A multi-dimensional taylor web-based air conditioning unit fault prediction method

Abstract

The application discloses a kind of air conditioning unit fault prediction methods based on multidimensional Taylor net, based on the respective fault feature corresponding to each type of fault specified in the preset air conditioning unit, the fault static threshold range of effective characteristic variable under normal state is obtained by exponential weighted moving average control chart calculation, then the prediction output value based on multidimensional Taylor net model is compared with static fault static threshold, the prediction of each type of fault specified in air conditioning unit is realized.The multidimensional fitting of multidimensional Taylor net BP-MTN model is used to realize the parallel prediction of each fault feature in the application, not only the fitting performance of original multidimensional Taylor net is enhanced, but also the prediction efficiency of fault feature is improved, so that the air conditioning unit fault prediction method based on multidimensional Taylor net can effectively improve the prediction efficiency of air conditioning unit fault.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of air conditioning unit fault detection, and particularly relates to an air conditioning unit fault prediction method based on a multi-dimensional Taylor network. BACKGROUND

[0002] In public buildings, heating, ventilation and air conditioning systems are the main energy-consuming equipment, accounting for about 50-60%. About 42% of the refrigeration energy consumption and 26% of the maintenance costs in heating, ventilation and air conditioning are caused by equipment failure. It is estimated that heating, ventilation and air conditioning fault prediction and diagnosis can reduce 10%-40% of energy consumption. Therefore, energy-saving and environmental protection treatment needs to be implemented to improve the scientificity and rationality of the application of heating, ventilation and air conditioning systems. As an important subsystem of heating, ventilation and air conditioning systems, air conditioning units are used to adjust indoor air to a suitable temperature. Moreover, the equipment in the air conditioning unit is coupled to each other, and through closed-loop control, the indoor temperature, humidity and the like are controlled to provide a comfortable environment for users. In this case, once a fault occurs in one of the equipment in the air conditioning unit, a chain reaction will be caused, resulting in the transmission and diffusion of the fault. Therefore, it is necessary to accurately predict the fault of the air conditioning unit under the condition of equipment coupling.

[0003] There are many existing technologies for fault prediction. The most common ones can be divided into: model-based fault prediction technology, data-driven fault prediction technology, and probability statistics-based fault prediction technology. However, the prediction accuracy of the existing prediction methods is limited, and the prediction efficiency is not high. SUMMARY

[0004] In view of the above problems, the present application provides an air conditioning unit fault prediction method based on a multi-dimensional Taylor network.

[0005] To achieve the purpose of the present application, an air conditioning unit fault prediction method based on a multi-dimensional Taylor network is provided, comprising the following steps:

[0006] s1: based on the preset fault features corresponding to the specified type of fault in the air conditioning unit, from the ASHRAE RP-1312 data set, collect fault feature values at l normal operation state time points in the historical time direction starting from t0 to construct an x multi-dimensional matrix; at the same time, collect fault feature values at l normal operation state time points in the historical time direction starting from (t0+t) to construct a y multi-dimensional matrix, to form a sample data set (x, y);

[0007] s2: performing data normalization operation update on the sample data set (x, y) to obtain sample data x * Multi-dimensional matrix and sample data y * Multi-dimensional matrix

[0008] s3: calculating the sample data y* a static fault threshold of each fault feature in the multi-dimensional matrix, and extracting the sample data y * a model estimated state of each fault feature in the next t time period as a multi-dimensional matrix a corresponding label vector;

[0009] s4: based on the sample data x * the dimensions of the multi-dimensional matrix, constructing a multi-dimensional Taylor network BP-MTN model with the same dimension input and output, and then inputting the sample data x * the multi-dimensional matrix as a model input feature, inputting the multi-dimensional Taylor network BP-MTN model and outputting a model estimated state then calculating the error of the label vector y * and the model estimated state , and finally completing the training of the multi-dimensional Taylor network BP-MTN model according to the back propagation method;

[0010] s5: based on the fault features corresponding to the specified type of fault in the air conditioning unit, from the ASHRAE RP-1312 data set, collecting l fault feature values at the time points containing normal operation state and fault state from t0 time to the historical time direction to construct X multi-dimensional matrix; at the same time, collecting l fault feature values at the time points containing normal operation state and fault state from (t0+t) time to the historical time direction to construct Y multi-dimensional matrix, to form a test data set (X, Y), and then performing a normalization operation on the test data set (X, Y) to obtain an updated test data set (X * , Y * );

[0011] s6: based on the static fault threshold of each fault feature calculated in step s3 and the trained multi-dimensional Taylor network BP-MTN model, determining whether the estimated state of each fault feature in the test data set (X * , Y * ) from (t0+t) time to the first t time period after that is abnormal, if it is abnormal, then determining that the corresponding fault feature is an abnormal fault feature, and further identifying the fault type corresponding to the abnormal fault feature based on the corresponding relationship between the fault feature and the fault type, so as to subsequently determine whether the fault actually occurs; otherwise, it is determined that the estimated state of the air conditioning unit from (t0+t) time to the first t time period after that is normal.

