CN116821796A - Self-adaptive water chilling unit fault diagnosis method based on online data - Google Patents
Self-adaptive water chilling unit fault diagnosis method based on online data Download PDFInfo
- Publication number
- CN116821796A CN116821796A CN202310780299.6A CN202310780299A CN116821796A CN 116821796 A CN116821796 A CN 116821796A CN 202310780299 A CN202310780299 A CN 202310780299A CN 116821796 A CN116821796 A CN 116821796A
- Authority
- CN
- China
- Prior art keywords
- data
- fault
- sensor
- model
- prediction
- 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
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 107
- 238000000034 method Methods 0.000 title claims abstract description 70
- 238000003745 diagnosis Methods 0.000 title claims abstract description 48
- 238000001514 detection method Methods 0.000 claims abstract description 37
- 230000002159 abnormal effect Effects 0.000 claims abstract description 20
- 238000012423 maintenance Methods 0.000 claims abstract description 17
- 238000012937 correction Methods 0.000 claims abstract description 7
- 239000013598 vector Substances 0.000 claims description 52
- 238000012549 training Methods 0.000 claims description 35
- 238000013528 artificial neural network Methods 0.000 claims description 34
- 238000012512 characterization method Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 11
- 238000001704 evaporation Methods 0.000 claims description 10
- 238000010977 unit operation Methods 0.000 claims description 10
- 238000013480 data collection Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 7
- 239000002826 coolant Substances 0.000 claims description 7
- 230000009466 transformation Effects 0.000 claims description 7
- 230000006978 adaptation Effects 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000004781 supercooling Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000009833 condensation Methods 0.000 claims description 4
- 230000005494 condensation Effects 0.000 claims description 4
- 238000000926 separation method Methods 0.000 claims description 4
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 3
- 230000008020 evaporation Effects 0.000 claims description 3
- 238000003064 k means clustering Methods 0.000 claims description 3
- 230000001105 regulatory effect Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000004941 influx Effects 0.000 claims description 2
- 238000003860 storage Methods 0.000 claims description 2
- 230000003044 adaptive effect Effects 0.000 claims 1
- 238000005259 measurement Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 6
- 239000003507 refrigerant Substances 0.000 description 5
- 238000004378 air conditioning Methods 0.000 description 3
- 230000036541 health Effects 0.000 description 3
- 238000010438 heat treatment Methods 0.000 description 3
- 230000003449 preventive effect Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000009423 ventilation Methods 0.000 description 3
- 239000000498 cooling water Substances 0.000 description 2
- 238000005265 energy consumption Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000005057 refrigeration Methods 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 1
- 239000010725 compressor oil Substances 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000007257 malfunction Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000003921 oil Substances 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F24—HEATING; RANGES; VENTILATING
- F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
- F24F11/00—Control or safety arrangements
- F24F11/30—Control or safety arrangements for purposes related to the operation of the system, e.g. for safety or monitoring
- F24F11/32—Responding to malfunctions or emergencies
- F24F11/38—Failure diagnosis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Probability & Statistics with Applications (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Chemical & Material Sciences (AREA)
- Combustion & Propulsion (AREA)
- Mechanical Engineering (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
A fault diagnosis method of a water chilling unit based on self-adaption of reliable line data collects real-time running data of the unit, performs null value prediction and abnormal value correction by using an orthogonal state iteration method, inputs complete unit data into an electric-thermal fault decoupling model to perform sensor state discrimination, eliminates fault sensor data and performs fault-free data prediction; and finally, judging the thermal state of the machine set by adopting a thermal fault detection model to the non-fault prediction data to obtain a fault prediction, a diagnosis result and an auxiliary maintenance decision. The invention realizes decoupling diagnosis when electrical faults such as empty value prediction and abnormal value correction of the water chilling unit data, measurement deviation of a sensor and the like coexist with thermal faults, accurate prediction of the thermal faults of the water chilling unit, reliable online data selection, online self-adaptive updating of a fault diagnosis model, establishes a maintenance decision database, greatly improves the intelligent level of the water chilling unit, and can be rapidly and efficiently applied to water chilling unit systems of other models.
Description
Technical Field
The invention relates to a technology in the field of building air conditioning, in particular to a water chilling unit fault diagnosis method based on an orthogonal state iteration method, electric-thermal fault decoupling and an on-line data self-adaptive strategy.
Background
The heating ventilation air conditioning system occupies more than 50% of the total energy consumption of the building. In a heating ventilation air conditioning system, the energy consumption of the water chilling unit is highest, and the operation reliability of the water chilling unit has direct influence on building energy conservation. The existing fault diagnosis method of the water chilling unit generally aims at single faults and alarms on hard faults affecting basic operation, and at present, a method for predicting and rapidly diagnosing a plurality of soft faults simultaneously exists is lacking, and particularly, a multi-fault decoupling diagnosis method capable of decoupling electric-thermal faults and rapidly generalizing the multi-fault decoupling diagnosis method to water chilling units of different types is lacking; meanwhile, the existing model does not have the capability of self-updating according to real-time data.
Disclosure of Invention
Aiming at the problems that the prediction deviation caused by abnormal data acquisition of a water chilling unit cannot be well processed in the prior art, the prediction accuracy is obviously reduced under the condition that the deviation between the operation working condition and the training is large, and the final diagnosis result is inconsistent with the actual result; the rapid decoupling diagnosis can not be carried out when a plurality of soft faults coexist, and the prediction accuracy can be obviously reduced when a plurality of faults coexist; the diagnosis model also has no self-updating problem according to real-time data, the prediction result cannot be corrected in real time, the prediction accuracy can be obviously reduced when the running working condition of the water chilling unit is inconsistent with the training working condition, and the like.
The invention is realized by the following technical scheme:
the invention relates to a fault diagnosis method of a water chilling unit based on self-adaption of on-line data, which is characterized by collecting real-time running data of the unit, carrying out null value prediction and abnormal value correction by using an orthogonal state iteration method, inputting complete unit data into an electric-thermal fault decoupling model to judge the state of a sensor, eliminating fault sensor data and predicting fault-free data; and finally, judging the thermal state of the machine set by adopting a thermal fault detection model to the non-fault prediction data to obtain a fault prediction, a diagnosis result and an auxiliary maintenance decision.
