Air conditioning system for refrigerant leakage diagnosis and control method thereof

Abstract

The present application provides an air conditioning system for refrigerant leakage diagnosis and a control method thereof, wherein the air conditioning system includes a unit and a sensor j (j=1, 2, ..., k), a test data sequence is obtained from the k sensors from the unit, a restoration deviation enhancement coefficient is calculated from the test data sequence, and several input sub-vectors are obtained; an output sub-vector corresponding to several input sub-vectors is obtained through an LSTM detection model, and then several diagnostic thresholds and enhanced restoration deviations are calculated; refrigerant leakage diagnosis is performed by comparing the obtained several diagnostic thresholds and enhanced restoration deviations. The control method of the present application can find sensitive features from a large number of system sensor parameters, and separate the factors of refrigerant leakage from other fault factors, so as to speed up the processing speed of the diagnostic method and reduce the processing burden of the system.

Description

Technical Field

[0001] The present application relates to a method for diagnosing an air-conditioning system, and in particular to a method for diagnosing refrigerant leakage in an air-conditioning system.

[0002] Background

[0003] A typical air conditioning system includes four components: a compressor, a condenser, a throttle valve, and an evaporator. The refrigerant forms a circulation loop in these four components to complete the heat exchange. Specifically, the refrigerant loop connects the compressor, condenser, throttle valve, and evaporator in sequence. The exhaust port of the compressor is in fluid communication with the inlet fluid of the condenser, the outlet of the condenser is in fluid communication with the inlet fluid of the throttle valve, the outlet of the throttle valve is in fluid communication with the inlet fluid of the evaporator, and the outlet of the evaporator is in fluid communication with the suction port of the compressor. The refrigerant is compressed into a high-pressure, high-temperature state in the compressor and then discharged into the condenser. Then the refrigerant exchanges heat with the ambient air in the condenser, releases heat and is condensed into a high-pressure, liquid state, and then discharged into the throttle valve. In the throttle valve, the refrigerant is expanded and throttled into a low-pressure two-phase state and then flows into the evaporator. Then, in the evaporator, heat is exchanged with the chilled water to absorb heat and evaporate into a low-pressure, gaseous state, and then returns to the compressor from the suction port of the compressor to complete the refrigerant cycle.

[0004] Therefore, in order to test the operating performance of the air-conditioning system and maintain the air-conditioning system, it is very important to perform a leak diagnosis function on the refrigerant that provides a refrigerant circulation path in the air-conditioning system. Summary of the invention

[0005] According to a first aspect of the present application, a control method for refrigerant leakage diagnosis of an air-conditioning system is provided, wherein the air-conditioning system includes a unit, and the control method for refrigerant leakage diagnosis of the air-conditioning system includes the following steps:

[0006] S01, in a combined test case, obtaining a test data sequence, obtaining restoration deviation enhancement coefficients of k sensors according to the test data sequence, obtaining a test threshold based on the restoration deviation enhancement coefficient, and obtaining a test threshold proportional coefficient according to the test threshold;

[0007] S02, in the absence of refrigerant leakage, obtaining a test data sequence, and obtaining a plurality of diagnostic thresholds based on the restoration deviation enhancement coefficient and the test threshold proportional coefficient;

[0008] S03, during the refrigerant leakage diagnosis process, a test data sequence is obtained, an enhanced restoration deviation is obtained based on the k restoration deviation enhancement coefficients, the enhanced restoration deviation is compared with the plurality of diagnosis thresholds, and a refrigerant leakage diagnosis is performed.

[0009] According to a first aspect of the present application, it is characterized in that: the air conditioning system includes a unit and a sensor j (j=1, 2, ..., k), the k sensors obtain state parameters from the unit, and the method for diagnosing refrigerant leakage of the air conditioning system includes the following steps:

[0010] In step S01, in the case of a combined test under different refrigerant charge amounts, the following steps are performed:

[0011] (i) Obtain all test data sequences Cj (j=1, 2, ..., k) for each sensor j among the k sensors, each test data sequence Cj comprising sequence elements Where i = 1, 2, ..., m, and m represents the number of data points obtained from the combined test according to the sampling period,

[0012] (ii) According to the test data sequence Cj, the restoration deviation enhancement coefficients β1, β2, ..., β of k sensors are obtained by multivariate linear regression. k ,

[0013] (iii) For each sensor j among the k sensors, obtain an n-dimensional input subvector from the test data sequence Cj Where i=1, 2, ..., n, where n represents the interception cycle time,

[0014] (iv) is the n-dimensional input subvector Obtain the subsequent n-dimensional input subvector from the test data sequence Cj where i = 1, 2, ..., n,

[0015] (v) corresponds to the n-dimensional input sub-vector Get an n-dimensional output subvector where i = 1, 2, ..., n,

[0016] (vi) Obtain k of the n-dimensional output sub-vectors and k of the n-dimensional input subvectors The respective reduction deviations,

[0017] (vii) based on the k restoration deviations, the restoration deviation enhancement coefficients β1, β2, ..., β k Obtaining test thresholds g0, g1, g2, g3 of refrigerant leakage from small to large, and obtaining test threshold proportional coefficients α1, α2, α3 according to the test thresholds g0, g1, g2, g3, wherein α1 = g1 / g0, α2 = g2 / g0, α3 = g3 / g0;

[0018] In step S02, when there is no refrigerant leakage, the following steps are performed:

[0019] (i) Obtain all test data sequences Cj (j=1, 2, ..., k) for each sensor j among the k sensors, each test data sequence Cj comprising sequence elements Where i=1, 2, ..., m, and m represents the number of data points obtained from the combined test according to the time period,

[0020] (ii) For each sensor j among the k sensors, obtain an n-dimensional input subvector from the test data sequence Cj Where i=1, 2, ..., n, where n represents the interception cycle time,

[0021] (iii) is the n-dimensional input subvector Obtain the subsequent n-dimensional input subvector from the test data sequence Cj where i = 1, 2, ..., n,

[0022] (iv) corresponding to the n-dimensional input sub-vector Get an n-dimensional output subvector where i = 1, 2, ..., n,

[0023] (v) Obtain k of the n-dimensional output sub-vectors and k of the n-dimensional input subvectors The respective reduction deviations,

[0024] (vi) based on the k restoration deviations, the restoration deviation enhancement coefficients β1, β2, ..., β k and the test threshold proportionality coefficients α1, α2, α3, to obtain several diagnostic thresholds G0, G1, G2, G3, wherein G1 = G0×α1, G2 = G0×α2, G3 = G0×α3;

[0025] In step S03, during the refrigerant leakage diagnosis process, the following steps are performed:

[0026] (i) Obtain all test data sequences Cj (j=1, 2, ..., k) for each sensor j among the k sensors, each test data sequence Cj comprising sequence elements Where i=1, 2, ..., m, and m represents the number of data points obtained from the combined test according to the time period,

[0027] (ii) For each sensor j among the k sensors, obtain an n-dimensional input subvector from the test data sequence Cj Where i=1, 2, ..., n, wherein n represents the cycle time,

[0028] (iii) is the n-dimensional input subvector Get the subsequent n-dimensional input subvector from the test data sequence Cj where i = 1, 2, ..., n,

[0029] (iv) corresponding to the n-dimensional input sub-vector Get an n-dimensional output subvector where i = 1, 2, ..., n,

[0030] (v) Obtain the n-dimensional output sub-vector and the n-dimensional input subvector The restoration deviation,

[0031] (vi) based on the calculated restoration deviations of the k sensors, the k restoration deviation enhancement coefficients β1, β2, ..., β k Get the enhanced restoration deviation,

[0032] (vii) Comparing the enhanced reduction deviation with the plurality of diagnostic thresholds G0, G1, G2, G3 to perform refrigerant leakage diagnosis.

[0033] According to the first aspect of the present application, it is characterized in that: the step S01 is performed under several different refrigerant filling amounts, and the several different refrigerant filling amounts include no leakage, a first leakage amount, a second leakage amount and a third leakage amount.

