A fault detection method based on markov process characteristic statistics
By employing a fault detection method based on Markov process characteristic statistics, data estimation and real-time anomaly analysis are performed on various target attributes of HVAC systems. This solves the problems of insufficient rapid diagnosis and interpretability in existing technologies, and achieves efficient and accurate fault detection.
Patent Information
- Application Number
- CN202311270885.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-28
- Publication Date
- 2026-11-03
- Estimated Expiration
- 2043-09-28
AI Technical Summary
Existing HVAC fault diagnosis methods are mainly offline, which makes it difficult to achieve rapid diagnosis, and deep learning methods lack interpretability, resulting in poor performance in practical applications.
A fault detection method based on Markov process characteristic statistics is adopted. By training a data estimation model on various target attributes of the target machine, the number of consecutive abnormal alarms is calculated in real time, and the fault is determined by confidence interval and probability statistics.
It improves the accuracy and efficiency of HVAC fault detection, reduces misjudgments, and enables rapid and accurate fault detection.
Smart Images
Figure CN117332232B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a fault detection method based on Markov process characteristic statistics, belonging to the field of data statistics technology. Background Technology
[0002] In HVAC systems, approximately 42% of cooling energy consumption and 26% of maintenance costs are caused by equipment failure. It is estimated that HVAC malfunctions lead to energy waste and failure to meet human comfort needs. Therefore, fault diagnosis in HVAC systems is crucial. Current fault diagnosis methods are primarily offline, which makes rapid fault diagnosis difficult. Furthermore, due to their high diagnostic accuracy, deep learning methods are widely used in HVAC fault diagnosis; however, these deep models are essentially data-driven and lack interpretation of diagnostic results, resulting in poor practical application performance. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a fault detection method based on Markov process characteristic statistics, which analyzes the errors of various target attributes related to the fault and uses the number of consecutive abnormal alarms as the basis to realize the detection of target machine faults.
[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention designs a fault detection method based on Markov process characteristic statistics. Based on various target attributes related to faults on the target machine, steps A to B are executed for each target attribute to obtain the data estimation model and the number of consecutive abnormal alarms corresponding to each target attribute. Then, steps i to v are executed in real time for each target attribute. If there is a target attribute that reaches the corresponding number of consecutive abnormal alarms, it is determined that the target machine has a fault; otherwise, it is determined that the target machine does not have a fault.
[0005] Step A. Based on the various analytical attributes related to the target attribute on the target machine, and using historical normal working data, train the preset neural network to obtain a data estimation model for calculating the normal estimation data of the target attribute;
[0006] Step B. Apply the data estimation model to obtain continuous historical normal estimation data of the target attribute, and obtain the continuous difference between it and the continuous historical normal detection data at the corresponding time. Combined with the preset false alarm probability upper limit, calculate the number of continuous abnormal alarms corresponding to the target attribute.
[0007] Step i. Initialize the target attribute exception count to 0, and proceed to step ii;
[0008] Step ii. Based on the data of each analysis attribute related to the target attribute at the current time, apply the data estimation model corresponding to the target attribute to process it, obtain the normal estimation data of the target attribute at the current time, and combine it with the detection data of the target attribute at the current time to obtain the difference corresponding to the target attribute at the current time, and then proceed to step iii.
[0009] Step iii. Based on the preset confidence estimation range value L, according to the difference of the target attribute from the previous time to the historical time direction L times, obtain the lower confidence limit and upper confidence limit corresponding to the target attribute at the current time, and form the corresponding confidence interval, and then proceed to step iv.
[0010] Step iv. Determine whether the difference corresponding to the target attribute at the current moment has jumped out of its corresponding confidence interval. If yes, increment the number of abnormalities of the target attribute by 1 and proceed to step v; otherwise, wait to enter the next moment and return to step i.
[0011] Step v. Determine whether the number of abnormal occurrences of the target attribute is greater than or equal to the number of consecutive abnormal alarms corresponding to the target attribute. If yes, determine that the target attribute has reached the corresponding number of consecutive abnormal alarms, and wait to enter the next moment, then return to step i; otherwise, wait to enter the next moment, then return to step ii.
[0012] As a preferred technical solution of the present invention: step A includes the following steps A1 to A2;
[0013] Step A1. Use correlation analysis to obtain the analysis attributes that are related to the target attribute among the preset attributes on the target machine, and then proceed to step A2;
[0014] Step A2. Obtain the analytical attribute data and target attribute data at each time point under the preset historical period corresponding to the normal operation of the target machine. Using the analytical attribute data as input and the target attribute data as output, train the preset neural network to obtain the data estimation model corresponding to the target attribute, which is used to calculate the normal estimation data of the target attribute.
