CN115773562A - Unified heating ventilation air-conditioning system fault detection method based on federal learning - Google Patents
Unified heating ventilation air-conditioning system fault detection method based on federal learning Download PDFInfo
- Publication number
- CN115773562A CN115773562A CN202211483557.6A CN202211483557A CN115773562A CN 115773562 A CN115773562 A CN 115773562A CN 202211483557 A CN202211483557 A CN 202211483557A CN 115773562 A CN115773562 A CN 115773562A
- Authority
- CN
- China
- Prior art keywords
- data
- fault
- fault detection
- model
- federal learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 77
- 238000004378 air conditioning Methods 0.000 title claims abstract description 35
- 238000010438 heat treatment Methods 0.000 title claims abstract description 35
- 238000009423 ventilation Methods 0.000 title claims abstract description 14
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 30
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 29
- 238000000034 method Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000000605 extraction Methods 0.000 claims abstract description 4
- 230000002776 aggregation Effects 0.000 claims description 7
- 238000004220 aggregation Methods 0.000 claims description 7
- 239000003507 refrigerant Substances 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000001816 cooling Methods 0.000 claims description 3
- 238000013459 approach Methods 0.000 claims description 2
- 238000009833 condensation Methods 0.000 claims description 2
- 230000005494 condensation Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 claims description 2
- 239000008236 heating water Substances 0.000 claims description 2
- 238000003745 diagnosis Methods 0.000 abstract description 31
- 230000000694 effects Effects 0.000 abstract description 20
- 230000006870 function Effects 0.000 description 9
- 210000002569 neuron Anatomy 0.000 description 9
- 238000011176 pooling Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000004913 activation Effects 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 2
- 238000012886 linear function Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000005057 refrigeration Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Images
Landscapes
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a unified heating ventilation air-conditioning system fault detection method based on federal learning. A novel federal learning architecture is presented. The Convolutional Neural Network (CNN) is combined with federal learning, the convolutional neural network is used as a general model, and better conditions are provided for the training of the model by utilizing the strong feature extraction capability of the convolutional neural network. And the federal learning ensures the privacy of the data, and simultaneously, the multi-party data is utilized to carry out common modeling, so that the generalization capability of the model is improved. The faults with different severity levels are combined by using federal learning, so that the detection and diagnosis accuracy of early mild fault levels is improved. The method has the advantages that cross-field fault detection and diagnosis are performed by means of federal learning, and the diagnosis effect of slight faults of the water chilling unit is improved. The water chiller group data and the air handling unit data are used for combined modeling, the two modeling data have different selected characteristics, and the unified combined model obtained through modeling can be used for simultaneously detecting the water chiller group fault and the air handling unit fault.
Description
Technical Field
The invention relates to the technical field of fault detection, in particular to a unified heating ventilation air conditioning system fault detection method based on federal learning.
Background
With the further development of the industry, modern industrial parks are gradually replacing traditional industrial parks. Heating, ventilation and air conditioning (HVAC) systems are an essential part of modern industrial parks. The energy consumption rate of the hvac system around the world is typically between 40% and 60% depending on the particular season and weather. This ratio will continue to increase as the industry level continues to increase.
The heating, ventilating and air conditioning system has a complex internal structure and mainly comprises a water chilling unit (Chiller) and an Air Handling Unit (AHU). Various failures of hvac systems result in significant energy waste. The time and economic costs of manually detecting and diagnosing these faults are often high. In order to reduce the waste of energy and manpower costs, a method for accurately detecting and diagnosing faults of the heating, ventilating and air conditioning system is required.
Industrial digitization has made machines that originally cooled ice into a lot of meaningful data. On the basis of the data, a large number of methods for detecting and diagnosing faults of the heating, ventilating and air conditioning system emerge. Existing fault detection and diagnosis methods, such as Support Vector Machines (SVMs), bayesian Networks (BNs), principal Component Analysis (PCA), extreme Learning Machines (ELMs), tree structure based methods, extreme gradient boost (XGBoost), and the like. However, the existing heating ventilation air conditioning system fault detection and diagnosis methods can only be applied to one fault level. The model is trained using fault data of a certain severity, and then faults of the same severity are detected by the model, without fully utilizing multi-level fault data. Most existing methods do not perform well for early minor-level failures. And is limited to exploring one of the chiller units or the air handling units alone, training a model using data from one system, and then detecting and diagnosing faults in the assembly, and is not capable of performing cross-system fault detection and diagnosis.