The real-time running data of the water chilling unit is obtained by reading the sensor data and the running parameters of the water chilling unit and recording the data window of the water chilling unit in real time, and the method specifically comprises the following steps: the system comprises an operating state, an alarm state, a compressor suction temperature, an exhaust temperature, suction pressure, condensing pressure, compressor current, coolant water inlet and outlet temperature, pressure difference between a condenser and an evaporator inlet and outlet, valve positions of an electronic expansion valve and a water quantity regulating valve, startup and shutdown and load increasing and load reducing temperature, superheat degree and supercooling degree data.
The orthogonal state iteration method is as follows: setting the length of a historical data collection time period to be alpha minutes, collecting real-time operation data of a unit, projecting the real-time operation data onto a Legendre orthogonal polynomial base, iteratively updating an orthogonal base coefficient according to a state space expression, reconstructing the unit operation data through the orthogonal base coefficient to obtain an operation data curve approximated by the Legendre orthogonal polynomial at the current operation moment, and filling empty values in data collection and abnormal values removed by hampel filtering with predicted data values.
The projection refers to: according to the definition of orthogonal projection, obtaining the coefficient of the running data of the water chilling unit under the base of the Legendre polynomial specifically comprises the following steps:
1) Acquiring the running number of the water chilling unit according to the historical data collection time period alpha minutesThe measure function of the projection space according to the data isAccording to the running time from [ -1,1]In-range conversion to [ t-alpha, t]Inside Legend orthogonal polynomial ++>Wherein: p (P) n The method is characterized in that the method is an nth order Legend orthogonal polynomial, x is certain operation data of the current water chilling unit, such as air suction temperature, air discharge temperature and the like, t is operation time, and the unit is minutes; polynomial p n (t, x) satisfy<p n (t) ,p m (t) >=δ n,m T is the runtime, n and m are the nth and mth order polynomials, respectively, delta n,m Delta when n=m to satisfy the condition n,m =1, the remaining cases δ n,m =0, symbol<·,·>Is the inner product of the polynomial in a given metric space.
2) Calculating coefficients of running data of water chilling unit projected to Legendre orthogonal polynomial basisWherein: />For the coefficients of the Legendre orthopolynomial of the nth order with respect to the chiller operation data vector x at time t, +.>Wherein: t (T) CI T is the suction temperature of the compressor CO For compressor discharge temperature, P RE For evaporating pressure, P RC For condensing pressure, sign<·,·> ω(t) To be the inner product of a polynomial under the measure function ω (t) in a given metric space, f ≤t Is a function of real operation data of the water chilling unit at the time less than or equal to t.
The state space expression refers to: calculation of base coefficients from Legendre orthonormal polynomialsFormula, obtaining an updated expression of the coefficient by solving the bias derivative of time t:wherein: the expression of the C matrix is: /> The expression of the D matrix is: />n is Legendre orthogonal polynomial order, j is the running data dimension of the chiller.
The iterative updating of the state space expression refers to: discretizing Legendre orthopolynomial base coefficients using bilinear methodCalculating a state update at time t: /> Wherein: i is an identity matrix.
The Legendre orthopolynomial basis approximation refers to: according to the orthogonal polynomial base coefficient obtained by projection, reconstructing the running data of the water chilling unit within the data collection time period alpha minutes by utilizing the Legendre orthogonal polynomial f (t), wherein the running data of the water chilling unit are specifically as follows: wherein: />Is Legend positiveThe cross polynomial coefficient vector, x is the current water chilling unit operation data vector, and n is the Legendre orthogonal polynomial order.
The Hampel filtering refers to: setting outlier screening windows of real-time running data of a water chilling unit system, calculating the number of bits and average absolute deviation of the data in each window, and eliminating the data when the deviation is more than 3 times of the standard deviation of the data window.
In the orthogonal state iteration method, if the acquired water chilling unit data has a null value or an abnormal value, a state space expression updating step of ignoring the Legendre orthogonal polynomial base coefficient at the current moment is performed, and when the operation data of the state space expression updating step is recovered to be normal, updating is performed; after Legend orthogonal polynomial base coefficients are updated, the running data of the water chilling unit in the time period alpha minutes are predicted by using the latest coefficients, the vacant data in the time period alpha minutes, which are directly collected by the water chilling unit, are covered, and the covered complete data are input into an electric-thermal fault decoupling model.
The electric-thermal fault decoupling model comprises: a sensor bias fault detection model and a sensor fault-free data prediction model, wherein: the sensor deviation fault detection model separates sensor faults from the real-time data of unit operation, and the sensor fault-free data prediction model eliminates the interference of the sensor faults to complete the decoupling of sensor-thermal faults and obtain fault-free prediction data; and training a sensor deviation fault detection model and a sensor fault-free data prediction model by adopting historical data in an offline stage, collecting real-time data of the water chilling unit for 20 minutes at the latest in the online stage, and inputting the data in the model training working condition parameter range into the sensor deviation fault detection model for detection to obtain a predicted sensor fault. If the output of the fault detection model is fault-free, directly inputting the thermal fault detection model to judge the thermal fault; if the output of the fault detection model is faulty, the sensor data predicted to be faulty is input into the sensor non-fault data prediction model to be predicted, and a sensor non-fault data prediction value is obtained; and the fault-free predicted value is used for replacing the latest 20 minutes of real-time data input thermal fault detection model of the water chilling unit so as to eliminate the interference of sensor faults on thermal fault diagnosis and carry out the diagnosis of thermal faults.
The sensor deviation fault detection model is a measurement learning model based on self-adaptive density discrimination, and the model is trained by the following modes:
s221: encoding the input sensor data and performing nonlinear transformation to form a characterization vector;
s222: for each sensor fault category, presetting K clustering centers, using K-means clustering to minimize the distance between the characterization vector and the clustering centers, and updating the clustering center vector;
s223: randomly sampling one cluster, obtaining M-1 clusters nearest to the cluster, and randomly sampling D sensor data vectors from each cluster;
s224: the loss function of the current iteration is calculated,
wherein: alpha is the size of the separation interval between clusters, < >> Representing the average value of the vectors for the m-th cluster of samples,/->For the mth cluster, the characterization vector of the mth sensor after being encoded by the artificial neural network, C (·) is the class label to which the current characterization vector belongs, <' >>
The variance of the characterization vector of the m-th cluster is determined, the subscript positive sign is a negative value, and only a positive value is taken;
s225: after a certain number of iterations, the training process is suspended, the clustering center vector is updated, and gradient descent training is started after updating.
