Method for detecting failure symptoms of an air conditioning system

Abstract

The application provides a fault symptom detection method of an air conditioning system, comprising: for the air conditioning system without historical operation data, adopting a fault symptom detection method based on statistical residual of operation performance parameters and a fault symptom detection method based on operation performance symptom rules; for the air conditioning system with historical operation data, simultaneously adopting the fault symptom detection method based on statistical residual of operation performance parameters, the fault symptom detection method based on operation performance symptom rules and a principal component based on air conditioning load mode matching similar historical operation data, a combined fault symptom detection method of the three methods; and performing credibility evaluation on the fault detection result to obtain a fault symptom credibility detection result of the air conditioning system fault. Through comprehensive application of multiple technical means, the application realizes automatic and intelligent processing of the air conditioning system fault symptom detection, improves the stability and reliability of the air conditioning system. Timely and accurate fault symptom detection information is provided for maintenance personnel, and intelligent operation performance symptom detection of the air conditioning system is realized.

Description

Technical Field

[0001] This invention relates to a method for detecting fault symptoms in an air conditioning system. Background Technology

[0002] Existing methods for detecting fault symptoms in air conditioning systems generally employ single statistical residuals or simple association rule methods. When historical operating data is available, machine learning or data mining algorithms are not used, and the effects of time-varying factors such as the delay and attenuation of the actual operating status of the air conditioning system as the external dynamic load changes are not considered. This often leads to problems with timely and accurate detection of fault symptoms, and makes it difficult to achieve intelligent fault symptom detection for air conditioning systems. Summary of the Invention

[0003] The purpose of this invention is to provide a method for detecting fault symptoms in an air conditioning system.

[0004] To address the above problems, this invention provides a method for detecting fault symptoms in an air conditioning system, comprising:

[0005] Based on key operating performance parameters, the steady-state or non-steady-state operation of the air conditioning system is determined. If the operating state of the air conditioning system is determined to be non-steady-state, the fault detection of the air conditioning system is terminated.

[0006] If the operating state of the air conditioning system is determined to be steady state, for air conditioning systems without historical operating data, a fault detection method based on statistical residuals of operating performance parameters and fault symptom rules based on operating performance is used to obtain fault detection results; for air conditioning systems with historical operating data, a combined fault symptom detection method combining statistical residuals of operating performance parameters, fault symptom rules based on operating performance, and principal component analysis based on air conditioning load pattern matching of similar historical operating data is used to obtain fault detection results.

[0007] To evaluate the reliability of fault detection results, two evaluation indicators, "probability of fault symptom detection" and "percentage of symptom occurrence time", were used to obtain the reliability of fault symptom detection results for air conditioning system faults.

[0008] Furthermore, in the above method, the key operating performance parameters include: supply air temperature and supply air setpoint, supply air flow rate and flow demand value, supply air pressure and supply air pressure setpoint, supply water temperature and return water temperature, and supply water flow rate and predicted water flow rate (the water flow rate predicted based on the water valve opening).

[0009] The criteria for judging steady-state operation include: the difference between the supply air temperature and the supply air setpoint is less than or equal to 3℃; the difference between the supply air flow rate and the flow demand is less than or equal to 10% of the maximum flow rate on the fan nameplate; the difference between the supply air pressure and the supply air pressure setpoint is less than or equal to 10% of the maximum static pressure on the fan nameplate; the difference between the supply water temperature and the return water temperature is less than or equal to 8℃; and the difference between the supply water flow rate and the predicted water flow rate (the water flow rate predicted based on the water valve opening) is less than or equal to 10% of the maximum flow rate on the water pump nameplate.

[0010] Furthermore, in the above methods, the fault symptom detection method based on statistical residuals of operating performance parameters includes:

[0011] Each operating performance parameter of the air conditioning system (e.g., temperature, humidity, flow rate, pressure, CO2 concentration, control signals, etc., collected on-site) is compared with its corresponding reference value (e.g., set value, previous measurement value, similar function measurement value, etc.), and the corresponding statistical residuals are calculated, including simple subtraction, normalization or standardization; different statistical residuals are compared with preset thresholds to identify different fault symptoms.

[0012] Furthermore, in the above methods, the fault symptom detection method based on operational performance symptom rules includes:

[0013] For systems without historical operating data, the set values ​​of existing similar air conditioning systems or fixed thresholds set based on expert experience are used as the discrimination thresholds. For systems with historical operating data, dynamic or adaptive thresholds obtained by training on historical operating data are used as the discrimination thresholds. Different discrimination thresholds correspond to different fault symptoms.

[0014] Furthermore, in the above method, the principal component fault symptom detection method based on matching similar historical operating data with air conditioning load patterns includes:

[0015] The operating performance data of the current time window and the filtered historical operating performance reference dataset without faults are used as input data for the principal component fault symptom detection model. The average prediction error and the control limit are calculated and compared. If the average prediction error is greater than the control limit, it is determined that the air conditioning system has fault symptoms at the current time. If the average prediction error is less than the control limit, it is determined that the air conditioning system has no fault symptoms at the current time.

[0016] Furthermore, in the above method, before using the current time window's operational performance data and the filtered, fault-free historical operational performance reference dataset as input data for the principal component fault symptom detection model, the following steps are also included:

[0017] Construct a dataset of cooling / heating load performance parameters inside and outside the air conditioning system within a sliding time window at the current moment;

[0018] The dataset of cooling / heating load performance parameters inside and outside the air conditioner in the current sliding time window is compared with the cooling / heating load performance parameters inside and outside the air conditioner in the fault-free historical operation performance database. The principal component similarity factor of these parameters is calculated to perform similarity matching. The fault-free historical operation performance data of the top n time windows with the largest similarity factor (n can be 5, 10, 15, 20, etc.) are selected as the reference dataset for the filtered fault-free historical operation performance parameters.

[0019] Furthermore, in the above method, the internal cooling / heating load parameters of the air conditioner include: supply air flow rate, supply air temperature, return air temperature, supply air humidity, return air humidity, heat generation of electrical equipment, heat generation of lighting, and heat dissipation from personnel.

[0020] The external cooling / heating load parameters of the air conditioner include: fresh air dry bulb temperature, fresh air humidity, fresh air enthalpy, fresh air volume, normal solar radiation intensity, and horizontal solar radiation intensity.

[0021] Furthermore, in the above method, the dataset of cooling / heating load performance parameters of the air conditioner's interior and exterior within the current sliding time window is compared with the cooling / heating load performance parameters of the air conditioner's interior and exterior in the fault-free historical operation performance database. Principal component similarity factors are calculated to perform similarity matching. The fault-free historical operation performance data of the top n time windows with the largest similarity factors (n can be 5, 10, 15, 20, etc.) are selected as the reference dataset for the filtered fault-free historical operation performance parameters, including:

[0022] The first load parameter matrix is ​​obtained by measuring the cooling / heating load performance parameters inside and outside the air conditioner in the current time window measurement data. Using a sliding time window (whose time dimension is equal to that of the current time window), the filtered reference dataset of fault-free historical operating performance parameters is divided into numerous fault-free historical data windows at a certain sliding speed. The cooling / heating load performance parameters inside and outside the air conditioner in each fault-free historical data window form a load parameter matrix of fault-free historical operating data, i.e., the second load parameter matrix. The principal component similarity factors between the first load parameter matrix and each second load parameter matrix are calculated one by one to obtain all principal component similarity factors between the first load parameter matrix and each second load parameter matrix.