[0034] According to a first aspect of the present application, it is characterized in that: when executing the step S01(vi), the step S02(v) and the step S03(v) to obtain the restoration deviation, the following formula is used for calculation:

[0035]

[0036] According to a first aspect of the present application, it is characterized in that: when executing the step S01(vii) to obtain the test thresholds g0, g1, g2, g3, the following steps are performed: the enhanced restoration deviation of each output sub-vector is calculated by the following formula:

[0037]

[0038] The non-leakage test data is selected from the test data set, and the obtained multiple enhancement and restoration deviations are sorted from high to low. The average of the first o enhancement and restoration deviations after sorting is used as the original test threshold g0. The specific calculation formula is as follows:

[0039]

[0040] Where i represents the sequence number of the first o enhanced restoration deviations after sorting,

[0041] The test data under the first leakage amount, the second leakage amount and the third leakage amount are screened out from the test data set, and the means of the first o enhanced restoration deviations after sorting under the first leakage amount, the second leakage amount and the third leakage amount are respectively used as the first test threshold g1, the second test threshold g2 and the third test threshold g3.

[0042] According to a first aspect of the present application, it is characterized in that: when executing step S01(ii), the β1, β2, ..., β k When the refrigerant filling amount is used as the dependent variable, other screening data are used as independent variables, and the function form is established as follows:

[0043]

[0044] Where i represents a sample, where i = 1, 2, ..., m, where m represents m time samples, Represents the refrigerant charge obtained by regression. The estimation method of multiple linear regression parameters usually adopts the least square method. For sample i, the difference between the actual value and the estimated value is:

[0045]

[0046] For all m time samples, the formula for the sum of squares of the differences between the measured values ​​and the estimated values ​​is as follows:

[0047]

[0048] The principle of the least squares method is to find a set of coefficients β0, β1, ..., β k Make Q minimum.

[0049] According to the first aspect of the present application, it is characterized in that: when executing the step S02 (vi) to obtain the plurality of diagnostic thresholds G0, G1, G2, G3, the following steps are performed: the largest one is selected from the enhanced restoration deviations of the plurality of output vector groups as the no-leakage diagnostic threshold G0, and the specific calculation formula is as follows:

[0050] G0=max(EnhancedRE i ), where i = 2, 3, 4...

[0051] According to a first aspect of the present application, it is characterized in that: when executing the step S03(vi) to obtain the enhanced restoration deviation, it is calculated by the following formula:

[0052] where i = 1, 2, 3, ...,

[0053] According to the first aspect of the present application, it is characterized in that it also includes the step of finding sensitive features from a large number of system sensor parameters to separate refrigerant leakage from other fault factors.

[0054] According to another aspect of the present application, an air-conditioning system is provided, which includes a unit, a sensor j (j=1, 2, ..., k) and a controller, wherein the sensor obtains vector parameters from the unit, and is characterized in that the controller controls the air-conditioning system according to the parameters obtained by the method described in the first aspect of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] These and other features and advantages of the present application may be better understood by reading the following detailed description with reference to the accompanying drawings, in which like reference numerals refer to like parts throughout, wherein:

[0056] Figure 1A A schematic component block diagram of an air conditioning system 100 of the present application is shown;

[0057] Figure 1B Shows Figure 1A Structural details of the compressor sensor 105;

[0058] Figure 1C Shows Figure 1A Structural details of the condenser sensor 107;

[0059] Figure 1D Shows Figure 1A Structural details of the mid-throttle sensor 109;

[0060] Figure 1E Shows Figure 1A Structural details of the evaporator sensor 111.

[0061] Figure 1F Shows Figure 1A Structural details of the middle environment sensor 113;

[0062] Figure 2A The following is a schematic diagram showing the working principle of the LSTM detection model used in this application;

[0063] Figure 2B Shows Figure 2A The format of input data in the LSTM detection model;

[0064] Figure 2C Shows Figure 2A The data format of the input sub-vector generated by the sensor in the LSTM detection model;

[0065] Figure 2D Shows Figure 2AThe data format of the output sub-vectors generated by the sensors in the LSTM detection model;

[0066] Figure 3 A flowchart 300 of a method for diagnosing a refrigerant leak in the air conditioning system 100 of the present application is shown;

[0067] Figure 4 Shows Figure 3 The specific process of step 304 in the flowchart 300 shown;

[0068] Figure 5 Shows Figure 3 The specific process of step 306 in the flowchart 300 shown;

[0069] Figure 6 Shows Figure 3 The specific process of step 308 in the flowchart 300;

[0070] Figure 7 The test steps under different refrigerant injection amounts are shown;

[0071] Figure 8 Shows Figure 7 The effectiveness of the implementation of the diagnostic process shown;

[0072] Fig. 9 Shows Figure 1A A block diagram of the controller 116 is shown in more detail. DETAILED DESCRIPTION

[0073] Figure 1A A schematic component block diagram of the air-conditioning system 100 of the present application is shown, which is used to illustrate the main components of the air-conditioning system of the present application.

[0074] like Figure 1A As shown, the air conditioning system 100 includes a compressor 104, a condenser 106, a condenser fan 117, a throttle valve 108, an evaporator 110, a chilled water pump 112, a heat load 114, and a controller 116. The condenser fan 117 cools the condenser 106 by wind. Figure 1A The air conditioning system 100 shown includes a refrigerant circulation circuit 101 and a chilled water circulation circuit 102 .

[0075] refer to Figure 1AThe refrigerant circulation loop 101 is connected in sequence to the compressor 104, the condenser 106, the throttle valve 108 and the evaporator 110. The exhaust port of the compressor 104 is in fluid communication with the inlet fluid of the condenser 106, the outlet of the condenser 106 is in fluid communication with the inlet fluid of the throttle valve 108, the outlet of the throttle valve 108 is in fluid communication with the inlet fluid of the evaporator 110, and the outlet of the evaporator 110 is in fluid communication with the suction port of the compressor 104. The refrigerant is compressed into a high-pressure, high-temperature state in the compressor 104 and then discharged into the condenser 106; then the refrigerant exchanges heat with the ambient air in the condenser 106 to release heat and condense into a high-pressure, liquid state; then it is discharged into the throttle valve 108; in the throttle valve 108, the refrigerant is expanded and throttled into a low-pressure two-phase state and flows into the evaporator 110; then it exchanges heat with the chilled water in the evaporator 110 to absorb heat and evaporate into a low-pressure, gaseous state and then returns to the compressor 104 from the air intake of the compressor 104 to complete the refrigerant cycle. The heat load 114 is set in the environment 118. The controller 116 is respectively connected to the compressor 104, the condenser 106, the throttle valve 108, the evaporator 110 and the chilled water pump 112, and is used to control the compressor 104, the condenser 106, the throttle valve 108, the evaporator 110 and the chilled water pump 112, and is used to collect parameters from sensors of the compressor 104, the condenser 106, the throttle valve 108 and the evaporator 110.

[0076] Still for reference Figure 1A , the chilled water circulation loop 102 is connected to the evaporator 110, the chilled water pump 112, and the heat load 114 in sequence. The chilled water pump 112 is used to drive the chilled water to circulate through the heat load and the evaporator, so that the chilled water exchanges heat with the refrigerant in the evaporator 110. In the chilled water circulation loop 101, the refrigerant absorbs heat and evaporates in the evaporator 110, while the chilled water releases heat and is cooled. The cooled chilled water flows through the heat load 114 and takes away the heat of the heat load, so that the heat load 114 is cooled and the ambient air is cooled.

[0077] In order to collect various parameters in the air conditioning system 100, the present application sets 5 types of sensors in the air conditioning system 100, including: (1) compressor sensor 105, (2) condenser sensor 107, (3) throttle sensor 109, (4) evaporator sensor 111, and (5) environmental sensor 113. In the present application, the output 125 of the compressor sensor 105, the output 127 of the condenser sensor 107, the output 129 of the throttle sensor 109, the output 121 of the evaporator sensor 111, and the output 123 of the environmental sensor 113 are connected to the controller 116 to provide the controller 116 with sensor parameters. The combined use of these sensor parameters enables the controller 116 to perform system refrigerant leakage diagnosis more effectively, more accurately, and more sensitively.