[0012] Further, in step s3, the sample data y * The formula for calculating the static fault threshold of each fault feature in the multi-dimensional matrix is as follows:

[0013]

[0014] Wherein, UCL represents the upper limit of static fault feature threshold, LCL represents the lower limit of static fault feature threshold, μ0 represents the current observation sample mean, σ is the standard deviation of the current observation sample, λ is a smoothing constant between 0 and 1, and L represents the control limit parameter.

[0015] Further, the multi-dimensional Taylor network BP-MTN model comprises an input layer, a data processing layer, an output layer, a full connection layer and an activation function layer, and the layers are sequentially connected in order, and the output of the previous layer is the input of the next layer.

[0016] Further, the formula of the weighted summation of the data processing layer is as follows:

[0017]

[0018] Wherein, T j represents the jth output node of the multi-dimensional Taylor network, m represents the number of the highest expansion term of the data processing layer, N(n,m) represents the number of terms of the polynomial after the nth input fault feature of the data processing layer is expanded by m power, q∈N(n,m) represents the qth polynomial after the power expansion of the data processing layer, w j,q represents the weight value before the qth polynomial of the jth output node of the multi-dimensional Taylor network, σ q,i represents the power of the variable x in the qth polynomial.

[0019] Further, in step s4, the multi-dimensional Taylor network BP-MTN model is fitted and predicted according to the following formula:

[0020]

[0021] Wherein, θ j represents the weight value vector of the jth node value of the full connection layer entering the activation function, and relu is the activation function.

[0022] Further, in step s4, the fitting error square is calculated by the back propagation algorithm according to the following formula:

[0023]

[0024] Further, in step s6,

[0025] It is judged whether the estimated state of each fault feature in the to-be-tested data set (X * , Y * ) in the first t time period after the time t0+t is abnormal or not, which includes: statistical identification by marking the fault node and the fault type.

[0026] Firstly, each fault feature is set as a fault feature node, when the fault feature is in a normal state, the node value is set as 0, when the fault feature is abnormal, the node value is set as 1, then a fault feature statistical vector [1, 0, 1, …, 0] is formed by 0 and 1, and the vector arrangement order is consistent with the node sorting order, at the same time, the fault type is also coded by 1 and 0, so that the fault type can be judged by comparing the state value statistics of the fault node and the fault coding.

[0027] Further, the preset specified types of faults in the air conditioning unit include: return air stuck at a fixed speed fault, return air fan completely damaged fault, outdoor air damper stuck at a full closed position fault, cooling coil valve stuck at a full open position fault, and cooling coil valve stuck at a fixed angle fault;

[0028] The fault features corresponding to the return air stuck at a fixed speed fault include: return air fan energy, air conditioner supply air temperature, supply air rate and return air rate;

[0029] The fault features corresponding to the return air fan completely damaged fault include: air conditioner supply air temperature, return air rate and mixed air temperature;

[0030] The fault features corresponding to the outdoor air damper stuck at a full closed position fault include: outdoor air flow rate, return air fan energy and supply air temperature;

[0031] The fault features corresponding to the cooling coil valve stuck at a full open position fault include: outdoor air flow rate, supply air temperature and return air temperature;

[0032] The fault features corresponding to the cooling coil valve stuck at a fixed angle fault include: outdoor air flow rate and return air rate.

[0033] Compared with the prior art, the air conditioning unit fault prediction method based on the multi-dimensional Taylor net has the following beneficial technical effects:

[0034] The air conditioning unit fault prediction method based on the multi-dimensional Taylor net designed in the application, based on each fault feature corresponding to each type of fault in the preset air conditioning unit, calculates the fault threshold range of the effective feature variable in the normal state, compares the multi-dimensional Taylor net prediction output value with the fault threshold range, realizes the prediction of the specified types of faults in the air conditioning unit, compared with other data-driven fault prediction technologies, the multi-dimensional Taylor net structure can express the complex nonlinear function with the mathematical combination of series terms, without a large amount of network training, the nonlinear system under the conventional condition can be fitted, and the multi-dimensional Taylor net has the advantages of simple network structure, strong network generalization ability, certain interpretability and stronger approximation performance; and further, the application can effectively improve the prediction efficiency of the air conditioning unit fault. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 FIG. 1 is a flow diagram of a multi-dimensional Taylor net-based air conditioning unit fault prediction method according to an embodiment;