[0023] All calculated principal component similarity factors are sorted from largest to smallest. Based on the pre-defined time dimension size of the fault-free historical operating performance reference dataset, which is an integer multiple of n (n can be 5, 10, 15, 20, etc.) of the sliding time window time dimension size (for example, the sliding time window time dimension size can be 30, 60, 90, 120, etc.), the top n second load parameter matrices with the largest principal component similarity factor values ​​are selected to form the fault-free historical operating performance reference dataset.

[0024] As redundant information, overlapping data in the fault-free historical performance reference dataset must be removed. At the positions of the removed overlapping data, the second loading parameter matrix of one or more subsequent principal component similarity factors, ordered sequentially, is then added until the entire fault-free historical performance reference dataset is filled.

[0025] Furthermore, in the above method, the reliability of the fault detection results is evaluated using two evaluation indicators: "probability of fault symptom detection" and "percentage of symptom occurrence time." This yields the reliability detection results of the fault symptoms for the air conditioning system faults, including:

[0026] The probability of detecting fault symptoms is defined as: the number of time points in a time window where fault symptoms are detected divided by the total number of time points in the entire time window, and then multiplied by 100%.

[0027] "Symptom occurrence time percentage" can be expressed as hourly symptom occurrence time percentage or daily symptom occurrence time percentage. The definitions are: the percentage of time a specific fault symptom was detected before the current moment, within one hour, or within one day, respectively. For example, hourly symptom occurrence time percentage is the percentage of time within 60 time points in one hour where fault symptoms were detected; daily symptom occurrence time percentage is the percentage of time within 1440 time points (60 × 24 = 1440 time points) in 24 hours where fault symptoms were detected.

[0028] For air conditioning systems without historical operating data, the probability of fault symptom detection obtained by the method of operating performance symptom rules is used as the final comprehensive fault symptom detection probability.

[0029] For air conditioning systems with historical operating data, the arithmetic mean of the following three probabilities is used as the final comprehensive fault symptom detection probability: the probability of fault symptom detection obtained by the statistical residual method, the probability of fault symptom detection obtained by the operating performance symptom rule method, and the probability of fault symptom detection obtained by the principal component detection method that matches similar historical operating data with air conditioning load patterns.

[0030] Based on the comprehensive fault symptom detection probability and the percentage of symptom occurrence time, the reliability detection results of the fault symptoms of the air conditioning system are obtained.

[0031] Furthermore, in the above method, based on the comprehensive fault symptom detection probability and the percentage of symptom occurrence time, the reliability detection result of the fault symptoms of the air conditioning system fault is obtained, including:

[0032] For the combined probability of fault symptom detection and the percentage of time when symptoms occur, if both are greater than or equal to the fault symptom discrimination threshold δ1, an alarm is triggered and a message "Fault symptom has occurred" is sent; if both are greater than or equal to the suspected fault symptom discrimination threshold δ2 and less than the fault symptom discrimination threshold δ1, an alarm is triggered and a message "Suspected fault symptom has occurred" is sent; if both are greater than or equal to the sub-healthy operating state discrimination threshold δ3 and less than the suspected fault symptom discrimination threshold δ2, an alarm is triggered and a message "Suspected fault symptom has occurred" is sent; if both are less than the sub-healthy operating state discrimination threshold δ3, no alarm is triggered; if one is greater than or equal to the suspected fault symptom discrimination threshold δ2, an alarm is triggered and a message "Sub-healthy operating state" is sent; if one is less than the suspected fault symptom discrimination threshold δ2 and the other is less than the sub-healthy operating state discrimination threshold δ3, no alarm is triggered.

[0033] Compared with existing technologies, this invention distinguishes between steady-state and non-steady-state operation based on key operating performance parameters. If the air conditioning system is judged to be non-steady-state, fault detection is terminated. If the air conditioning system is judged to be steady-state, for systems without historical operating data, a fault detection method based on statistical residuals of operating performance parameters and fault symptom rules based on operating performance is used to obtain fault detection results. For systems with historical operating data, a combined fault symptom detection method combining statistical residuals of operating performance parameters, fault symptom rules based on operating performance, and principal component analysis based on matching similar historical operating data with air conditioning load patterns is used to obtain fault detection results. The reliability of the fault detection results is evaluated using two indicators: "fault symptom detection probability" and "symptom occurrence time percentage," to obtain the reliability of the fault symptom detection results for air conditioning system faults. This invention is applicable to constant air volume, variable air volume, and 100% fresh air air conditioning systems for fault symptom detection and suspected fault symptom prediction. Through the comprehensive application of multiple technologies, this invention achieves automated and intelligent processing of air conditioning system fault symptom detection, improving system stability and reliability. Provide maintenance personnel with timely and accurate fault symptom detection information to achieve intelligent system operation performance symptom detection. Attached Figure Description

[0034] Figure 1This is an implementation flowchart of a combined fault symptom detection method based on sliding time window parameter statistical residual, performance symptom rules and principal components, according to an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram illustrating the applicable categories of three methods for fault symptom detection according to an embodiment of the present invention;

[0036] Figure 3 This is an implementation flowchart of constructing the operational performance symptom characteristics of an air conditioning system according to an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram of a fault symptom detection method based on operational performance symptom rules without historical operational data, according to an embodiment of the present invention.

[0038] Figure 5 This is a schematic diagram of a fault symptom detection method based on operational performance symptom rules with historical operational data, according to an embodiment of the present invention.

[0039] Figure 6 This is a flowchart of the training process for a dynamic threshold based on runtime performance symptom rules according to an embodiment of the present invention.

[0040] Figure 7 This is a schematic diagram of a principal component fault symptom detection method based on similar historical operating data matching air conditioning load patterns according to an embodiment of the present invention;

[0041] Figure 8 This is a schematic diagram of a fault-free historical operation performance database segmentation based on a sliding time window according to an embodiment of the present invention;

[0042] Figure 9 This is a schematic diagram illustrating the calculation principle of the principal component similarity factor of the current air conditioning load and the fault-free historical air conditioning load based on a sliding time window according to an embodiment of the present invention.

[0043] Figure 10 This is a flowchart illustrating the construction process of a reference dataset for matching the principal component similarity factor of air conditioning load with fault-free historical operating performance parameters according to an embodiment of the present invention.