[0078] As an embodiment of the present application, the positions of various sensors are set as follows: (1) the compressor sensor 105 is set inside or near the compressor 104; (2) the condenser sensor 107 is set inside or near the condenser 106; (3) the throttle valve sensor 109 is set inside or near the throttle valve 108; (4) the evaporator sensor 111 is set inside or near the evaporator 110; (5) the environmental sensor 113 is set in the use environment 118 of the air-conditioning system 100.

[0079] Still reference Figure 1A The air conditioning system 100 further includes a refrigerant pump 182 and a refrigerant container 184, wherein the refrigerant container 184 is used to store the refrigerant and provide or recover the refrigerant to the refrigerant circulation loop 101 through the refrigerant pump 182. At the same time, the controller 116 can control the opening and closing of the refrigerant pump 182, so that the refrigerant container 184 can provide different refrigerant filling amounts to the air conditioning system 100 under the control of the controller 116.

[0080] Figures 1B-1F Shows Figure 1A Structural details of the compressor sensor 105 , the condenser sensor 107 , the throttle valve sensor 109 , the evaporator sensor 111 , and the ambient sensor 113 .

[0081] like Figure 1B As shown, the compressor sensor 105 includes: a compressor speed sensor 130, a compressor suction temperature sensor 132, a compressor suction pressure sensor 134, a compressor exhaust temperature sensor 136, a compressor exhaust pressure sensor 138, a compressor motor voltage sensor 140, and a compressor motor current sensor 142. The compressor speed sensor 130, the compressor suction temperature sensor 132, the compressor suction pressure sensor 134, the compressor exhaust temperature sensor 136, the compressor exhaust pressure sensor 138, the compressor motor voltage sensor 140, and the compressor motor current sensor 142 respectively have a sensor output 131, a sensor output 133, a sensor output 135, a sensor output 137, a sensor output 139, a sensor output 141, and a sensor output 143 connected to the controller 116. For the convenience of illustration, the sensor output 131, the sensor output 133, the sensor output 135, the sensor output 137, the sensor output 139, the sensor output 141, and the sensor output 143 are collectively labeled as the sensor output 125.

[0082] like Figure 1CAs shown, the condenser sensor 107 includes: a condenser refrigerant inlet temperature sensor 150, a condenser refrigerant outlet temperature sensor 152, a condenser refrigerant pressure sensor 154, and a condenser wind speed sensor 156. The condenser refrigerant inlet temperature sensor 150, the condenser refrigerant outlet temperature sensor 152, the condenser refrigerant pressure sensor 154, and the condenser wind speed sensor 156 respectively have a sensor output 151, a sensor output 153, a sensor output 155, and a sensor output 157 connected to the controller 116. For the convenience of illustration, the sensor output 151, the sensor output 153, the sensor output 155, and the sensor output 157 are collectively labeled as the sensor output 127.

[0083] like Figure 1D As shown, the throttle valve sensor 109 includes: a throttle valve opening sensor 144 and a throttle valve flow sensor 146. The throttle valve opening sensor 144 and the throttle valve flow sensor 146 respectively have a sensor output 145 and a sensor output 147 connected to the controller 116. For the convenience of illustration, the sensor output 145 and the sensor output 147 are collectively identified as the sensor output 129.

[0084] like Figure 1E As shown, the evaporator sensor 111 includes: an evaporator refrigerant inlet temperature sensor 160, an evaporator refrigerant outlet temperature sensor 162, an evaporator refrigerant pressure sensor 164, an evaporator water flow sensor 166, an evaporator water side inlet temperature sensor 168, and an evaporator water side outlet temperature sensor 170. The evaporator refrigerant inlet temperature sensor 160, the evaporator refrigerant outlet temperature sensor 162, the evaporator refrigerant pressure sensor 164, the evaporator water flow sensor 166, the evaporator water side inlet temperature sensor 168, and the evaporator water side outlet temperature sensor 170 respectively have a sensor output 161, a sensor output 163, a sensor output 165, a sensor output 167, a sensor output 169, and a sensor output 171 connected to the controller 116. For the convenience of illustration, the sensor output 161, the sensor output 163, the sensor output 165, the sensor output 167, the sensor output 169, and the sensor output 171 are collectively identified as the sensor output 121.

[0085] like Figure 1F As shown, the environmental sensor 113 includes an environmental temperature sensor 182 and an environmental temperature sensor 184. The environmental temperature sensor 182 and the environmental temperature sensor 184 respectively have a sensor output 183 and a sensor output 185 connected to the controller 116. For the convenience of illustration, the sensor output 183 and the sensor output 185 are collectively identified as the sensor output 123.

[0086] Figure 2AA schematic block diagram showing the working principle of the LSTM detection model used in this application is shown.

[0087] LSTM is the abbreviation of Long Short-Term Memory Detection Model. Hoang Duy Trinh et al. described this model in a 2019 IEEE paper (HD Trinh et al.: Detecting Mobile Traffic Anomalies Through Physical Control Channel, IEEE Access, VOLUME 7, 2019). Here, the applicant incorporates this document into the present patent application document by reference.

[0088] like Figure 2A As shown, the LSTM detection model includes: an input vector group module 202, an LSTM encoding network module 204, a latent vector module 206, an LSTM decoding network module 208, and an output vector group module 210.

[0089] The input vector group module 202 converts the data collected by k sensors into an input vector group. The LSTM encoding network module 204 compresses the input vector group into a hidden state vector. The hidden state vector module 206 is an intermediate vector between the LSTM encoding network and the LSTM decoding network. The LSTM encoding network module 204 compresses the input vector group into a determined hidden state vector, and the LSTM decoding network module 208 uses the determined hidden state vector as input and restores it to a second input vector group, that is, an adjacent successor vector group of the first vector group, as an output vector group module 210.

[0090] Figure 2B Shows Figure 2A The format of input data in the LSTM detection model includes the test data sequence, the segmentation of sub-vectors, and the time rolling characteristics of sub-vectors.

[0091] exist Figure 2B In the LSTM detection model shown, we assume that one of the air conditioning systems 100 in the present application has k sensors, and each sensor j (j = 1, 2, ..., k) collects unit data according to a time period. Therefore, after m unit time periods (for example, 10,000 unit time periods), the m unit data are arranged into a data test data sequence Cj arranged in a periodic manner. Among them, each test data sequence Cj includes sequence elements Wherein i=1, 2, ..., m, and m represents the number of data points obtained from the combined test according to the sampling period.

[0092] To use the LSTM detection model to diagnose refrigerant leakage in the air conditioning system of the present application, the present application rolls and intercepts a number of input sub-vectors from the test data sequence Cj corresponding to sensor k at periodic time intervals, that is Each input sub-vector has n vector elements (or an n-dimensional vector, where n < m). Each of the n-dimensional input sub-vectors has i (i = 1, 2,..., n) vector elements, that is It should be noted that for the input sub-vector, i (i = 1, 2,..., n) has two meanings: (1) i is the second subscript value, indicating the periodic time position of each vector element in the x sub-vector as input in the test data sequence Cj, and (2) the position of the i subscript value also indicates the position of each vector element in the input sub-vector.

[0093] After the input vector group module 202 receives the n-dimensional input sub-vector the output vector group module 210 generates a number of output sub-vectors, that is It should be noted that for the output sub-vector, i (i = 1, 2,..., n) has two meanings: (1) i is the second subscript value, indicating the periodic time position of each vector element in the x sub-vector as output in the test data sequence Cj, and (2) the position of the i subscript value also indicates the position of each vector element in the output x sub-vector.

[0094] As Figure 2B shown, the comparison relationship between the input sub-vector and the output sub-vector is as follows:

[0095] Compare ( and ), ( and ), ( and ), …, and obtain the deviation between them, and then use these deviations to calculate the reduction deviation in the calculation (the specific calculation method will be described below).

[0096] Figure 2C shows Figure 2A the data format of the input sub-vector generated by sensor j (j = 1, 2,..., k) in the LSTM detection model in

[0097] Figure 2D shows Figure 2A the data format of the output sub-vector generated by sensor j (j = 1, 2,..., k) in the LSTM detection model in

[0098] As Figure 2C shown, the vector elements in a subsequent input vector Vector elements relative to the previous input vector Scrolled back one cycle.

