) to determine the operating rules for air conditioning and fresh air systems. This step also clarifies the load distribution ratio between fresh air and air conditioning in different scenarios (e.g., fresh air dominates in normal mode, air conditioning dominates in extreme weather), avoiding the limitations of single-parameter decision-making. The failure rate is reduced by 30% in high pollution scenarios and by 75% in heavy rain scenarios, significantly improving system adaptability.
[0098] In some embodiments, step 300 includes pollutant data such as PM2.5 concentration (Cpm, μg / m³), collected by an outdoor PM2.5 sensor (every 10 minutes, accuracy ±5 μg / m³). Weather data includes real-time rainfall (Rreal, mm / h), collected by an outdoor rain gauge (accuracy ±0.1 mm / h). System status data includes current air conditioning power (Pac, kW) and current fresh air volume (Qvent, m³ / h), read by the device controller.
[0099] like Figure 3 In some embodiments, step 300 specifically includes: generating a first mode command and first mode parameters when the corrected enthalpy difference is less than a negative dynamic threshold and the PM2.5 concentration is less than or equal to a preset pollution threshold (step 301); generating a second mode command and second mode parameters when the corrected enthalpy difference is less than a negative dynamic threshold, the PM2.5 concentration is greater than the preset pollution threshold, and the real-time rainfall is less than or equal to a preset rainfall threshold (step 302); generating a third mode command and third mode parameters when the real-time rainfall is greater than the preset rainfall threshold (step 303); and generating a fourth mode command and fourth mode parameters when the corrected enthalpy difference is greater than or equal to a negative dynamic threshold (step 304). Figure 3 As shown, steps 301-304 are performed sequentially according to their priority. That is, the first mode command and the first mode parameters are used to prioritize the use of fresh air for energy saving in the control system, with air conditioning as an auxiliary regulator; the second mode command and the second mode parameters are used to balance the use of fresh air and air quality in the control system, and control the entry of pollutants; the third mode command and the third mode parameters are used to ensure the safety of the control system and prevent rainwater backflow or a sudden increase in humidity; the fourth mode command and the fourth mode parameters are used to make the air conditioning the main cooling system in the control system, with fresh air only maintaining basic ventilation.
[0100] In some embodiments, a dynamic threshold where the corrected enthalpy difference is less than negative indicates that the combined value of "outdoor air enthalpy + equipment heat generation effect" is lower than that of the indoor air, meaning the fresh air has cooling potential and introducing fresh air can reduce the air conditioning load. The preset rainfall threshold is 50 mm / h. When real-time rainfall exceeds this threshold, it corresponds to a rainstorm level, requiring restriction of fresh air to prevent rainwater from entering the equipment room and causing short circuits. The air pollution threshold is 75 μg / m³. 3 When the PM2.5 concentration exceeds this threshold, it corresponds to the national level II standard for exceeding the standard, and enhanced filtration is required to avoid deterioration of indoor air quality.
[0101] In some embodiments, the first mode parameters include a first fresh air load ratio, a first air conditioning descent ratio, and a first protection level. The second mode parameters include a second fresh air load ratio, a second air conditioning descent ratio, and a second protection level. The third mode parameters include a third fresh air load ratio, a third air conditioning descent ratio, and a third protection level. The fourth mode parameters include a fourth fresh air load ratio, a fourth air conditioning descent ratio, and a fourth protection level. The fresh air load ratio is the ratio of the cooling load borne by the fresh air to the total cooling load. The air conditioning descent ratio is the ratio of the actual power of the air conditioner to its rated power. The protection level quantifies the filter performance of the fresh air system through resistance and efficiency.
[0102] Specifically, the first fresh air load accounts for β1=0.7 (accounting for 70% of the cooling load); the first air conditioning load reduction ratio γ1=0.3 (only 30% of the rated power); the first protection level F1=primary (resistance ≤50Pa).
[0103] Specifically, the second fresh air load ratio β2 = 0.5; the second air conditioning load reduction ratio γ2 = 0.5; and the second protection level F2 = high efficiency (resistance ≤ 200Pa, filtration efficiency ≥ 99.97%).
[0104] Specifically, the third fresh air load ratio β3=0.1 (maintaining only the minimum air exchange rate of 2 times / h); the third air conditioning load reduction ratio γ3=0.9 (operating at 90% full load); and the third air valve protection level P3=IP54 (splash-proof).
[0105] Specifically, the fourth fresh air load ratio β4 = 0.2; the fourth air conditioning load reduction ratio γ4 = 0.8; and the fourth protection level F4 = primary.