[0044] Figure 11 This is a definition and downgrade processing of the reliability level of fault symptom detection according to an embodiment of the present invention. Detailed Implementation

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] like Figure 1 As shown, the present invention provides a method for detecting fault symptoms of an air conditioning system, comprising:

[0047] Step S1: Based on key operating performance parameters, determine whether the operating state is steady-state or unsteady-state. If the operating state of the air conditioning system is determined to be unsteady-state, terminate the fault detection of the air conditioning system; if the operating state of the air conditioning system is determined to be steady-state, proceed to step S2.

[0048] Preferably, step S1 involves determining whether the operating state is steady-state or unsteady-state based on key operating performance parameters, including:

[0049] Step S1.1, the determination of whether the key operating performance parameters are in a steady state or not, is achieved by judging whether the key operating performance parameters of the air conditioning system are stable within a reasonable steady-state operation determination threshold range;

[0050] The key operating performance parameters include: supply air temperature and supply air setpoint, supply air flow rate and flow demand (applicable to constant air volume variable air volume air conditioning systems), supply air pressure and supply air pressure setpoint (applicable to constant static pressure variable air volume air conditioning systems), supply water temperature and return water temperature, supply water flow rate and water flow rate predicted based on water valve opening.

[0051] Preferably, the recommended values ​​for the discrimination thresholds in the steady-state operation criterion correlation are: the difference between the supply air temperature and the supply air setpoint is less than or equal to 3℃; the difference between the supply air flow rate and the flow demand value is less than or equal to 10% of the maximum flow rate on the fan nameplate; the difference between the supply air pressure and the supply air pressure setpoint is less than or equal to 10% of the maximum static pressure on the fan nameplate; the difference between the supply water temperature and the return water temperature is less than or equal to 8℃; and the difference between the supply water flow rate and the predicted water flow rate (the water flow rate predicted based on the water valve opening) is less than or equal to 10% of the maximum flow rate on the water pump nameplate.

[0052] Step S2, selection of fault symptom detection method: For air conditioning systems without historical operating data, a fault symptom detection method based on statistical residuals of operating performance parameters and a fault symptom detection method based on operating performance symptom rules is used to obtain the detection results; for air conditioning systems with historical operating data, a combined fault symptom detection method is used, which combines the three methods: a fault symptom detection method based on statistical residuals of operating performance parameters, a fault symptom detection method based on operating performance symptom rules, and a fault symptom detection method based on principal component analysis of similar historical operating data matching air conditioning load patterns, to obtain the detection results.

[0053] Specifically, in order to determine whether the operating status of an air conditioning system is abnormal, it is necessary to extract features from the operating performance parameters of the air conditioning system and construct performance symptom features that characterize the operating status of the air conditioning system. These performance symptom features include: statistical residuals, performance symptom rules, and information matrix models.

[0054] Statistical residuals are calculated by comparing on-site collected data on the air conditioning system's operating performance variables, such as temperature, humidity, flow rate, pressure, CO2 concentration, and control signals, with their corresponding reference values ​​(e.g., setpoints, previous measurements, or measurements with similar functions). These residuals are obtained through simple subtraction, normalization, or standardization. Different statistical residuals are compared with preset thresholds to identify different fault symptoms.

[0055] The performance symptom rules are constructed based on the law of conservation of energy (i.e., the first law of thermodynamics), the steady flow energy equation of flow rate and pressure (i.e., Bernoulli's equation), the professional knowledge of air conditioning and control systems, the expert experience accumulated after long-term operation of equipment (such as the difference between the controlled parameter and the set value being within a certain range), actual operating performance characteristics (such as the blower power being within a certain range of its predicted value), and on-site inspections (such as dirt and blockage on the air side of the heat exchanger).

[0056] like Figure 3 As shown, the construction of the information matrix model is based on multiple conservation laws, including: the law of conservation of energy (i.e., the first law of thermodynamics), the constant flow energy equation of flow rate-pressure (i.e., Bernoulli's equation), etc.; then, a data-driven mathematical algorithm is used to solve the information matrix model to detect fault symptoms; wherein, the mathematical algorithm includes: data mining and machine learning, etc.

[0057] Preferably, step S2.1, as follows Figure 2 As shown, the fault symptom detection method based on statistical residuals of operating performance parameters includes:

[0058] Each operating performance parameter of the air conditioning system is compared with its corresponding reference value (such as set value, measured value at the previous moment, measured value with similar function, etc.) to obtain the corresponding statistical residual. Different statistical residuals are compared with preset thresholds to identify different fault symptoms.

[0059] like Figure 3 , 4 As shown in Figure 5, in step S2.2, the fault symptom detection method based on operational performance symptom rules,

[0060] The specific value of the discrimination threshold based on operational performance symptom rules is determined as follows: For systems without historical operational data, a fixed threshold is set by referencing the setpoints of existing similar air conditioning systems or based on expert experience; for systems with historical operational data, a dynamic threshold is obtained by training on the historical operational data (e.g., ...). Figure 6 (as shown) or an adaptive threshold, used as the discrimination threshold; different discrimination thresholds correspond to different fault symptoms;

[0061] like Figure 7 and8 As shown, step S2.3, the principal component fault symptom detection method based on matching similar historical operating data of air conditioning load patterns, includes:

[0062] The current time-window operating performance data and the filtered, fault-free historical operating performance reference dataset are used as input data for the principal component fault symptom detection model. The average prediction error and the control limit are calculated and compared. If the average prediction error is greater than the control limit, the air conditioning system is determined to have fault symptoms at the current time; if the average prediction error is less than the control limit, the air conditioning system is determined not to have fault symptoms at the current time.

[0063] Preferably, the principal component fault symptom detection model is an information matrix model, and the principal component analysis method is used.

[0064] Specifically, Principal Component Analysis (PCA) is a multivariate data analysis method, commonly used for dimensionality reduction and feature extraction. PCA can also be used for fault detection; its basic principle is to detect faults in a system by monitoring the main directions of change in the data. Below is the calculation flow of the PCA fault detection method:

[0065] 1. Calculation of covariance matrix:

[0066] Calculate the covariance matrix on the preprocessed data. The covariance matrix describes the linear relationship between the data and is the basis of principal component analysis.

[0067] 2. Calculation of eigenvalues ​​and eigenvectors:

[0068] The covariance matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and corresponding eigenvectors.

[0069] 3. Extraction of principal components (principal eigenvectors):

[0070] Based on the magnitude of the eigenvalues, the eigenvectors corresponding to the first few largest eigenvalues ​​are selected as principal components. Typically, the first few principal components with larger eigenvalues ​​are chosen to describe most of the variation in the data.

[0071] 4. Data projection:

[0072] The original data is projected onto the selected principal components to obtain the projected values ​​along the principal component directions. This is equivalent to mapping high-dimensional data into a low-dimensional space (principal component space).

[0073] 5. Calculate the squared prediction error:

[0074] Compare the original data with its projection in the principal component space, and calculate the squared prediction error (the difference between the original data and the projection).

[0075] 6. Determine control limits:

[0076] Calculate the load matrix of the filtered, fault-free historical parameter dataset, and then calculate the control limit as the threshold for fault detection.

[0077] 7. Fault detection:

[0078] The squared prediction error is compared with the control limit. If the squared prediction error exceeds the control limit, it indicates that the air conditioning system may have malfunction symptoms.