[0099] like Figure 2D As shown, the vector elements in a subsequent output vector Relative to the vector element in the previous output vector Scrolled back one cycle.

[0100] In the LSTM model, the input subvector and the output subvector The time periods are aligned.

[0101] Figure 3 A flowchart 300 of a method for diagnosing a refrigerant leak in the air conditioning system 100 of the present application is shown.

[0102] like Figure 3 As shown, at step 304 , the system processor 116 performs an operation of obtaining a restoration deviation enhancement coefficient and a test threshold proportional coefficient.

[0103] In the present application, the functions of step 304 include: enhancing the sensitivity of parameters related to refrigerant leakage, and separating refrigerant leakage from other faults. It is characterized by including combined tests under different refrigerant filling amounts, data collection, data cleaning, normalization, correlation analysis, sensitivity analysis, input quantity screening, multivariate linear regression, training LSTM model, and obtaining test thresholds and test threshold ratio coefficients.

[0104] Specifically, the operating structure of the air conditioning system of the present application is: a device that realizes refrigeration or heating through a vapor compression cycle. Refrigerant is the medium filled in the air conditioning circuit system. Commonly used refrigerants include R22, R32, R410a, R134a, R1234ze, R515B, etc. This application is applicable to the leakage detection of all refrigerants. There are two technical problems encountered when detecting refrigerant leakage, as follows:

[0105] (1) In the air conditioning system, changes in the refrigerant charge will have a global impact. For example, when the refrigerant is overcharged, the static (base) pressure in the circuit is too high, which in turn causes the phase change temperature of the evaporator and condenser to drift, affecting the cooling / heating and performance. At the same time, the high suction pressure will also cause the compressor operating point to shift in the efficiency diagram. In addition, changes in component performance will cause the system control logic to drive them to make new adjustments, which will ultimately have a greater impact. After a refrigerant leak, the air conditioning system usually has problems such as reduced cooling / heating capacity, high power consumption, low suction pressure, and even suction pressure alarms.

[0106] (2) When the performance of the air conditioning system deteriorates, refrigerant leakage is only one of the potential factors. Other factors may also include excessive fouling coefficient in the condenser or evaporator, dirt blockage on the wind side of the air-cooled heat exchanger (such as the accumulation of willow catkins in early spring in northern regions), abnormal electronic expansion valve, etc., or even multiple faults occurring simultaneously.

[0107] The purpose of step 304 of the present application includes solving the above two problems: one is to find sensitive features from a large number of system sensor parameters, and the other is to be able to separate refrigerant leakage from other fault factors. In actual operation, for all sensor parameters, the reduction deviation enhancement coefficients β1, β2, ..., β k The steps of the reduction deviation enhancement coefficients β1, β2, …, β k Some coefficients are very close to zero, that is, they are irrelevant to the refrigerant leakage. For irrelevant proportional coefficients, removing them is equivalent to retaining the sensor parameters related to the refrigerant leakage.

[0108] After completing the operation of step 304 , the operation proceeds to step 306 .

[0109] In step 306, the system processor 116 performs the operation of collecting and training data without refrigerant. Step 306 is performed by collecting air conditioning system parameters without refrigerant leakage diagnosis, training the collected air conditioning system parameters, and diagnosing the threshold value according to the obtained restoration deviation enhancement coefficient.

[0110] After completing the operation of step 306 , the operation proceeds to step 308 .

[0111] In step 308, the system processor 116 performs an online refrigerant leakage diagnosis operation. In executing step 308, the system processor 116 first obtains the enhanced reduction deviation, and then provides the leakage condition of the diagnostic system according to the enhanced reduction deviation and sends a diagnostic report.

[0112] After completing the operation of step 308, the operation proceeds to step 310 and the process 300 ends.

[0113] Figure 4 Shows Figure 3 The specific process of step 304 in the flowchart 300 is used to illustrate more specific steps of step 304 .

[0114] like Figure 4 As shown, at step 402, the system processor 116 performs test operations under different combinations of refrigerant charging amounts and / or different refrigeration conditions (GB refrigeration condition, AHRI condition, I PLV refrigeration condition).

[0115] To obtain the restoration deviation enhancement coefficient, we need to set up the experiment, namely:

[0116] (1) Each new model of air conditioner needs to be tested at 100% refrigerant charge according to the AHRI and / or GB tables. Therefore, the existing test scheme makes full use of the test data when obtaining IPLV (Integrated Part-Load Value).

[0117] (2) Only 100% filling volume cannot meet the requirements for obtaining the reduction deviation enhancement factor. Therefore, we also added tests under 95%, 90% and 85% refrigerant filling volume. This setting also makes full use of the design techniques for partial load conditions in GB standard (GB / T10870) and AHRI standard (AHRI Standard 550 / 590). These two standards propose four gears based on the statistical analysis of the actual operating conditions of a large number of customer groups. Combining these four gears can better estimate the actual operating status of the customer. Therefore, conducting refrigerant leakage simulation under these four gears is equivalent to using fewer test resources to obtain state parameters that are closer to the user's actual application (that is, parameters that reflect the unit status, such as temperature, pressure, flow, power, etc.). In other words, a more effective reduction deviation enhancement factor can be obtained.

[0118] As an embodiment, in step 402, there are several (for example, 4) options for the combination of refrigerant filling amount changes, and the several different refrigerant filling amounts include no leakage, a first leakage amount, a second leakage amount and a third leakage amount, and the corresponding values ​​are: 100%, 95%, 90%, 85%, wherein 100% indicates that the unit has no leakage and the refrigerant filling amount is in the factory state; 95% indicates that the refrigerant has leaked 5% of the standard filling amount; 90% indicates that the refrigerant has leaked 10% of the standard filling amount; 85% indicates that the refrigerant has leaked 15% of the standard filling amount.

[0119] After completing the operation of step 402 , the operation proceeds to step 404 .

[0120] In step 404, the controller 116 performs data acquisition operations. Specifically, step 404 performs data acquisition operations for 4 refrigerant charging amounts, and for 4 working state combinations in 2 refrigeration standards (taking AHRI refrigeration conditions and GB refrigeration standards as examples) in each charging amount, a total of 16 test points are combined. Based on these 16 test points, predictive data acquisition is performed on these 16 test points through the corresponding sensors in Figure 1. The following Tables 1 and 2 each show 4 gear states, as follows:

[0121] Table 1. AHRI standard cold working conditions

[0122]

[0123] Table 2. GB standard cooling conditions

[0124]

[0125] In the AHRI standard, the standard refrigeration condition only explicitly mentions that the evaporator outlet water temperature is 44℉ (6.7℃), and the inlet temperature is determined according to the designed 100% refrigeration capacity. Therefore, in Table 1, the inlet temperature is assumed to be A, so A is a value to be determined that needs to be calculated.

[0126] Now combine Figure 2B The data collection operation of step 404 is described below.

[0127] In step 404, we assume that there are j sensors, where j = 1, 2, ..., k. Each sensor j will collect a Figure 2B The experimental data sequence Cj (j=1,2,…,k) is shown.

[0128] After completing the operation of step 404 , the operation proceeds to step 406 .

[0129] In step 406, the system processor 116 performs a data cleaning operation on the data collected by the sensor. The purpose of the data cleaning operation is to remove the response delay caused by short-term fluctuations in the prediction data collected by the sensor, because these delayed data will affect the effectiveness of restoring the deviation enhancement coefficient.

[0130] After completing the operation of step 406 , the operation proceeds to step 408 .

[0131] After completing data cleaning at step 408 , the system processor 116 performs a data normalization operation on the prediction data collected by the sensor.

[0132] The purpose of data normalization operation is: The benefit of normalizing the sensor collected prediction data is to avoid the interference of digital units or magnitudes on the stability of subsequent data calculation and analysis.

[0133] As an example, the normalization method is maximum value normalization, that is, each data is adjusted to [0,1] using the maximum and minimum values ​​in the collected data set. n ] element x i , get the maximum normalized result x ′ The calculation method of i is as follows:

[0134]

[0135] After completing the operation of step 408 , the operation proceeds to step 410 .