[0036] Figure 2 FIG. 2 is a network structure diagram of a multi-dimensional Taylor net adding a full connection layer and an activation function layer according to an embodiment;

[0037] Figure 3 FIG. 3 is a fault feature node identification statistical diagram according to an embodiment;

[0038] Figure 4 FIG. 4 is a waveform diagram illustrating a cooling coil valve stuck in the fully open position fault diagnosis principle according to an embodiment. DETAILED DESCRIPTION

[0039] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0040] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor does it necessarily refer to a particular embodiment that is "preferred" over other embodiments. It will be explicitly understood that the embodiments described herein can be combined with other embodiments.

[0041] The present application designs a multi-dimensional Taylor net-based air conditioning unit fault prediction method, which is based on each fault feature corresponding to a specified type of fault in a preset air conditioning unit, and realizes prediction of the specified type of fault in the air conditioning unit. In actual application, as shown in FIG. 1, the following steps are performed in real time. Figure 1

[0042] Step 1. Since the ASHRAE RP-1312 dataset stores 95 characteristic parameters collected during the operation of the air conditioning unit, including operation data in the normal state and operation data in the fault state, for each fault feature involved in a specified type of fault in a preset air conditioning unit, a plurality of time fault feature values in a historical time direction from a current time t0are preset from the ASHRAE RP-1312 dataset, to construct an x multi-dimensional matrix; a plurality of time fault feature values in a historical time direction from a current time (t0+t) are preset, to construct a y multi-dimensional matrix, to obtain a sample dataset (x, y), and then proceed to step 2.

[0043] ​Step 2. The multi-dimensional matrix x, y is subjected to data normalization operation according to the following formula:

[0044]

[0045] Wherein: min(x i ) represents the minimum value in the input i-dimensional fault feature state vector in the sample data set x multi-dimensional matrix, max(x i ) represents the maximum value in the input i-dimensional fault feature state vector in the sample data set x multi-dimensional matrix; min(y i ) represents the minimum value of the label value in the i-dimensional fault feature state vector in the sample data set y multi-dimensional matrix, max(y i ) represents the maximum value of the label value in the i-dimensional fault feature state vector in the sample data set y multi-dimensional matrix; thus the normalized sample data x * , y * are obtained as network input; then step 3 is entered.

[0046] Step 3. The fault threshold of each fault feature in y * multi-dimensional matrix is calculated, and the formula is as follows:

[0047]

[0048] Wherein, UCL is the upper limit of the static fault feature threshold, LCL is the lower limit of the static fault feature threshold, μ0 is the current observation sample mean, σ is the standard deviation of the current observation sample, λ is a smoothing constant between 0 and 1, L is a control limit parameter, usually set to 1.5; then step 4 is entered.

[0049] Step 4. According to the dimension of the sample input data x*, the BP-MTN model is built, wherein the MTN structure is divided into input layer, data processing layer and output layer, wherein the nodes of the input and output layers are equal to the dimension of x*, and the data processing layer is used to realize the weighted summation of each power product term of the input variable, and the formula is as follows:

[0050]

[0051] Wherein, m represents the number of the highest expansion term of the multi-dimensional Taylor network data processing layer, N(n, m) represents the number of polynomials after n input fault features of the multi-dimensional Taylor network data processing layer are expanded by m power, q ∈ N(n, m) represents the qth polynomial after the data processing layer is expanded by power, T j represents the jth output node of the multi-dimensional Taylor network, w j,q represents the weight before the qth polynomial of the jth output node of the multi-dimensional Taylor network, σ q,i represents the power of the variable x * in the qth polynomial; then step 5 is entered.

[0052] Step 5. As shown in FIG. 5, after the multi-dimensional Taylor network structure built in step 4, a fully connected layer and an activation function layer are added, and the multi-dimensional fitting prediction is as follows: Figure 2

[0053]

[0054] wherein, θ j represents the weight vector of the jth node value of the fully connected layer entering the activation function, and relu is the activation function; then step 6 is entered.

[0055] Step 6. According to the principle of back propagation algorithm, the fitting error square is calculated as follows:

[0056]

[0057] Taking the output of the jth node as an example, the network-related weight gradient calculation is as follows:

[0058]

[0059] The average gradient of each weight is calculated from the gradient vector obtained from formula (6) The weight update is performed according to the following formula:

[0060]

[0061] Then step 7 is entered.

