A temperature control method based on air conditioning
By employing iterative optimization and multi-dimensional fresh air treatment strategies, combined with airflow short-circuit detection and sensor fault reconstruction mechanisms, the problems of cold aisle airflow short-circuit and sensor faults in the air conditioning system were solved, achieving efficient and stable temperature control of the air conditioning system.
Patent Information
- Application Number
- CN202611270714.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing air conditioning systems suffer from energy waste and localized overheating due to cold aisle airflow short circuits in high heat density environments. Furthermore, single-point sensor faults are difficult to identify, and existing solutions lack effective online verification methods, resulting in insufficient control accuracy.
By iteratively optimizing the chilled water valve control temperature, combined with a multi-dimensional fresh air treatment strategy, utilizing an airflow short-circuit detection and compensation mechanism, and combining segmented coordinated adjustment of the fan and chilled pump, an online sensor fault detection and reconstruction mechanism based on deviation value, variance, and spatial correlation coefficient is introduced to achieve on-demand matching of cooling capacity and real-time repair of sensor faults.
It significantly reduces the total energy consumption of the air conditioning system, improves temperature control stability and sensor fault tolerance, and ensures the continuous reliability and accuracy of control input, making it suitable for high heat density scenarios such as data centers.
Smart Images

Figure CN123124283A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of temperature control technology, and in particular to a temperature control method based on air conditioning. Background Technology
[0002] Precision air conditioning systems are widely used for temperature control in high-heat-density environments such as data centers. Traditional control methods typically use the air conditioner return air temperature as the sole feedback signal, adjusting fan speed and chilled water valve opening via PID control or simple logic. However, this approach ignores the impact of cold aisle airflow organization on return air temperature. When the cold aisle is poorly sealed, hot air exhausted from the server racks can directly short-circuit and flow back to the air conditioner return air vent, resulting in artificially high return air temperatures. The controller then increases cooling capacity accordingly, leading to both energy waste and localized overheating. Furthermore, single-point sensor failures (drift, disconnection) can completely distort the control baseline, and existing systems lack online verification methods for sensor reliability. Once abnormal temperature signals enter the control loop, they are difficult to correct.
[0003] Another type of improvement attempt to enhance control accuracy through multi-point temperature comparison or historical data modeling, such as using statistical differences in the return air temperatures of adjacent air conditioners to identify anomalies. However, this type of method has limitations when dealing with simultaneous anomalies from multiple sensors: when a sudden increase in regional load causes multiple air conditioners to rise in return air temperature simultaneously, simple deviation comparison cannot distinguish whether this is a genuine change in heat load or a collective sensor failure. Furthermore, existing solutions do not adequately consider the nonlinear relationship between temperature and load; fixed threshold judgments are prone to misjudgment under extreme conditions of light or heavy loads, making it difficult to balance the accuracy of identification across different operating stages. Summary of the Invention
[0004] This application provides a temperature control method based on air conditioning, which achieves on-demand matching of cooling capacity by iteratively optimizing the temperature control of the chilled water valve, and fully utilizes natural cold sources by combining a multi-dimensional fresh air treatment strategy, thereby significantly reducing the total energy consumption of the air conditioning system and improving the stability of temperature control.
[0005] This application provides a temperature control method based on air conditioning, including: S1, acquire the power consumption of IT cabinets in each cold aisle of the air conditioning cabinet and the control information of the air conditioner, and collect the inlet air temperature, exhaust air temperature and air conditioner return air temperature of each cabinet in real time; S2, the temperature difference between the exhaust air of the computer cabinet and the return air of the air conditioner and the temperature difference between the intake air of the cabinet and the supply air of the air conditioner, the airflow short circuit triggers the compensation mode to correct the return air temperature based on the two temperature differences, and after the airflow short circuit fault is repaired, it returns to using the actual measured return air temperature by the sensor for control. S3. Based on the cabinet's energy consumption information, the air conditioner's control information, the temperature corresponding to the air conditioner's minimum operating state, and the temperature corresponding to the air conditioner's maximum operating state, determine the chilled water valve control temperature. S4, based on the real-time current temperature and the determined chilled water valve control temperature, adjust the fan air volume and water valve opening in segments according to the relationship between the real-time temperature and the minimum, control, and maximum operating temperatures; S5 executes fresh air treatment strategies based on the outdoor fresh air temperature, real-time current temperature, and preset supply air temperature setpoint collected by the outdoor side sensor.
[0006] Preferably, the conditions for determining airflow short circuit are: the first temperature difference is less than 2°C and the second temperature difference is greater than 5°C; after determining airflow short circuit, an alarm is sent to the operation and maintenance system to check the cold aisle sealing; after the airflow short circuit fault is repaired, the system automatically resumes using the sensor to measure the return air temperature.
[0007] Preferably, S3 specifically includes: calculating the minimum air volume of the fan based on the cabinet energy consumption, real-time current temperature, supply air temperature, air density, and air specific heat capacity; determining whether to calculate the minimum opening of the water valve based on the ratio of the current air volume of the fan to the rated air volume; if the ratio is less than the air supply ratio threshold, calculating the minimum opening of the water valve based on the cabinet energy consumption, supply and return water temperature, liquid density, and the flow rate corresponding to the current water valve opening; setting an initial chilled water valve control temperature between the minimum and maximum operating temperatures of the air conditioner; generating a first mapping relationship between fan air volume and temperature and a second mapping relationship between water valve opening and temperature; performing control based on the real-time current temperature and measuring the total energy consumption of the fan and chilled water pump; iteratively adjusting the chilled water valve control temperature until the total energy consumption no longer decreases; and determining the temperature at this point as the final chilled water valve control temperature.