[0079] Preferably, step S2.3, the principal component fault symptom detection method based on matching similar historical operating data of air conditioning load patterns, includes:

[0080] Step S2.3.1: Construct a dataset of cooling / heating load performance parameters inside and outside the air conditioning system within the sliding time window at the current moment;

[0081] Here, a dataset of operating performance parameters in a sliding time window is used to address the lag or delayed response of the air conditioning system's operating status and performance parameters to dynamic changes in the building's internal and external cooling / heating loads.

[0082] Specifically, the internal cooling / heating load parameters of the air conditioner include: supply air flow rate, supply air temperature, return air temperature, supply air humidity, return air humidity, heat generation of electrical equipment, heat generation of lighting, and heat dissipation from personnel.

[0083] The external cooling / heating load parameters of the air conditioner include: fresh air dry bulb temperature, fresh air humidity, fresh air enthalpy, fresh air volume, normal solar radiation intensity, and horizontal solar radiation intensity.

[0084] Step S2.3.2, as follows Figure 7 As shown, the dataset of cooling / heating load performance parameters of the air conditioner inside and outside in the sliding time window at the current moment is compared with the cooling / heating load performance parameters of the air conditioner inside and outside in the fault-free historical operation performance database. The principal component similarity factor is calculated to perform similarity matching (calculate the principal component similarity factor). The fault-free historical operation performance data of the top n time windows with the largest similarity factor (n can be 5, 10, 15, 20, etc.) are selected as the reference dataset for the filtered fault-free historical operation performance parameters.

[0085] Figure 7 In this context, SPE represents the squared prediction error; δ represents the control limit; δ1 represents the fault symptom discrimination threshold; δ2 represents the suspected fault symptom discrimination threshold; and δ3 represents the sub-health operation status discrimination threshold.

[0086] First, the fault-free historical operating performance database contains: internal air conditioning load parameters characterizing the building load status (such as: supply air dry-bulb temperature, supply air relative humidity, supply air volume, return air dry-bulb temperature, return air relative humidity, chilled water supply flow rate, chilled water supply and return water temperature difference, heat generation of electrical equipment, heat generation of lighting, and heat dissipation of people, etc.) and external air conditioning load parameters (such as: fresh air dry-bulb temperature, fresh air relative humidity, fresh air volume, normal solar radiation intensity, horizontal solar radiation intensity, solar radiation heat gain, and solar radiation heat dissipation, etc.).

[0087] Secondly, the fault-free historical operating performance database also includes: operating performance parameters that characterize the operating status of the air conditioning system and are measured on-site, including parameters such as temperature, humidity, flow rate, pressure, power, current, setpoint, control signal, actuator feedback signal, switch signal, meteorological parameters, and operation and maintenance records.

[0088] The fault-free historical performance database is divided into multiple fault-free historical data windows according to a sliding time window of a certain time dimension (which can be the same as the time dimension of the current time window, and can be selected as 15, 30, 60 or 90 data points) and a sliding speed (which can be selected as 5, 10, 15 or 20 data points).

[0089] In reality, the current operating status and performance parameters of an air conditioning system are often delayed and attenuated by the dynamic changes in the cooling / heating load inside and outside the building. In other words, the operating performance parameters of the air conditioning system measured at the current moment do not reflect the dynamic changes in the cooling / heating load inside and outside the building at the current moment, but rather reflect the dynamic changes in the cooling / heating load inside and outside the building several minutes or even tens of minutes ago, and their values ​​are usually lower than the maximum value of the dynamic load change.

[0090] like Figure 8 and 9 As shown, to minimize the delayed or attenuated response of the measured parameters to changes in cooling / heating load, all data uses a dataset spanning a time window, rather than a set of data at a specific point in time. Here, a time window is defined as the temporal dimension of the dataset. Performance parameters characterizing the current operating state of the air conditioning system are represented by the temporal dimension of the current time window dataset (the set of measured data at the current point in time and the set of measured data from the immediately preceding time window). This time dimension can be selected as 15, 30, 60, or 90 data points. The sampling period for each data point can be 0.5 min, 1 min, or 5 min.

[0091] The time dimension of the fault-free historical operating performance reference dataset (i.e., measurement data within a time window, not a single point in time) is defined as a sliding time window size that is an integer multiple of n (n can be 5, 10, 15, 20, etc.). For example, the sliding time window size could be 30, 60, 90, 120, etc. For instance, assuming a sliding time window has 60 time points, and the sampling period of the air conditioning system's measuring equipment is 1 minute, then one time point corresponds to 1 minute, and 60 time points in a sliding time window correspond to 60 minutes. If the time dimension of the fault-free historical operating performance reference dataset is defined as an integer n = 15 times the sliding time window size, then the fault-free historical operating performance reference dataset contains 60 × 15 = 900 time points.

[0092] The preferred detailed steps for constructing a fault-free historical performance reference dataset are as follows:

[0093] Ⅰ) such as Figure 9 , 10 As shown, the principal component similarity factor matching is performed on the fault-free historical operating performance reference dataset: The load parameter matrix at the current time window is obtained by measuring the cooling / heating load performance parameters inside and outside the air conditioner in the current time window data, i.e., the first load parameter matrix. Using a sliding time window (whose time dimension is equal to the time dimension of the current time window), the filtered fault-free historical operating performance parameter reference dataset is divided into numerous fault-free historical data windows at a certain sliding speed. The cooling / heating load performance parameters inside and outside the air conditioner in each fault-free historical data window form a load parameter matrix of fault-free historical operating data, i.e., the second load parameter matrix. The principal component similarity factor between the first load parameter matrix and each second load parameter matrix is ​​calculated one by one to obtain all principal component similarity factors between the first load parameter matrix and each second load parameter matrix.

[0094] Ⅱ) Sort all the calculated principal component similarity factors from largest to smallest. Based on the time dimension size of the pre-set fault-free historical operation performance reference dataset, that is, the time dimension size of the sliding time window that is an integer multiple of n (n can be 5, 10, 15, 20, etc.), select the top n second load parameter matrices with the largest principal component similarity factor values ​​(i.e. the highest similarity) to form the fault-free historical operation performance reference dataset.

[0095] III) As redundant information, overlapping data in the fault-free historical operating performance reference dataset obtained in II) above must be removed. At the positions of the removed overlapping data, sequentially supplement the dataset with the second load parameter matrix of one or more of the largest subsequent principal component similarity factors, until the entire fault-free historical operating performance reference dataset is filled.

[0096] Specifically, if the internal and external cooling / heating loads of an air-conditioned area are similar, then the operating performance of the air conditioning system should be close to or similar. Based on this principle, the principal component similarity factor is calculated between the current time-window measurement data and the air conditioning load parameters in each segmented fault-free historical data window, thus obtaining the degree of similarity between these data. All calculated principal component similarity factors are sorted from largest to smallest, and the operating performance data from the fault-free historical data window with the largest similarity factor is selected to construct a filtered fault-free historical operating performance reference dataset. This filtered fault-free historical operating performance dataset, used as reference data, is input together with the current time-window measurement data into the principal component fault symptom detection model.