[0106] This step generates four system operation modes, quantifies the load distribution between air conditioning and fresh air, and saves energy. For example, at 12:00, multi-dimensional data shows: baseline data: ∆h' (36.62kJ / kg) ≥ -ɛ(t) (-2.486kJ / kg). Other data: Rreal=20mm / h≤50, Cpm=60μg / m³≤75. The process generates the fourth mode instruction, which is the default: "Fresh air maintains minimum airflow (meeting 2 air changes / h, i.e., Qvent=300m³ / h), air conditioning operates at full capacity of Qpred=90kW (Pac=90 / COP=90 / 3=30kW, COP is the air conditioning energy efficiency ratio)."
[0107] Some embodiments provide step 400 specifically including: generating an execution command based on the mode command and the collected third real-time data, wherein the third real-time data includes temperature gradient, wind speed and wind pressure, and the execution command includes air conditioning adjustment command and fresh air adjustment command.
[0108] This step receives mode commands and parameters, combines them with real-time flow field data (temperature gradient, wind speed), calculates execution commands such as supply air velocity and damper opening, and drives the air conditioning and fresh air equipment to operate. This step optimizes airflow organization to ensure that the load distribution in the mode commands is evenly distributed spatially, eliminating localized hotspots. It transforms "macro-level mode commands" into "micro-level execution commands," avoiding the problem of "overall load meeting standards but localized overheating." After this step, the computer room temperature gradient decreased from ±3℃ to ±1℃, the hotspot occurrence rate dropped to 0, and equipment operational stability improved by 40%.
[0109] like Figure 4 Some embodiments provide step 400, which includes: converting a first mode instruction into a first execution instruction based on a first mode parameter and third real-time data (step 401); converting a second mode instruction into a second execution instruction based on a second mode parameter and third real-time data (step 402); converting a third mode instruction into a third execution instruction based on a third mode parameter and third real-time data (step 403); and converting a fourth mode instruction into a fourth execution instruction based on a fourth mode parameter and third real-time data (step 404).
[0110] Specifically, based on the first fresh air load ratio, the baseline wind speed and temperature gradient of the computer room air, the first mode command is transformed into a first execution command that includes controlling the air supply speed. Based on the first protection level, the first mode command is transformed into a first execution command that includes the opening degree of the air damper. Based on the first air conditioning load reduction ratio, the first mode command is transformed into a first execution command that includes controlling the frequency of the air conditioning compressor.
[0111] For example, the first fresh air load percentage β1 is 0.7, the first air conditioning load reduction percentage γ1 = 0.3, and the first protection level F1 = primary (resistance ≤ 50Pa). Real-time flow field data: reference wind speed vbase = 3m / s, reference damper opening θbase = 100%, temperature gradient ∇T, air conditioning rated frequency fbase = 50Hz. The first mode command needs to control the fresh air supply velocity of the executing equipment to v1 = vbase × β1. If the temperature gradient ∇T ≤ 3℃ (no obvious hot spots), and the first fresh air load percentage β1 is 0.7, then the supply velocity v1 = 3 × 0.7 = 2.1m / s; if the temperature gradient ∇T > 3℃ (e.g., ∇T is 4℃ in a certain cabinet area), then the supply velocity v1 = 2.1 × 1.2 = 2.52m / s, to increase by 20% to enhance heat dissipation. If the first protection level is F1, then the damper opening θ1 is 100%. Fully opening the air valve reduces resistance loss and ensures that the fresh air volume reaches the target value (Qvent = A × v1, where A is the air outlet area); reducing the opening in the vortex zone can change the airflow direction and avoid wasting cooling capacity. The air conditioner compressor frequency f1 = fbase × γ1 = 50 × 0.3 = 15Hz. The air conditioner compressor frequency is directly proportional to the air conditioner's cooling capacity (15Hz corresponds to 30% of the rated cooling capacity), supplementing only the cooling load not covered by fresh air, thus achieving energy saving.
[0112] Specifically, based on the second fresh air load ratio, the baseline wind speed of the computer room air, the temperature gradient, and the actual filter resistance, the second mode command is transformed into a second execution command that includes controlling the air supply speed. Based on the actual filter resistance, the second mode command is transformed into a second execution command that includes controlling the air valve opening; based on the second air conditioning load reduction ratio, the second mode command is transformed into a second execution command that includes controlling the air conditioning compressor frequency.