[0008] Preferably, before step S2, the system further includes: S21, sequentially numbering all air conditioners serving the same cold aisle or the same area within the air conditioning group control system, and establishing a relational mapping for the uncorrected return air temperature of each air conditioner; S22, calculating the average return air temperature of all air conditioners, recording the difference between the return air temperature of each air conditioner and the average value as the deviation value, and calculating the variance of the deviation value of each air conditioner within a historical window time; S23, collecting historical data from 24 consecutive hours of operation during the air conditioning group control phase, calculating the Pearson correlation coefficient between the return air temperatures of every two adjacent air conditioners, and forming a correlation coefficient matrix as a fault determination benchmark; S24, monitoring the deviation value and variance of each air conditioner in real time, updating the correlation coefficient matrix between the current air conditioners every half window time, and determining whether the return air temperature sensor of the air conditioner is faulty; S245, immediately triggering a reconstruction mode when a sensor fault is determined, selecting the set of air conditioners that are spatially closest and not faulty, calculating the estimated return air temperature using the distance-weighted average method, and using this estimated value to replace the measured value of the faulty sensor as the real-time current temperature.
[0009] Preferably, the fault determination criteria specifically include: retrieving the historical minute-by-minute return air temperature records of the group control system during continuous and stable operation, removing data points that are in the airflow short-circuit compensation period, sensor fault reconstruction period, and outliers whose temperature jump amplitude exceeds a preset threshold, and retaining valid data segments with stable operating conditions; if the total duration of valid data is less than the preset duration, the collection is extended until the requirements are met; pre-defining the adjacent objects of each air conditioner according to the air conditioning space layout, traversing all adjacent air conditioner pairs, and calculating the Pearson correlation coefficient by taking the return air temperature sequence after cleaning of the two; constructing a square matrix with the air conditioner number as the row and column index, setting the values of non-adjacent pairs to zero, and setting the values of the diagonal to 1, to obtain the benchmark correlation coefficient matrix; verifying the correlation coefficient values of all adjacent air conditioner pairs in the matrix, if all of them meet the preset correlation threshold, they are fixed as the fault determination criteria, otherwise the historical data collection time is extended and recalculated until the threshold is met.
[0010] Preferably, the judgment conditions for determining whether the return air temperature sensor of the air conditioner is faulty include: Condition 1, the absolute value of the deviation value exceeds the preset deviation threshold for 5 consecutive samplings, and the variance exceeds 3 times the normal variance of the deviation value of the air conditioner during the historical stable period; Condition 2, the correlation coefficient between the monitored air conditioner and the adjacent air conditioner at the current moment decreases from above 0.9 to below 0.7 compared with the historical benchmark R.
[0011] Preferably, the triggered reconstruction mode further includes: using the faulty air conditioner as a reference, selecting several air conditioners whose spatial distance meets preset conditions as candidates according to the pre-stored topology of the space to which it belongs; verifying the status of each candidate air conditioner, and including candidate air conditioners that have not been judged as sensor faults and whose current deviation value and variance are within the normal range into the available set; if the number of air conditioners in the available set is less than the preset number, expanding the selection range to make up for it; reading the distance between the faulty air conditioner and each air conditioner in the available set from the preset spatial distance table, calculating the distance weight of each air conditioner, and calculating the estimated value of the return air temperature of the faulty air conditioner by using the current return air temperature of each air conditioner in the available set according to the distance weighted average method; switching the real-time current temperature of the faulty air conditioner from the sensor measured value to the estimated value, and marking the air conditioner as being in sensor fault reconstruction in the status table to shield duplicate judgments.
[0012] Preferably, determining whether the return air temperature sensor of the air conditioner is faulty further includes: S241, obtaining the average power consumption of all cabinets in the cold aisle corresponding to each air conditioner serving the same cold aisle or the same area, and establishing a regression model for each air conditioner with the average power consumption of the cabinet as the independent variable and the return air temperature as the dependent variable; S242, using the regression model to calculate the predicted temperature of each air conditioner based on the current power consumption, and then calculating the difference between the measured temperature and the predicted temperature as the load deviation; at the same time, calculating the sliding correlation coefficient between adjacent air conditioners based on the temperature data of the most recent ten minutes; S243, monitoring the load deviation and correlation coefficient of all air conditioners in real time, if three air conditioners are simultaneously... If the absolute value of the load deviation of an air conditioner of 2 degrees Celsius or higher exceeds two degrees Celsius and the correlation coefficient with the adjacent air conditioner is less than 0.7, then a multi-sensor simultaneous abnormal event is determined; S244, after entering the multi-fault handling mode, candidate faulty sensors are initially screened and determined using the load deviation; S245, for the candidate faulty sensors, the direction of change of their correlation coefficient with adjacent sensors and the consistency of the load deviation direction are analyzed; S246, for the set of sensors determined to be faulty, the temperature at the location of the faulty sensor is estimated using the radial basis function interpolation method based on the current temperature of the surrounding non-faulty sensors, and the measured temperature is replaced with the interpolated estimated value as the control input.