[0097] Step S3: Evaluate the reliability of the fault detection results by using two evaluation indicators: "probability of fault symptom detection" and "percentage of symptom occurrence time" to obtain the reliability of the fault symptom detection results of the air conditioning system fault.

[0098] Here, the credibility detection result, i.e., the credibility level of the occurrence of fault symptoms, adopts two evaluation indicators: the probability of fault symptom detection and the percentage of time when symptoms appear. The credibility level of the occurrence of fault symptoms can be divided into four categories: "fault symptoms appear", "suspected fault symptoms appear", "sub-healthy operating state", and "no fault symptoms".

[0099] The confidence level of a fault symptom is defined based on the probability of fault symptom detection. The probability of fault symptom detection is defined as follows: regardless of whether the fault symptom detection method uses statistical residuals based on operating performance parameters, operating performance symptom rules, or principal component analysis of similar historical operating data based on air conditioning load patterns, the probability of fault symptom detection is obtained by dividing the number of time points in a time window that detect fault symptoms by the total number of time points in the entire time window, and then multiplying by 100%. This is the probability of a fault symptom occurring. If the probability of detecting fault symptoms is greater than or equal to the fault symptom discrimination threshold δ1 (e.g., its value is 80%), it is defined as "fault symptoms appear". If the probability of detecting fault symptoms is greater than or equal to the suspected fault symptom discrimination threshold δ2 (e.g., its value is 60%) and less than the fault symptom discrimination threshold δ1 (e.g., its value is 80%), it is defined as "suspected fault symptoms appear". If the probability of detecting fault symptoms is greater than or equal to the sub-health operation state discrimination threshold δ3 (e.g., its value is 40%) and less than the suspected fault symptom discrimination threshold δ2 (e.g., its value is 60%), it is defined as "sub-health operation state". If the probability of detecting fault symptoms is less than the sub-health operation state discrimination threshold δ3 (e.g., its value is 40%), it is defined as "no fault symptoms" or "no fault symptoms appear".

[0100] For air conditioning systems without historical operating data, the probability of fault symptom detection obtained by the operation performance symptom rule method will be used as the final comprehensive fault symptom detection probability. For air conditioning systems with historical operating data, the arithmetic mean of the probability of fault symptom detection obtained by the statistical residual method, the probability of fault symptom detection obtained by the operation performance symptom rule method, and the probability of fault symptom detection obtained by the principal component detection method that matches similar historical operating data with air conditioning load patterns will be used as the final comprehensive fault symptom detection probability.

[0101] The confidence level of a malfunction symptom is defined based on the percentage of time the symptom occurs. This can be achieved using either hourly or daily percentages of symptom occurrence.

[0102] The percentage of time a symptom appears is defined as the percentage of time before the current moment, within one hour, when a certain fault symptom is detected. For example, the percentage of time within 60 time points in one hour when a fault symptom is detected.

[0103] The daily symptom occurrence percentage is defined as the percentage of a specific fault symptom detected within a day prior to the current moment. For example, it represents the percentage of time within 1440 time points (60 x 24 = 1440) during which a fault symptom was detected.

[0104] If the percentage of time during which symptoms occur hourly is greater than or equal to the fault symptom discrimination threshold δ1 (e.g., 80%), it is defined as "fault symptom present". If the percentage of time during which symptoms occur hourly is greater than or equal to the suspected fault symptom discrimination threshold δ2 (e.g., 60%) and less than the fault symptom discrimination threshold δ1 (e.g., 80%), it is defined as "suspected fault symptom present". If the percentage of time during which symptoms occur hourly is greater than or equal to the sub-healthy operating state discrimination threshold δ3 (e.g., 40%) and less than the suspected fault symptom discrimination threshold δ2 (e.g., 60%), it is defined as "sub-healthy operating state". If the percentage of time during which symptoms occur hourly is less than the sub-healthy operating state discrimination threshold δ3 (e.g., 40%), it is defined as "no fault symptoms". The daily percentage of time during which symptoms occur is pushed to users as a statistical value within a day for reference.

[0105] like Figure 11 As shown, when used in combination: the probability of detecting a comprehensive fault symptom and the percentage of time when the symptom occurs must simultaneously meet the same level of confidence level for the occurrence of the fault symptom in order to send the confidence level of the fault symptom at that level. If only one of them meets the confidence level, but the other is lower than the confidence level of this fault symptom, then a downgrade is performed, reducing the confidence level of the fault symptom, while also considering the confidence level of the fault symptom at a lower level.

[0106] For example, regarding the combined probability of detecting fault symptoms and the percentage of time when symptoms occur, if both are simultaneously greater than or equal to the fault symptom discrimination threshold δ1 (e.g., its value is 80%), an alarm is triggered and a message "Fault symptoms have occurred" is sent; if both are simultaneously greater than or equal to the suspected fault symptom discrimination threshold δ2 (e.g., its value is 60%) and less than the fault symptom discrimination threshold δ1 (e.g., its value is 80%), an alarm is triggered and a message "Suspected fault symptoms have occurred" is sent; if both are simultaneously greater than or equal to the sub-health operation state discrimination threshold δ3 (e.g., its value is 40%) and less than the suspected fault symptom discrimination threshold δ2, an alarm is triggered and a message "Suspected fault symptoms have occurred" is sent; If the fault symptom discrimination threshold δ2 (e.g., its value is 60%), an alarm will be triggered and a message "Suspected fault symptom has appeared" will be sent. If both thresholds are less than the sub-healthy operating state discrimination threshold δ3 (e.g., its value is 40%), no alarm will be triggered. If one threshold is greater than or equal to the suspected fault symptom discrimination threshold δ2 (e.g., its value is 60%), an alarm will be triggered and a message "Sub-healthy operating state" will be sent. If one threshold is less than the suspected fault symptom discrimination threshold δ2 (e.g., its value is 60%) and the other threshold is less than the sub-healthy operating state discrimination threshold δ3 (e.g., its value is 40%), no alarm will be triggered.

[0107] In summary, this invention relates to a combined fault symptom detection method based on three elements: statistical residuals of operating performance parameters using a sliding time window, symptom rules of operating performance characteristics, and principal component analysis of similar historical operating data based on load patterns. Specific technical measures include: performance symptom rules characterizing the operating state of the air conditioning system; online training of fixed and dynamic thresholds; discrimination of steady-state or non-steady-state operating states of key operating performance parameters; a sliding time window to address the lag or delayed response of air conditioning operating states and performance parameters to dynamic changes in internal and external cooling / heating loads; a reference dataset of fault-free historical operating performance parameters matched with principal component similarity factors of air conditioning loads; a fault symptom detection method based on operating performance symptom rules; a principal component fault symptom detection method based on similar historical operating data based on air conditioning load patterns; and a reliability level of the fault symptom detection results.