Disclosure of Invention
The invention provides a unified heating ventilation air-conditioning system fault detection method based on federal learning, aiming at overcoming the limitations that the existing method is difficult to carry out multi-level fault detection and diagnosis on a heating ventilation air-conditioning system, find early slight level faults and carry out fault detection and diagnosis across systems.
In order to achieve the purpose, the invention adopts the following technical scheme:
the utility model provides a unified heating ventilation air conditioning system fault detection method based on federal study, includes the step:
s1, each client acquires fault data of each heating, ventilating and air conditioning system;
s2, each client locally trains a CNN neural network model serving as a general model and sent by a central server based on the acquired fault data, and sends the trained general model parameters to the central server;
s3, the central server aggregates the universal model parameters sent by the clients, generates global fault detection model parameters and sends the global fault detection model parameters to the clients;
and S4, each client continuously utilizes the local fault data to repeatedly train the general model according to the received global model parameters until the global fault detection model is converged and the model training is finished.
Preferably, the method further comprises, after step S4, the steps of:
and S5, deploying the trained global fault detection model at each client, inputting the acquired operation data of the heating, ventilating and air conditioning system into the global fault detection model by each client, and predicting and outputting the fault type of the heating, ventilating and air conditioning system by the model.
Preferably, the hvac system operating data for diagnosing the hvac system fault includes chiller characteristic data of the hvac system, and/or air handling unit characteristic data.
Preferably, an SVC-based embedded feature selection technology (EIFS) selects 8 chiller feature data from 65 features in a chiller data set as basic data for diagnosing a fault of the hvac system, where the 8 chiller feature data are: oil supply pressure, condenser approach temperature, evaporator valve position, oil line pressure differential, condenser water flow rate, evaporator water flow rate, condensing temperature, thermal balance tolerance.
Preferably, 8 air handling unit feature data are selected from 102 features in an AHU data set as basic data for diagnosing the faults of the heating, ventilating and air conditioning system through a cost-sensitive sequential feature selection (CSSFS) algorithm, wherein the 8 air handling unit feature data are respectively as follows: supply fan power, supply air flow, return air flow, supply air temperature, mix air temperature, outdoor air temperature, heating water coil discharge air temperature, cooling coil load.
Preferably, the fault types of the water chilling unit comprise 7 types, namely F1 (CF), F2 (EO), F3 (NCR), F4 (RCW), F5 (REW), F6 (RL) and F7 (RO), and each fault type is divided into 4 severity levels.
Preferably, each of said clients locally trains said generic model using the same or different said fault data of different severity levels produced by the chiller.
Preferably, each of said clients trains said generic model locally using different said fault data of the same severity level generated by a chiller.
Preferably, each of said clients trains said generic model locally using the same or different said fault data of different severity levels produced by the chiller units and/or said fault data produced by the air handling units.
Preferably, each of said clients trains said generic model locally using different said fault data of the same severity level generated by a chiller unit, and/or said fault data generated by an air handling unit.
Preferably, the network structure of the general model comprises an input layer, a hidden layer and an output layer, wherein the input layer is used for inputting fault data of the heating, ventilating and air conditioning system into the hidden layer, and the hidden layer performs characteristic convolution extraction on the fault data and then outputs the extracted data characteristics to the output layer for data classification;
the hidden layer comprises a convolution layer and a full connection layer, the convolution layer is used for extracting the data characteristics of the fault data, and the full connection layer is used for integrating the data characteristics extracted by the convolution layer and outputting the data characteristics to the output layer.
Preferably, in step S3, a FedAVG algorithm is used for parameter aggregation, and the specific method is expressed by the following formula (4):
whereinIs that the central server is atThe global fault detection model parameters are sent to the clients after aggregation in round of federal learning;
The invention has the following beneficial effects:
1. a novel federal learning architecture is presented. The Convolutional Neural Network (CNN) is combined with federal learning, serves as a general model, and provides better conditions for model training by utilizing the powerful feature extraction capability of the convolutional neural network. And the federal learning ensures the privacy of the data, and simultaneously, the multi-party data is utilized to carry out the common modeling, thereby improving the generalization capability of the model.
2. The faults with different severity levels are combined by using federal learning, so that the detection and diagnosis accuracy of early mild fault levels is improved. And joint modeling is performed by utilizing fault data with different severity levels, so that the trained joint model can better cope with early slight faults.