[0136] At step 410 , the system processor 116 performs a data dependency analysis operation.

[0137] The purpose of the data correlation analysis operation is to select parameters with better correlation with refrigerant leakage from the predicted parameters collected by the sensor. In one embodiment, the screening results on the York brand water-cooled air conditioner include: 1) compressor suction temperature, 2) compressor exhaust temperature, 3) ambient temperature, 4) evaporator inlet water temperature, 5) evaporator outlet water temperature, 6) expansion valve opening, 7) fan speed, 8) compressor frequency, 9) compressor current, 10) condenser outlet refrigerant temperature, 11) evaporator inlet and outlet water temperature difference, 12) compressor suction and exhaust temperature difference, 13) condenser inlet and outlet refrigerant temperature difference.

[0138] After completing the operation of step 410 , the operation proceeds to step 412 .

[0139] In step 412, the system processor 116 performs an input quantity screening operation. The purpose of the input quantity screening operation is to find parameters that are sensitive to refrigerant changes among all sensor parameters. Due to the large number of sensors and the high frequency of data acquisition, directly storing and calculating a large amount of data consumes a lot of storage space and computing resources, and the cost is too high. Therefore, by screening the input quantity, the input can be streamlined, the data storage cost can be reduced, and the computing speed can be improved.

[0140] It should be noted that the operation steps 406-412 are performed on the sensor j (j=1, 2, ..., k) collected in step 404. Figure 2B The test data sequence Cj (j=1, 2, ..., k) shown is optimized. Even if steps 406-412 are not executed and step 404 is directly transferred to step 413, this operation method can also obtain improved results and is also within the scope of protection claimed in this application.

[0141] After completing the operation of step 412 , the operation proceeds to step 413 .

[0142] In step 413, the system processor 116 performs a multiple linear regression operation. The multiple linear regression operation is used to obtain the regression coefficient of each sensor, and the regression coefficient is used as the reduction deviation enhancement coefficient. The calculation method of the reduction deviation enhancement coefficient is characterized in that the normalized refrigerant filling amount is used as the dependent variable, and the normalized other screening data is used as the dependent variable, and then the function form is established as follows:

[0143]

[0144] Among them, i represents a sample (such as 0s, 10s, 20s, 30s...), k represents k independent variables (such as the number of sensors), Indicates the predicted value of the refrigerant charge amount at time i.

[0145] The method for estimating the parameters of multiple linear regression usually adopts the least squares method, the principle of which is to minimize the sum of the squares of the differences between the measured values ​​and the estimated values. i Represents the actual value of the refrigerant charge at time i. For sample i, the difference between the actual value and the estimated value is:

[0146]

[0147] For all n samples, the formula for the sum of squares of the differences between the observed values ​​and the estimated values ​​is as follows:

[0148]

[0149] The principle of the least squares method is to find a set of coefficients β0, β1,…, β k Make Q minimum. Matlab, python scikit-learn have corresponding package functions that can be used to calculate the multivariate regression coefficients. The regression coefficients β1, β2, …, β k This is the restoration deviation enhancement coefficient β to be used later. The restoration deviation enhancement coefficient obtained in one embodiment is shown in Table 3 below:

[0150] Table 3

[0151]

[0152]

[0153] in,

[0154] Tcomp_suc Compressor suction temperature

[0155] Tcomp_dis Compressor discharge temperature

[0156] Tedb environment temperature

[0157] Twater_in Evaporator inlet water temperature

[0158] Twater_out Evaporator outlet water temperature

[0159] EXV expansion valve opening

[0160] Speed_fan fan speed

[0161] Freq_comp compressor frequency

[0162] I_comp_motor compressor current

[0163] Tcond_out Condenser outlet refrigerant temperature

[0164] Twater_delta Evaporator inlet and outlet water temperature difference

[0165] Tcomp_delta compressor suction and discharge temperature difference

[0166] Tcond_deltaCondenser inlet and outlet refrigerant temperature difference

[0167] The predicted value for each input and the measured value y i ; We calculate the mean absolute error (MAE) of this input quantity:

[0168]

[0169] Obviously, for k inputs there are k MAEs, characterized by the fact that the enhancement-reduction bias is defined as:

[0170]

[0171] Among the parameters calculated in this application, the reduction deviation enhancement coefficient β represents the contribution of each reduction deviation, but β1,β2,…,β k This is different from the weighted algorithm, because the general weighted algorithm configures a number of coefficients whose sum is 1, in order to map a set of vectors with a larger sum of values ​​to the average value; and here β1, β2, …, β k The contribution is to improve the sensitivity by comparing the deviation that can better reflect the degree of refrigerant leakage according to the sensitivity, where β1, β2, …, β k The sum of may not be equal to 1.

[0172] As an example, we take three sensors as an example to calculate the reduction deviation enhancement coefficient (β1, β2, β3) of each sensor. As shown in Table 4 below, the actual value of the refrigerant charge in the first 1000 seconds (s) is 100 kg, that is, y = 100 kg, and each sensor takes 101 (one sample parameter is taken every 10 seconds) sample parameters (i = 101, j = 1, 2, 3).

[0173] Table 4

[0174]

[0175] At this time, the actual value for each time sample is:

[0176]

[0177] The difference between the actual value and the estimated value is:

[0178]

[0179] For 101 samples, the formula for the sum of squares of the differences between the measured values ​​and the estimated values ​​is as follows:

[0180]

[0181] Using the least squares method, find a set of coefficients β1, β2, β3 that minimizes Q.

[0182] The above embodiment performs calculations for three sensors, but the calculation method is also applicable to the calculation of the restoration deviation enhancement coefficients of any k sensors.

[0183] After completing the operation of step 413 , the operation proceeds to step 414 .

[0184] In step 414, the system processor 116 performs an input data encoding operation. Figure 2A and 2B To describe the input data encoding operation.

[0185] After optimization, Figure 2B As shown, in step 414, the data test data sequence Cj (j = 1, 2, ..., k) is intercepted into several input sub-vectors according to the rolling cycle time, that is, Each input sub-vector has n vector elements (or n-dimensional vectors). Each of the n-dimensional input sub-vectors has i (i = 1, 2 ..., n) vector elements, that is: For example, we analyze the data of the past month (long time series data), and the starting point and end point of the data time correspond to the beginning and end of the month. Based on the above long time series data, we indent and truncate it into several sub-vectors.

[0186] After completing the operation of step 414 , the operation proceeds to step 415 .

[0187] In step 415, the system processor 116 performs a training LSTM model operation. Figure 2A and 2B To describe the operation of training LSTM model.

[0188] In the input vector group module 202 (see Figure 2B ) receives the n-dimensional input sub-vector After that, the output vector group module 210 (see Figure 2B ) generates several output sub-vectors Right now

[0189] After completing the operation of step 415 , the operation proceeds to step 416 .

[0190] In step 416 , the controller 116 obtains a test threshold.

[0191] Now, an example is given with three sensors (ie, k=3).

[0192] For the first sensor, the elements of the first output vector group Y1 are The corresponding elements of the first target vector group X2 are The restored deviation for the first output vector group is:

[0193]

[0194] For the second sensor, the elements of the first output vector group Y2 are The corresponding elements of the first target vector group X2 are The restored deviation for the first output vector group is:

[0195]

[0196] For the third sensor, the elements of the first output vector group Y3 are The corresponding elements of the first target vector group X3 are The restored deviation for the first output vector group is:

[0197]

[0198] Under the assumption of three input sensors, there are three restoration deviation enhancement coefficients β1, β2, β3, so the enhanced restoration deviation of the first output vector group is calculated as follows:

[0199]

[0200] The “2” on the left side of the equal sign indicates that the start and end times correspond to the second input vector group.