[0108] To mitigate the lag and delay in the response of the air conditioning system to external and internal loads, all data in this application are collected within a sliding time window (e.g., 15 min, 30 min, 60 min, 90 min, or others).

[0109] For air conditioning systems without historical operating data, a fault symptom detection method based on statistical residuals of operating performance parameters and fault symptom rules based on operating performance is adopted. For existing air conditioning systems with historical operating data, a combined fault symptom detection method based on statistical residuals of operating performance parameters, fault symptom detection method based on operating performance symptom rules, and principal component analysis based on air conditioning load pattern matching of similar historical operating data is adopted.

[0110] In the fault symptom detection method based on operational performance symptom rules in this application, the specific value of the discrimination threshold used to determine whether a fault symptom has occurred is determined as follows: for systems without historical operational data, the set value of existing similar air conditioning systems is referenced, or a fixed threshold is set based on expert experience; for systems with historical operational data, a dynamic threshold or an adaptive threshold is obtained by training on the historical operational data.

[0111] In the principal component fault symptom detection method of this application, the reference dataset of fault-free historical operating performance parameters is matched by the principal component similarity factor of the air conditioning load. The obtained fault-free historical operating performance reference dataset and the time window measurement data at the current moment are input together into the principal component fault symptom detection model. The magnitude of the prediction average error and the control limit are calculated and compared to detect fault symptoms.

[0112] This invention is applicable to air conditioning systems with constant air volume, variable air volume, and 100% fresh air intake, and is used for fault symptom detection and suspected fault symptom prediction. Through the comprehensive application of multiple technologies, this invention achieves automated and intelligent processing of air conditioning system fault symptom detection, improving system stability and reliability. It provides maintenance personnel with timely and accurate fault symptom detection information, realizing intelligent detection of system operational performance symptoms.

[0113] The following is combined Figure 1 , 2 Sections 3, 4, 5, 6, 7, 8, 9, 10, and 11 further elaborate on the implementation of the present invention.

[0114] Combination Figure 1 As shown, the extraction of operational performance symptom features of the air conditioning system employs statistical residuals of operational performance parameters, operational performance symptom rules, and an information matrix model. Operational data of the air conditioning system is collected, including key operational performance parameters and characteristic indicators such as temperature, humidity, flow rate, pressure, CO2 concentration, and control signals. Data processing, filtering, and noise reduction methods are used to extract the operational performance symptom features of the air conditioning system.

[0115] Combination Figure 1 As shown, the steady-state / non-steady-state discrimination of the air conditioning system involves determining whether the changes in key operating performance parameters under different operating conditions fall within a reasonable steady-state discrimination threshold range, thereby determining whether the current operating state of the air conditioning system is steady. Fault symptom detection can only be performed when the current operating state of the air conditioning system is steady.

[0116] In this embodiment, combined with Figure 1 , Figure 2 and Figure 3 As shown, statistical residuals are calculated statistically (including mean, standard deviation, maximum, minimum, etc.), in the frequency domain (e.g., power spectral density), and time domain (e.g., autocorrelation function) of air conditioning system operating performance variables collected on-site, such as temperature, humidity, flow rate, pressure, CO2 concentration, and control signals. A baseline statistical model is established by fitting a probability distribution model (e.g., Gaussian distribution) or using nonparametric methods (e.g., K-nearest neighbors algorithm) to obtain the probability distribution function of each feature variable. The collected real-time data is compared with the baseline model to calculate the residuals. The residuals are the differences between the actual observed values ​​and the predicted values ​​of the baseline model. Various methods can be used to calculate the residuals, including simple subtraction, normalization, or standardization.

[0117] In this embodiment, combined with Figure 1 and Figure 3As shown, statistical residuals based on operating performance parameters are mainly used for sensor fault identification, including sensor measurement signal limit out-of-bounds faults, sensor measurement signal stagnation faults, and sensor measurement signal deviation faults. Sensor measurement signal limit out-of-bounds faults are based on reasonable ranges of operating performance parameters collected from the user side. For stagnation faults, the sensor measurement signal shows no change within a certain period (e.g., 30–120 minutes), but the maximum or minimum limit values ​​should be excluded.

[0118] In this embodiment, combined with Figure 1 and Figure 3 As shown, the operational performance symptom rules employ a complete set of operational performance feature selection and feature extraction methods. Based on the law of conservation of energy (i.e., the first law of thermodynamics), the steady flow energy equation of flow rate and pressure (i.e., Bernoulli's equation), professional knowledge of air conditioning and control systems, expert experience accumulated after long-term operation of equipment (such as the difference between the controlled parameter and the set value being within a certain range), actual operational performance characteristics (such as the blower power being within a certain range of its predicted value), and on-site inspections (such as dirt and blockage on the air side of the heat exchanger), etc., the corresponding operational performance symptom rules are constructed by integrating multiple operational performance parameters.

[0119] In this embodiment, combined with Figure 1 and Figure 4 As shown, for systems without historical operational data, fault symptom detection is performed using statistical residuals based on operational performance parameters and operational performance symptom rules; combined with... Figure 1 and Figure 4 As shown, for systems with historical operating data, a combined approach is used to detect fault symptoms, which combines three methods: statistical residuals based on operating performance parameters, rules based on operating performance symptoms, and principal component analysis based on matching similar historical operating data with air conditioning load patterns.

[0120] In this embodiment, combined with Figure 4 , Figure 5 and Figure 6 As shown, the threshold for judging operational performance symptoms should be set according to different situations. For systems without historical operational data, the set values ​​of existing similar air conditioning systems can be referenced, or a fixed threshold can be set based on expert experience; for systems with historical operational data, dynamic thresholds or adaptive thresholds can be obtained by training on historical operational data.

[0121] In this embodiment, combined with Figure 1 and Figure 7 As shown, in order to mitigate the lag or delay in the response of the air conditioning system's operating status and performance parameters to the dynamic changes in the building's internal and external cooling / heating loads, all data are datasets within a time window, i.e., sliding time windows, rather than data sets at a specific point in time.

[0122] In this embodiment, combined with Figure 1 and Figure 7 As shown, in the principal component analysis (PCA) fault symptom detection based on similar historical operating data of air conditioning load pattern matching, similar historical operating data of pattern matching with a sliding time window is used as the input matrix of the PCA fault symptom detection calculation model based on similar historical operating data of air conditioning load pattern matching. To minimize the lag and delay (generally less than 10 minutes) of the dynamic response of the air conditioning system's operating performance parameters to internal and external load parameters, a parameter matrix composed of a sliding time window (the recommended window size, i.e., the time dimension size, is 10 minutes to 60 minutes) is used. The historical reference data is selected from historical operating data similar to the time window measurement data at the current moment.