[0013] Preferably, the process of determining candidate fault sensors includes: for all air conditioners, substituting the current average power consumption of the cabinet into their respective temperature and power consumption regression models to calculate the predicted return air temperature for each air conditioner, and then calculating the difference between the measured return air temperature and the predicted return air temperature as the load deviation; checking whether the absolute value of the load deviation of each air conditioner exceeds a preset load deviation threshold, which is set as a preset multiple of the standard deviation of the regression model residuals; if the absolute value of the load deviation is within the normal range, it is determined that the current temperature matches the predicted value of the regression model and is not temporarily listed as a candidate fault sensor; if the absolute value of the load deviation exceeds the threshold, it is marked as a candidate fault sensor.
[0014] Preferably, the regression model adopts a piecewise nonlinear form: the power consumption range is divided into several intervals, and a linear regression is fitted in each interval to ensure the continuity of the interval boundaries. The regression coefficients, residual standard deviations, and sample sizes of each interval are recorded as confidence weights. In real-time operation, the corresponding regression model is selected according to the current power consumption interval to calculate the predicted temperature. For sparse sample intervals, the adjacent intervals are extrapolated and the load deviation threshold is amplified. The load deviation threshold is dynamically adjusted according to the sample density of the current power consumption interval. The threshold for dense sample intervals is set to a smaller value, and the threshold for sparse sample intervals is set to a larger value.
[0015] One or more technical solutions provided in this application have at least the following technical effects or advantages: By employing an airflow short-circuit detection and compensation mechanism, temperature misjudgment caused by poor cold aisle sealing is effectively avoided, ensuring accurate and reliable control benchmarks. An iterative optimization method is used to determine the chilled water valve control temperature, and combined with the segmented coordinated adjustment of the fan and chilled pump, the cooling capacity is precisely matched on demand, significantly reducing the total energy consumption of the air conditioning system. At the same time, a multi-dimensional fresh air handling strategy is introduced to make full use of natural cold sources while ensuring the safe operation of equipment, further improving overall energy efficiency and temperature control stability, making it suitable for high heat density scenarios such as data centers.
[0016] An online sensor fault detection and reconstruction mechanism based on deviation value, variance, and spatial correlation coefficient is introduced to effectively identify and isolate abnormal temperature signals, avoiding the inaccuracy of the entire control loop due to the failure of a single sensor. The reliable return air temperature is reconstructed in real time using data from adjacent normal sensors through a distance-weighted average method, ensuring continuous and reliable control input, significantly improving the fault tolerance capability for sensor faults, reducing the risk of false alarms and misadjustments, and thus ensuring the long-term stable operation and temperature control accuracy of the air conditioning group control system.
[0017] By introducing a piecewise nonlinear regression model based on the average power consumption of the cabinet, the prediction of return air temperature is made more consistent with the actual load variation pattern, significantly improving the identification accuracy of simultaneous abnormal events from multiple sensors. Utilizing dynamically adjusted load deviation thresholds and directional consistency joint judgment, the system effectively distinguishes between real-area temperature changes and sensor malfunctions, significantly reducing false alarm and missed alarm rates. Simultaneously, radial basis function interpolation is used to reconstruct fault signals, ensuring continuous and reliable control input, further enhancing the system's adaptability and anti-interference capabilities under complex operating conditions, and improving overall temperature control stability and energy efficiency. Attached Figure Description
[0018] Figure 1 This is a schematic flowchart of a temperature control method based on air conditioning according to an embodiment of the present invention. Detailed Implementation
[0019] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.
[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the terminology used herein includes any and all combinations of one or more of the associated listed items.
[0021] Example 1: Figure 1 This is a schematic flowchart of a temperature control method based on air conditioning according to an embodiment of the present invention.
[0022] like Figure 1 As shown, a temperature control method based on air conditioning includes the following steps: S1 acquires the power consumption of IT cabinets in each cold aisle of the air conditioning cabinet and the control information of the air conditioner, and collects the inlet air temperature, exhaust air temperature and air conditioner return air temperature of each cabinet in real time (collected through the air conditioner return air vent sensor).
[0023] The control information specifically includes: fan operation status (return air temperature, supply air temperature, current supply air volume); chilled water pump operation status (supply water temperature, return water temperature); heat exchanger operation status (current water valve opening); and fresh air handling unit operation status (bypass valve opening, heat recovery efficiency, whether the preheating device is on, and the mixing ratio of fresh air and return air).
[0024] S2, the temperature difference between the exhaust air of the computer cabinet and the return air of the air conditioner and the temperature difference between the intake air of the cabinet and the supply air of the air conditioner, the airflow short circuit triggers the compensation mode to correct the return air temperature based on the two temperature differences, and after the airflow short circuit fault is repaired, the control is restored to using the actual measured return air temperature by the sensor.
[0025] The trigger compensation mode determines that an airflow short circuit occurs if the return air temperature difference is less than the preset normal heat dissipation temperature difference threshold (less than 2 degrees Celsius) and the supply air temperature difference is greater than the preset normal supply air temperature difference threshold (greater than 5 degrees Celsius).
[0026] When an airflow short circuit is detected, the compensation mode is triggered by correcting the air conditioning return air temperature to the average of the exhaust air temperatures of all cabinets. This corrected return air temperature replaces the sensor's measured value and is used as the real-time current temperature in all subsequent control steps. Simultaneously, an alarm is sent to the maintenance system, prompting a check of the cold aisle's sealing. Once the airflow short circuit fault is repaired, control automatically reverts to using the sensor's measured return air temperature.