[0113] For example, the second fresh air load ratio β2 is 0.5, the second air conditioning load reduction ratio γ2=0.5, and the second protection level F2=high efficiency (resistance ≤200Pa). Real-time flow field data: reference wind speed vbase=3m / s, reference damper opening θbase=100%, temperature gradient ∇T, actual filter resistance Pfilter=180Pa. The second mode command needs to control the fresh air supply velocity of the executing equipment to be v2=vbase×β2×(1+Pfilter / 200), which is calculated to be v2=3×0.5×(1+180 / 200)=1.5×1.9=2.85m / s. If the temperature gradient ∇T = 2℃ (no obvious hot spots), the air supply velocity v2 remains at 2.85 m / s; if the temperature gradient ∇T = 3.5℃, the air supply velocity v2 = 2.85 × 1.2 = 3.42 m / s, increasing the velocity to offset the airflow reduction caused by filter resistance (Qvent ∝ v), ensuring that the actual cooling capacity contribution of the fresh air still reaches the target of 50%. If Pfilte exceeds 180, the damper opening θ2 is 80%. The damper opening in areas with high heat generation equipment is increased by an additional 10% (θ2′ = 90%), prioritizing the delivery of cooling capacity. The air conditioning compressor frequency f2 = fbase × γ2 = 50 × 0.5 = 25 Hz. 25 Hz corresponds to 50% of the rated cooling capacity, sharing the total load with the fresh air. Simultaneously, due to the reduced fresh air volume, the air conditioner needs to compensate for more cooling capacity (at a higher frequency than the first mode).
[0114] Specifically, based on the proportion of the third fresh air load, the baseline wind speed of the computer room air, and the indoor-outdoor pressure difference, the third mode command is transformed into a third execution command that includes controlling the air supply speed. Based on rainfall, the third mode command is transformed into a third execution command that includes controlling the opening degree of the air damper. Based on the third air conditioning load reduction ratio, the third mode command is transformed into a third execution command that includes controlling the frequency of the air conditioning compressor.
[0115] For example, the third fresh air load ratio β3=0.1, the third air conditioning load reduction ratio γ3=0.9, and the third protection level P3=IP54 (to prevent water splashing from the air valve). Real-time flow field data: vbase=3m / s, rainfall R=60mm / h (heavy rain), indoor-outdoor pressure difference ΔP=5Pa (positive pressure needs to be maintained to prevent rainwater from entering). The supply air velocity v3=vbase×β3×(1+ΔP / 10) (low speed + positive pressure compensation to prevent rainwater from entering with the airflow); the calculated result is v3=3×0.1×(1+5 / 10)=0.3×1.5=0.45m / s; the extremely low wind speed reduces the risk of rainwater being brought in by the fresh air, and the positive pressure compensation (ΔP increase) can prevent outdoor rainwater from entering through gaps. If the rainfall is greater than or equal to 60mm / h, the air valve opening θ3=30%, and the side air supply outlets are closed, leaving only the top air supply outlet (θtop=30%) open to prevent rainwater from directly entering. A 30% opening and top air supply create a "narrow channel airflow," satisfying minimum air exchange (2 times / hour) while minimizing rainwater intrusion. The calculated air conditioner compressor frequency f3 is f3 = 50 × 0.9 = 45Hz. 45Hz is close to full load, quickly offsetting the heat generated by the equipment (at this point, the contribution of fresh air cooling is only 10%), while high-frequency operation can increase the indoor temperature (indirectly reducing relative humidity and preventing condensation).
[0116] Specifically, based on the proportion of the fourth fresh air load and the baseline air velocity in the computer room, the fourth mode command is transformed into a fourth execution command that includes controlling the air supply speed. Based on the protection level, the fourth mode command is transformed into a fourth execution command that includes controlling the opening degree of the air damper. Based on the fourth air conditioning load reduction ratio, the fourth mode command is transformed into a fourth execution command that includes controlling the frequency of the air conditioning compressor.
[0117] For example, the fourth fresh air load ratio β4 = 0.2, the fourth air conditioning load reduction ratio γ4 = 0.8, and the fourth protection level F4 = primary; real-time flow field data: vbase = 3 m / s, reference damper opening θbase = 100%, temperature gradient ∇T℃ (temperature distribution is relatively uniform). The supply air velocity v4 = vbase × β4 = 3 × 0.2 = 0.6 m / s (the fresh air load ratio is low, so high speed is not required); low-speed fresh air only meets ventilation needs (diluting CO2) and does not bear the main cooling load, avoiding "cold-heat cancellation" with air conditioning. According to the fourth protection level, the damper opening θ4 is 100% to ensure uniform distribution of fresh air (avoiding local oxygen deficiency); the return air vent opening is simultaneously maintained at 100% to form a "supply and return balance". Fully opening the dampers reduces ventilation resistance, allowing fresh air to diffuse evenly to all areas of the computer room, working in conjunction with air conditioning to form "overall environmental control". The air conditioner compressor frequency f4=50×0.8=40Hz. 40Hz corresponds to 80% of the rated cooling capacity and bears the main cooling load. Because the outdoor enthalpy value is high (Δh′≥−ε(t)), the fresh air has no energy-saving value, so the air conditioner dominates the control.