[0015] As a preferred embodiment of the present invention: step B includes steps B1 to B7;
[0016] Step B1. Based on the various analytical attributes related to the target attribute, and the values of each analytical attribute at each consecutive time point under the preset historical period corresponding to the normal operation of the target machine, apply the data estimation model to calculate the normal estimated data of the target attribute at each consecutive time point, and combine it with the normal detection data of the target attribute at each consecutive time point to obtain the difference of the target attribute at each consecutive time point, thus forming the continuous difference corresponding to the target attribute, and then proceed to step B2.
[0017] Step B2. Based on the preset information estimation range value L, initialize the definition j = L + 1, and proceed to step B3;
[0018] Step B3. According to the following formula:
[0019]
[0020]
[0021] Calculate the jLth difference x in the corresponding order of the target attributes j-L x to the (j-1)th difference in sequence j-1 average variance and according to Calculate the j-th difference x corresponding to the target attribute in the correct order. j The corresponding confidence lower bound LCL j Confidence Upper Limit UCL j Where k represents a preset coefficient;
[0022] Similarly, based on the (j-L+1)th difference value x in the corresponding order of the target attributes j-L+1 x up to the j-th difference in sequence j average variance Calculate the (j+1)th difference x corresponding to the target attribute in the correct order. j+1 The corresponding confidence lower bound LCL j+1 Confidence Upper Limit UCL j+1 Then proceed to step B4;
[0023] Step B4. According to the following formula:
[0024]
[0025] The calculation yields the j-th difference in the order corresponding to the target attribute, which lies within its confidence interval [LCL]. j UCL j Event A j The probability of occurrence P(A) j If the j-th difference value corresponding to the target attribute in the given order jumps out of its confidence interval [LCL], then... j UCL j Event A j probability of occurrence Similarly, the (j+1)th difference value of the target attribute corresponding to its corresponding order is located within its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 The probability of occurrence P(A) j+1 Then proceed to step B5;
[0026] Step B5. According to the following formula:
[0027]
[0028] The calculation yields the j-th difference in the order corresponding to the target attribute, which lies within its confidence interval [LCL]. j UCL j Event A j The occurrence of the (j+1)th difference in the order of the target attribute is within its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 The probability of occurrence P(A) j+1 A j Then proceed to step B6;
[0029] Step B6. According to the following formula:
[0030]
[0031] The calculation yields the result when the j-th difference in the corresponding order of the target attribute jumps out of its confidence interval [LCL]. j UCL j Event A j If this occurs, the (j+1)th difference in the target attribute's sequence will also jump out of its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 probability of occurrence Then proceed to step B7;
[0032] Step B7. Update the value of j by incrementing it by 1, using the Markov property to obtain... And judge If the result is less than the preset false alarm probability upper limit, return to step B3; otherwise, take time j as the time when the system malfunctions and K as the number of consecutive abnormal alarms corresponding to the target attribute.
[0033] As a preferred embodiment of the present invention: in step iii, based on the differences between the target attributes corresponding to L times from the previous time point towards the historical time direction, the average of these L differences is first obtained. variance Then press Calculate the lower confidence limit (LCL) and upper confidence limit (UCL) corresponding to the target attribute; finally, construct the confidence interval [LCL, UCL] corresponding to the target attribute, where k represents a preset coefficient.
[0034] As a preferred embodiment of the present invention: the target machine is an air conditioning unit, and the various target attributes related to the fault on the air conditioning unit include the supply air humidity W.a,sup Return air humidity W a,rn chilled water flow rate V chwCC Enthalpy of mixed air E a,mix PR of return air fan rf Return air fan speed Mixed air temperature T a,mix Mixed air humidity W a,mix Mixed air flow rate M a,mix and fresh air flow rate V a,oa .