[0201] For the output vector groups of the second, third, and other subsequent cycle times, the present application repeats the above calculation process to obtain the subsequent EnhancedRE 3 ,EnhancedRE 4 ,…,EnhancedRE n+1 . For several enhanced reduction deviations EnhancedRE 3 ,EnhancedRE 4 ,…,EnhancedRE n+1 Sort from high to low, and take the average of the first o (10 as an embodiment) enhanced restoration deviations as the test threshold. The specific formula for obtaining the diagnostic threshold coefficient is as follows:

[0202]

[0203] Where i represents the sequence number of the first o enhanced reduction deviations after sorting, and g0 corresponds to the test data without refrigerant leakage. Then, the test data under the first leakage amount, the second leakage amount, and the third leakage amount are screened out from the test data set, and step 418 is repeated in the same way to obtain the first test threshold g1, the second test threshold g2, and the third test threshold g3.

[0204] After completing the operation of step 416 , the operation proceeds to step 418 .

[0205] In step 418 , the controller 116 obtains a test threshold scaling factor.

[0206] The test threshold proportionality coefficients α1, α2, α3 are calculated according to the test thresholds g0, g1, g2, g3 obtained in step 416, wherein α1 = g1 / g0; α2 = g2 / g0; α3 = g3 / g0.

[0207] At this point, the processing of the test data set (Test Data) has been completed.

[0208] Figure 5 Shows Figure 3 The specific process of step 306 in the flowchart 300 is used to illustrate more specific steps of step 306 .

[0209] like Figure 5 As shown, in step 502, under the condition that there is no refrigerant leakage in the air-conditioning system, the system processor 116 performs a data collection operation.

[0210] In step 502, the system processor 116 performs data acquisition operations similar to step 402. After completing the operation of step 502, the operation proceeds to step 504.

[0211] In step 504 , the system processor 116 performs a data cleaning operation.

[0212] In step 504, the system processor 116 performs a data cleaning operation similar to step 406. After completing the operation of step 504, the operation proceeds to step 506.

[0213] In step 506, the system processor 116 performs a data input volume screening operation.

[0214] In step 506, the system processor 116 performs a data input volume screening operation similar to step 412. After completing the operation of step 506, the operation proceeds to step 508.

[0215] It should be noted that operation steps 504-506 are performed on the sensor j (j=1, 2, ..., k) collected in step 502. Figure 2B The test data sequence Cj (j=1, 2, ..., k) shown is optimized. Even if steps 504-506 are not executed and step 502 is directly transferred to step 508, this operation method can also obtain improved results and is also within the scope of protection claimed in this application.

[0216] In step 508, the system processor 116 performs an input data encoding operation. The system processor 116 performs an input data encoding operation similar to step 414. After completing the operation of step 508, the operation proceeds to step 510.

[0217] In step 510, the system processor 116 performs the operation of training the LSTM model. The system processor 116 performs the input data encoding operation similar to step 415. After completing the operation of step 510, the operation proceeds to step 512.

[0218] In step 512 , the system processor 116 performs an operation of calculating a restoration deviation.

[0219] Similar to the method in step 416, taking three sensors as an example, for the first sensor, the elements of the first output vector group Y1 are The corresponding elements of the first target vector group X2 are The restored deviation for the first output vector group is:

[0220]

[0221] For the second sensor, the elements of the first output vector group Y2 are The corresponding elements of the first target vector group X2 are The restored deviation for the first output vector group is:

[0222]

[0223] For the third sensor, the elements of the first output vector group Y3 are The corresponding elements of the first target vector group X3 are The restored deviation for the first output vector group is:

[0224]

[0225] After completing the operation of step 512 , the operation proceeds to step 514 .

[0226] In step 514 , the system processor 116 performs an operation of calculating an enhanced restoration deviation.

[0227] Under the assumption of three input sensors, there are three restoration deviation enhancement coefficients β1, β2, β3, so the enhanced restoration deviation of the first output vector group is calculated as follows:

[0228]

[0229] For the output vector groups of the second, third, and other subsequent cycle times, the present application repeats the above calculation process to obtain the enhanced restoration bias EnhancedRE of the subsequent output vector groups. 3 ,EnhancedRE 4 ,…,EnhancedRE n+1 .

[0230] After completing the operation of step 514 , the operation proceeds to step 516 .

[0231] In step 516, the system processor 116 selects the largest one from the enhanced restoration deviations of the plurality of output vector groups obtained from the above operation as the no-leakage diagnosis threshold:

[0232] G0=max(EnhancedRE i ), where i=2,3,4…

[0233] At this point, the operation of collecting and training data without refrigerant has been completed. After completing the operation of step 516, the operation proceeds to step 308 (see Figure 3 ).

[0234] Figure 6 Shows Figure 3 The specific process of step 308 in the flowchart 300 is used to illustrate more specific steps of step 308 .

[0235] like Figure 6 As shown, in step 602, in order to diagnose a refrigerant leak, the system processor 116 performs a data acquisition (or online data acquisition) operation.

[0236] In step 602, the system processor 116 performs data collection (or online data collection) similar to step 404. After completing the operation of step 602, the operation proceeds to step 604.

[0237] At step 604 , the system processor 116 performs a data cleaning operation.

[0238] In step 604, the system processor 116 performs a data cleaning operation similar to step 406. After completing the operation of step 604, the operation proceeds to step 606.

[0239] In step 606, the system processor 116 performs a designated input quantity operation. The designated input quantity operation is a process of screening sensors, that is, selecting k sensor data from all sensor data as time series data to be analyzed.

[0240] It should be noted that operation steps 604-606 are performed on the sensor j (j=1, 2, ..., k) collected in step 602. Figure 2B The test data sequence Cj (j=1, 2, ..., k) shown is optimized. Even if steps 604-606 are not executed and step 602 is directly transferred to step 608, this operation method can also obtain improved results and is also within the scope of protection claimed in this application.

[0241] After completing the operation of step 606 , the operation proceeds to step 608 .

[0242] In step 608, the processor 116 performs an input data encoding operation, namely, sub-vector truncation. The system processor 116 performs an input data encoding operation similar to step 414. After completing the operation of step 608, the operation proceeds to step 610.

[0243] At step 610, the system processor 116 performs a prediction operation using the LSTM decoder. The prediction operation refers to the process of inputting a latent state vector into the LSTM decoder and generating an output vector group.

[0244] After completing the operation of step 610 , the operation proceeds to step 612 .

[0245] In step 612, the system processor 116 performs an operation of calculating the enhanced restoration deviation. When executing step 612, the system processor 116 obtains the enhanced restoration deviation according to the restoration deviation enhancement coefficient and the restoration deviation. The formula for obtaining the enhanced restoration deviation is as follows:

[0246] where i = 1, 2, 3, ...,

[0247] After completing the operation of step 612 , the operation proceeds to step 613 .

[0248] In step 613, the first diagnostic threshold G1, the second diagnostic threshold G2 and the third diagnostic threshold G3 are calculated according to the test threshold proportional coefficients α1, α2, α3 calculated in step 418 and the no-leakage diagnostic threshold G0 calculated in step 516. The specific calculation formula is as follows:

[0249] G1=G0×α1,

[0250] G1=G0×α2,

[0251] G1=G0×α3.

[0252] After completing the operation of step 613 , the operation proceeds to step 614 .

[0253] In step 614 , the system processor 116 reads the four levels of diagnostic thresholds G0 , G1 , G2 , and G3 . After completing the reading operation, the operation flow goes to step 615 .

[0254] The diagnostic threshold is a set of values. The first value is the maximum enhanced reduction deviation in the data without refrigerant leakage, and each subsequent value is greater than the previous value. A preferred embodiment is that the diagnostic threshold includes four diagnostic thresholds G0, G1, G2, and G3, wherein G0 is the diagnostic threshold in the data without leakage, G1 = α1xG0, G2 = α2xG0, and G3 = α3xG0. The results of one embodiment are as follows:

[0255] G0=0.234, α1=1.5, α2=3, α3=6,

[0256] G1=1.5×0.2334≈0.35,

[0257] G2=3×0.2334≈0.7,

[0258] 63=6×0.2334≈1.4.

[0259] At step 614, the system processor 116 compares the enhanced restoration deviation with the no-leakage threshold. If the enhanced restoration deviation is not greater than the no-leakage threshold, the system processor 116 proceeds to step 615 to report system safety and then proceeds to step 310 (see Figure 3 ), stop the diagnostic operation process; if the restored deviation is greater than the no-leakage threshold, go to step 616.