3. And the method utilizes federal learning to detect and diagnose the cross-field faults, and improves the diagnosis effect of the slight faults of the water chilling unit. The water chilling unit data and the air handling unit data are used for combined modeling, the two modeling data have different selected characteristics, the combined model obtained through modeling can be used for simultaneously detecting the faults of the water chilling unit and the air handling unit, and the diagnosis effect of slight faults in the water chilling unit is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments of the present invention will be briefly described below. It is obvious that the drawings described below are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
Fig. 1 is a diagram illustrating implementation steps of a unified hvac system fault detection method based on federal learning according to an embodiment of the present invention;
FIG. 2 is a hierarchical structure diagram of a convolutional neural network with improved structure adopted in the present embodiment;
FIG. 3 is a diagram of the internal structure of a multilayer convolution;
FIG. 4 is a Federal learning framework diagram;
FIG. 5 is a view of the internal structure of the chiller;
FIG. 6 is an internal block diagram of an air handling unit;
FIG. 7 is a comparison of F1-score curves for different types of chiller faults for 7 different methods;
FIG. 8 is a schematic diagram of the FL2 chiller diagnostic results;
FIG. 9 is a schematic representation of the FL2 air handling unit diagnostic results.
Detailed Description
The technical scheme of the invention is further explained by the specific implementation mode in combination with the attached drawings.
Wherein the showings are for the purpose of illustration only and not for the purpose of limiting the same, the same is shown by way of illustration only and not in the form of limitation; to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product; it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The same or similar reference numerals in the drawings of the embodiments of the present invention correspond to the same or similar components; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "inner", "outer", etc. are used for indicating the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, it is only for convenience of description and simplification of description, but it is not indicated or implied that the referred device or element must have a specific orientation, be constructed in a specific orientation and be operated, and therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes and are not to be construed as limitations of the present patent, and the specific meanings of the terms may be understood by those skilled in the art according to specific situations.
In the description of the present invention, unless otherwise explicitly specified or limited, the term "connected" or the like, if appearing to indicate a connection relationship between components, is to be understood broadly, for example, as being either fixedly connected, detachably connected, or integrated; can be mechanically or electrically connected; they may be directly connected or indirectly connected through intervening media, or may be connected through one or more other components or may be in an interactive relationship with one another. The specific meanings of the above terms in the present invention can be understood in a specific case to those of ordinary skill in the art.
The unified heating ventilation air conditioning system fault detection method based on federal learning provided by the embodiment of the invention is shown in figure 1, and comprises the following steps:
s1, each client acquires fault data of each heating, ventilating and air conditioning system;
s2, each client locally trains a CNN neural network model serving as a general model and sent by the central server based on the acquired fault data, and sends the trained general model parameters to the central server;
s3, the central server aggregates the universal model parameters sent by the clients, generates global fault detection model parameters and sends the global fault detection model parameters to the clients;
s4, each client continuously utilizes local fault data to repeatedly train the general model according to the received global model parameters until the global fault detection model is converged and the model training is finished;
and S5, deploying the trained global fault detection model at each client, inputting the acquired operation data of the heating, ventilating and air conditioning system into the global fault detection model by each client, and predicting and outputting the fault type of the heating, ventilating and air conditioning system by the model.
How the method provided by the embodiment realizes multi-level and cross-system fault detection of the heating, ventilating and air conditioning system is specifically explained from two aspects of relevant network algorithm selection, combined model training and model performance evaluation.
1. Correlation network algorithm selection and joint model training
1. The Convolutional Neural Network (CNN) with the improved network structure shown in FIG. 2 is used as a general model for Federal learning. Convolutional neural networks were originally used for image processing. The main features of CNNs include shared weights and sparse connections. Weight sharing in CNN effectively avoids overfitting of the algorithm, while sparse connections reduce the number of training parameters. CNN models can identify certain features in the data well by convolution and then form more complex information at a high level. In the present invention, the CNN model is used as a general model for federal learning, and as shown in fig. 2, the general model is generally divided into three parts: an input layer, a hidden layer, and an output layer. The hidden layers of the CNN model generally consist of convolutional layers, pooling layers, and fully-connected layers. Convolutional and pooling layers are used to extract high-level features, and fully-connected layers further integrate features for classification of output layers. However, because the data dimensionality of the heating, ventilation and air conditioning system is small, the down-sampling is not performed by using the pooling layer in the embodiment, and the efficiency of fault detection is favorably improved by reducing the application of the pooling layer.
1.1 convolution layer
Convolutional Layers (CL) are layers that perform most of the computation. The purpose of the convolutional layer is to extract features from the input data, which is the first layer of the network to follow the input layer. The convolutional layer is composed of a plurality of convolutional kernels and is used for calculating different feature maps. The input is first convolved with a convolution kernel, and the convolution result uses an activation function to obtain a new feature map. To generate each feature map, all spatial locations of the input share a convolution kernel. The complete feature map is obtained by using several different convolution kernels, and the convolution formula can be expressed as:
whereinThe definition is as an input to the system,is defined asThe first of the convolutional layerThe bias of the individual neurons is such that,is the firstFirst of the layerThe output of each of the plurality of neurons,is the firstFirst of a layerFirst neuron to second neuronFirst of the layerThe nucleus of individual neurons. conv1D represents performing a one-dimensional convolution.