[0035] The fault detection method based on Markov process characteristic statistics described in this invention has the following technical advantages compared with existing technologies:
[0036] This invention presents a fault detection method based on Markov process characteristic statistics. It targets various fault-related attributes on a target machine, using the error of each attribute as a fault feature. First, it obtains a corresponding data estimation model and a confidence interval based on real-time analysis. The probability of events exceeding the confidence interval is accumulated to obtain the number of consecutive abnormal alarms that meet a preset upper limit for false alarm probability. During actual detection of the target machine, the fault features of each attribute are used as the object of anomaly analysis under the updated confidence interval in real time. Combined with the continuous accumulation statistical method, if a target attribute reaches the corresponding number of consecutive abnormal alarms, the target machine is determined to be faulty. The entire technical solution comprehensively considers the expected value of the target attribute, the real-time confidence interval, and introduces probability statistics based on the cumulative occurrence of events, as well as anomaly misjudgment analysis, thereby improving the accuracy and efficiency of actual fault detection of the target machine. Attached Figure Description
[0037] Figure 1 The fault characteristic ΔT corresponding to the mixing air temperature in this embodiment is... a,mix A diagram illustrating the changes in its confidence interval;
[0038] Figure 2 This is a schematic diagram illustrating the concept behind the fault detection method based on Markov process characteristic statistics designed in this invention. Detailed Implementation
[0039] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.
[0040] The present invention designs a fault detection method based on Markov process characteristic statistics. Based on various target attributes related to faults on the target machine, the following steps A to B are performed for each target attribute to obtain the data estimation model and the number of consecutive abnormal alarms K corresponding to each target attribute, thus completing the initialization operation.
[0041] Step A. Based on the various analytical attributes related to the target attribute on the target machine, and using historical normal working data, train a preset neural network to obtain a data estimation model for calculating normal estimation data of the target attribute.
[0042] In practical applications, step A above includes steps A1 to A2.
[0043] Step A1. Use correlation analysis to obtain the analysis attributes that are related to the target attribute among the preset attributes on the target machine, and then proceed to step A2.
[0044] Step A2. Obtain the analytical attribute data and target attribute data at each time point under the preset historical period corresponding to the normal operation of the target machine. Using the analytical attribute data as input and the target attribute data as output, train the preset neural network to obtain the data estimation model corresponding to the target attribute, which is used to calculate the normal estimation data of the target attribute.
[0045] Step B. Apply the data estimation model to obtain continuous historical normal estimation data of the target attribute, and obtain the continuous difference between it and the continuous historical normal detection data at the corresponding time. Combined with the preset false alarm probability upper limit, calculate the number of continuous abnormal alarms K corresponding to the target attribute.
[0046] In practical applications, step B above includes steps B1 to B7.
[0047] Step B1. Based on the various analytical attributes related to the target attribute, and the values of each analytical attribute at each consecutive time point under the preset historical period corresponding to the normal operation of the target machine, apply the data estimation model to calculate the normal estimated data of the target attribute at each consecutive time point, and combine it with the normal detection data of the target attribute at each consecutive time point to obtain the difference of the target attribute at each consecutive time point, thus forming the continuous difference corresponding to the target attribute, and then proceed to step B2.
[0048] Step B2. Based on the preset range value L, initialize the definition j = L + 1, and proceed to step B3.
[0049] Step B3. According to the following formula:
[0050]
[0051]
[0052] Calculate the jLth difference x in the corresponding order of the target attributes j-L x to the (j-1)th difference in sequence j-1 average variance and according to Calculate the j-th difference x corresponding to the target attribute in the correct order. j The corresponding confidence lower bound LCL j Confidence Upper Limit UCL j Where k represents a preset coefficient;
[0053] Similarly, based on the (j-L+1)th difference value x in the corresponding order of the target attributes j-L+1 x up to the j-th difference in sequence j average variance Calculate the (j+1)th difference x corresponding to the target attribute in the correct order. j+1 The corresponding confidence lower bound LCL j+1 Confidence Upper Limit UCL j+1 Then proceed to step B4.
[0054] Step B4. According to the following formula:
[0055]
[0056] The calculation yields the j-th difference in the order corresponding to the target attribute, which lies within its confidence interval [LCL]. j UCL j Event A j The probability of occurrence P(A) j If the j-th difference value corresponding to the target attribute in the given order jumps out of its confidence interval [LCL], then... j UCL j Event A j probability of occurrence Similarly, the (j+1)th difference value of the target attribute corresponding to its corresponding order is located within its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 The probability of occurrence P(A) j+1 Then proceed to step B5.
[0057] Step B5. According to the following formula:
[0058]
[0059] The calculation yields the j-th difference in the order corresponding to the target attribute, which lies within its confidence interval [LCL]. j UCL j Event A j The occurrence of the (j+1)th difference in the order of the target attribute is within its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 The probability of occurrence P(A) j+1A j Then proceed to step B6.