[0118] In this step, the first execution command will control the air conditioning and fresh air supply to maintain the computer room in normal mode. The air supply speed is proportional to the proportion of fresh air load, with accelerated speed for gradient over-limit conditions. The air valve opening is fully open and fine-tuned in conjunction with the vortex zone. The air conditioning compressor frequency is proportional to the load reduction ratio to achieve a low-frequency energy-saving, fresh air-dominated mode and eliminate temperature stratification. The second execution command will control the air conditioning and fresh air supply to maintain the computer room in high-pollution mode. The air supply speed is accelerated to compensate for filter resistance, with further acceleration for gradient over-limit conditions. The air valve opening is reduced to balance resistance, and locally increased in high-load areas. The air conditioning compressor frequency is at a medium frequency to balance insufficient fresh air. This command mainly restricts pollution intrusion and maintains cooling supply. The third execution command will control the air conditioning and fresh air supply to maintain the computer room in extreme weather mode. The air supply speed is controlled through low-speed positive pressure compensation and top-only air supply mode. The air valve opening is significantly reduced to prevent rainwater intrusion. The air conditioning compressor frequency reaches high frequency and full load to ensure heat dissipation and positive pressure. This command mainly prioritizes safety and prevents equipment damage. The generated fourth execution command will control the air conditioning and fresh air to keep the computer room in the air conditioning main control mode, meet ventilation through low speed, not participate in the main cooling mode to control the air supply speed, fully open the air valve to ensure uniform ventilation, and achieve high-frequency dominance of cooling by the air conditioning compressor frequency to stabilize the environment. This command is mainly to enable the air conditioning to bear the main load and the fresh air to assist ventilation.
[0119] Some embodiments provide step 500, which specifically includes: acquiring real-time feedback data and adjusting the parameters of the prediction model based on the real-time feedback data.
[0120] This step collects actual operational data from the execution layer, calculates the deviation from previous prediction / decision data, and calibrates model parameters (such as LSTM weights and enthalpy correction coefficients), forming a closed loop of "sensing-prediction-execution-correction." This step uses actual data to correct the model and parameters, eliminating deviations caused by factors such as sensor drift and equipment aging. It ensures long-term stable system operation and avoids a decrease in control accuracy due to "model solidification." Through this step, after one year of system operation, the prediction error can still be maintained within ±8% (compared to over 15% with traditional methods), and the annual energy efficiency degradation rate is ≤5%.
[0121] like Figure 5 In some embodiments, step 500 includes: calculating the prediction deviation based on the actual cooling load and the predicted cooling load (step 501); adjusting the hidden layer weights of the prediction model based on the prediction deviation (step 502); calculating the correction deviation based on the actual enthalpy difference and the corrected enthalpy difference, and recalibrating the correction coefficient based on the correction deviation (step 503); calculating the execution deviation based on the actual temperature gradient, and correcting the drag coefficient of the flow field model based on the execution deviation (step 504).
[0122] For example, real-time feedback data includes: actual cooling load (Qreal, unit kW), actual temperature gradient (∇Treal, calculated by distributed sensors), actual enthalpy difference (∆hreal, calculated based on indoor and outdoor temperature and humidity after execution), predicted cooling load Qpred output in step 100, and ∆h' output in step 200.
[0123] Prediction bias: Epred = |Qreal - Qpred| / Qpred × 100%. If Epred > 10%, adjust the hidden layer weights of the LSTM model (through backpropagation) to enhance the learning of recent load trends.
[0124] Correction deviation: Ecorr = |∆hreal - ∆h'| / ∆h' × 100%. If Ecorr > 5%, recalibrate the correction coefficient k (knew = k × ∆hreal / ∆h').
[0125] Execution deviation: Eexec = |∇Treal-1| / 1×100% (target gradient 1℃). If Eexec > 20%, correct the drag coefficient of the flow field model (based on the deviation between actual and predicted wind speed).
[0126] The calibrated model parameters (such as LSTM weights and know) are fed back to steps 100 and 200 to improve the accuracy of subsequent control.