The specific operation between the multi-layer convolutions is shown in FIG. 3, with the output of the convolution middle layer beingCan be prepared byThe activation function f (x) is input for representation. There are three types of activation functions that are popular today, namely the sigmoid function, the tanh function, and the rectifying linear unit (ReLU). The specific representation of the ReLU function is:
whereinIs the firstLayer oneInput to individual neurons. ReLU is a piece-wise linear function that marks the negative part as zero and retains the positive part. The ReLU function has the advantages of accelerating learning speed and alleviating gradient disappearance compared with other two activation functions.
1.2 full connection layer
The fully-connected layer (FC) is an important component in convolutional neural networks. After stacking of multiple convolutional and pooling layers, most architectures add one or more fully connected layers to further process the features. The fully-connected layer connects all neurons in the previous layer to each neuron of the current layer, and then transforms the feature map using a non-linear function.
It should be emphasized that, in this embodiment, since the data dimension of the hvac system is small, the pooling layer is not used for down-sampling, but the full connection layer directly connects all the neurons in the previous convolution layer to each neuron of the current layer, and the efficiency of fault detection is improved by reducing the application of the pooling layer.
1.3 output layer
The output layer uses the previously extracted features for final decision making. Among the multi-classification problems of fault detection and diagnosis, the normalized exponential function (Softmax) has become one of the most popular choices in classification tasks due to its validity. The Softmax operation can be described as:
whereinDenotes that z belongs toThe probability of a class is determined by the probability of the class,,representing all kinds of quantities;is shown asAnd (4) class.
1.4 Federal learning
Federated learning is a macro distributed learning. Distributed learning divides a huge machine learning task into a plurality of smaller tasks, then distributes the smaller tasks to devices in various places, and uses a plurality of devices to improve the efficiency of machine learning. Compared with the traditional distributed learning, the federal learning does not need data transmission, and the privacy of data is ensured.
The specific steps of the invention for carrying out federal learning on a CNN neural network model as a general model are shown in FIG. 4, and the specific steps are as follows:
first, the central server sends the generic model to each client.
And secondly, each client receives the universal model sent by the central server, trains the model by using own private data and updates the weight of the model. In order to enable the general global model to be trained better, the data features are selected, and a specific selection method is described in the following content.
And thirdly, the trained client sends the updated model weight to the central server. Because each client uses different private data for model training, the model weights uploaded by each client are different.
And fourthly, the central server receives the updated weight from each client, and then performs aggregation operation to construct a global fault detection model.
In this example, a FedAVG algorithm was used for parameter aggregation:
whereinIs that the central server is in (A)) The update weights sent in the round-robin communication,is a clientWeights sent to the central server. Parameter(s)Is the total number of data points used to train the global fault detection model,is a clientTotal number of data points used for training.
And fifthly, after the federal learning is carried out for many times, the global fault detection model gradually converges to reach an expected target, and the central server sends the trained global fault detection model to the client.
And sixthly, the client side uses the latest model received from the central server to detect and diagnose the faults of the heating, ventilating and air conditioning.
1.5 feature selection
Currently, the operating state of the hvac system is acquired through various types of sensors installed in various structures of the hvac system. However, most manufacturers of heating, ventilating and air conditioning systems do not place too many sensors in the experiment due to the expensive sensors, and usually only place the sensors on key structural parts to monitor the operation state of the key parts. Therefore, whether data selection is reasonable in federal learning becomes a key to ensure model fault detection performance.
The water chiller group data used in this example was collected from the american society of heating, refrigeration and air conditioning engineers (ASHRAE) RP-1043 project. This project collected fault data for a centrifugal air conditioning chiller weighing approximately 90 tons. As shown in fig. 5, the project installed sensors at different locations of the air conditioner, recording 65 signatures every 10 seconds. These 65 characteristics include temperature sensors (TCA, TWCD, etc.), valve sensors (VE, VC, etc.), flow rate sensors (FWC, FEW, etc.), and other sensor values.
In this example, 7 typical air cooler failures were intensively studied, including F1) Condenser Fouling (CF), F2) Excess Oil (EO), F3) refrigerant non-condensation (NCR), (F4) condenser water flow Reduction (RCW), F5) evaporator water flow Reduction (REW), F6) Refrigerant Leakage (RL) and F7) refrigerant excess (RO) and normal conditions shown in table 1. Various fault classifications for water chillers are shown in fig. 5. Each fault type is classified into 4 severity classes as explained in table 1.