[0060] Step B6. According to the following formula:
[0061]
[0062] The calculation yields the result when the j-th difference in the corresponding order of the target attribute jumps out of its confidence interval [LCL]. j UCL j Event A j If this occurs, the (j+1)th difference in the target attribute's sequence will also jump out of its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 probability of occurrence Then proceed to step B7.
[0063] Step B7. Update the value of j by incrementing it by 1, using the Markov property to obtain... And judge If the result is less than the preset false alarm probability upper limit, return to step B3; otherwise, take time j as the time when the system malfunctions and K as the number of consecutive abnormal alarms corresponding to the target attribute.
[0064] Based on the above, steps A to B are executed for each target attribute to obtain the data estimation model and the number of consecutive abnormal alarms K corresponding to each target attribute. In the actual fault detection of the target machine, steps i to v are executed in real time for each target attribute, and analysis is performed based on the execution of this series of steps. If there is a target attribute that reaches the corresponding number of consecutive abnormal alarms K, it is determined that the target machine has a fault; otherwise, it is determined that the target machine does not have a fault.
[0065] Step i. Initialize the target attribute exception count to 0 and proceed to step ii.
[0066] Step ii. Based on the data of each analysis attribute related to the target attribute at the current time, apply the data estimation model corresponding to the target attribute to process it, obtain the normal estimation data of the target attribute at the current time, and combine it with the detection data of the target attribute at the current time to obtain the difference value corresponding to the target attribute at the current time, and then proceed to step iii.
[0067] Step iii. Based on the preset information estimation range value L, according to the target attribute, the difference of each of the L time points from the previous time point towards the historical time direction is obtained first. variance Then press Calculate the lower confidence limit (LCL) and upper confidence limit (UCL) corresponding to the target attribute; finally, construct the confidence interval [LCL, UCL] corresponding to the target attribute, and then proceed to step iv, where k represents the preset coefficient.
[0068] Step iv. Determine whether the difference corresponding to the target attribute at the current moment has jumped out of its corresponding confidence interval. If yes, increment the number of abnormalities of the target attribute by 1 and proceed to step v; otherwise, wait to enter the next moment and return to step i.
[0069] Step v. Determine whether the number of abnormal occurrences of the target attribute is greater than or equal to the number of consecutive abnormal alarms K corresponding to the target attribute. If yes, determine that the target attribute has reached the corresponding number of consecutive abnormal alarms K, and wait to enter the next moment, then return to step i; otherwise, wait to enter the next moment, then return to step ii.
[0070] Applying the above design scheme to practice, the readings of various sensors involved in the main equipment of the air conditioning unit, including the cooling coils and the blower fan, as well as the chilled water flow rate M, are measured. chw The supply temperature of chilled water, T chw,sup The return water temperature T of chilled water chw,rn Mixed air flow rate M a,mix Airflow rate (M) a,sup Mixed air temperature T a,mix Supply air temperature T a,sup The power Q of the air supply fan sf ; and certain equipment parameters, including the pressure drop ΔP of the air supply fan. sf,des The set value T of the supply air temperature sup,spt Design value of air supply volume M a,des Supply air humidity W a,sup Return air humidity W a,rn chilled water flow rate V chwCC Enthalpy of mixed air E a,mix PR of return air fan rf Return air fan speed Mixed air temperature T a,mix Mixed air humidity W a,mix Mixed air flow rate M a,mix and fresh air flow rate V a,oa .
[0071] Based on the readings of the aforementioned sensors, and according to the physical mechanism, various target attributes related to the fault were selected, including supply air humidity (W). a,sup Return air humidity W a,rn chilled water flow rate V chwCC Enthalpy of mixed air E a,mix PR of return air fan rf Return air fan speed Mixed air temperature Ta,mix Mixed air humidity W a,mix Mixed air flow rate M a,mix and fresh air flow rate V a,oa Next, for the ten target attributes related to the fault, we will first execute design steps A to B for each target attribute to obtain the data estimation model and the number of consecutive abnormal alarms K corresponding to each target attribute, and complete the initialization operation of each target attribute.
[0072] After obtaining the data estimation model and the number of consecutive abnormal alarms K corresponding to each target attribute, steps i to v are executed in real time for each target attribute, and analysis is performed simultaneously. If there is a target attribute that reaches the corresponding number of consecutive abnormal alarms K, it is determined that the target machine has a fault; otherwise, it is determined that the target machine does not have a fault.