[0123] In this embodiment, combined with Figure 8 As shown, the sliding time window used has the same time dimension size as the current time window, selected as 15, 30, 60, and 90 data points. In the fault-free historical performance database, the sliding time window moves forward at a sliding rate w (e.g., 5, 10, 15, and 20 data points), dividing the database into n fault-free historical performance data windows of equal time dimension size: S1, S2, ..., S... i ... S n The sliding rate *w* can be understood as: for every *w* data points the sliding time window moves forward, that is, the sliding time window skips *w* time point observations. The value of the sliding rate *w* directly affects both the computational complexity of the reference dataset for matching the fault-free historical operating performance parameters of the principal component similarity factor of the air conditioning load and the fault symptom detection accuracy of the principal component fault symptom detection method based on similar historical operating data of air conditioning load pattern matching. An excessively large or small sliding rate *w* may lead to a high rate of missed alarms and false alarms in the fault symptom detection results.

[0124] In this embodiment, combined with Figure 8 and Figure 9 As shown, in the fault-free historical performance database segmentation based on a sliding time window, the sliding time window slides across the historical database, with the sliding rate w selected as 1 / 12 to 1 / 6 of the sliding window size. Each time the window slides, a similarity factor is calculated, and all similarity factors are sorted in descending order. The window with the largest similarity factor is selected to form the historical reference data. If duplicate data appears in the historical reference data, it must be deleted and new data added. New data is then selected sequentially from windows with larger similarity factor values ​​to replace the duplicate data.

[0125] In this embodiment, combined with Figure 8 and Figure 9 As shown, based on pattern matching and similarity in historical operational data, the principal component similarity factor is used to characterize the principal component similarity factor between the air conditioning load measurement data of the current time window and the air conditioning load database obtained after segmentation from the fault-free historical operational performance data for each sliding time window. The sliding time window divides the air conditioning load database in the fault-free historical operational performance into n data windows with the same time dimension as the current time window. The principal component similarity factor between the current time window and each sliding time window of the fault-free historical operational performance is calculated.

[0126] In this embodiment, combined with Figure 10 As shown, in the reference dataset for matching the principal component similarity factors of air conditioning load with fault-free historical operating performance parameters, all principal component similarity factors are sorted from largest to smallest to obtain the similarity ranking of historical data windows. The operating performance data of the top n fault-free historical sliding time windows corresponding to the maximum values ​​of the top n principal component similarity factors are selected to form the reference dataset for fault-free historical operating performance parameters. If duplicate data appears in the reference dataset for fault-free historical operating performance parameters, it must be deleted and new data added. New data is selected sequentially from windows with larger similarity factors to replace duplicate data.

[0127] In this embodiment, combined with Figure 10 As shown, the principal component similarity factor is an important metric parameter for the reference dataset of fault-free historical operating performance parameters for matching air conditioning loads. It characterizes the similarity between the current time-window air conditioning load measurement data and the fault-free historical sliding time-window air conditioning load data. The current time-window air conditioning load measurement data is defined as S, and the fault-free historical sliding time-window air conditioning load data is defined as H, both consisting of n variables and m data points. Assuming k1 and k2 are the principal components of S and H respectively, the first k principal components are selected to form eigenvector matrices with spaces L and M, and the principal component similarity factor is calculated.

[0128]

[0129] In the formula: S PCA —Principal component similarity factor;

[0130] trace — used to find the trace of a two-dimensional square matrix, that is, the sum of the diagonal elements of the matrix;

[0131] L—The principal component space of the current measurement data;

[0132] M—Principal component space of historical operational reference data;

[0133] k — number of principal components.

[0134] In this embodiment, combined with Figure 10 As shown, air conditioning load data mainly includes internal and external loads. Internal load parameters mainly include cooling loads generated by occupants' heat dissipation, cooling loads generated by lighting, and equipment loads; external load parameters mainly include outdoor air enthalpy, outdoor air temperature, fresh air volume, and solar radiation intensity. Due to limitations in actual measurement, parameters reflecting air conditioning load, such as horizontal / normal solar radiation intensity, equipment or lamp power, and the number of building users, may not be available. Therefore, external parameters can be selected as fresh air volume, fresh air dry-bulb temperature, and fresh air relative humidity. Internal parameters can be selected as temperature, humidity, and air volume; the enthalpy of moist air can be calculated from the dry-bulb temperature and relative humidity.

[0135] In this embodiment, combined with Figure 11 As shown, both the probability of detecting a fault symptom and the percentage of time the symptom occurs must simultaneously meet the same level of confidence for the fault symptom to be sent. If only one meets the confidence level, but the other is lower, a downgrade is performed, lowering the confidence level of the fault symptom and considering a lower-level confidence level. For example, if both the probability of detecting a fault symptom and the percentage of time the symptom occurs are greater than or equal to 80%, an alarm is triggered and a "fault symptom detected" message is sent; if both are greater than or equal to 60% and less than 80%, an alarm is triggered and a "suspected fault symptom detected" message is sent; if both are greater than or equal to 40% and less than 60%, an alarm is triggered and a "suspected fault symptom detected" message is sent; if both are less than 40%, no alarm is triggered. If one is greater than or equal to 60%, an alarm is triggered and a "sub-healthy operating state" message is sent; if one is less than 60% and the other is less than 40%, no alarm is triggered.

[0136] 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.

[0137] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0138] Obviously, those skilled in the art can make various modifications and variations to the invention without departing from the spirit and scope of the invention. Therefore, if these modifications and variations fall within the scope of the claims of the invention and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A method for detecting fault symptoms in an air conditioning system, characterized in that, include: Based on key operating performance parameters, the steady-state or non-steady-state operation of the air conditioning system is determined. If the operating state of the air conditioning system is determined to be non-steady-state, the fault detection of the air conditioning system is terminated. If the operating state of the air conditioning system is determined to be steady state, for air conditioning systems without historical operating data, a fault detection method based on statistical residuals of operating performance parameters and fault symptom rules based on operating performance is used to obtain fault detection results; for air conditioning systems with historical operating data, a combined fault symptom detection method combining statistical residuals of operating performance parameters, fault symptom rules based on operating performance, and principal component analysis based on air conditioning load pattern matching of similar historical operating data is used to obtain fault detection results. To evaluate the reliability of the fault detection results, two evaluation indicators, "probability of fault symptom detection" and "percentage of symptom occurrence time", were used to obtain the reliability of the fault symptom detection results of the air conditioning system fault. The reliability of fault detection results is evaluated using two indicators: "probability of fault symptom detection" and "percentage of symptom occurrence time." The resulting reliability of fault symptom detection for the air conditioning system includes: The "probability of detecting fault symptoms" is defined as: the number of time points in a time window where fault symptoms are detected, divided by the total number of time points in the entire time window, and then multiplied by 100%. "Symptom occurrence time percentage" can be expressed as hourly symptom occurrence time percentage or daily symptom occurrence time percentage. The definitions of the two are: the percentage of time before the current moment, within one hour, or within one day when a certain fault symptom was detected. For air conditioning systems without historical operating data, the probability of fault symptom detection obtained by the method of operating performance symptom rules is used as the final comprehensive fault symptom detection probability. For air conditioning systems with historical operating data, the arithmetic mean of the following three probabilities is used as the final comprehensive fault symptom detection probability: the probability of fault symptom detection obtained by the statistical residual method, the probability of fault symptom detection obtained by the operating performance symptom rule method, and the probability of fault symptom detection obtained by the principal component detection method that matches similar historical operating data with air conditioning load patterns. Based on the comprehensive fault symptom detection probability and the percentage of symptom occurrence time, the reliability detection results of the fault symptoms of the air conditioning system are obtained. Based on the comprehensive fault symptom detection probability and symptom occurrence time percentage, the reliability detection results of the fault symptoms of the air conditioning system are obtained, including: For a combination of the probability of detecting fault symptoms and the percentage of time when symptoms occur, if both are greater than or equal to the fault symptom discrimination threshold... δ If 1, an alarm is triggered and a message "Fault symptoms have occurred" is sent; if both are greater than or equal to the suspected fault symptom discrimination threshold, an alarm is triggered. δ 2. And less than the fault symptom discrimination threshold. δ If 1, an alarm is triggered and a message "Suspected fault symptoms have appeared" is sent; if both are simultaneously greater than or equal to the threshold for determining a sub-healthy operating state... δ 3. And it is less than the threshold for judging suspected fault symptoms. δ 2. If an alarm is triggered and a message indicating "suspected fault symptoms have appeared" is sent, then if both are below the threshold for determining a sub-healthy operating state... δ If 3 is true, no alarm will be triggered; if one of them is greater than or equal to the suspected fault symptom discrimination threshold, no alarm will be triggered. δ 2. If an alarm is triggered and a "sub-healthy operating status" message is sent, then if one of the conditions is less than the suspected fault symptom discrimination threshold... δ 2. And another threshold is less than the threshold for judging a sub-healthy operating state. δ 3. In this case, no alarm will be triggered.