TABLE 1 quantitative measurement of four severity levels (from 1 to 4) for seven faults
In this embodiment, the SVC-based embedded feature selection technique (EIFS) selects 8 features described in the following table 2 from 65 features in the water chiller group data set as feature data representing the operation state of the water chiller group:
table 2 water chilling unit 8 important characteristic data
The air handling unit data set used in this example was collected by ASHRAE with item number 1312-RP. In this project, the basic structure of the AHU is shown in fig. 6, and is composed of a supply fan, a return fan, an exhaust damper, a cooling coil, a heating coil, a coil control valve, and the like. The project uses two identical AHUs, wherein one AHU simulates the condition of fault occurrence, and the other AHU simulates the condition of normal operation of equipment. In the present embodiment, 8 kinds of faults are selected, and the 8 kinds of fault data are used for joint model learning, and the 8 kinds of fault types are specifically described in the following table 3:
TABLE 3 class 8 failure of air handling unit
In this embodiment, 8 features described in the following table 4 are selected from 102 features in the AHU dataset as feature data representing the operating state of the air handling unit by a Cost Sensitive Sequential Feature Selection (CSSFS) algorithm:
table 4 8 important characteristic data of air handling unit
2. Global fault detection model performance evaluation
2.1 FL1 and FL2 federal learning experiments
The general model provided in this embodiment employs a three-layer one-dimensional convolution structure as shown in fig. 2. In the first federal learning experiment (FL 1), 8 pieces of characteristic data of the chiller set described in table 2 were selected and divided into data of four fault levels, and then chiller set data of different fault levels were trained in different clients. After the training is finished, 4 general model parameters trained by the four clients are transmitted to the central server for summarizing, and the central server returns the aggregated parameters for training again. This operation is repeated, and when the global fault detection model converges to a desired value, the global fault detection model is transmitted to 4 local clients for fault diagnosis of the hvac system.
In a second federal learning experiment (FL 2), federal learning was performed using chiller and air handler data. To verify the ability of federal learning in cross-domain fault detection and diagnosis, FL2 was divided into four sections. After feature selection and data division, the data of the air handling unit and the data of the four fault levels of the water chilling unit are subjected to combined learning respectively. For example, data of a certain fault level of the water chilling unit and data of the air handling unit are trained in different clients, after the training is finished, two general model parameters are transmitted to the central server to be aggregated, and the central server returns the aggregated global model parameters to perform local training again. This operation is repeated until the global fault detection model converges, and then fault detection and diagnosis are performed using the global fault detection model.
2.2 Evaluation index
Currently, for a deep learning model, accuracy (Accuracy) is generally adopted as an evaluation index of an experimental result, and this embodiment adopts a more comprehensive index F1-score, which is defined as follows:
where Precision is called Precision. N is a radical of TP Is the number of predicted positives, which are actually positive, N FP Is the number that is predicted to be positive and actually negative. Recall is called precision. N is a radical of FN Is the number of positive, in fact, predicted negative.
2.3 Analysis of FL1 Experimental results
The performance of a global fault detection model (Fed _ CNN model) obtained by joint training is compared with other traditional models, and the global fault detection model comprises a CNN, a long-short term memory network (LSTM), a nested long-short term memory Network (NLSTM), a bidirectional long-short term memory network (BILST), a gradient Lifter (LGBM) and a recurrent neural network (GRU). The Fed _ CNN model performs combined modeling by using data of four fault levels of the water chilling unit, and can detect and diagnose faults of the four fault levels simultaneously. As shown in table 5 below, the Fed _ CNN performed well in fault detection and diagnosis for four fault classes. In failure Level 1 (Level _ 1), fed _ CNN has an F1-score of 0.9052, which is the only model in excess of 0.9 among many models. Fed _ CNN performs unexpectedly well at failure Level 2 (Level _ 2), 6% better than the GRU model. The global fault detection model based on the federal learning also performs well in the fault Level three (Level _ 3) and the fault Level four (Level _ 4), although the best effect is not achieved, the difference between the diagnosis effects of the global fault detection model and other models is not large, and the diagnosis effect is over 0.99.
TABLE 5 comparison of chiller fault detection and diagnosis by different methods
The effect of different methods on the detection of different types of faults can also be seen in fig. 7. Overall, three faults, exOil, refleak, refOver, of the four fault levels are difficult to detect, especially in the light fault level. The global fault detection model trained in the embodiment has great advantages in the mild fault Level for the three faults which are difficult to diagnose, and in Level3 and Level4, the global fault detection model provided by the embodiment has slightly inferior diagnosis effect on the three faults to other models, which is caused by weight aggregation in the federal learning process.