[0027] S3 determines the chilled water valve control temperature based on the cabinet's energy consumption information, the air conditioner's control information, the temperature corresponding to the air conditioner's minimum operating state, and the temperature corresponding to the air conditioner's maximum operating state.
[0028] The specific method for determining the chilled water valve control temperature is as follows: Based on the cabinet energy consumption, real-time current temperature (return air temperature after S2 correction), supply air temperature, air density, and air specific heat capacity, calculate the minimum air volume of the fan. This minimum air volume represents the minimum air volume required to use all cooling capacity for cooling under the existing energy consumption.
[0029] Based on the ratio of the current air volume to the rated air volume of the fan, determine whether it is necessary to calculate the minimum opening of the water valve. If the ratio is less than the preset air supply ratio threshold, then calculate the minimum opening of the water valve based on the cabinet energy consumption, supply water temperature, return water temperature, liquid density, and the liquid flow rate corresponding to the current water valve opening.
[0030] An initial chilled water valve control temperature is set, falling between the temperatures corresponding to the minimum and maximum operating states of the air conditioner. Based on this initial control temperature, the minimum fan airflow, and the minimum water valve opening, a first mapping relationship between fan airflow and temperature, and a second mapping relationship between water valve opening and temperature are generated. A control operation is performed based on the real-time current temperature, and the total energy consumption of the fan and chilled water pump is measured. The chilled water valve control temperature is then adjusted, and the above process is repeated until the total energy consumption no longer decreases. This final temperature is determined as the final chilled water valve control temperature.
[0031] S4, based on the real-time current temperature and the determined chilled water valve control temperature, adjusts the fan air volume and water valve opening in segments according to the relationship between the real-time temperature and the minimum, control, and maximum operating temperatures.
[0032] Specifically, based on the chilled water valve control temperature determined in S3, cabinet energy consumption information, air conditioning control information, the temperature corresponding to the minimum operating state of the air conditioner, the temperature corresponding to the maximum operating state of the air conditioner, and the real-time current temperature corrected in S2, the air conditioner fan and chilled water pump are controlled in two stages: When the real-time temperature is lower than or equal to the temperature corresponding to the minimum operating state of the air conditioner, the fan air volume is set to the minimum fan air volume, and the water valve opening is set to the minimum water valve opening. When the real-time temperature is between the temperature corresponding to the minimum operating state of the air conditioner and the chilled water valve control temperature, the fan air volume gradually increases from the minimum fan air volume to the maximum fan air volume, and the water valve opening remains unchanged at the minimum water valve opening. When the real-time temperature is between the chilled water valve control temperature and the temperature corresponding to the maximum operating state of the air conditioner, the fan air volume remains at the maximum fan air volume, and the water valve opening gradually increases from the minimum water valve opening to the maximum water valve opening. When the real-time temperature is higher than the temperature corresponding to the maximum operating state of the air conditioner, the fan air volume is the maximum fan air volume, and the water valve opening is the maximum water valve opening.
[0033] S5 executes fresh air treatment strategies based on the outdoor fresh air temperature, real-time current temperature, and preset supply air temperature setpoint collected by the outdoor side sensor.
[0034] The specific fresh air handling strategies are as follows: Comfort and Energy Saving Strategy: If the difference between the outdoor fresh air temperature and the real-time current temperature exceeds a set threshold, the opening of the heat recovery bypass valve or the heat recovery efficiency is adjusted to ensure that the fresh air temperature after heat recovery is as close as possible to the real-time current temperature. Anti-Condensation and Anti-Icing Strategy: If the outdoor fresh air temperature is lower than the condensation / icing risk temperature, the preheating device is activated or the heat recovery bypass is adjusted to ensure that the fresh air temperature entering the air conditioning unit is not lower than the safe temperature. Precision Temperature Control Strategy: Based on the supply air temperature setpoint, the opening of the fresh air valve and return air valve is adjusted to ensure that the mixed air temperature equals the setpoint. Global Energy Efficiency Optimization Strategy: Based on indoor temperature control requirements or the air conditioning operating status, the supply air temperature setpoint or operating mode of the fresh air unit is dynamically adjusted. The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By employing an airflow short-circuit detection and compensation mechanism, temperature misjudgment caused by poor cold aisle sealing is effectively avoided, ensuring accurate and reliable control benchmarks. An iterative optimization method is used to determine the chilled water valve control temperature, and combined with the segmented coordinated adjustment of the fan and chilled pump, the cooling capacity is precisely matched on demand, significantly reducing the total energy consumption of the air conditioning system. At the same time, a multi-dimensional fresh air handling strategy is introduced to make full use of natural cold sources while ensuring the safe operation of equipment, further improving overall energy efficiency and temperature control stability, making it suitable for high heat density scenarios such as data centers.
