CN117093896A - Heat pump system fault diagnosis method based on combined data driving model - Google Patents
Heat pump system fault diagnosis method based on combined data driving model Download PDFInfo
- Publication number
- CN117093896A CN117093896A CN202310984152.9A CN202310984152A CN117093896A CN 117093896 A CN117093896 A CN 117093896A CN 202310984152 A CN202310984152 A CN 202310984152A CN 117093896 A CN117093896 A CN 117093896A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- trained
- time sequence
- heat pump
- 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
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000003745 diagnosis Methods 0.000 title claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 65
- 238000012360 testing method Methods 0.000 claims abstract description 41
- 239000002245 particle Substances 0.000 claims abstract description 37
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 18
- 238000012545 processing Methods 0.000 claims abstract description 15
- 230000006870 function Effects 0.000 claims description 43
- 238000004364 calculation method Methods 0.000 claims description 22
- 230000000694 effects Effects 0.000 claims description 18
- 230000000737 periodic effect Effects 0.000 claims description 15
- 230000008859 change Effects 0.000 claims description 14
- 238000011156 evaluation Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 12
- 238000006073 displacement reaction Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 6
- 230000006403 short-term memory Effects 0.000 claims description 6
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 4
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 claims description 3
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 claims description 3
- 230000001133 acceleration Effects 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000000354 decomposition reaction Methods 0.000 claims description 3
- 238000001704 evaporation Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 239000002131 composite material Substances 0.000 abstract description 4
- 238000012544 monitoring process Methods 0.000 abstract 1
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 3
- 238000004880 explosion Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000001932 seasonal effect Effects 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention relates to the technical field of heat pump system fault diagnosis, and particularly discloses a heat pump system fault diagnosis method based on a combined data driving model, which comprises the following steps: acquiring time sequence data acquired by each sensor on the heat pump equipment, and dividing the time sequence data into a training set and a testing set after processing the time sequence data; respectively inputting the data in the training set into a Prophet model and an LSTM model optimized by a particle swarm algorithm, simultaneously carrying out model training, and obtaining an optimal weight coefficient to obtain a trained combined reference model; inputting the data in the test set into the trained combined reference model for fault prediction so as to output a fault prediction result of the test set; verifying the fitting degree of the combined reference model; and deploying a verified combined reference model, and monitoring the heat pump system in real time. The invention can better analyze the time series data characteristics, has good composite characteristic capturing capability for a complex heat pump system, and can improve fault diagnosis precision.
Description
Technical Field
The invention relates to the technical field of heat pump system fault diagnosis, in particular to a heat pump system fault diagnosis method based on a combined data driving model.
Background
A heat pump system is a highly complex thermodynamic system with large hysteresis, strong coupling. When the heat pump system actually operates, the system is unstable due to the change of control logic and environmental factors, and is from one temporary steady state to another temporary steady state, and at the moment, each parameter of the system has different hysteresis, so that difficulty is brought to system reference modeling. To build a reference model, an accurate fit system hysteresis is required, its influencing factors are related to environmental variables and control variables, wherein the mutual coupling between the two is the most difficult, so that a reference model suitable for a heat pump system needs to be built to realize fault diagnosis of the unsteady heat pump system.
The modeling method of the current heat pump system can be divided into: the physical model and the data driving model are built based on thermodynamic basic principles, energy conservation and mass conservation. Depending on the large number of data formulas, the disadvantage is: the calculated amount is large, and the precision is low; the data-driven model is divided into: multiple regression and machine learning regression models, wherein the multiple regression is mainly based on statistical regression modeling by using polynomials and the like, and the method is simple and has the defects that: the residual error data introduced in the later stage can cause larger errors, so that the precision is poor; machine learning regression is commonly used with: support vector regression, decision tree regression, random forest regression, neural network regression. Compared with the former three, the neural network has the advantages that: large capacity and strong generalization. However, in the current method for diagnosing faults of the heat pump system, a single model is mostly adopted, the time sequence in the heat pump system is usually nonlinear, the requirement of high-speed fault diagnosis is difficult to meet, and the single model has defects on the composite characteristics of the time sequence.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a heat pump system fault diagnosis method based on a combined data driving model, which utilizes the combined model to exert respective advantages and has better fault diagnosis advantage compared with a single model.