2. The method for detecting fault symptoms of an air conditioning system as described in claim 1, characterized in that, The key operating performance parameters include: supply air temperature and supply air setpoint, supply air flow rate and flow demand, supply air pressure and supply air pressure setpoint, supply water temperature and return water temperature, and supply water flow rate and predicted water flow rate. The steady-state criteria for judging the operating status of the air conditioning system include: the difference between the supply air temperature and the supply air setpoint is less than or equal to 3°C; the difference between the supply air flow rate and the flow demand is less than or equal to 10% of the maximum nameplate fan flow rate; the difference between the supply air pressure and the supply air pressure setpoint is less than or equal to 10% of the maximum static pressure of the fan nameplate; the difference between the supply water temperature and the return water temperature is less than or equal to 8°C; and the difference between the supply water flow rate and the predicted water flow rate is less than or equal to 10% of the maximum flow rate of the water pump nameplate.

3. The method for detecting fault symptoms of an air conditioning system as described in claim 1, characterized in that, Fault symptom detection methods based on statistical residuals of operating performance parameters include: Each operating performance parameter of the air conditioning system is compared with its corresponding reference value, and the corresponding statistical residuals are calculated, including simple subtraction, normalization or standardization. Different statistical residuals are compared with preset thresholds to identify different fault symptoms.

4. The method for detecting fault symptoms of an air conditioning system as described in claim 1, characterized in that, Fault symptom detection methods based on runtime performance symptom rules include: For systems without historical operating data, the set values ​​of existing similar air conditioning systems or fixed thresholds set based on expert experience are used as the discrimination thresholds. For systems with historical operating data, dynamic or adaptive thresholds obtained by training on historical operating data are used as the discrimination thresholds. Different discrimination thresholds correspond to different fault symptoms.

5. The method for detecting fault symptoms of an air conditioning system as described in claim 1, characterized in that, Principal component analysis (PCA) fault symptom detection methods based on matching similar historical operating data with air conditioning load patterns include: The current time window's operating performance data and the filtered, fault-free historical operating performance reference dataset are used as input data for the principal component fault symptom detection model. The average prediction error and the control limit are calculated and compared. If the average prediction error is greater than the control limit, the air conditioning system is determined to have fault symptoms at the current time. If the average prediction error is less than the control limit, the air conditioning system is determined not to have fault symptoms at the current time.

6. The method for detecting fault symptoms of an air conditioning system as described in claim 5, characterized in that, Before using the current time window's operational performance data and the filtered, fault-free historical operational performance reference dataset as input data for the principal component fault symptom detection model, the following steps are also included: Construct a dataset of cooling / heating load performance parameters inside and outside the air conditioning system within a sliding time window at the current moment; The dataset of cooling / heating load performance parameters of the air conditioner's internal and external systems within the current sliding time window is compared with the cooling / heating load performance parameters of the air conditioner's internal and external systems in the fault-free historical operation performance database. Principal component similarity factors are calculated to perform similarity matching, and the dataset with the largest similarity factor is selected. n The historical operational performance data without faults within a time window is used as a reference dataset for the filtered historical operational performance parameters without faults.

7. The method for detecting fault symptoms of an air conditioning system as described in claim 6, characterized in that, The internal cooling / heating load parameters of the air conditioner include: supply air flow rate, supply air temperature, return air temperature, supply air humidity, return air humidity, heat generation of electrical equipment, heat generation of lighting, and heat dissipation from personnel. The external cooling / heating load parameters of the air conditioner include: fresh air dry bulb temperature, fresh air humidity, fresh air enthalpy, fresh air volume, normal solar radiation intensity, and horizontal solar radiation intensity.

8. The method for detecting fault symptoms of an air conditioning system as described in claim 6, characterized in that, The dataset of cooling / heating load performance parameters of the air conditioner's internal and external systems within the current sliding time window is compared with the cooling / heating load performance parameters of the air conditioner's internal and external systems in the fault-free historical operation performance database. Principal component similarity factors are calculated to perform similarity matching, and the dataset with the largest similarity factor is selected. n The historical operational performance data without failures within a certain time window is used as a reference dataset for the filtered historical operational performance parameters without failures, including: The first load parameter matrix is ​​obtained by measuring the cooling / heating load performance parameters inside and outside the air conditioner in the current time window measurement data. A sliding time window is used, with its time dimension equal to that of the current time window. The filtered dataset of fault-free historical operating performance parameters is divided into numerous fault-free historical data windows at a certain sliding speed. The cooling / heating load performance parameters inside and outside the air conditioner in each fault-free historical data window form a second load parameter matrix. The principal component similarity factors between the first load parameter matrix and each second load parameter matrix are calculated one by one to obtain all principal component similarity factors between the first load parameter matrix and each second load parameter matrix. All calculated principal component similarity factors are sorted from largest to smallest, and then the time dimension of the reference dataset for fault-free historical operation performance is set to an integer value. n The sliding time window is times the size of the time dimension, and the components with the largest principal component similarity factor values ​​are selected. n A second load parameter matrix is ​​used to form a fault-free historical operating performance reference dataset. As redundant information, overlapping data in the fault-free historical operating performance reference dataset are removed; the second loading parameter matrix of one or more subsequent principal component similarity factors, which are ranked next to the removed overlapping data, is then added in order until the entire fault-free historical operating performance reference dataset is filled.

Images