[0260] At step 616, the system processor 116 compares the enhanced reduction deviation with the first threshold value. If the enhanced reduction deviation is not greater than the first threshold value, the system processor 116 proceeds to step 617 to report a minor leakage risk, and then proceeds to step 310 (see Figure 3 ), stop the diagnostic operation process; if the restored deviation is greater than the first threshold, go to step 618.

[0261] At step 618, the system processor 116 compares the enhanced reduction deviation with the second threshold value. If the enhanced reduction deviation is not greater than the second threshold value, the system processor 116 proceeds to step 619 to report a moderate leakage risk and then proceeds to step 310 (see Figure 3 ), stop the diagnostic operation process; if the restored deviation is greater than the second threshold, go to step 620.

[0262] In step 620, the system processor 116 compares the enhanced restoration deviation with the third threshold value. If the enhanced restoration deviation is not greater than the third threshold value, the system processor 116 proceeds to step 621 to report a moderate leakage risk and then proceeds to step 310 (see Figure 3 ), stop the diagnostic operation process; if the enhanced reduction deviation is greater than the third threshold, go to step 622, report the risk of severe leakage, and then go to step 310 (see Figure 3 ) to stop the diagnostic process.

[0263] Figure 7 , 8 The implementation process of verifying the effectiveness of the model using a piece of experimental data is shown.

[0264] Figure 7 The test steps under different refrigerant injection amounts are shown: Step 1, conduct several working condition combination tests under the standard refrigerant injection amount of 1.6kg; Step 2, discharge part of the refrigerant to adjust the injection amount to 1.36, simulate a 15% leakage state, and conduct several working condition combination tests; Step 3, continue to discharge part of the refrigerant to adjust the injection amount to 1.12, simulate a 30% leakage state, and conduct several working condition combination tests. The first 1661 data are data under the standard refrigerant injection amount. This set of data is used to train the model parameters of the LSTM encoding network and decoding network. After training, the no-leakage threshold G0 can be calculated, and the threshold proportional coefficient (not shown) can be used to calculate the first diagnostic threshold G1, the second diagnostic threshold G2 and the third diagnostic threshold G3.

[0265] exist Figure 7 In the figure, the horizontal axis is the time series number, that is, the change of the refrigerant filling amount during the whole test process is recorded, and the vertical axis is the refrigerant filling amount. The first 1661 data are used as non-leakage data to train the model parameters of the LSTM encoding network and decoding network, and calculate the threshold array of refrigerant leakage. The following 5400 data are used as online monitoring verification data. Different thresholds are used to obtain the calculation and restoration enhancement deviation results, and refrigerant leakage diagnosis is performed.

[0266] Figure 8 Shows Figure 7 The implementation effect of the diagnostic process shown. The trained model is used to process the data with filling amounts of 1.36kg, 1.12kg and 1.6kg. The first threshold, the second threshold and the third threshold here correspond to the enhanced reduction deviation under 5% leakage, 10% leakage and 15% leakage, respectively, and the part exceeding the threshold is marked in red on the display screen, so the application can identify whether a corresponding degree of refrigerant leakage occurs from a period of time series data.

[0267] Fig. 9 Shows Figure 1AFIG. 1 is a block diagram of the controller 116 in more detail, showing the major components of the controller 116. The controller 116 is capable of storing and executing Figure 3-6 The procedures, storage and calling shown in the flow chart are as follows Figure 3-6 Parameters required for the process shown.

[0268] like Fig. 9 As shown, the controller 116 includes a bus 902, a processor 904, a memory 906, an input interface 908, and an output interface 910. The processor 904, the memory 906, the input interface 908, and the output interface 910 are connected to the bus 902. The processor 904 can read a program (or instruction) from the memory 906 and execute the program (or instruction) to process data; the processor 904 can also write data or a program (or instruction) into the memory 906. The memory 906 can store programs (instructions) or data. By executing the instructions in the memory 906, the processor 904 can control the memory 906, the input interface 908, and the output interface 910. In the present application, the memory 906 is capable of storing instructions for executing Figure 3-6 The program of the process shown in and the operating parameters required to execute the program.

[0269] The input interface 908 is configured to receive sensor parameters from the condenser sensor, evaporator sensor, compressor sensor, throttle valve sensor and environmental sensor through connection lines 127 , 121 , 125 , 129 , 123 , respectively, and convert the data of these parameters into signals recognizable by the processor 904 and store them in the memory 906 .

[0270] The processor 904 is configured to calculate the diagnostic report related parameters according to the program stored in the memory 906 .

[0271] The output interface 910 is configured to receive diagnostic report related parameters from the processor 904, convert the related parameters into a readable diagnostic report, and output the generated diagnostic report from the output interface 910 through the connection line 190. The output interface 910 is also configured to receive condenser control parameters, throttle valve control parameters, compressor control parameters, evaporator control parameters, heat load (water machine) control parameters, refrigerant pump control parameters and cold water pump control parameters from the processor 904, and output the generated condenser control signal, throttle valve control signal, compressor control signal, evaporator control signal, heat load (water machine) control signal, refrigerant pump control signal and cold water pump control signal from the output interface 910 through the connection lines 191, 192, 193, 194, 195, 196, 197.

[0272] Since it is difficult for us to design a completely sealed air-conditioning system, refrigerant leakage is the most common air-conditioning failure. When a refrigerant leak occurs in an air-conditioning system, it usually manifests as a deterioration in cooling or heating capacity (corresponding to a heat pump) and an increase in power consumption. After a refrigerant leak occurs, in order to maintain the set temperature of the room and maintain the cooling capacity, the air-conditioning system has to run at a higher current, thereby increasing the probability of failure. Compared with the air-conditioning control system based on data-driven or machine learning diagnostic methods in the prior art, the air-conditioning control system of the present application includes but is not limited to the following advanced technical effects:

[0273] First, the method of the present application is a weakly supervised learning method, that is, a small amount of refrigerant leakage test is used to obtain the threshold ratio and the reduction deviation enhancement coefficient. In a product series, only a small amount of testing is required on one model of air conditioner, and the rest of the training and diagnosis process is completed entirely on the client group.

[0274] Second, the diagnostic results of the method of the present application are highly sensitive and stable. The sensitivity means that a refrigerant leakage within 5% can be identified. Usually, when a refrigerant leakage within 5% occurs in an air-conditioning unit, the state parameters of the system change very slightly. The present application can improve the sensitivity by using the reduction deviation enhancement coefficient. The stability means that the present application can make reliable diagnoses under complex variable working conditions.

[0275] Third, the present application performs real-time diagnosis on the newly collected status data and gives the diagnosis results, so that the customer or maintenance personnel can find the leak in the first place and make repairs, thereby minimizing the customer's operating costs and the losses caused by downtime due to unexpected air conditioning failures.

[0276] Fourth, this application is applicable to all air conditioners that use refrigerant vapor compression cycle for cooling or heating.

[0277] Fifth, the present application can manage the diagnostic thresholds in a graded manner to achieve leakage risk grading including (i) safety, (ii) reporting slight leakage risk, (iii) reporting slight leakage risk, (iv) reporting moderate leakage risk and (v) reporting severe leakage risk.

[0278] Sixth, the control method of the present application can find sensitive features from a large number of system sensor parameters and separate the factors of refrigerant leakage from other fault factors to speed up the processing speed of the diagnostic method and reduce the processing burden of the system.

[0279] Although the present application has been described in conjunction with the examples of the embodiments summarized above, various alternatives, modifications, changes, improvements and / or substantially equivalent solutions, whether known or currently or foreseeable in the near future, may be apparent to those skilled in the art. In addition, the technical effects and / or technical problems described in this specification are exemplary rather than restrictive; so the disclosure in this specification may be used to solve other technical problems and have other technical effects and / or may solve other technical problems. Therefore, the examples of the embodiments of the present application as stated above are intended to be illustrative rather than restrictive. Various changes may be made without departing from the spirit or scope of the present application. Therefore, the present application is intended to include all known or earlier developed alternatives, modifications, changes, improvements and / or substantially equivalent solutions.