[0035] Example 2: In Example 1, the correction of return air temperature relies solely on airflow short-circuit detection (temperature difference condition). However, in actual operation, the return air temperature may also exhibit abnormal deviations due to factors such as local load fluctuations, individual air conditioner performance differences, or environmental interference. These deviations do not meet the criteria for airflow short-circuit detection, causing uncorrected abnormal temperatures to directly enter subsequent control stages, resulting in inaccurate adjustments of the fan and chilled water pump. Since the degree of temperature deviation and stability varies significantly among different air conditioners under different operating conditions, a single airflow short-circuit compensation mechanism is insufficient to comprehensively identify and correct various abnormal temperature signals. To achieve a more granular assessment of the reliability of return air temperature, a fault diagnosis and reconstruction method based on the statistical characteristics and spatial correlation of multiple air conditioner temperatures needs to be introduced to further improve the accuracy of control input.
[0036] In some embodiments, the airflow short-circuit trigger compensation mode is determined based on two temperature differences, and step S2 further includes: S21, number all air conditioners serving the same cold aisle or the same area in the air conditioning group control system sequentially, and assign a relationship to the uncorrected return air temperature of each air conditioner.
[0037] The sampling frequency is no less than once per minute, and the return air temperature in this step is the original measured value directly collected from the sensor in step S1 before correction.
[0038] S22, calculate the average return air temperature of all air conditioners, and record the difference between the return air temperature of each air conditioner and the average value as the deviation value. Calculate the variance of the deviation value of each air conditioner within the historical window time.
[0039] The window time is 10 minutes.
[0040] S23. During the air conditioning group control operation phase, collect historical data for 24 consecutive hours of operation, calculate the Pearson correlation coefficient between the return air temperature of each two adjacent air conditioners, and form a correlation coefficient matrix as a fault judgment benchmark.
[0041] Specifically, this step is performed on the premise that the air conditioning group control system has been put into operation, all return air temperature sensors have been calibrated, there are currently no active airflow short circuit alarms, no identified sensor faults, and the system is in a stable operating condition.
[0042] Retrieve minute-by-minute return air temperature records from the group control system during 24 hours of continuous and stable operation, corresponding to all air conditioners numbered above. For each air conditioner, the return air temperature sequence must have at least 1440 valid sampling points. Data points within the 24-hour sequence that are in the airflow short-circuit compensation period or sensor fault reconstruction period are removed. Outliers with single-point temperature fluctuations exceeding ±5 degrees Celsius are also removed, retaining only valid data segments with stable operating conditions. If the total valid data duration is less than 24 hours, collection continues until the total valid data duration reaches 24 hours.
[0043] Based on the cold aisle / area spatial layout of the air conditioning service, predefine the adjacent objects of each air conditioner (air conditioners within the same cold aisle are adjacent to each other, and for cross-aisle pairs, select the 2-3 closest air conditioners). Iterate through all adjacent air conditioner pairs. For each pair of adjacent air conditioners, take the return air temperature sequences after cleaning and calculate the Pearson correlation coefficient, calculated as the covariance of the two sequences divided by the product of their respective standard deviations. Construct a square matrix using the air conditioner number as the row and column index (the number of rows and columns is consistent with the total number of air conditioners). The value in the i-th row and j-th column of the matrix is the correlation coefficient between air conditioner numbered i and air conditioner numbered j (values for non-adjacent air conditioner pairs are set to zero), and the diagonal values are 1, resulting in the baseline correlation coefficient matrix.
[0044] If the correlation coefficient values of all adjacent air conditioner pairs in the verification matrix are ≥0.9, the matrix is solidified as the fault determination benchmark. If there is an adjacent pair with a correlation coefficient <0.9, the historical data collection time is extended to 48 hours and recalculated until the threshold is met before solidification. After the benchmark is solidified, the subsequent real-time deviation monitoring and fault determination steps begin.
[0045] S24 monitors the deviation and variance of each air conditioner in real time, updates the correlation coefficient matrix between air conditioners every half window time, and determines whether the return air temperature sensor of the air conditioner is faulty.
[0046] There are two criteria for determining whether the return air temperature sensor of the air conditioner is faulty. Criterion 1: The absolute value of the deviation exceeds the preset deviation threshold (3℃) for 5 consecutive samplings, and the variance exceeds 3 times the normal variance of the deviation value of the air conditioner during the historical stable period (the normal variance is the variance of the deviation value of the air conditioner during the historical stable period). Criterion 2: The correlation coefficient between the monitored air conditioner and 2 to 3 adjacent air conditioners at the current moment decreases from above 0.9 to below 0.7 compared with the historical benchmark R.
[0047] S245: When a sensor fault is detected, the reconfiguration mode is immediately triggered. The set of air conditioners that are closest in space and not faulty is selected, and the estimated return air temperature is calculated using the distance-weighted average method. This estimated value is used to replace the measured value of the faulty sensor as the real-time current temperature.
[0048] Specifically, taking the faulty air conditioner i as the baseline, and according to the pre-stored spatial topology of its cold aisle / zone, the 2-3 air conditioners with the closest spatial distance are selected as candidates; the status of each candidate air conditioner j is checked—if j itself has not been determined to be a sensor fault, and j's current deviation value... Dⁿ and variance If all are within the normal range, then j is added to the available set S; if there are fewer than 2 available air conditioners in S, then the set is expanded outwards to include the next closest air conditioners. The distance between the faulty air conditioner i and each air conditioner j in set S is read from the system's preset air conditioner spatial distance table. (Units can be consistent; calculations can be based on rack row / cold aisle number or actual length in meters.) >0.