[0073] This invention takes into account the possibility of false alarms, i.e., brief jumps out of the confidence interval, such as... Figure 1 As shown, the difference between the model estimate and the measured value of the mixed air temperature is used as the fault characteristic ΔT. a,mix The fault characteristics are as follows: Figure 1 The circled lines represent the upper limit of the confidence interval (UCL), as shown by the dashed line, and the lower limit of the confidence interval (LCL), as shown by the solid line. Figure 1 The data points from 200 to 600 are all fault data, and those from 600 to 700 are all normal data. As you can see, when a fault occurs, the fault characteristics jump outside the [LCL, UCL] range, and when the fault returns to normal, the fault characteristics return to the confidence interval [LCL, UCL].
[0074] Regarding the above situation, in practical applications, such as Figure 2 As shown, the dots represent the L differences preceding the j-th difference, including the (j-1)-th difference. These dots, falling within the corresponding confidence limits, indicate normal differences. These points can be used to generate the confidence limits (LCLs) corresponding to the j-th difference. j and UCL j The (j+1)th to (j+K+2)th differences are marked with triangles, which represent anomalies. At the (j+K-1)th difference position, a fault is detected because K consecutive differences fall outside the corresponding confidence limits. A square indicates that a fault has been confirmed at the (j+K-1)th difference position. This design incorporates a consecutive anomaly alarm count K; only after a series of consecutive anomalies can a fault be determined.
[0075] The fault detection method based on Markov process characteristic statistics designed in this invention targets various target attributes related to faults on the target machine, using the error of the target attribute as the fault feature. First, it executes the corresponding data estimation model and, based on the confidence interval under real-time analysis, accumulates the probability of events falling out of the confidence interval to obtain the number of consecutive abnormal alarms K that meets the preset upper limit of false alarm probability. Thus, in the actual detection process of the target machine, taking the fault features of each target attribute as the object, it performs anomaly analysis under the real-time updated confidence interval, and combines the continuous accumulation statistical method. If there is a situation where the target attribute reaches the corresponding number of consecutive abnormal alarms K, it is determined that the target machine has a fault. The entire technical solution design comprehensively considers the expected value of the target attribute, the real-time confidence interval factor, and introduces the probability statistics of the cumulative occurrence of events and the anomaly misjudgment analysis, thereby improving the actual detection accuracy and work efficiency of the target machine fault.
[0076] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. A fault detection method based on Markov process characteristic statistics, characterized in that: Based on the various target attributes related to the fault on the target machine, for each target attribute, the following steps A to B are executed to obtain the data estimation model and the number of consecutive abnormal alarms corresponding to each target attribute; then, for each target attribute, steps i to v are executed in real time. If there is a target attribute that reaches the corresponding number of consecutive abnormal alarms, it is determined that the target machine has a fault; otherwise, it is determined that the target machine does not have a fault. Step A. Based on the various analytical attributes related to the target attribute on the target machine, and using historical normal working data, train the preset neural network to obtain a data estimation model for calculating the normal estimation data of the target attribute; Step B. Apply the data estimation model to obtain continuous historical normal estimation data of the target attribute, and obtain the continuous difference between it and the continuous historical normal detection data at the corresponding time. Combined with the preset false alarm probability upper limit, calculate the number of continuous abnormal alarms corresponding to the target attribute. Step i. Initialize the target attribute exception count to 0, and proceed to step ii; Step ii. Based on the data of each analysis attribute related to the target attribute at the current time, apply the data estimation model corresponding to the target attribute to process it, obtain the normal estimation data of the target attribute at the current time, and combine it with the detection data of the target attribute at the current time to obtain the difference corresponding to the target attribute at the current time, and then proceed to step iii. Step iii. Based on the preset confidence estimation range value L, according to the difference of the target attribute from the previous time to the historical time direction L times, obtain the lower confidence limit and upper confidence limit corresponding to the target attribute at the current time, and form the corresponding confidence interval, and then proceed to step iv. Step iv. Determine whether the difference corresponding to the target attribute at the current moment has jumped out of its corresponding confidence interval. If yes, increment the number of abnormalities of the target attribute by 1 and proceed to step v; otherwise, wait to enter the next moment and return to step i. Step v. Determine whether the number of abnormal occurrences of the target attribute is greater than or equal to the number of consecutive abnormal alarms corresponding to the target attribute. If yes, determine that the target attribute has reached the corresponding number of consecutive abnormal alarms, and wait to enter the next moment, then return to step i; otherwise, wait to enter the next moment, then return to step ii.