[0127] Based on the control method described in steps 100-500 above, this invention also provides a control system for computer room air conditioning and fresh air based on enthalpy. This system includes a perception layer, a decision-making layer, and an execution layer.
[0128] like Figure 6 This sensing layer includes temperature sensors, humidity sensors, power sensors, PM2.5 sensors, rain gauges, wind speed sensors, wind pressure sensors, and a data acquisition module. The temperature, humidity, power, PM2.5, rain, wind speed, and wind pressure sensors are all connected to the data acquisition module. Each sensor collects physical quantities (such as temperature, humidity, and power values) in real time and transmits analog or digital signals to the data acquisition module. The data acquisition module summarizes and preprocesses the raw data (such as analog-to-digital conversion and format standardization) to form uniformly formatted sensing data, preparing it for uploading to the decision-making layer.
[0129] like Figure 6The decision-making layer includes an edge controller with PLC functionality, an industrial server for deploying the LSTM model, and a time-series database server. The edge controller and industrial server are connected via Gigabit Ethernet. The time-series database server and industrial server are connected via fiber optic cable. Every 15 minutes, the edge controller sends pre-processed historical data uploaded from the perception layer to the industrial server via Gigabit Ethernet, and the industrial server then stores the data in the time-series database server. Every 6 hours, the industrial server sends "look-ahead data" (such as predicted cooling load and outdoor enthalpy) to the edge controller via Gigabit Ethernet to provide a basis for the edge controller's decision-making. The industrial server reads historical data from the time-series database server for LSTM model training, generates look-ahead data, and then feeds it back to the edge controller.
[0130] like Figure 6 The execution layer includes an air conditioning inverter driver, an air conditioning compressor, a fresh air inverter driver, a damper actuator, a damper, a coarse filter, a medium-efficiency filter, and a high-efficiency filter. The air conditioning inverter driver is directly connected to the air conditioning compressor via a data line, and the driver outputs control signals to the compressor. The fresh air inverter driver is also directly connected to the damper actuator via a data line, and the driver outputs control signals to the damper actuator. The damper actuators are divided into three groups, each connected to the damper inlet of the coarse, medium, and high-efficiency filters via mechanical linkage or control lines, enabling individual control. The edge controller connects to the execution layer devices: the air conditioning inverter driver, the fresh air inverter driver, and the damper actuators are all connected to the relay output module of the edge controller via control cables, forming a control link of "edge controller → execution device".
[0131] The edge controller sends execution commands (such as compressor frequency and target values for damper opening) to the air conditioning inverter driver and the fresh air inverter driver via the ModbusTCP protocol, with a refresh cycle of 1 second. The air conditioning inverter driver (feedback of the current compressor frequency) and the fresh air inverter driver (feedback of the current damper actuator opening) feed back actual operating data to the edge controller every 0.5 seconds via control cables, forming a closed-loop control system. The air conditioning inverter driver adjusts the compressor frequency according to the commands, and the damper actuator adjusts the opening (open, close, or proportionally adjust) of the corresponding filter inlet damper according to the commands, thereby controlling the filtration level and fresh air volume.
[0132] The data acquisition module of the perception layer transmits the preprocessed perception data to the edge controller of the decision layer through data lines (such as shielded cables), providing raw input for the data analysis and judgment of the decision layer.
[0133] The edge controller at the decision-making level sends control commands (based on the Modbus TCP protocol) to the air conditioning inverter driver and fresh air inverter driver at the execution level through the relay output module and control cable. The commands include compressor frequency, air valve opening, etc.
[0134] The air conditioning inverter driver and fresh air inverter driver in the execution layer feed back the actual execution parameters (such as the current frequency and opening degree) to the edge controller in the decision layer in real time through control cables. The edge controller judges the execution deviation based on the feedback data and makes real-time corrections when necessary.
[0135] In one embodiment, step 100 primarily operates at the decision-making layer. The time-series database server reads historical data from the past two years, such as device power, temperature, humidity, and rainfall data uploaded by the perception layer, which is then preprocessed by the edge controller and stored. The industrial server loads an LSTM model, inputting historical data and real-time updated daily data (e.g., obtained from the edge controller every 15 minutes), trains and generates a predicted cooling load value Qpred, an outdoor enthalpy value hout,pred, and a load sensitivity coefficient α′ for the next 6 hours. The industrial server packages the prospective data (in JSON format) and sends it to the edge controller via Ethernet (once every 6 hours).