The trained sensor deviation fault detection model calculates the category in the prediction stage
Wherein: r is (r) n For the nth sensor data sample x n Characterization vector, mu obtained by network transformation l For the mean vector of the L nearest clusters, σ is +.>C (·) is the class label to which the current token vector belongs.
The data in the model training working condition parameter range refers to: and when the unit operates under the model training working condition, judging whether the value of the sensor meets a preset threshold condition. And when the threshold value is exceeded, diagnosing the sensor fault, and when the threshold value is not exceeded or the unit does not run under the model training working condition, inputting data into a sensor deviation fault detection model for diagnosis. And when the fault is diagnosed, outputting the fault of the sensor, and inputting the fault-free data prediction model of the sensor for prediction.
The sensor fault-free data prediction model is an artificial neural network based on an on-line data self-adaptive strategy, and the network minimizes the mean square error of prediction data and real sensor data through gradient descent in a training stage; and in the prediction stage, according to the fault sensor data, the numerical value of the sensor in the absence of faults is predicted through coding.
The data self-adaptive strategy capable of leaning on the line refers to that: according to the data output after the decoupling of the electric-thermal faults, firstly removing online data with the prediction entropy of the artificial neural network model larger than a set threshold; and secondly, setting a parameter disturbance range, and minimizing the maximum model prediction entropy in the disturbance range. According to the training mode, training an artificial neural network model based on an on-line data self-adaptive strategy, judging the thermal state of unit operation by using the model, and storing the judging result into a diagnosis result database; the strategy only intervenes in the prediction stage of the model, the information entropy of an output result is obtained according to the fault probability distribution output by the training model on line, and when the information entropy is larger, the model tends to give an error classification result. Therefore, in the online updating stage, the model can carry out gradient descent updating according to entropy loss, and the entropy of the model in the prediction stage is reduced. When the on-line data sharpness sensing and adaptation strategy is applied, the model only updates the normalization layer of the lower layer, namely only updates the model: to reduce the impact on the stability of the model prediction results. The statistical normalization parameters are estimated according to real-time data, and the affine parameters gamma and beta are obtained by gradient descent calculation according to the entropy loss of the real-time data.
Further, in order to alleviate the problem of degradation of prediction accuracy caused by the problem of unbalance of categories, the problem of batch size, the problem of batch data distribution and the like in real-time data batch in the online data self-adaption process, samples with entropy loss smaller than a set threshold value need to be selected in the gradient descent updating process of entropy loss. Selecting samples with small entropy loss reduces the number of samples with large gradients, so that the updating process is more stable, and the occurrence of model collapse is reduced. Meanwhile, the optimization of the entropy loss selects an extremely small and extremely large optimization mode, the disturbance amplitude of the model parameters is given, and an extremely small and extremely large strategy is used, namely, the entropy of an output result under disturbance is maximized firstly, and then the output entropy is minimized. This training encourages the model to enter flat areas of the entropy-losing surface, allowing the model to have better generalization ability.
The sensor fault-free data prediction model only adjusts the artificial neural network based on the online data self-adaptive strategyThe standardized layer in (3) is used for training, and specifically comprises the following steps: the input parameters are parameters of suction temperature, exhaust temperature, suction pressure, condensing pressure, compressor current value, coolant water inlet and outlet temperature, pressure difference between an inlet and an outlet of a condenser and an evaporator, load increasing and reducing temperature, superheat degree and supercooling degree; the normalized parameter updating in the L layer is specifically as follows: mu≡E f L (T CI ,T CO ,P RE ,P RC …;Θ)],Wherein: t (T) CI T is the suction temperature of the compressor CO For compressor discharge temperature, P RE For evaporating pressure, P RC For condensing pressure, μ is the mean vector in the normalized layer, σ 2 To normalize the variance vector in the layer, f L (. Cndot.) is the coding function of the L layer before the network, E (-). Cndot.; Θ) is the prediction entropy of the artificial neural network model when the network parameter is Θ.
Entropy of output result of trained artificial neural network: optimizing network parameters by minimizing loss L (x): l (T) CI ,T CO ,P RE ,P RC …)=minS(T CI ,T CO ,P RE ,P RC …)E R (T CI ,T CO ,P RE ,P RC …;Θ),Wherein: index functionT CI T is the suction temperature of the compressor CO For compressor discharge temperature, P RE For evaporating pressure, P RC For condensing pressure, E is the disturbance quantity of neural network parameters, ρ is the disturbanceThe upper limit of the quantity, f (& gt) is a coding function of the network, E (x; Θ) is the predictive entropy of the artificial neural network model when the network parameter is Θ, and the fault-free data vector x of the water chilling unit is input; e (E) 0 Is the upper limit of the set entropy threshold. The gamma, beta parameters are updated as follows: />Wherein: gamma is the scaling parameter of the normalization layer, beta is the offset parameter of the normalization layer, and L is the network loss.
The thermodynamic fault diagnosis model is an artificial neural network based on an on-line data self-adaptive strategy, and the network obtains fault class probability through softmax conversion after nonlinear conversion according to inlet and outlet pressure, evaporation pressure and condensation pressure of a compressor, and recommends maintenance suggestions and auxiliary decisions according to a prestored unit thermodynamic state discrimination result and a current unit state.
The diagnosis result comprises: health, sub-health and malfunction, wherein: when the system refrigerant leakage, the condenser filth blockage, the evaporator filth blockage and the sensor fixed deviation occur, the system is prompted to be in a sub-health or early warning state, and the electronic expansion valve is blocked, the system exhaust high pressure, the system exhaust low pressure and the exhaust high temperature, the compressor oil leakage, the refrigerant water and cooling water low pressure difference, the sensor broken line and the like are prompted to be in an alarm state.