Claims

1. A control method for refrigerant leakage diagnosis of an air conditioning system, the air conditioning system comprising a unit, characterized in that: The control method for refrigerant leakage diagnosis of an air conditioning system comprises the following steps: S01, in the case of combined tests under different refrigerant filling amounts, obtain a test data sequence, and obtain the reduction deviation enhancement coefficients (β1, β2, …, β k ); obtain an n-dimensional input subvector from the test data sequence and obtain an n-dimensional output subvector for the n-dimensional input subvector, obtain respective restoration deviations according to the n-dimensional input subvector and the n-dimensional output subvector, and based on the respective restoration deviations and the restoration deviation enhancement coefficients (β1, β2, …, β k ) sequentially determine the test thresholds (g0, g1, g2, g3) of the test data, and obtain the test threshold proportional coefficients (α1, α2, α3) according to the test thresholds (g0, g1, g2, g3); S02, in the case of no refrigerant leakage, obtain a test data sequence, obtain the n-dimensional input sub-vector from the test data sequence and obtain the n-dimensional output sub-vector for the n-dimensional input sub-vector, obtain the respective restoration deviations in the state of no leakage according to the n-dimensional output sub-vector, and based on the restoration deviation enhancement coefficients (β1, β2, …, β k ), obtaining several diagnostic thresholds (G0, G1, G2, G3) by using respective restoration deviations and test threshold proportional coefficients (α1, α2, α3) in the state without leakage; S03, in the refrigerant leakage diagnosis process, obtaining a test data sequence, obtaining the n-dimensional input sub-vector from the test data sequence and obtaining the n-dimensional output sub-vector for the n-dimensional input sub-vector, obtaining the respective restoration deviations in the refrigerant leakage diagnosis process according to the n-dimensional output sub-vector, and based on the respective restoration deviations in the refrigerant leakage diagnosis process, the restoration deviation enhancement coefficients (β1, β2, …, β k ) to obtain the enhanced restoration deviation, and compare the enhanced restoration deviation with the plurality of diagnostic thresholds (G0, G1, G2, G3) to perform refrigerant leakage diagnosis.

2. The control method for refrigerant leakage diagnosis of an air conditioning system according to claim 1, characterized in that: The air conditioning system comprises a unit and k sensors, wherein the k sensors obtain state parameters from the unit, and the method for diagnosing refrigerant leakage in the air conditioning system comprises the following steps: In step S01, in the case of a combined test under different refrigerant charge amounts, the steps are performed: (i) obtaining all test data sequences (Cj) for each sensor (j) among the k sensors, each test data sequence (Cj) comprising sequence elements (ii) According to the test data sequence (Cj), the restoration deviation enhancement coefficients (β1, β2, ..., β k ), (iii) for each sensor (j) among the k sensors, obtain an n-dimensional input subvector from the test data sequence (Cj) (iv) corresponding to the n-dimensional input sub-vector Get an n-dimensional output subvector (v) Obtain k of the n-dimensional output sub-vectors The respective reduction deviations, (vi) based on the k restoration deviations, the restoration deviation enhancement coefficients (β1, β2, ..., β k ) obtain the test thresholds (g0, g1, g2, g3) of refrigerant leakage from small to large, and obtain the test threshold proportional coefficients (α1, α2, α3) according to the test thresholds (g0, g1, g2, g3); In step S02, when there is no refrigerant leakage, the following steps are performed: (i) obtaining all test data sequences (Cj) for each sensor (j) among the k sensors, each test data sequence (Cj) comprising sequence elements (ii) for each sensor (j) among the k sensors, obtain an n-dimensional input subvector from the test data sequence (Cj) (iii) corresponding to the n-dimensional input sub-vector Get an n-dimensional output subvector (iv) Obtain k of the n-dimensional output sub-vectors The respective reduction deviations, (v) Based on the k restoration deviations, the restoration deviation enhancement coefficients (β1, β2, …, β k ) and the test threshold proportional coefficient (α1, α2, α3), to obtain several diagnostic thresholds (G0, G1, G2, G3); In step S03, during the refrigerant leakage diagnosis process, the following steps are performed: (i) obtaining all test data sequences (Cj) for each sensor j among the k sensors, each test data sequence (Cj) comprising sequence elements (ii) For each sensor (j) of the k sensors, obtain an n-dimensional input subvector from the test data sequence (Cj) (iii) corresponding to the n-dimensional input sub-vector Get an n-dimensional output subvector (iv) Obtaining the n-dimensional output sub-vector The restoration deviation, (v) based on the calculated restoration deviations of the k sensors and the k restoration deviation enhancement coefficients (β1, β2, ..., β k ) to obtain the enhanced restoration deviation, (vi) comparing the enhanced reduction deviation with the plurality of diagnostic thresholds (G0, G1, G2, G3) to perform refrigerant leakage diagnosis.

3. The control method for refrigerant leakage diagnosis of an air conditioning system according to claim 2, characterized in that: The step S01 is performed under several different refrigerant filling amounts, and the several different refrigerant filling amounts include no leakage, a first leakage amount, a second leakage amount, and a third leakage amount.

4. The control method for refrigerant leakage diagnosis of an air conditioning system according to claim 2, characterized in that: When executing step S01(v), step S02(iv) and step S03(iv) to obtain the restoration deviation, the following formula is used for calculation:

5. The control method for refrigerant leakage diagnosis of an air conditioning system according to claim 2, characterized in that: When executing step S01(vi) to obtain the test threshold (g0, g1, g2, g3), perform the following steps: The augmented restoration bias for each output subvector is calculated by the following formula: Filter out the non-leakage test data from the test data set, and sort the obtained multiple enhanced restoration deviations from high to low. The mean of the first o enhanced restoration deviations after sorting is taken as the original test threshold g0. The specific calculation formula is as follows: Where i represents the sequence number of the first o enhanced restoration deviations after sorting, Test data under the first leakage amount, the second leakage amount and the third leakage amount are screened out from the test data set, and the means of the first o enhanced restoration deviations after sorting under the first leakage amount, the second leakage amount and the third leakage amount are respectively used as the first test threshold (g1), the second test threshold (g2) and the third test threshold (g3).

6. The control method for refrigerant leakage diagnosis of an air conditioning system according to claim 2, characterized in that: When executing step S01(ii), the restoration deviation enhancement coefficient (β1, β2, ..., β k ), perform the following steps: Taking the refrigerant charge as the dependent variable and other screening data as the independent variables, the function form is established as follows: Where i represents a sample, where i = 1, 2, ..., m, and m represents m time samples. Represents the refrigerant charge obtained by regression. The estimation method of multivariate linear regression parameters usually adopts the least square method. For sample i, the difference between the actual value and the estimated value is: For all m time samples, the formula for the sum of squares of the differences between the measured values ​​and the estimated values ​​is as follows: The principle of the least squares method is to find a set of coefficients β0, β1,…, β k Make Q minimum.

7. The control method for refrigerant leakage diagnosis of an air conditioning system according to claim 2, characterized in that: When executing step S02(v) to obtain the plurality of diagnostic thresholds (G0, G1, G2, G3), the following steps are performed: The largest one is selected from the enhanced restoration deviations of multiple output vector groups as the no-leakage diagnosis threshold (G0). The specific calculation formula is as follows: G0=max(EnhancedRE i ), where i=2,3,4….

8. The control method for refrigerant leakage diagnosis of an air conditioning system according to claim 2, characterized in that: When executing step S03(v) to obtain the enhanced restoration deviation, the calculation is performed using the following formula: Where i=1,2,3….

9. The control method for refrigerant leakage diagnosis of an air conditioning system according to claim 2, characterized in that The following steps are also included: Find sensitive features from a large number of system sensor parameters, and Separate refrigerant leakage from other fault factors.

10. An air conditioning system, comprising a unit, a plurality of sensors and a controller, wherein the sensor obtains vector parameters from the unit, characterized in that: The controller controls the air conditioning system according to the parameters obtained by the method according to any one of claims 1-9.

Images