As a first aspect of the present invention, there is provided a heat pump system fault diagnosis method based on a combined data driving model, the heat pump system including a heat pump apparatus, the heat pump system fault diagnosis method based on a combined data driving model including the steps of:
step S1: acquiring time sequence data acquired by each sensor on the heat pump equipment, wherein the time sequence data of the heat pump equipment comprises historical time sequence data and current time sequence data;
step S2: processing the historical time sequence data and the current time sequence data respectively to correspondingly obtain processed historical time sequence data and processed current time sequence data, and dividing the processed historical time sequence data into a training set and a testing set;
step S3: respectively inputting the data in the training set into a Prophet model and an LSTM model optimized by a particle swarm algorithm, and simultaneously carrying out model training to obtain a trained Prophet model and a trained LSTM model, obtaining optimal weight coefficients according to failure prediction results in respective training, and obtaining a trained combined reference model according to the trained Prophet model and the corresponding optimal weight coefficients and the trained LSTM model and the corresponding optimal weight coefficients;
step S4: inputting the data in the test set into the trained combined reference model for fault prediction so as to output a fault prediction result of the test set;
step S5: judging whether the precision of the trained combined reference model meets the requirement according to the fault prediction result of the test set and the fault actual result of the test set, if so, entering a step S6, otherwise, returning to the step S3;
step S6: and inputting the processed current time sequence data into the trained combined reference model for fault diagnosis so as to output the current fault diagnosis result of the heat pump system.
Further, the time sequence data collected by each sensor comprises: compressor surface temperature, condensing temperature, evaporating temperature, discharge temperature, suction pressure, outlet water temperature and return water temperature.
Further, the processing the historical time sequence data and the current time sequence data respectively to obtain processed historical time sequence data and processed current time sequence data correspondingly, and dividing the processed historical time sequence data into a training set and a testing set, and further includes:
preprocessing the historical time sequence data and the current time sequence data respectively to obtain preprocessed historical time sequence data and preprocessed current time sequence data;
respectively carrying out normalization processing on the preprocessed historical time sequence data and the preprocessed current time sequence data to correspondingly obtain normalized historical time sequence data and normalized current time sequence data;
and (3) the normalized historical time sequence data is processed according to 8:2 are divided into training and testing sets.
Further, the preprocessed historical time sequence data and the preprocessed current time sequence data are normalized, and a normalization processing formula is as follows:
wherein x is i Is the original dataCollection, x min Is the minimum value in the original data set, x max X is the maximum in the original dataset * To normalize the processed data set, x * The value range of (1) is (0, 1).
Further, the step of inputting the data in the training set into the propset model and the LSTM model optimized by the particle swarm algorithm respectively and performing model training simultaneously to obtain a trained propset model and a trained LSTM model, further includes:
decomposing data y (t) in the training set into three parts using the propset model: the trend change function g (t), the holiday effect function h (t) and the periodic function s (t) have the following calculation formulas:
y(t)＝g(t)+s(t)+h(t)+ε(t)
wherein a trend change function g (t) processes the aperiodic changes of the data in the training set; a periodic function s (t) processes the periodic variation of the data in the training set; the holiday effect function h (t) is directed at the effect of holidays on the data in the training set; epsilon (t) is an error fluctuation function of the model;
and training the Prophet model by utilizing the data y (t) in the training set, obtaining model parameters by continuous evaluation and adjustment, and determining three decomposition functions to obtain a fault prediction result p (t) of the Prophet model.
Further, the method further comprises the following steps:
(1) The calculation formula of the trend change function g (t) is as follows:
wherein C is the capacity of the Prophet model, k is the growth rate of the trend, n is the offset parameter, and t is the time used for trend change;
(2) Approximating the periodic function s (t) with a fourier series, the periodic function s (t) having a calculation formula:
wherein s is the period of the time sequence, 2N is the number of periods expected in the model, N is the order of Fourier transformation, a and b are estimated parameters;
(3) The calculation formula of h (t) of the holiday effect function is as follows:
wherein k is i I is holiday information, t is time required by a window period, and L is the length of training set data.