[0049] For each air conditioner j in set S, calculate the weight. The total weights are obtained by summing them up. Take the current return air temperature of each air conditioner j in set S. (Verified as a reliable value via S23 / S24), by Calculate the estimated return air temperature of the faulty air conditioner i. If the system has a pre-stored linear regression model (coefficients) between air conditioner i and each air conditioner in set S, based on historical normal operation data... If the intercept (b) is used, the regression estimate can be calculated in parallel. The distance weighted value is compared with the distance weighted value. If the deviation between the two is within the threshold, the distance weighted value is used. If the deviation exceeds the limit, the average of the two values is used or an alarm is triggered for manual intervention.
[0050] The real-time current temperature of the faulty air conditioner i is measured by the sensor. Switch to The estimated value is used in the calculation for subsequent S2 airflow short circuit detection and S3 chilled water valve control. At the same time, the air conditioner i is marked as sensor fault reconstruction in the system status table. This mark synchronously shields the repeated judgment of this air conditioner by S24 (to avoid dead loop).
[0051] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: An online sensor fault detection and reconstruction mechanism based on deviation value, variance, and spatial correlation coefficient is introduced to effectively identify and isolate abnormal temperature signals, avoiding the inaccuracy of the entire control loop due to the failure of a single sensor. The reliable return air temperature is reconstructed in real time using data from adjacent normal sensors through a distance-weighted average method, ensuring continuous and reliable control input, significantly improving the fault tolerance capability for sensor faults, reducing the risk of false alarms and misadjustments, and thus ensuring the long-term stable operation and temperature control accuracy of the air conditioning group control system.
[0052] Example 3: In Example 1, self-testing was performed on the same power line using the voltage differences between different electricity meters to obtain a sequence of abnormal probability values for the electricity meters. However, although power lines with similar overall electricity usage patterns were grouped together by power line grouping, different electricity meters on the same power line may have significant differences due to different types of electricity usage or usage habits. This leads to reduced accuracy due to inconsistent reference standards. For example, different users in the same building may have different types and patterns of electricity usage. Therefore, when comparing the voltage values with other electricity meters on the same power line, the standard of the comparison object is extremely important. Self-testing comparisons between electricity meters with similar usage habits have higher credibility and reliability. Therefore, in order to differentiate the minor differences between different electricity meters on the same power line (to compensate for local errors in the clustering of overall electricity usage behavior on the power line), the selection of the reference electricity meter in the self-testing process is extremely crucial.
[0053] In some embodiments, determining whether the return air temperature sensor of the air conditioner is faulty in step S24 further includes: S241, obtain the average power consumption of all cabinets in the cold aisle corresponding to each air conditioner serving the same cold aisle or the same area, and establish a regression model for each air conditioner with the average power consumption of the cabinet as the independent variable and the return air temperature as the dependent variable.
[0054] The regression model also includes: 1A. In the initial modeling phase, collect at least 14 days of normal operation data and plot a scatter plot of the return air temperature of each air conditioner versus the average power consumption of the cabinet.
[0055] 1B. Piecewise linear regression or locally weighted regression is used to fit the nonlinear relationship curve between temperature and power consumption. Specifically, the power consumption range is divided into several intervals, and a linear regression is fitted within each interval, ensuring continuity at the interval boundaries.
[0056] 1C records the regression coefficients and residual standard deviations for each interval, and also records the sample size for each interval as the confidence weight; in real-time operation, based on the interval where the current average power consumption of the cabinet is located, the corresponding regression model is selected to calculate the predicted temperature and load deviation.
[0057] 1D, if the current power consumption falls in the sparse sample range, then the model extrapolation of the adjacent range is used, and the threshold of the load deviation is appropriately increased.
[0058] 1E, the fault determination process is the same as in Example 3, but the threshold of load deviation is dynamically adjusted according to the current power consumption range: the threshold is smaller in the range with dense samples and larger in the range with sparse samples, and the piecewise nonlinear model is refitted periodically.
[0059] S242 uses a regression model to calculate the predicted temperature of each air conditioner based on the current power consumption, and then calculates the difference between the measured temperature and the predicted temperature as the load deviation; at the same time, it calculates the sliding correlation coefficient between adjacent air conditioners based on the temperature data of the most recent ten minutes.
[0060] 243. Monitor the load deviation and correlation coefficient of all air conditioners in real time. If the absolute value of the load deviation of three or more air conditioners exceeds two degrees Celsius and the correlation coefficient with the adjacent air conditioner is less than 0.7, it is determined that a multi-sensor simultaneous abnormal event has occurred.
[0061] S244, after entering the multi-fault handling mode, the candidate faulty sensor is initially screened and determined by the load deviation.
[0062] Specifically, for all air conditioners, based on the current average power consumption of the rack, the predicted return air temperature for each air conditioner is calculated by substituting it into their respective temperature and power consumption regression models. The difference between the measured and predicted return air temperatures for each air conditioner is then calculated, representing the load deviation.
[0063] Each air conditioner is checked individually to see if its absolute load deviation exceeds a preset load deviation threshold. This threshold is typically set as a multiple of the standard deviation of the regression model residuals, such as 1 or 2 times the standard deviation. If the absolute load deviation of an air conditioner is within the normal range (i.e., less than or equal to the threshold), its current temperature is considered to match the regression model's predicted value, and it is not temporarily listed as a candidate fault sensor, even if its correlation coefficient with adjacent sensors is low, as this may be due to actual temperature changes. If the absolute load deviation of an air conditioner exceeds the threshold, it is marked as a candidate fault sensor, awaiting further joint determination in subsequent steps.