The auxiliary maintenance decision comprises: preventive maintenance assistance decisions and fault maintenance assistance decisions, wherein: the preventive maintenance auxiliary decision function sets timing, and the user is reminded of performing preventive maintenance every time a fixed maintenance time node is reached; the fault maintenance auxiliary decision is linked to a corresponding maintenance information management library according to a maintenance suggestion decision tree generated by the corresponding historical fault information of the water chilling unit, so that the fault information is provided with the maintenance auxiliary decision, and a user is reminded to execute corresponding maintenance operation.
The invention relates to a system for realizing the method, which comprises the following steps: the water chilling unit operation data acquisition and storage unit, the fault diagnosis unit and the upper computer communication unit, wherein: the method comprises the steps that a water chiller operation data acquisition unit acquires water chiller PLC acquisition data and stores the water chiller PLC acquisition data into a Mysql or Influx database, a fault diagnosis unit acquires water chiller real-time operation data of approximately 20 minutes from the database, an abnormal value and a null value are replaced through an orthogonal state iteration method, electric-thermal fault decoupling is carried out, and then a thermal fault judgment is carried out on a decoupling result through an artificial neural network based on an on-line data self-adaption strategy, so that a thermal fault prediction result is obtained; and the upper computer communication unit packages the fault prediction result and the auxiliary decision content and sends the packaged fault prediction result and the auxiliary decision content to the upper computer.
Technical effects
The invention discloses a cold water unit sensor null value prediction and abnormal value correction method based on orthogonal state iteration, which is an artificial neural network model based on an on-line data self-adaptive strategy and used for cold water unit sensor fault-free data prediction and system thermal fault diagnosis. Compared with the prior art, after the cold water unit sensor null prediction and abnormal value correction method based on orthogonal state iteration is used, the fault diagnosis system can obtain more accurate sensor data prediction values, and compared with the direct use of data interpolation or abnormal values, the diagnosis accuracy is respectively improved by 3.3% and 7.7%. After the artificial neural network model based on the self-adaptive strategy of the reliable line data is used for predicting the data of the fault-free sensor, the accuracy rate of the data of the fault-free sensor is improved by 9.8 percent compared with that of the data of the fault-free sensor which is directly used, and compared with that of the traditional artificial neural network model which is directly used, the accuracy rate of the data of the fault-free sensor is improved by 4.3 percent when the working condition of the online data is greatly different from that of the training data.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a flowchart of a method for predicting empty values and correcting abnormal values of a water chilling unit sensor based on orthogonal state iteration;
FIG. 3 is a block diagram of a sensor bias fault detection model;
FIG. 4 is a block diagram of a sensor fault-free data prediction model;
FIG. 5 is a diagram of a thermal fault detection model;
FIG. 6 is a diagram of a fault simulation experiment system;
FIG. 7 is a graph of an orthorhombic iterative method for predicting inlet water temperature of a No. 2 condenser;
FIG. 8 is a schematic diagram of the diagnosis of the fixed deviation of the inlet water temperature sensor of the electric-thermal decoupling module 1# condenser;
fig. 9 is a graph of predicted values of failure-free data of the outlet water temperature sensor of the electric-thermal decoupling module # 1 condenser.
Detailed Description
As shown in fig. 1, this embodiment relates to a heating ventilation air conditioner and a method for intelligent control and health management of a water chiller thereof, including the following steps:
s1: the system reads the sensor data and the operation parameters of the water chilling unit in real time through RS485 communication, records the data window of the water chilling unit in real time for approximately 20 minutes, predicts the null value and corrects the abnormal value by using an orthogonal state iteration method, and stores the corrected data into a diagnosis database for reading in the diagnosis process. The implementation of this step is described in detail below:
the data transmitted through the RS485 communication mainly comprises the running state of a water chilling unit system, an alarm state, the air suction temperature of a compressor, the air discharge temperature, the air suction pressure, the condensing pressure, the compressor current value, the coolant water inlet and outlet water temperature, the pressure difference between the inlet and outlet of a condenser and an evaporator, the valve position of an electronic expansion valve and a water quantity regulating valve, the starting and stopping temperature, the load increasing and reducing temperature, the superheat degree and the supercooling degree. Mapping and converting the numerical value read by the RS485, storing the mapped result into a database after the current reading is finished, starting the next RS485 inquiry process by the master station at fixed intervals, and if a plurality of water chilling units are configured, simultaneously reading the data of the water chilling units in one conversion process.
In the orthogonal state iteration method, if the acquired water chilling unit data has null value or abnormal value, the state space expression updating step of ignoring the Legendre orthogonal polynomial base coefficient at the current moment is performed, and when the operation data is recovered to be normal, the state space expression updating step is performed again. After Legendre's orthogonal polynomial base coefficient is updated, the latest coefficient is utilized to predict the running data of the water chilling unit within the time period of alpha minutes, and the running data is coveredAnd (5) covering the vacant data in the time period alpha minutes directly acquired by the water chiller, and outputting the covered complete data. The updating formula of the Legendre orthogonal polynomial base coefficient in the orthogonal state iteration is as follows: wherein: n is Legendre orthogonal polynomial order, j is the running data dimension of the chiller. The operation data reconstruction process expression of the water chilling unit is as followsWherein->The vector is Legendre orthogonal polynomial coefficient vector, x is current chiller operation data vector, t is operation time, and n is Legendre orthogonal polynomial order.
The abnormal values are removed through a hampel filtering method, the hampel filtering method mainly selects the running data of the water chilling unit in approximately one minute as the window size for screening the abnormal values, and for the data in each window, the data in each window are calculated, wherein: the number of bits and the average absolute deviation are replaced by a null value if the deviation is greater than 3 times the standard deviation of the data window. The window is moved to the next piece of data and the process is repeated.