[0136] In one embodiment, step 200 primarily operates within the PLC core module of the edge controller. Input data includes the look-ahead data from step 100 (α′ sent by the industrial server) and real-time data from the sensing layer (cabinet power Preal collected by the power sensor, Tin,real, Tout,real collected by the temperature and humidity sensor, etc., transmitted to the edge controller via the data acquisition module). The edge controller calculates the real-time heat generation Qdevice=∑(Preal⋅η), the corrected enthalpy difference Δh′=Δh+k⋅Qdevice, and the dynamic threshold ε(t) according to the formula. The edge controller stores Δh′ and ε(t) in its local cache as input for step 300.
[0137] In one embodiment, step 300 primarily operates within the logic control module of the edge controller. Input data includes Δh′ and ε(t) from step 200 and real-time data from the sensing layer (Cpm from the PM2.5 sensor and Rreal from the rain gauge). The edge controller determines the triggering conditions according to priority (e.g., triggering the third mode command if Rreal > 50 mm / h), and generates corresponding mode parameters (e.g., third mode parameters β3 = 0.1, γ3 = 0.9). The edge controller packages the mode command (e.g., "third mode") and parameters into structured data (e.g., Modbus register values) and sends it to the local execution control module (the execution carrier of step 4).
[0138] In one embodiment, step 400 primarily operates within the execution control module of the edge controller. Input data includes the mode commands and parameters from step 300, and real-time flow field data from the sensing layer (vbase from the wind speed sensor, ∇T calculated by the distributed temperature sensor). The execution control module calculates execution parameters according to the mode parameters (e.g., in the third mode, v3=0.45m / s, θ3=30%, f3=45Hz). The edge controller sends electrical signals to the execution devices via the relay output module (e.g., a 4-20mA analog signal to control the opening of the damper, and a pulse signal to control the frequency of the inverter driver), driving the fresh air fan, air conditioning compressor, and damper to operate.
[0139] In one embodiment, step 500 primarily operates on the edge controller (feedback module) and the industrial server (model calibration module). Input data includes execution layer feedback data (the actual frequency *freal* uploaded by the frequency converter and the actual opening *θreal* uploaded by the damper actuator) and perception layer feedback data (the executed data *Qreal* and *∇Treal*). The edge controller calculates the execution deviation (e.g., *∣θreal−θ3∣ / θ3*). If the deviation > 5%, it fine-tunes the execution parameters in real time (e.g., correcting the damper opening from 30% to 32%). The industrial server receives deviation data uploaded by the edge controller every 24 hours (e.g., *epred=∣Qreal−Qpred∣ / Qpred*). If the deviation > 10%, it retrains the prediction model (adjusting the weights) and sends the calibrated model parameters back to the edge controller.
[0140] Throughout the system's operation, data flow follows three main logical directions. First, there's the data uplink direction: sensors in the perception layer collect data, which is then sequentially transmitted to the data acquisition module, then to the edge controller, and finally to the industrial server or database. This process provides the foundation for data input in steps 100-200. Second, there's the command downlink direction: based on the look-ahead data from step 100, the industrial server transmits commands to the edge controller. After processing in steps 200-400, the edge controller relays the commands to the actuators, which then execute the operation according to the parameters from step 4. Finally, there's the feedback loop direction: information generated by the actuators or sensors is fed back to the edge controller, which performs real-time corrections in step 500. This information is then transmitted back to the industrial server, which performs model calibration in step 500, generating updated models or parameters. These updated values are then transmitted back to the edge controller to optimize subsequent control.
[0141] The system first establishes physical connections—cables and pipes—to enable inter-device communication. Building upon this, it utilizes industrial buses or Ethernet to achieve real-time data exchange. This data interaction and device linkage allows steps 100-500 to form a complete closed loop, encompassing the entire process of "perception-prediction-decision-execution-correction." Through this comprehensive process, the intelligent collaborative operation of the data center's air conditioning and fresh air system is ultimately achieved.
[0142] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0143] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A control method for computer room air conditioning and fresh air based on enthalpy value, characterized in that, include: Acquire equipment layer data, environmental layer data, and historical data inside and outside the computer room. Based on the constructed prediction model, the equipment layer data, environmental layer data, and historical data, obtain forward-looking data, which includes predicted cooling load, predicted outdoor enthalpy value, and load sensitivity coefficient correction parameters. Based on the prospective data and the collected first real-time data, a benchmark judgment data is generated. The first real-time data includes the real-time equipment power, real-time equipment heat generation, real-time indoor and outdoor temperature, and real-time indoor and outdoor humidity in the computer room. The benchmark judgment data includes the corrected enthalpy difference and dynamic threshold. The model instruction is constructed using the benchmark judgment data and the collected second real-time data, wherein the second real-time data includes pollutant data, weather data, and equipment status data of the atmosphere outside the computer room; An execution command is generated based on the mode command and the collected third real-time data, wherein the third real-time data includes temperature gradient, wind speed and wind pressure, and the execution command includes air conditioning adjustment command and fresh air adjustment command. Obtain real-time feedback data and adjust the parameters of the prediction model based on the real-time feedback data.