2. The fault detection method based on Markov process characteristic statistics according to claim 1, characterized in that: Step A includes the following steps A1 to A2; Step A1. Use correlation analysis to obtain the analysis attributes that are related to the target attribute among the preset attributes on the target machine, and then proceed to step A2; Step A2. Obtain the analytical attribute data and target attribute data at each time point under the preset historical period corresponding to the normal operation of the target machine. Using the analytical attribute data as input and the target attribute data as output, train the preset neural network to obtain the data estimation model corresponding to the target attribute, which is used to calculate the normal estimation data of the target attribute.
3. The fault detection method based on Markov process characteristic statistics according to claim 1, characterized in that: Step B includes steps B1 to B7; Step B1. Based on the various analytical attributes related to the target attribute, and the values of each analytical attribute at each consecutive time point under the preset historical period corresponding to the normal operation of the target machine, apply the data estimation model to calculate the normal estimated data of the target attribute at each consecutive time point, and combine it with the normal detection data of the target attribute at each consecutive time point to obtain the difference of the target attribute at each consecutive time point, thus forming the continuous difference corresponding to the target attribute, and then proceed to step B2. Step B2. Based on the preset information estimation range value L, initialize the definition j = L + 1, and proceed to step B3; Step B3. According to the following formula: Calculate the jLth difference x in the corresponding order of the target attributes j-L x to the (j-1)th difference in sequence j-1 average variance and according to Calculate the j-th difference x corresponding to the target attribute in the correct order. j The corresponding confidence lower bound LCL j Confidence Upper Limit UCL j Where k represents a preset coefficient; Similarly, based on the (j-L+1)th difference value x in the corresponding order of the target attributes j-L+1 x up to the j-th difference in sequence j average variance Calculate the (j+1)th difference x corresponding to the target attribute in the correct order. j+1 The corresponding confidence lower bound LCL j+1 Confidence Upper Limit UCL j+1 Then proceed to step B4; Step B4. According to the following formula: The calculation yields the j-th difference in the order corresponding to the target attribute, which lies within its confidence interval [LCL]. j UCL j Event A j The probability of occurrence P(A) j If the j-th difference value corresponding to the target attribute in the given order jumps out of its confidence interval [LCL], then... j UCL j [The event] probability of occurrence Similarly, the (j+1)th difference value of the target attribute corresponding to its corresponding order is located within its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 The probability of occurrence P(A) j+1 Then proceed to step B5; Step B5. According to the following formula: The calculation yields the j-th difference in the order corresponding to the target attribute, which lies within its confidence interval [LCL]. j UCL j Event A j The occurrence of the (j+1)th difference in the order of the target attribute is within its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 The probability of occurrence P(A) j+1 A j Then proceed to step B6; Step B6. According to the following formula: The calculation yields the result when the j-th difference in the corresponding order of the target attribute jumps out of its confidence interval [LCL]. j UCL j Event A j If this occurs, the (j+1)th difference in the target attribute's sequence will also jump out of its confidence interval [LCL]. j+1 UCL j+1 Event A j+1 probability of occurrence Then proceed to step B7; Step B7. Update the value of j by incrementing it by 1, using the Markov property to obtain... And judge If the result is less than the preset false alarm probability upper limit, return to step B3; otherwise, take time j as the time when the system malfunctions and K as the number of consecutive abnormal alarms corresponding to the target attribute.
4. The fault detection method based on Markov process characteristic statistics according to claim 1, characterized in that: In step iii, based on the differences between the target attributes and the L time intervals from the previous time interval towards the historical time interval, the average of these L differences is first obtained. variance Then press Calculate the lower confidence limit (LCL) and upper confidence limit (UCL) corresponding to the target attribute; finally, construct the confidence interval [LCL, UCL] corresponding to the target attribute, where k represents a preset coefficient.
5. A fault detection method based on Markov process characteristic statistics according to any one of claims 1 to 4, characterized in that: The target machine is an air conditioning unit, and the various target attributes related to the fault on the air conditioning unit include the supply air humidity W. a,sup Return air humidity W a,rn chilled water flow rate V chwCC Enthalpy of mixed air E a,mix PR of return air fan rf Return air fan speed Mixed air temperature T a,mix Mixed air humidity W a,mix Mixed air flow rate M a,mix and fresh air flow rate V a,oa .