[0064] S245, for candidate faulty sensors, analyze the consistency of their correlation coefficient change direction and load deviation direction with those of adjacent sensors: if the correlation coefficient of a group of adjacent candidate sensors decreases and the load deviation direction is consistent (all positive or all negative), it is determined to be a real area temperature change and the signal is not reconstructed; if the directions are inconsistent, it is determined to be a multi-sensor fault.
[0065] S246, for the set of sensors determined to be faulty, based on the current temperature of the surrounding non-faulty sensors, the radial basis function interpolation method is used to estimate the temperature at the location of the faulty sensor, and the measured temperature is replaced with the interpolated estimate as the control input.
[0066] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By introducing a piecewise nonlinear regression model based on the average power consumption of the cabinet, the prediction of return air temperature is made more consistent with the actual load variation pattern, significantly improving the identification accuracy of simultaneous abnormal events from multiple sensors. Utilizing dynamically adjusted load deviation thresholds and directional consistency joint judgment, the system effectively distinguishes between real-area temperature changes and sensor malfunctions, significantly reducing false alarm and missed alarm rates. Simultaneously, radial basis function interpolation is used to reconstruct fault signals, ensuring continuous and reliable control input, further enhancing the system's adaptability and anti-interference capabilities under complex operating conditions, and improving overall temperature control stability and energy efficiency.
[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A temperature control method based on air conditioning, characterized in that, include: S1, acquire the power consumption of IT cabinets in each cold aisle of the air conditioning cabinet and the control information of the air conditioner, and collect the inlet air temperature, exhaust air temperature and air conditioner return air temperature of each cabinet in real time; S2, the temperature difference between the exhaust air of the computer cabinet and the return air of the air conditioner and the temperature difference between the intake air of the cabinet and the supply air of the air conditioner, the airflow short circuit triggers the compensation mode to correct the return air temperature based on the two temperature differences, and after the airflow short circuit fault is repaired, it returns to using the actual measured return air temperature by the sensor for control. S3. Based on the cabinet's energy consumption information, the air conditioner's control information, the temperature corresponding to the air conditioner's minimum operating state, and the temperature corresponding to the air conditioner's maximum operating state, determine the chilled water valve control temperature. S4, based on the real-time current temperature and the determined chilled water valve control temperature, adjust the fan air volume and water valve opening in segments according to the relationship between the real-time temperature and the minimum, control, and maximum operating temperatures; S5 executes fresh air treatment strategies based on the outdoor fresh air temperature, real-time current temperature, and preset supply air temperature setpoint collected by the outdoor side sensor.
2. The temperature control method based on air conditioning according to claim 1, characterized in that, The conditions for determining an airflow short circuit are: the first temperature difference is less than 2°C and the second temperature difference is greater than 5°C; after determining an airflow short circuit, an alarm is sent to the operation and maintenance system to check the cold aisle sealing; after the airflow short circuit fault is repaired, the system automatically resumes using the sensor to measure the return air temperature.
3. The temperature control method based on air conditioning according to claim 1, characterized in that, S3 specifically includes: calculating the minimum air volume of the fan based on the cabinet energy consumption, real-time current temperature, supply air temperature, air density, and air specific heat capacity; determining whether to calculate the minimum opening of the water valve based on the ratio of the current air volume of the fan to the rated air volume; if the ratio is less than the air supply ratio threshold, calculating the minimum opening of the water valve based on the cabinet energy consumption, supply and return water temperature, liquid density, and the flow rate corresponding to the current water valve opening; setting an initial chilled water valve control temperature between the minimum and maximum operating temperatures of the air conditioner; generating a first mapping relationship between fan air volume and temperature and a second mapping relationship between water valve opening and temperature; performing control based on the real-time current temperature and measuring the total energy consumption of the fan and chilled water pump; iteratively adjusting the chilled water valve control temperature until the total energy consumption no longer decreases; and determining the temperature at this point as the final chilled water valve control temperature.
4. The temperature control method based on air conditioning according to claim 1, characterized in that, Before S2, the system further includes: S21, sequentially numbering all air conditioners serving the same cold aisle or the same area within the air conditioning group control system, and establishing a relational mapping for the uncorrected return air temperature of each air conditioner; S22, calculating the average return air temperature of all air conditioners, recording the difference between the return air temperature of each air conditioner and this average as the deviation value, and calculating the variance of the deviation value of each air conditioner within a historical window time; S23, collecting historical data from 24 consecutive hours of operation during the air conditioning group control phase, calculating the Pearson correlation coefficient between the return air temperatures of every two adjacent air conditioners, and forming a correlation coefficient matrix as a fault determination benchmark; S24, monitoring the deviation value and variance of each air conditioner in real time, updating the correlation coefficient matrix between the current air conditioners every half window time, and determining whether the return air temperature sensor of the air conditioner is faulty; S245, immediately triggering the reconstruction mode when a sensor fault is determined, selecting the set of air conditioners that are spatially closest and not faulty, calculating the estimated return air temperature using the distance-weighted average method, and using this estimated value to replace the measured value of the faulty sensor as the real-time current temperature.