2.4 Analysis of FL2 Experimental results
In the embodiment, four fault level data of a Chiller (Chiller) and Air Handling Unit (AHU) data are subjected to combined modeling respectively, and cross-system federal learning is performed. In the aspect of the water chilling unit, the detection and diagnosis effects on four fault levels of the water chilling unit are not reduced, and the setting is slightly improved. As can be seen from FIG. 8, after Federal learning is performed by using the air handling unit data and the chiller unit fault level one data, the diagnosis effect of RefOver fault is obviously improved, and is improved by 14.4% compared with the traditional CNN, the diagnosis effects of two faults of Refleak and Exoil are improved by 2% -4^ and the diagnosis effects of other faults are not reduced. After the data of the air handling unit and the data of the fault level two of the water chilling unit are subjected to federal learning, the diagnosis effect of RefOver and Refleak faults is improved by about 4% -7%. However, the detection of normal data has slightly decreased, about 1%. In the third fault level and the fourth fault level of the water chilling unit, the detection and diagnosis effect of the federal learning model on a single fault is slightly reduced, but the diagnosis effect of each fault exceeds 0.99, and the whole fault detection effect is equivalent to that of the traditional CNN model.
In terms of the air handling unit, it can be seen from fig. 9 that the failure diagnosis effect of F5 is most reduced by about 1.5% through federal learning. After the AHU data and the water chiller fault level 2 data are subjected to combined learning, the diagnosis effect of F5 is reduced most, and is reduced by about 1.5%, and F8 is reduced by about 0.8%. After the AHU data and the four fault level data of the water chilling unit are subjected to combined learning, the whole fault diagnosis effect is slightly reduced to about 0.3% -0.5%, but the whole fault diagnosis effect is still kept about 0.99, and the influence on practical application is small.
It should be understood that the above-described embodiments are merely preferred embodiments of the invention and the technical principles applied thereto. It will be understood by those skilled in the art that various modifications, equivalents, changes, and the like can be made to the present invention. However, such variations are within the scope of the invention as long as they do not depart from the spirit of the invention. In addition, certain terminology used in the description and claims of the present application is not limiting, but is used for convenience only.
Claims (12)
1. A unified heating ventilation air-conditioning system fault detection method based on federal learning is characterized by comprising the following steps:
s1, each client acquires fault data of each heating, ventilating and air conditioning system;
s2, each client locally trains a CNN neural network model serving as a general model and sent by a central server based on the acquired fault data, and sends the trained general model parameters to the central server;
s3, the central server aggregates the universal model parameters sent by the clients, generates global fault detection model parameters and sends the global fault detection model parameters to the clients;
and S4, each client continuously utilizes the local fault data to repeatedly train the general model according to the received global model parameters until the global fault detection model is converged and the model training is finished.
2. The unified heating, ventilating and air-conditioning system fault detection method based on federal learning as claimed in claim 1, wherein the method further comprises the following steps after the step S4:
and S5, deploying the trained global fault detection model at each client, inputting the acquired operation data of the heating, ventilating and air conditioning system into the global fault detection model by each client, and predicting and outputting the fault type of the heating, ventilating and air conditioning system by the model.
3. The unified hvac system fault detection method according to claim 2, wherein the hvac system operating data for diagnosing the hvac system fault comprises chiller characteristic data and/or air handling unit characteristic data of an hvac system.
4. The unified hvac system fault detection method based on federal learning of claim 3, wherein an embedded feature selection technology based on SVC selects 8 chiller feature data from a chiller group data set as basic data for diagnosing the hvac system fault, and the 8 chiller feature data are respectively: oil supply pressure, condenser approach temperature, evaporator valve position, oil line pressure differential, condenser water flow rate, evaporator water flow rate, condensing temperature, thermal balance tolerance.
5. The unified hvac system fault detection method based on federal learning of claim 3, wherein 8 pieces of air handling unit feature data are selected from an AHU data set as basic data for diagnosing the hvac system fault through a cost-sensitive sequential feature selection algorithm, and the 8 pieces of air handling unit feature data are respectively: supply fan power, supply air flow, return air flow, supply air temperature, mix air temperature, outdoor air temperature, heating water coil discharge air temperature, cooling coil load.