2. The control method according to claim 1, characterized in that, The equipment layer data includes the real-time power of the cabinet and the operating time of the equipment; the environmental layer data includes indoor temperature, indoor humidity, outdoor temperature, outdoor humidity, and rainfall; the historical data includes equipment layer data and environmental layer data from the past two years, stored in a time-series database.
3. The control method according to claim 1, characterized in that, The steps for constructing the prediction model include: Multiple samples are constructed based on the sliding window method and the historical data. Each sample includes the input features of the first 24 time steps, the label of the 25th time step, and the simulation error. Multiple samples are used to construct a training set and a validation set. The prediction model is trained using the training set and validated using the validation set. A test set is constructed using multiple samples, and the validated model is then tested using the test set.
4. The control method according to claim 1, characterized in that, The model command is constructed using the aforementioned benchmark judgment data and the collected second real-time data. The second real-time data includes pollutant data, weather data, and equipment status data from the atmosphere outside the computer room, specifically including: When the corrected enthalpy difference is less than the negative dynamic threshold and the PM2.5 concentration is less than or equal to the preset pollution threshold, the first mode command and first mode parameters are generated. When the corrected enthalpy difference is less than the negative dynamic threshold, the PM2.5 concentration is greater than the preset pollution threshold, and the real-time rainfall is less than or equal to the preset rainfall threshold, the second mode command and second mode parameters are generated. When the real-time rainfall exceeds the preset rainfall threshold, a third-mode command and third-mode parameters are generated. When the correction enthalpy difference is greater than or equal to a negative dynamic threshold, a fourth mode instruction and fourth mode parameters are generated.
5. The control method according to claim 1, characterized in that, The parameters for the first mode include the first fresh air load ratio, the first air conditioning load reduction ratio, and the first protection level. The second mode parameters include the second fresh air load ratio, the second air conditioning load reduction ratio, and the second protection level; The third mode parameters include the third fresh air load ratio, the third air conditioning load reduction ratio, and the third protection level; The fourth mode parameters include the fourth fresh air load ratio, the fourth air conditioning load reduction ratio, and the fourth protection level.
6. The control method according to claim 5, characterized in that, The first fresh air load accounts for 0.7%, the first air conditioning load reduction ratio is 0.3%, the first protection level is primary, and the resistance is ≤50Pa; In some embodiments of the present invention, the second fresh air load ratio is 0.5; the second air conditioning load reduction ratio is 0.5; the second protection level is high efficiency, with resistance ≤200Pa and filtration efficiency ≥99.97%; In some embodiments of the present invention, the third fresh air load ratio is 0.1, the third air conditioning load reduction ratio is 0.9, and the third air valve protection level is IP54. In some embodiments of the present invention, the fourth fresh air load ratio is 0.2, the fourth air conditioning load reduction ratio is 0.8, and the fourth protection level is primary.
7. The control method according to claim 5, characterized in that, Based on the first mode parameters and the third real-time data, the first mode instruction is converted into a first execution instruction; based on the second mode parameters and the third real-time data, the second mode instruction is converted into a second execution instruction; based on the third mode parameters and the third real-time data, the third mode instruction is converted into a third execution instruction; based on the fourth mode parameters and the third real-time data, the fourth mode instruction is converted into a fourth execution instruction.