5. The temperature control method based on air conditioning according to claim 4, characterized in that, The fault determination criteria specifically include: retrieving the historical minute-by-minute return air temperature records of the group control system during continuous and stable operation, removing data points that are in the airflow short-circuit compensation period, sensor fault reconstruction period, and outliers whose temperature jump amplitude exceeds a preset threshold, and retaining valid data segments with stable operating conditions; if the total duration of valid data is less than the preset duration, the collection is extended until the requirements are met; pre-defining the adjacent objects of each air conditioner according to the air conditioning spatial layout, traversing all adjacent air conditioner pairs, and calculating the Pearson correlation coefficient by taking the return air temperature sequence after cleaning of the two; constructing a square matrix with the air conditioner number as the row and column index, setting the values of non-adjacent pairs to zero, and setting the values of the diagonal to 1, to obtain the benchmark correlation coefficient matrix; verifying the correlation coefficient values of all adjacent air conditioner pairs in the matrix, if all of them meet the preset correlation threshold, they are fixed as the fault determination criteria, otherwise the historical data collection time is extended and recalculated until the threshold is met.
6. The temperature control method based on air conditioning according to claim 4, characterized in that, The conditions for determining whether the return air temperature sensor of the air conditioner is faulty include: Condition 1, the absolute value of the deviation value exceeds the preset deviation threshold for 5 consecutive samplings, and the variance exceeds 3 times the normal variance of the deviation value of the air conditioner during the historical stable period; Condition 2, the correlation coefficient between the monitored air conditioner and the adjacent air conditioner at the current moment decreases from above 0.9 to below 0.7 compared with the historical benchmark R.
7. The temperature control method based on air conditioning according to claim 4, characterized in that, The triggering reconstruction mode further includes: using the faulty air conditioner as a benchmark, selecting several air conditioners whose spatial distance meets preset conditions as candidates according to the pre-stored topology of the space to which it belongs; verifying the status of each candidate air conditioner, and including candidate air conditioners that have not been judged as sensor faults and whose current deviation value and variance are within the normal range into the available set; if the number of air conditioners in the available set is less than the preset number, expanding the selection range to make up for it; reading the distance between the faulty air conditioner and each air conditioner in the available set from the preset spatial distance table, calculating the distance weight of each air conditioner, and calculating the estimated value of the return air temperature of the faulty air conditioner by using the current return air temperature of each air conditioner in the available set according to the distance weighted average method; switching the real-time current temperature of the faulty air conditioner from the sensor measured value to the estimated value, and marking the air conditioner as being in sensor fault reconstruction in the status table to shield duplicate judgments.
8. The temperature control method based on air conditioning according to claim 4, characterized in that, The determination of whether the return air temperature sensor of the air conditioner is faulty further includes: S241, obtaining the average power consumption of all cabinets in the cold aisle corresponding to each air conditioner serving the same cold aisle or the same area, and establishing a regression model for each air conditioner with the average power consumption of the cabinet as the independent variable and the return air temperature as the dependent variable; S242, using the regression model to calculate the predicted temperature of each air conditioner based on the current power consumption, and then calculating the difference between the measured temperature and the predicted temperature as the load deviation; at the same time, calculating the sliding correlation coefficient between adjacent air conditioners based on the temperature data of the most recent ten minutes; S243, monitoring the load deviation and correlation coefficient of all air conditioners in real time, if three or more air conditioners are simultaneously affected... If the absolute value of the load deviation of the upper air conditioner exceeds two degrees Celsius and the correlation coefficient with the adjacent air conditioner is less than 0.7, then a multi-sensor simultaneous abnormal event is determined; S244, after entering the multi-fault handling mode, the candidate faulty sensors are initially screened and determined using the load deviation; S245, for the candidate faulty sensors, the direction of change of their correlation coefficient with the adjacent sensors and the consistency of the load deviation direction are analyzed; S246, for the set of sensors determined to be faulty, the temperature at the location of the faulty sensor is estimated using the radial basis function interpolation method based on the current temperature of the surrounding non-faulty sensors, and the measured temperature is replaced with the interpolated estimated value as the control input.
9. The temperature control method based on air conditioning according to claim 8, characterized in that, The process of identifying candidate fault sensors includes: for all air conditioners, substituting the current average power consumption of the cabinet into their respective temperature and power consumption regression models to calculate the predicted return air temperature for each air conditioner, and then calculating the difference between the measured return air temperature and the predicted return air temperature as the load deviation; checking whether the absolute value of the load deviation for each air conditioner exceeds a preset load deviation threshold, which is set as a preset multiple of the standard deviation of the regression model residuals; if the absolute value of the load deviation is within the normal range, it is determined that the current temperature matches the predicted value of the regression model and is not listed as a candidate fault sensor for the time being; if the absolute value of the load deviation exceeds the threshold, it is marked as a candidate fault sensor.
10. The temperature control method based on air conditioning according to claim 9, characterized in that, The regression model adopts a piecewise nonlinear form: the power consumption range is divided into several intervals, and a linear regression is fitted in each interval to ensure the continuity of the interval boundaries. The regression coefficients, residual standard deviations, and sample sizes of each interval are recorded as confidence weights. In real-time operation, the corresponding regression model is selected according to the current power consumption interval to calculate the predicted temperature. For sparse sample intervals, the adjacent intervals are extrapolated and the load deviation threshold is amplified. The load deviation threshold is dynamically adjusted according to the sample density of the current power consumption interval. The threshold is set to a smaller value for dense sample intervals and a larger value for sparse sample intervals.