6. A unified heating, ventilating and air-conditioning system fault detection method based on federal learning as claimed in any one of claims 3-5, wherein the fault types of the chiller are 7, which are respectively condenser fouling, excessive oil, non-condensation of refrigerant, reduction of condenser flow, reduction of evaporator water flow, leakage of refrigerant and excessive refrigerant, and each fault type is classified into 4 severity levels.
7. The unified HVAC system fault detection method according to claim 6, wherein each of the clients locally trains the generic model using the same or different fault data of different severity levels generated by chiller.
8. The unified HVAC system fault detection method according to claim 6, wherein each of the plurality of clients trains the generic model locally using different fault data of the same severity level generated by a chiller.
9. The unified HVAC system fault detection method according to claim 6, wherein each of the clients locally uses the same or different fault data generated by chiller units and/or the fault data generated by air handling units to train the generic model.
10. The unified HVAC system fault detection method according to claim 6, wherein each of the plurality of clients trains the generic model locally using different fault data generated by a chiller unit with the same severity level and/or the fault data generated by an air handling unit.
11. The unified heating, ventilation and air conditioning system fault detection method based on federal learning as claimed in claim 1, wherein the network structure of the general model comprises an input layer, a hidden layer and an output layer, wherein the input layer is used for inputting fault data of the heating, ventilation and air conditioning system into the hidden layer, and after the hidden layer performs feature convolution extraction on the fault data, the extracted data features are output to the output layer for data classification;
the hidden layer comprises a convolution layer and a full connection layer, the convolution layer is used for extracting the data characteristics of the fault data, and the full connection layer is used for integrating the data characteristics extracted by the convolution layer and outputting the data characteristics to the output layer.
12. The unified HVAC system fault detection method according to claim 1, wherein in step S3, fedAVG algorithm is used for parameter aggregation, and the specific method is expressed by the following formula (4):
whereinIs that the central server is atThe global fault detection model parameters are sent to the clients after aggregation in round of federal learning;
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211483557.6A CN115773562A (en) | 2022-11-24 | 2022-11-24 | Unified heating ventilation air-conditioning system fault detection method based on federal learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211483557.6A CN115773562A (en) | 2022-11-24 | 2022-11-24 | Unified heating ventilation air-conditioning system fault detection method based on federal learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115773562A true CN115773562A (en) | 2023-03-10 |
Family
ID=85390137
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211483557.6A Pending CN115773562A (en) | 2022-11-24 | 2022-11-24 | Unified heating ventilation air-conditioning system fault detection method based on federal learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115773562A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115952442A (en) * | 2023-03-09 | 2023-04-11 | 山东大学 | Global robust weighting-based federal domain generalized fault diagnosis method and system |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2011265563A1 (en) * | 2010-12-24 | 2012-07-12 | Commonwealth Scientific And Industrial Research Organisation | System and method for detecting and/or diagnosing faults in multi-variable systems |
WO2015158198A1 (en) * | 2014-04-17 | 2015-10-22 | 北京泰乐德信息技术有限公司 | Fault recognition method and system based on neural network self-learning |
WO2021059607A1 (en) * | 2019-09-26 | 2021-04-01 | 富士フイルム株式会社 | Machine learning system and method, integration server, information processing device, program, and inference model generation method |
CN112611080A (en) * | 2020-12-10 | 2021-04-06 | 浙江大学 | Intelligent air conditioner control system and method based on federal learning |
KR20210079052A (en) * | 2019-12-19 | 2021-06-29 | 한전케이디엔주식회사 | Underground facility monitoring and diagnosis system and method using autonomous-federated learning enable robot |
CN113240184A (en) * | 2021-05-21 | 2021-08-10 | 浙江大学 | Building space unit cold load prediction method and system based on federal learning |
CN113283532A (en) * | 2021-06-09 | 2021-08-20 | 重庆大学 | Water chilling unit fault diagnosis method based on deep migration learning |
CN114413409A (en) * | 2022-01-05 | 2022-04-29 | 青岛海尔智能技术研发有限公司 | Detection method and device for air conditioner fault probability and intelligent air conditioner |
CN114484731A (en) * | 2021-12-27 | 2022-05-13 | 浙江英集动力科技有限公司 | Method and device for diagnosing faults of central air conditioner based on stacking fusion algorithm |
CN115311205A (en) * | 2022-07-07 | 2022-11-08 | 上海工程技术大学 | Industrial equipment fault detection method based on pattern neural network federal learning |