S2: separating the sensor faults by utilizing a sensor deviation fault detection model according to the unit operation real-time data stored in the database, eliminating the interference of the sensor faults by using a sensor fault-free data prediction model, and completing the decoupling of the sensor-thermal faults to obtain fault-free prediction data, wherein the method specifically comprises the following steps of:
s21: reading historical data from a database, performing steady state discrimination on the historical data, extracting steady state data, and performing training of a sensor deviation fault detection model;
s22: training a sensor deviation fault detection model and a sensor fault-free data prediction model according to the historical data;
s23: reading real-time data from a database to obtain real-time data of the water chiller in the last 20 minutes;
s24: inputting data in the model training working condition parameter range into a sensor deviation fault detection model to detect, so as to obtain a predicted sensor fault;
s25: inputting the sensor data predicted to be faulty into a sensor non-fault data prediction model to detect, so as to obtain a sensor non-fault data prediction value;
s26: after the sensor fault is detected, the non-fault predicted value is used for replacing the sensor value to be input into a thermal fault detection model so as to eliminate the interference caused by the sensor fault and diagnose the thermal fault.
The sensor deviation fault detection model in the step S22 adopts a measurement learning model based on self-adaptive density discrimination, and the training steps comprise:
s221: encoding input sensor data by using an artificial neural network, and enabling the input sensor data to be converted into a characterization vector through nonlinear transformation;
s222: for each sensor fault category, presetting K clustering centers, using K-means clustering to minimize the distance between the characterization vector and the clustering centers, and updating the clustering center vector;
s223: randomly sampling one cluster, obtaining M-1 clusters nearest to the cluster, and randomly sampling D sensor data vectors from each cluster;
s224: the loss function of the sample in the current iteration is calculated according to the following equation, wherein: alpha is the size of the separation interval between clusters,representing the average value of the vectors for the m-th cluster of samples,/->For the mth cluster, the characterization vector of the mth sensor after being encoded by the artificial neural network, C (·) is the class label to which the current characterization vector belongs, <' >> The variance of the characterization vector of the m-th cluster is determined, the subscript positive sign is a negative value, and only a positive value is taken; />
S225: after a certain number of iterations, the training process is suspended, the clustering center vector is updated, and gradient descent training is started after updating.
The sensor deviation fault detection model is determined according to the following formula in the prediction stage, wherein: r is (r) n For the nth sensor data sample x n Characterization vector, mu obtained by network transformation l For the mean vector of the L nearest clusters, σ is in trainingC (·) is the class label to which the current token vector belongs, the last output class +.>
The data in the model training working condition parameter range refers to: whether the unit operates in the model training working condition is judged firstly, namely, whether the value of the sensor meets the preset threshold value condition is judged when the unit operates in the model training working condition. If the threshold value is exceeded, diagnosing the sensor as a fault, and if the threshold value is not exceeded or the unit does not run under the model training working condition, inputting data into a sensor deviation fault detection model for diagnosis. And if the sensor is diagnosed as faulty, outputting the sensor fault and inputting the sensor fault-free data prediction model to predict.
The sensor fault-free data prediction model adopts an artificial neural network model based on an on-line data self-adaptive strategy. In the training phase, the predicted mean square error is minimized by gradient descent. In the prediction stage, given fault sensor data, the numerical value of the prediction sensor in the absence of faults is transmitted to a thermal fault detection model through artificial neural network coding.
S3: and judging the operation thermodynamic state of the unit by utilizing a thermodynamic fault diagnosis model according to the fault-free prediction data, and storing a judging result into a diagnosis result database.
The thermodynamic fault diagnosis model adopts an artificial neural network based on an on-line data self-adaptive strategy, and the network is transformed by the artificial neural network model according to input characteristics such as inlet and outlet pressure of a compressor, evaporation pressure, condensation pressure and the like, and finally obtains fault class probability through softmax transformation.
The thermal fault diagnosis model updates model parameters according to real-time data, namely gradient descent update is carried out according to entropy loss, and entropy of the model in a prediction stage is reduced. When the on-line data sharpness sensing and adaptation strategy is applied, the model only updates the normalized feature modulation layer of the lower layer, as shown in fig. 5, and specifically includes:
(1) the input parameters of the model are parameters of suction temperature, exhaust temperature, suction pressure, condensing pressure, compressor current value, coolant water inlet and outlet temperature, pressure difference between an inlet and an outlet of a condenser and an evaporator, load increasing and reducing temperature, superheat degree and supercooling degree. The normalized parameter update in the L layer is obtained by the L layer output and the following formula, wherein: t (T) CI T is the suction temperature of the compressor CO For compressor discharge temperature, P RE For evaporating pressure, P RC For condensing pressure, the updating mode of the normalization parameters in the normalization layer is specifically as follows: mu≡E f L (T CI ,T CO ,P RE ,P RC …;Θ)],
(2) According to the trained network, obtaining entropy of a network output result: optimizing network parameters by minimizing loss L (x): l (T) CI ,T CO ,P RE ,P RC …)=minS(T CI ,T CO ,P RE ,P RC …)E SA (T CI ,T CO ,P RE ,P RC …;Θ),Wherein: index function S (T) CI ,T CO ,P RE ,P RC …) can be obtained from the following formula: />The gamma, beta parameters are updated as follows: />
And in the gradient descent updating process of entropy loss, selecting samples with entropy loss smaller than a set threshold value. Given the model parameter disturbance amplitude, a minimisation maximisation strategy is used, i.e. the entropy of the output result under disturbance is maximised first, and then the output entropy is minimised.
The diagnosis results stored in the diagnosis result database are mainly divided into 3 states, namely health, sub-health and faults, wherein: when the system refrigerant leakage, the condenser filth blockage, the evaporator filth blockage and the sensor fixed deviation occur, the system is prompted to be in a sub-health or early warning state, and the system is prompted to be in an alarm state such as failure of an electronic expansion valve, high pressure and low pressure and high temperature of exhaust of the system, oil leakage of a compressor, low pressure difference of coolant and cooling water, disconnection of a sensor and the like.