[0062] Step 7. According to the normal threshold interval of each fault feature data, it is judged whether there is an abnormality in the prediction state of the next time corresponding to the current time of each fault feature, if yes, the fault feature of the prediction state abnormality is determined as each abnormal fault feature, and the corresponding relationship between each type of fault and each fault feature involved in the pre-set air conditioning unit is further determined to determine that the air conditioning unit corresponding to the prediction state of the next time corresponding to the current time exists each specified type of fault corresponding to each abnormal fault feature; otherwise, it is determined that the prediction state of the air conditioning unit corresponding to the next time corresponding to the current time is normal. As shown in FIG. 7, n identification statistical nodes corresponding to the input fault feature are established to distinguish the corresponding relationship between the fault feature and the fault type, and the above fault type prediction is performed. Figure 3

[0063] In actual application, the specified type of fault includes return air stuck at fixed speed fault, return air fan completely damaged fault, outdoor air damper stuck at full closed position fault, cooling coil valve stuck at full open position fault, and cooling coil valve stuck at fixed angle fault.

[0064] ​​Among them, the fault characteristics corresponding to the fixed speed fault of the return air block include: return air fan energy, air conditioner supply air temperature, supply air rate, and return air rate.

[0065] The fault characteristics corresponding to a complete failure of the return air fan include: air conditioner supply air temperature, return air rate, and mixed air temperature.

[0066] The characteristics of an outdoor air damper stuck in the fully closed position include: outdoor air velocity, return air fan energy, and supply air temperature.

[0067] The characteristics of a cooling coil valve stuck in the fully open position include: outdoor air velocity, supply air temperature, and return air temperature.

[0068] The characteristics of a cooling coil valve stuck at a fixed angle include: outdoor air velocity and return air rate.

[0069] In practical applications, based on the fault characteristics corresponding to each of the aforementioned types of faults in the air conditioning unit, prediction of these fault types is achieved. The air conditioning unit fault prediction method based on a multidimensional Taylor network, designed in the above technical solution, calculates the fault threshold range of effective feature variables under normal conditions based on the preset fault characteristics corresponding to each specified type of fault in the air conditioning unit. By comparing the predicted output value of the multidimensional Taylor network with the fault threshold range, prediction of the specified types of faults in the air conditioning unit is achieved. Figure 4 (a), (b), and (c) are waveform diagrams illustrating the fault diagnosis principle of a cooling coil valve stuck in the fully open position. Compared to other data-driven fault prediction techniques, the multidimensional Taylor network structure can express complex nonlinear functions using a mathematical combination of series terms. It can fit nonlinear systems under normal conditions without requiring extensive network training. Furthermore, the multidimensional Taylor network has advantages such as simple network structure, strong generalization ability, certain interpretability, and more powerful approximation performance. Therefore, the air conditioning unit fault prediction method designed in this invention can effectively improve the prediction efficiency of air conditioning unit faults.

[0070] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] It should be noted that the terms "first", "second", "third" and the like in the description and in the claims of the present application are used for distinguishing between similar elements and not necessarily for describing a sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances and embodiments of the application can operate in other sequences than described or illustrated herein.

[0072] The terms "comprise", "comprising", "include", "including", "have", "having" and any variations thereof in the present application are intended to cover a non-exclusive inclusion. For example, a process, method, article, or apparatus that comprises a list of steps or components does not necessarily comprise only those steps or components but can include additional steps or components not expressly listed or inherent to such process, method, article, or apparatus.