8. The control method according to claim 7, characterized in that, Based on the first fresh air load ratio, the baseline wind speed and temperature gradient of the computer room air, the first mode command is transformed into a first execution command that includes controlling the air supply speed; based on the first protection level, the first mode command is transformed into a first execution command that includes the opening degree of the air valve; based on the first air conditioning load reduction ratio, the first mode command is transformed into a first execution command that includes controlling the frequency of the air conditioning compressor. Optionally, based on the proportion of the second fresh air load, the reference wind speed of the computer room air, the temperature gradient, and the actual resistance of the filter, the second mode command is transformed into a second execution command that includes controlling the air supply speed. Based on the actual resistance of the filter, the second mode command is transformed into a second execution command that includes controlling the opening of the air valve; based on the second air conditioning load reduction ratio, the second mode command is transformed into a second execution command that includes controlling the frequency of the air conditioning compressor. Optionally, based on the proportion of the third fresh air load, the reference wind speed of the air in the computer room, and the indoor-outdoor pressure difference, the third mode command is transformed into a third execution command that includes controlling the air supply speed; based on the rainfall, the third mode command is transformed into a third execution command that includes controlling the opening of the air valve; based on the third air conditioning load reduction ratio, the third mode command is transformed into a third execution command that includes controlling the frequency of the air conditioning compressor. Optionally, based on the fourth fresh air load ratio and the reference wind speed of the computer room air, the fourth mode command is transformed into a fourth execution command that includes controlling the air supply speed; based on the protection level, the fourth mode command is transformed into a fourth execution command that includes controlling the opening degree of the air valve; based on the fourth air conditioning load reduction ratio, the fourth mode command is transformed into a fourth execution command that includes controlling the frequency of the air conditioning compressor.
9. The control method according to claim 1, characterized in that, Obtaining real-time feedback data and adjusting the parameters of the prediction model based on the real-time feedback data specifically includes: Calculate the prediction deviation based on the actual and predicted cooling loads; adjust the hidden layer weights of the prediction model based on the prediction deviation. The correction deviation is calculated based on the actual enthalpy difference and the corrected enthalpy difference, and the correction coefficient is recalibrated based on the correction deviation. The execution deviation is calculated based on the actual temperature gradient obtained, and the drag coefficient of the flow field model is corrected based on the execution deviation.
10. A control system for computer room air conditioning and fresh air based on enthalpy value, characterized in that, include: The layers are: perception layer, decision-making layer, and execution layer. This includes a temperature sensor, humidity sensor, power sensor, PM2.5 sensor, rain gauge, wind speed sensor, wind pressure sensor, and data acquisition module. The temperature sensor, humidity sensor, power sensor, PM2.5 sensor, rain gauge, wind speed sensor, and wind pressure sensor are all connected to the data acquisition module. This decision-making layer includes an edge controller, an industrial server, and a time-series database server. The edge controller and the industrial server are connected via gigabit Ethernet. The database server and the industrial server are connected via fiber optic cable. This decision-making layer includes an edge controller with PLC functionality, an industrial server for deploying LSTM models, and a time-series database server. The edge controller and the industrial server are connected via gigabit Ethernet. The execution layer includes air conditioner inverter driver, air conditioner compressor, fresh air inverter driver, damper actuator, damper, coarse filter, medium efficiency filter and high efficiency filter; The air conditioner inverter driver is directly connected to the air conditioner compressor via a data line, and the fresh air inverter driver is directly connected to the damper actuator via a data line. The damper actuator is divided into 3 groups, which are connected to the dampers at the inlet of the coarse filter, medium filter, and high efficiency filter via mechanical linkage or control lines to achieve individual control. The data acquisition module is connected to the edge controller. The air conditioning inverter driver, fresh air inverter driver, and air valve actuator are all connected to the relay output module of the edge controller through control cables. The system acquires equipment layer data, environmental layer data, and historical data from inside and outside the computer room. Based on the constructed prediction model, the equipment layer data, environmental layer data, and historical data, it obtains forward-looking data. The forward-looking data includes steps such as predicting cooling load, predicting outdoor enthalpy, and correcting load sensitivity coefficient parameters, which are implemented at the decision layer. Based on the prospective data and the collected first real-time data, a benchmark judgment data is generated. The first real-time data includes the real-time equipment power, real-time equipment heat generation, real-time indoor and outdoor temperature, and real-time indoor and outdoor humidity in the computer room. The benchmark judgment data includes steps for correcting the enthalpy difference and dynamic threshold, which are implemented in the PLC core module of the edge controller. The steps of constructing a mode instruction using the benchmark judgment data and the collected second real-time data, where the second real-time data includes pollutant data, weather data, and equipment status data of the atmosphere outside the computer room, are executed in the logic control module of the edge controller. The execution instructions are generated based on the mode instructions and the collected third real-time data, including temperature gradient, wind speed and wind pressure. The execution instructions include steps for adjusting the air conditioning and adjusting the fresh air, which are executed in the execution control module of the edge controller. The system acquires real-time feedback data and adjusts the parameters of the prediction model based on this data, running the system on an edge controller and an industrial server.
Citation Information
Cited By
PID self-adaptive intelligent control method and system for constant-temperature and constant-humidity air conditioner
CN121297194A