Through specific practical experiments, fault simulation experiments are carried out on a water-cooled chiller. The system consists of two screw compressors, two shell-and-tube condensers, a platen heat exchanger, an electronic expansion valve and the like, and the unit comprises two sets of refrigeration circulation loops, wherein the evaporator is shared by the two sets of refrigeration circulation loops. The unit refrigerant is R22, the nominal filling amount is 100kg, the rated refrigerating capacity is 400kW, the rated power is 107kW, the system structure diagram is shown in FIG. 6, the fault simulation items are shown in the following table, wherein the electric faults comprise fixed deviations of a condenser water inlet and outlet temperature and a pressure sensor, the evaporator water inlet and outlet temperature and the pressure sensor, and the electric-thermal coupling faults comprise two-by-two combinations of all electric faults and refrigerant leakage, condenser water flow reduction and evaporator water flow reduction.
And the method adopts an orthogonal state iteration method to predict the null value and correct the abnormal value of the running data of the water chilling unit, compared with the method without prediction correction, the average accuracy is improved by 7.7 percent, compared with the method of interpolation by using a numerical value near the current moment, the average accuracy is improved by 3.3 percent, and compared with the method of using a traditional artificial neural network model, the average accuracy is improved by 4.1 percent. The water inlet temperature curve of the No. 2 condenser in the experimental system is predicted by using an orthogonal state iterative method as shown in FIG. 7.
Compared with the method without sensor fault separation and fault-free data prediction, the method adopts an electric-thermal fault decoupling model, and improves the average accuracy by 9.8%. The diagnosis of the fixed deviation result of the water inlet temperature sensor of the No. 1 condenser in the electric-thermal decoupling module is shown in fig. 8, and the predicted value of the water outlet temperature sensor of the No. 1 condenser without fault data is shown in fig. 9.
The thermal fault detection model adopts an artificial neural network based on an on-line data self-adaptive strategy, compared with a support vector machine, the average accuracy is improved by 6.8%, compared with a random forest, the average accuracy is improved by 8.5%, and compared with a traditional artificial neural network model, the average accuracy is improved by 4.3%.
The foregoing embodiments may be partially modified in numerous ways by those skilled in the art without departing from the principles and spirit of the invention, the scope of which is defined in the claims and not by the foregoing embodiments, and all such implementations are within the scope of the invention.
Claims (10)
1. A fault diagnosis method of a water chilling unit based on self-adaption of reliable line data is characterized by collecting real-time running data of the unit, carrying out null value prediction and abnormal value correction by using an orthogonal state iteration method, inputting complete unit data into an electric-thermal fault decoupling model to carry out sensor state discrimination, eliminating fault sensor data and carrying out fault-free data prediction; finally, carrying out unit thermodynamic state discrimination on the fault-free prediction data by adopting a thermodynamic fault detection model to obtain a fault prediction, a diagnosis result and an auxiliary maintenance decision;
the real-time running data of the water chilling unit is obtained by reading the sensor data and the running parameters of the water chilling unit and recording the data window of the water chilling unit in real time, and the method specifically comprises the following steps: the operation state, the alarm state, the air suction temperature, the air discharge temperature, the air suction pressure, the condensation pressure, the compressor current, the coolant water inlet and outlet temperature, the pressure difference between the inlet and outlet of the condenser and the evaporator, the valve position of the electronic expansion valve and the water quantity regulating valve, the start-up and stop and load increasing and load reducing temperature, the superheat degree and the supercooling degree data;
the electric-thermal fault decoupling model comprises: a sensor bias fault detection model and a sensor fault-free data prediction model, wherein: the sensor deviation fault detection model separates sensor faults from the real-time data of unit operation, and the sensor fault-free data prediction model eliminates the interference of the sensor faults to complete the decoupling of sensor-thermal faults and obtain fault-free prediction data;
the thermodynamic fault diagnosis model is an artificial neural network based on an on-line data self-adaptive strategy, and the network obtains fault class probability through softmax conversion after nonlinear conversion according to inlet and outlet pressure, evaporation pressure and condensation pressure of a compressor, and recommends maintenance suggestions and auxiliary decisions according to a prestored unit thermodynamic state discrimination result and a current unit state.
2. The method for diagnosing faults of a water chiller based on online data adaptation according to claim 1, wherein the orthogonal state iteration method is as follows: setting the length of a historical data collection time period to be alpha minutes, collecting real-time operation data of a unit, projecting the real-time operation data onto a Legendre orthogonal polynomial base, iteratively updating an orthogonal base coefficient according to a state space expression, reconstructing the unit operation data through the orthogonal base coefficient to obtain an operation data curve approximated by the Legendre orthogonal polynomial at the current operation moment, and filling empty values in data collection and abnormal values removed by hampel filtering with predicted data values;
the projection refers to: according to the definition of orthogonal projection, obtaining the coefficient of the running data of the water chilling unit under the base of the Legendre polynomial specifically comprises the following steps:
1) According to the historical data collection time period alpha minutes, the measure function of the projection space for obtaining the running data of the water chilling unit is as followsAccording to the running time from [ -1,1]In-range conversion to [ t-alpha, t]Internal Legend orthogonal polynomialsWherein: p (P) n The method is characterized in that the method is an nth order Legend orthogonal polynomial, is certain operation data of the current water chilling unit, t is operation time, and the unit is minutes; polynomial p n (t, x) satisfy<p n (t) ,p m (t) >=δ n,m T is the runtime, n and m are the nth and mth order polynomials, respectively, delta n,m Delta when n=m to satisfy the condition n,m =1, the remaining cases δ n,m =0, symbol<·,·>Is the inner product of a polynomial in a given metric space;
2) Calculating coefficients of running data of water chilling unit projected to Legendre orthogonal polynomial basisWherein: />To be the coefficients of the nth order legendre orthogonalization polynomial with respect to chiller operation data vector at time t,wherein: t (T) CI T is the suction temperature of the compressor CO For compressor discharge temperature, P RE For evaporating pressure, P RC For condensing pressure, sign<·,·> ω(t) To be the inner product of a polynomial under the measure function ω (t) in a given metric space, f ≤t Is a function of real operation data of the water chilling unit at the time less than or equal to t.
3. The online data adaptation-based water chilling unit fault diagnosis method according to claim 2, which comprises the following steps ofCharacterized in that the state space expression refers to: according to the calculation formula of Legendre orthogonal polynomial base coefficients, calculating bias derivative at time t to obtain an updated expression of the coefficients:wherein: the expression of the C matrix is: /> The expression of the D matrix is: />n is Legendre orthogonal polynomial order, j is the running data dimension of the water chilling unit;
the iterative updating of the state space expression refers to: discretizing Legendre orthopolynomial base coefficients using bilinear methodCalculating a state update at time t: /> Wherein: i is an identity matrix.