[0073] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. A fault prediction method for air conditioning units based on multidimensional Taylor networks, characterized in that, Includes the following steps: s1: Based on the fault characteristics corresponding to each type of fault in the pre-defined air conditioning unit, from the ASHRAE RP-1312 dataset, collect the fault feature values ​​at l normal operating time points in the historical time direction starting from time t0 to construct the x multidimensional matrix; at the same time, collect the fault feature values ​​at l normal operating time points in the historical time direction starting from time (t0+t) to construct the y multidimensional matrix, thus forming the sample dataset (x,y). s2: Perform data normalization on the sample dataset (x,y) to update the sample data x. * Multidimensional matrix and sample data y * Multidimensional matrix; s3: Calculate the sample data y * The static fault thresholds of each fault feature in the multidimensional matrix are then extracted from the sample data y. * The multidimensional matrix serves as the model's predicted state of each fault feature in the next time period t. The corresponding label vector; s4: Based on the sample data x * The dimensions of the multidimensional matrix are used to construct a multidimensional Taylor network BP-MTN model with the same dimension of input and output. Then, the sample data x is used. * The multidimensional matrix serves as the model input feature, inputting into the multidimensional Taylor network BP-MTN model and outputting the model's predicted state. Next, the label vector y * and model predicted state Error calculation is performed, and finally the training of the multidimensional Taylor network BP-MTN model is completed using the backpropagation method. s5: Based on the fault characteristics corresponding to each type of fault in the pre-defined air conditioning unit, l fault feature values ​​containing normal operating state and fault state time points are collected from the ASHRAE RP-1312 dataset from time t0 along the historical time direction to construct the X multidimensional matrix; simultaneously, l fault feature values ​​containing normal operating state and fault state time points are collected from time (t0+t) along the historical time direction to construct the Y multidimensional matrix, thus forming the test dataset (X,Y). Then, the test dataset (X,Y) is normalized to obtain the updated test dataset (X). * ,Y * ); s6: Based on the static fault thresholds of each fault feature calculated in step s3 and the trained multidimensional Taylor network BP-MTN model, determine the test dataset (X) * ,Y * If there is an anomaly in the estimated state of each fault feature in the first t time period after time (t0+t), the corresponding fault feature is determined to be an abnormal fault feature. Based on the correspondence between fault features and fault types, the fault type corresponding to the abnormal fault feature is further determined so as to determine whether the fault has actually occurred. Otherwise, the estimated state of the air conditioning unit in the first t time period after time (t0+t) is determined to be normal.

2. The air conditioning unit fault prediction method based on a multidimensional Taylor network according to claim 1, characterized in that, In step s3, the sample data y * The formulas for calculating the static fault thresholds of each fault feature in the multidimensional matrix are as follows: Where UCL represents the upper limit of the static fault characteristic threshold, LCL represents the lower limit of the static fault characteristic threshold, μ0 represents the mean of the current observed sample, σ is the standard deviation of the current observed sample, λ is the smoothing constant between 0 and 1, and L represents the control limit parameter.

3. The air conditioning unit fault prediction method based on a multidimensional Taylor network according to claim 2, characterized in that, The multidimensional Taylor network BP-MTN model includes an input layer, a data processing layer, an output layer, a fully connected layer, and an activation function layer. The layers are connected sequentially, and the output of the previous layer becomes the input of the next layer.

4. The air conditioning unit fault prediction method based on a multidimensional Taylor network according to claim 3, characterized in that, The formula for the weighted summation of the data processing layer is as follows: Among them, T j Let represent the j-th output node of the multidimensional Taylor network, m represent the exponent of the highest power expansion term in the data processing layer, N(n,m) represent the number of terms in the polynomial after the n input fault features of the data processing layer are expanded to the power of m, q∈N(n,m) represent the q-th polynomial after the power expansion of the data processing layer, and w j,q σ represents the weights before the q-th polynomial at the j-th output node of the Viytale network. q,i This represents the power of the variable x in the q-th polynomial.

5. The air conditioning unit fault prediction method based on a multidimensional Taylor network according to claim 4, characterized in that, In step s4, the multidimensional Taylor network BP-MTN model performs multidimensional fitting and prediction according to the following formula: Where, θ j This represents the weight vector of the j-th node value in the fully connected layer after entering the activation function, where ReLU is the activation function.

6. The air conditioning unit fault prediction method based on a multidimensional Taylor network according to claim 5, characterized in that, In step s4, the backpropagation algorithm calculates the squared fitting error according to the following formula:

7. The air conditioning unit fault prediction method based on a multidimensional Taylor network according to claim 6, characterized in that, In step s6 Determine the dataset to be tested (X) * ,Y * The process of determining whether the predicted state of each fault feature in the first time interval t after time (t0+t) is abnormal includes: statistical identification by marking fault nodes and fault types.

8. The air conditioning unit fault prediction method based on a multidimensional Taylor network according to claim 7, characterized in that, The specified types of faults in the preset air conditioning unit include: return air stuck at a fixed speed fault, return air fan completely damaged fault, outdoor air valve stuck in the fully closed position fault, cooling coil valve stuck in the fully open position fault, and cooling coil valve stuck at a fixed angle fault. The fault characteristics corresponding to the fixed speed fault of the return air card include: return air fan energy, air conditioner supply air temperature, supply air rate, and return air rate. The fault characteristics corresponding to the complete failure of the return air fan include: air conditioner supply air temperature, return air rate, and mixed air temperature. The fault characteristics corresponding to the outdoor air valve being stuck in the fully closed position include: outdoor air velocity, return air fan energy, and supply air temperature. The fault characteristics corresponding to the cooling coil valve being stuck in the fully open position include: outdoor air velocity, supply air temperature and return air temperature. The fault characteristics corresponding to the cooling coil valve being stuck at a fixed angle include: outdoor air flow rate and return air rate.

Images