Further, the step of inputting the data in the training set into the propset model and the LSTM model optimized by the particle swarm algorithm respectively and performing model training simultaneously to obtain a trained propset model and a trained LSTM model, further includes:
the LSTM model is composed of a plurality of unit structures, including three gates: input gate i t Forgetting door f t And an output gate o t The calculation formula is as follows:
wherein i is t 、f t 、o t Respectively an input door, a forget door and an output door; x is x t Input at time t; h is a t-1 Short term memory at time t; h is a t Short term memory at time t+1; sigma is a sigmoid function; tanh is a hyperbolic tangent function; c t-1 For the previous cell state, c after updating t ；Is a point-by-point product; w (w) i 、w f 、w o A weighting matrix for each threshold; b i 、b f 、b o A conversion deviation value for each threshold;
optimizing an LSTM model by using a particle swarm algorithm PSO, initializing various parameters of the particle swarm, and performing iterative computation to obtain the parameters of the optimized LSTM model, wherein the formula of the particle swarm algorithm PSO is as follows:
wherein w is an inertial weight for controlling weight distribution of the particles in local and global optima; c 1 、c 2 Adjusting the flight step length as an acceleration factor; r is (r) 1 、r 2 A random number between 0 and 1; the speed of the lower particle, the displacement of the lower particle, the local optimal solution of the position, the global optimal solution and the displacement of the optimal solution at the corresponding moment are respectively determined;the speed of the lower particle and the displacement of the lower particle at the next moment are respectively corresponding;
and training the optimized LSTM model by utilizing the data in the training set to obtain a trained LSTM model and a fault prediction result L (t) of the optimized LSTM model.
Further, the method includes the steps of obtaining optimal weight coefficients according to failure prediction results during respective training, and obtaining a trained combined reference model according to the trained propset model and the corresponding optimal weight coefficients thereof and the trained LSTM model and the corresponding optimal weight coefficients thereof, and further includes:
assuming that at time t, the failure prediction result of the propset model is p (t), the failure prediction result of the optimized LSTM model is L (t), and a dynamic weight value w is respectively given to the propset model and the optimized LSTM model 1 And w 2 According to p (t) and L (t)Parameter optimizing to obtain optimal weight coefficient w 1 And w 2 According to the trained Prophet model and the corresponding optimal weight coefficient w 1 And the trained LSTM model and the corresponding optimal weight coefficient w 2 Obtaining a trained combined reference model;
the method comprises the steps of carrying out linear weighting on a fault prediction result p (t) of the propset model and a fault prediction result L (t) of the optimized LSTM model according to a corresponding optimal weight coefficient to obtain an optimal fault diagnosis result R (t), wherein the calculation formula is as follows:
R(t)＝w 1 P(t)+w 2 L(t)
wherein w is 1 、w 2 For the optimal weight coefficient, the values satisfy the following conditions: w (w) 1 +w 2 =1, determined by the least squares method.
Further, the method judges whether the precision of the trained combined reference model meets the requirement according to the fault prediction result of the test set and the fault actual result of the test set, if yes, the method proceeds to step S6, otherwise, the method returns to step S3 and further includes:
and evaluating the trained combined reference model by using two evaluation indexes of Root Mean Square Error (RMSE) and average absolute percentage error (MAPE), wherein the calculation formulas of the two evaluation indexes are as follows:
wherein p is i 、a i Respectively obtaining a fault prediction result of the test set at the moment i and a fault actual result of the test set, wherein n is sample data in the test set;
if the evaluation index of the trained combined reference model meets the requirement, the parameters and the structure of the current combined reference model are stored, and the step S6 is carried out; if the requirement is not satisfied, the process returns to step S3.
Further, the step of inputting the processed current time sequence data into the trained combined reference model for fault diagnosis to output the current fault diagnosis result of the heat pump system further includes:
deploying the trained combined reference model on a cloud system or a local system;
and inputting the processed current time sequence data into the trained combined reference model to perform real-time fault diagnosis on the heat pump system, and performing alarm pushing on the diagnosed real-time fault information by using a cloud system or a local system and storing the information into a database.
The heat pump system fault diagnosis method based on the combined data driving model has the following beneficial effects:
(1) Compared with a single model, the combined model can better analyze time series data characteristics, has good composite characteristic capturing capability for a complex heat pump system, and can improve fault diagnosis precision;
(2) The LSTM model can be used for solving the problems of gradient explosion and disappearance of a simple cyclic neural network, and the LSTM is optimized by a particle swarm algorithm, so that the hysteresis problem of a heat pump system can be well solved, and the problems of long training time and poor diagnosis effect caused by inaccurate parameters selected empirically can be solved.
Drawings
The accompanying drawings are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate the invention and together with the description serve to explain, without limitation, the invention.
Fig. 1 is a flowchart of a heat pump system fault diagnosis method based on a combined data driving model provided by the invention.