4. The method for diagnosing faults of a water chiller based on self-adaption of reliable line data according to claim 2, wherein the Legendre orthopolynomial basis approximation is as follows: according to the orthogonal polynomial base coefficient obtained by projection, reconstructing the running data of the water chilling unit within the data collection time period alpha minutes by utilizing the Legendre orthogonal polynomial f (t), wherein the running data of the water chilling unit are specifically as follows: wherein: />The vector is Legendre orthogonal polynomial coefficient vector, is the current water chiller operation data vector, and n is Legendre orthogonal polynomial order.
5. The method for diagnosing faults of a water chiller based on online data adaptation according to claim 2, wherein the hampel filtering is as follows: setting outlier screening windows of real-time running data of a water chilling unit system, calculating the number of bits and average absolute deviation of the data in each window, and eliminating the data when the deviation is more than 3 times of the standard deviation of the data window.
6. The method for diagnosing faults of a chiller based on self-adaption of reliable line data according to any one of claims 2 to 5, wherein in the orthogonal state iterative method, if the collected chiller data has null value or abnormal value, updating the state space expression of the Legendre orthogonal polynomial base coefficient at the current moment is ignored, and when the running data is recovered to be normal, updating is performed; after Legend orthogonal polynomial base coefficients are updated, the running data of the water chilling unit in the time period alpha minutes are predicted by using the latest coefficients, the vacant data in the time period alpha minutes, which are directly collected by the water chilling unit, are covered, and the covered complete data are input into an electric-thermal fault decoupling model.
7. The method for diagnosing faults of the water chilling unit based on self-adaption of reliable line data according to claim 1 is characterized in that the electric-thermal fault decoupling model adopts historical data to train a sensor deviation fault detection model and a sensor fault-free data prediction model in an off-line stage, collects real-time data of the water chilling unit for 20 minutes recently in an on-line stage, and inputs the data in a model training working condition parameter range into the sensor deviation fault detection model to detect so as to obtain predicted sensor faults; if the output of the fault detection model is fault-free, directly inputting the thermal fault detection model to judge the thermal fault; if the output of the fault detection model is faulty, the sensor data predicted to be faulty is input into the sensor non-fault data prediction model to be predicted, and a sensor non-fault data prediction value is obtained; and the fault-free predicted value is used for replacing the latest 20 minutes of real-time data input thermal fault detection model of the water chilling unit so as to eliminate the interference of sensor faults on thermal fault diagnosis and carry out the diagnosis of thermal faults.
8. The method for diagnosing faults of a water chiller based on self-adaption of reliable line data according to claim 1 or 7, wherein the sensor deviation fault detection model is a metric learning model based on self-adaption density discrimination, and the model is trained by the following modes:
s221: encoding the input sensor data and performing nonlinear transformation to form a characterization vector;
s222: for each sensor fault category, presetting K clustering centers, using K-means clustering to minimize the distance between the characterization vector and the clustering centers, and updating the clustering center vector;
s223: randomly sampling one cluster, obtaining M-1 clusters nearest to the cluster, and randomly sampling D sensor data vectors from each cluster;
s224: the loss function of the current iteration is calculated,wherein: alpha is the size of the separation interval between clusters, < >>Representing the average value of the vectors for the m-th cluster of samples,/->For the mth cluster, the characterization vector of the mth sensor after being encoded by the artificial neural network, C (·) is the class label to which the current characterization vector belongs, <' >>The variance of the characterization vector of the m-th cluster is determined, the subscript positive sign is a negative value, and only a positive value is taken;
s225: after a certain number of iterations, suspending the training process, updating the clustering center vector, and starting gradient descent training after updating;
the trained sensor deviation fault detection model calculates the category in the prediction stage
Wherein: r is (r) n For the nth sensor data sample n Characterization vector, mu obtained by network transformation l For the mean vector of the L nearest clusters, σ is +.>C (·) is the class label to which the current token vector belongs.
9. The fault diagnosis method of the water chilling unit based on the self-adaption of the leaning line data according to claim 1 or 7, wherein the sensor fault-free data prediction model is an artificial neural network based on the self-adaption strategy of the leaning line data, and the network minimizes the mean square error of prediction data and real sensor data through gradient descent in a training stage; according to the fault sensor data, the numerical value of the sensor in the absence of faults is predicted through coding in a prediction stage;
the data self-adaptive strategy capable of leaning on the line refers to that: according to the data output after the decoupling of the electric-thermal faults, firstly removing online data with the prediction entropy of the artificial neural network model larger than a set threshold; secondly, setting a parameter disturbance range, and minimizing the maximum model prediction entropy in the disturbance range; according to the training mode, training an artificial neural network model based on an on-line data self-adaptive strategy, judging the thermal state of unit operation by using the model, and storing the judging result into a diagnosis result database; the strategy only intervenes in the prediction stage of the model, the information entropy of an output result is obtained according to the fault probability distribution output by the training model on line, and when the information entropy is larger, the model tends to give an error classification result; therefore, in the online updating stage, the model can carry out gradient descent updating according to entropy loss, so that the entropy of the model in the prediction stage is reduced; when the on-line data sharpness sensing and adaptation strategy is applied, the model only updates the normalization layer of the lower layer, namely only updates the model: the statistical normalization parameters and affine parameters of the model are used for reducing the influence on the stability of the model prediction result; the statistical normalization parameters are estimated according to real-time data, and affine parameters gamma and beta are obtained by gradient descent calculation according to entropy loss of the real-time data;
the sensor fault-free data prediction model only adjusts a standardized layer in an artificial neural network to train based on an on-line data self-adaptive strategy, and specifically comprises the following steps: the input parameters are parameters of suction temperature, exhaust temperature, suction pressure, condensing pressure, compressor current value, coolant water inlet and outlet temperature, pressure difference between an inlet and an outlet of a condenser and an evaporator, load increasing and reducing temperature, superheat degree and supercooling degree; the normalized parameter updating in the L layer is specifically as follows: mu≡E f L (T CI ,T CO ,P RE ,P RC ...;Θ)],Wherein: t (T) CI T is the suction temperature of the compressor CO For compressor discharge temperature, P RE For evaporating pressure, P RC Mu is in the normalized layer for condensing pressureMean vector, sigma of 2 To normalize the variance vector in the layer, f L (. Cndot.) is the coding function of the front L layer of the network, E (-). Cndot.; Θ) is the prediction entropy of the artificial neural network model when the network parameter is Θ;
entropy of output result of trained artificial neural network: optimizing network parameters by minimizing loss L (x): l (T) CI ,T CO ,P RE ,P RC …)=minS(T CI ,T CO ,P RE ,P RC …)E R (T CI ,T CO ,P RE ,P RC ...;Θ),Wherein: index functionT CI T is the suction temperature of the compressor CO For compressor discharge temperature, P RE For evaporating pressure, P RC For condensing pressure, E is the disturbance quantity of the neural network parameter, ρ is the upper limit of the disturbance quantity, f (& gt) is the coding function of the network, E (x; Θ) is the predictive entropy of the artificial neural network model when the network parameter is Θ, and the input chiller unit has no fault data vector ; e (E) 0 An upper limit of the set entropy threshold; the gamma and beta parameters are updated as follows: />Wherein: gamma is the scaling parameter of the normalization layer, beta is the offset parameter of the normalization layer, and L is the network loss.