-
2022
- 2022-11-24 CN CN202211483557.6A patent/CN115773562A/en active Pending
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2011265563A1 (en) * | 2010-12-24 | 2012-07-12 | Commonwealth Scientific And Industrial Research Organisation | System and method for detecting and/or diagnosing faults in multi-variable systems |
WO2015158198A1 (en) * | 2014-04-17 | 2015-10-22 | 北京泰乐德信息技术有限公司 | Fault recognition method and system based on neural network self-learning |
WO2021059607A1 (en) * | 2019-09-26 | 2021-04-01 | 富士フイルム株式会社 | Machine learning system and method, integration server, information processing device, program, and inference model generation method |
KR20210079052A (en) * | 2019-12-19 | 2021-06-29 | 한전케이디엔주식회사 | Underground facility monitoring and diagnosis system and method using autonomous-federated learning enable robot |
CN112611080A (en) * | 2020-12-10 | 2021-04-06 | 浙江大学 | Intelligent air conditioner control system and method based on federal learning |
CN113240184A (en) * | 2021-05-21 | 2021-08-10 | 浙江大学 | Building space unit cold load prediction method and system based on federal learning |
CN113283532A (en) * | 2021-06-09 | 2021-08-20 | 重庆大学 | Water chilling unit fault diagnosis method based on deep migration learning |
CN114484731A (en) * | 2021-12-27 | 2022-05-13 | 浙江英集动力科技有限公司 | Method and device for diagnosing faults of central air conditioner based on stacking fusion algorithm |
CN114413409A (en) * | 2022-01-05 | 2022-04-29 | 青岛海尔智能技术研发有限公司 | Detection method and device for air conditioner fault probability and intelligent air conditioner |
CN115311205A (en) * | 2022-07-07 | 2022-11-08 | 上海工程技术大学 | Industrial equipment fault detection method based on pattern neural network federal learning |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115952442A (en) * | 2023-03-09 | 2023-04-11 | 山东大学 | Global robust weighting-based federal domain generalized fault diagnosis method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhao et al. | Artificial intelligence-based fault detection and diagnosis methods for building energy systems: Advantages, challenges and the future | |
Kocyigit | Fault and sensor error diagnostic strategies for a vapor compression refrigeration system by using fuzzy inference systems and artificial neural network | |
Kim et al. | A review of fault detection and diagnostics methods for building systems | |
Li et al. | An explainable one-dimensional convolutional neural networks based fault diagnosis method for building heating, ventilation and air conditioning systems | |
Najjar et al. | Optimal sensor selection and fusion for heat exchanger fouling diagnosis in aerospace systems | |
Zhou et al. | A comparison study of basic data-driven fault diagnosis methods for variable refrigerant flow system | |
CN108931387B (en) | Fault diagnosis method based on multi-sensor signal analysis and capable of providing accurate diagnosis decision | |
Singh et al. | A comprehensive review: Fault detection, diagnostics, prognostics, and fault modeling in HVAC systems | |
CN114484731A (en) | Method and device for diagnosing faults of central air conditioner based on stacking fusion algorithm | |
CN110555479A (en) | fault feature learning and classifying method based on fusion of 1DCNN and GRU | |
US7899652B2 (en) | Linear programming support vector regression with wavelet kernel | |
CN102175282B (en) | Method for diagnosing fault of centrifugal air compressor based on information fusion | |
Shi et al. | An efficient VRF system fault diagnosis strategy for refrigerant charge amount based on PCA and dual neural network model | |
Zhu et al. | Fault diagnosis based operation risk evaluation for air conditioning systems in data centers | |
Li et al. | Review on fault detection and diagnosis feature engineering in building heating, ventilation, air conditioning and refrigeration systems | |
Zhang et al. | Fault detection and diagnosis for the screw chillers using multi-region XGBoost model | |
CN110276247A (en) | A kind of driving detection method based on YOLOv3-Tiny | |
Zhang et al. | Imbalanced data based fault diagnosis of the chiller via integrating a new resampling technique with an improved ensemble extreme learning machine | |
CN115773562A (en) | Unified heating ventilation air-conditioning system fault detection method based on federal learning | |
Zhang et al. | Deep learning in fault detection and diagnosis of building HVAC systems: A systematic review with meta analysis | |
Li et al. | An efficient online wkNN diagnostic strategy for variable refrigerant flow system based on coupled feature selection method | |
Zhou et al. | Fault diagnosis method for building VRF system based on convolutional neural network: Considering system defrosting process and sensor fault coupling | |
Li et al. | Across working conditions fault diagnosis for chillers based on IoT intelligent agent with deep learning model | |
Li et al. | Machine learning based diagnosis strategy for refrigerant charge amount malfunction of variable refrigerant flow system | |
Aqlan et al. | Integrating artificial neural networks and cluster analysis to assess energy efficiency of buildings |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230310 |