Fig. 2 is a flowchart of a specific implementation manner of a heat pump system fault diagnosis method based on a combined data driving model provided by the invention.
Fig. 3 is a training flowchart of the propset model provided by the invention.
Fig. 4 is a flowchart of the optimization of the LSTM model by the particle swarm algorithm provided by the present invention.
Detailed Description
In order to further describe the technical means and effects adopted by the invention to achieve the preset aim, the following is a detailed description of specific implementation, structure, characteristics and effects of a heat pump system fault diagnosis method based on a combined data driving model according to the invention with reference to the accompanying drawings and preferred embodiments. It will be apparent that the described embodiments are some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to fall within the scope of the invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate in order to describe the embodiments of the invention herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
In this embodiment, there is provided a heat pump system fault diagnosis method based on a combined data driving model, as shown in fig. 1, where the heat pump system includes a heat pump device, and the heat pump system fault diagnosis method based on a combined data driving model includes the following steps:
step S1: acquiring time sequence data acquired by each sensor on the heat pump equipment, wherein the time sequence data of the heat pump equipment comprises historical time sequence data and current time sequence data;
preferably, the time sequence data collected by each sensor includes: compressor surface temperature, condensing temperature, evaporating temperature, discharge temperature, suction pressure, outlet water temperature and return water temperature.
Step S2: processing the historical time sequence data and the current time sequence data respectively to correspondingly obtain processed historical time sequence data and processed current time sequence data, and dividing the processed historical time sequence data into a training set and a testing set;
preferably, as shown in fig. 2, the processing the historical time series data and the current time series data respectively to obtain processed historical time series data and processed current time series data correspondingly, and dividing the processed historical time series data into a training set and a testing set, further includes:
preprocessing the historical time sequence data and the current time sequence data respectively to obtain preprocessed historical time sequence data and preprocessed current time sequence data; for example, outliers and duplicates in the historical and current timing data are deleted, and missing values in the historical and current timing data are patched.
Because the difference of different data value ranges is large, normalization processing is further needed to be carried out on the preprocessed historical time sequence data and the preprocessed current time sequence data respectively so as to correspondingly obtain the normalized historical time sequence data and the normalized current time sequence data;
and (3) the normalized historical time sequence data is processed according to 8:2 are divided into training and testing sets.
Specifically, the preprocessed historical time sequence data and the preprocessed current time sequence data are normalized, and a normalization processing formula is as follows:
wherein x is i X is the original data set min Is the minimum value in the original data set, x max Is the original numberAccording to the maximum value in the set, x * For the normalized dataset, the range of x values is (0, 1).
Step S3: respectively inputting the data in the training set into a Prophet model and an LSTM model optimized by a Particle Swarm Optimization (PSO) to perform model training simultaneously so as to obtain a trained Prophet model and a trained LSTM model, acquiring optimal weight coefficients according to failure prediction results during respective training, and acquiring a trained combined reference model according to the trained Prophet model and the corresponding optimal weight coefficients and the trained LSTM model and the corresponding optimal weight coefficients;
preferably, the data in the training set is respectively input into a propset model and an LSTM model optimized by a particle swarm algorithm to perform model training at the same time, so as to obtain a trained propset model and a trained LSTM model, and the method further includes:
the Prophet model can process time sequence characteristics such as periodicity, festival effect, trend change and the like of data, and has good effect on robustness of abnormal values and missing values. Decomposing data y (t) in the training set into three parts using the propset model: the trend change function g (t), the holiday effect function h (t) and the periodic function s (t) have the following calculation formulas:
y(t)＝g(t)+s(t)+h(t)+ε(t)
wherein a trend change function g (t) processes the aperiodic changes of the data in the training set; a periodic function s (t) processes the periodic variation of the data in the training set; the holiday effect function h (t) is directed at the effect of holidays on the data in the training set; epsilon (t) is an error fluctuation function of the model;
as shown in fig. 3, the propset model is trained by using the data y (t) in the training set, model parameters are obtained by continuous evaluation and adjustment, and the three decomposition functions are determined, so that a fault prediction result p (t) of the propset model is obtained.