10. A system for implementing the online data-based adaptive chiller fault diagnosis method of any of claims 1-9 comprising: the water chilling unit operation data acquisition and storage unit, the fault diagnosis unit and the upper computer communication unit, wherein: the method comprises the steps that a water chiller operation data acquisition unit acquires water chiller PLC acquisition data and stores the water chiller PLC acquisition data into a Mysql or Influx database, a fault diagnosis unit acquires water chiller real-time operation data from the database, an abnormal value and a null value are replaced through an orthogonal state iteration method, electric-thermal fault decoupling is carried out, and then a thermal fault judgment is carried out on a decoupling result through an artificial neural network based on an on-line data self-adaption strategy, so that a thermal fault prediction result is obtained; and the upper computer communication unit packages the fault prediction result and the auxiliary decision content and sends the packaged fault prediction result and the auxiliary decision content to the upper computer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310780299.6A CN116821796A (en) | 2023-06-29 | 2023-06-29 | Self-adaptive water chilling unit fault diagnosis method based on online data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310780299.6A CN116821796A (en) | 2023-06-29 | 2023-06-29 | Self-adaptive water chilling unit fault diagnosis method based on online data |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116821796A true CN116821796A (en) | 2023-09-29 |
Family
ID=88115120
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310780299.6A Pending CN116821796A (en) | 2023-06-29 | 2023-06-29 | Self-adaptive water chilling unit fault diagnosis method based on online data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116821796A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117074628A (en) * | 2023-10-17 | 2023-11-17 | 山东鑫建检测技术有限公司 | Multi-sensor air quality detection equipment fault positioning method |
CN117928918A (en) * | 2024-01-24 | 2024-04-26 | 南通大学 | Cold water unit fault detection method based on local abnormal nuclear entropy component analysis |
-
2023
- 2023-06-29 CN CN202310780299.6A patent/CN116821796A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117074628A (en) * | 2023-10-17 | 2023-11-17 | 山东鑫建检测技术有限公司 | Multi-sensor air quality detection equipment fault positioning method |
CN117074628B (en) * | 2023-10-17 | 2024-01-09 | 山东鑫建检测技术有限公司 | Multi-sensor air quality detection equipment fault positioning method |
CN117928918A (en) * | 2024-01-24 | 2024-04-26 | 南通大学 | Cold water unit fault detection method based on local abnormal nuclear entropy component analysis |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN116821796A (en) | Self-adaptive water chilling unit fault diagnosis method based on online data | |
CN111275288B (en) | XGBoost-based multidimensional data anomaly detection method and device | |
Wang et al. | Enhanced chiller fault detection using Bayesian network and principal component analysis | |
CN112527788A (en) | Method and device for detecting and cleaning abnormal value of transformer monitoring data | |
CN113792762A (en) | Water chilling unit fault diagnosis method, system and medium based on Bayesian optimization LightGBM | |
Chen et al. | A fault detection technique for air-source heat pump water chiller/heaters | |
CN114484731A (en) | Method and device for diagnosing faults of central air conditioner based on stacking fusion algorithm | |
JPH1144449A (en) | System for monitoring outdoor heat exchanger coil | |
CN112283876A (en) | Air conditioner fault prediction method and air conditioner | |
CN106971027B (en) | Water chilling unit fault feature selection method based on DR-BN model | |
Zeng et al. | A hybrid deep forest approach for outlier detection and fault diagnosis of variable refrigerant flow system | |
WO2019097822A1 (en) | Performance diagnosis device and performance diagnosis method for air conditioner | |
CN111723925A (en) | Method, device, equipment and medium for fault diagnosis of on-road intelligent train air conditioning unit | |
CN117272212B (en) | Method for selecting optimal operation parameters of refrigeration system based on real-time data | |
CN111649449B (en) | Air conditioner fault sensing method based on user side ubiquitous power Internet of things | |
CN112598144A (en) | CNN-LSTM burst fault early warning method based on correlation analysis | |
CN115127192A (en) | Semi-supervised water chilling unit fault diagnosis method and system based on graph neural network | |
CN109711049B (en) | Hybrid subway air conditioner water chilling unit energy efficiency estimation method | |
CN112560339B (en) | Method for predicting guide bearing bush temperature of hydroelectric generating set by utilizing machine learning | |
CN116594446B (en) | Temperature control method and system for big data center | |
CN117091273A (en) | Control method and device for self-adaptive learning data model of central air conditioning system | |
CN113934191A (en) | Fault diagnosis system for water chilling unit | |
CN117013520A (en) | Flexible load adjustable power evaluation method based on incremental learning | |
JP4434905B2 (en) | Process management device | |
CN117111661B (en) | Centralized control system and method for production workshops |
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 |