Specifically, the method further comprises the following steps:
(1) The calculation formula of the trend change function g (t) is as follows:
wherein C is the capacity of the Prophet model, k is the growth rate of the trend, n is the offset parameter, and t is the time used for trend change;
(2) Approximating the periodic function s (t) by a fourier series from seasonal characteristics, the periodic function s (t) having a calculation formula:
wherein s is the period of the time sequence, 2N is the number of periods expected in the model, N is the order of Fourier transformation, a and b are estimated parameters;
(3) Since holidays are independent for different time points, the calculation formula of h (t) of the holiday effect function is as follows:
wherein k is i I is holiday information, t is time required by a window period, and L is the length of training set data.
Preferably, the data in the training set is respectively input into a propset model and an LSTM model optimized by a particle swarm algorithm to perform model training at the same time, so as to obtain a trained propset model and a trained LSTM model, and the method further includes:
because of the over fitting of the diagnosis results of the single model, the combined diagnosis is needed by combining with the LSTM model, but because of the gradient disappearance and explosion problems of the LSTM processing time sequence data, the LSTM model is optimized by introducing a particle swarm algorithm PSO, and the optimal super-parameter combination is found; the specific implementation is as follows:
the LSTM model is composed of a plurality of unit structures, including three gates: input gate i t Forgetting door f t And output ofDoor o t The calculation formula is as follows:
wherein i is t 、f t 、o t Respectively an input door, a forget door and an output door; x is x t Input at time t; h is a t-1 Short term memory at time t; h is a t Short term memory at time t+1; sigma is a sigmoid function, tanh is a hyperbolic tangent function, both of which are activation functions; c t-1 For the previous cell state, c after updating t ；Is a point-by-point product; w (w) i 、w f 、w o A weighting matrix for each threshold; b i 、b f 、b o A conversion deviation value for each threshold;
the LSTM model is optimized by using a particle swarm algorithm PSO, as shown in fig. 4, each parameter of the particle swarm is initialized first, and then iterative calculation is performed to obtain the optimized parameter of the LSTM model, and the formula of the particle swarm algorithm PSO is as follows:
wherein w is an inertial weight for controlling weight distribution of the particles in local and global optima; c 1 、c 2 For the acceleration factor, adjusting the flight step (non-negative value); r is (r) 1 、r 2 A random number between 0 and 1;the speed of the lower particle, the displacement of the lower particle, the local optimal solution of the position, the global optimal solution and the displacement of the optimal solution at the corresponding moment are respectively determined; />Respectively isThe speed of the lower particles and the displacement of the lower particles at the next moment correspond to each other;
and training the optimized LSTM model by utilizing the data in the training set to obtain a trained LSTM model and a fault prediction result L (t) of the optimized LSTM model.
Preferably, the obtaining an optimal weight coefficient according to the failure prediction result during respective training, and then obtaining a trained combined reference model according to the trained propset model and the corresponding optimal weight coefficient thereof, and the trained LSTM model and the corresponding optimal weight coefficient thereof, further includes:
assuming that at time t, the failure prediction result of the propset model is p (t), the failure prediction result of the optimized LSTM model is L (t), and a dynamic weight value w is respectively given to the propset model and the optimized LSTM model 1 And w 2 Parameter optimization is carried out according to p (t) and L (t) to obtain an optimal weight coefficient w 1 And w 2 According to the trained Prophet model and the corresponding optimal weight coefficient w 1 And the trained LSTM model and the corresponding optimal weight coefficient w 2 Obtaining a trained combined reference model;
the method comprises the steps of carrying out linear weighting on a fault prediction result p (t) of the propset model and a fault prediction result L (t) of the optimized LSTM model according to a corresponding optimal weight coefficient to obtain an optimal fault diagnosis result R (t), wherein the calculation formula is as follows:
R(t)＝w 1 P(t)+w 2 L(t)
wherein w is 1 、w 2 For the optimal weight coefficient, the values satisfy the following conditions: w (w) 1 +w 2 =1, determined by the least squares method.
Step S4: inputting the data in the test set into the trained combined reference model for fault prediction so as to output a fault prediction result of the test set;
step S5: verifying the accuracy of the fault prediction result of the test set, judging whether the accuracy of the trained combined reference model meets the requirement according to the fault prediction result of the test set and the fault actual result of the test set, if so, entering a step S6, otherwise, returning to the step S3;
preferably, the method includes judging whether the accuracy of the trained combined reference model meets the requirement according to the failure prediction result of the test set and the failure actual result of the test set, if so, entering step S6, otherwise, returning to step S3, and further including:
in order to verify the fitting degree of the trained combined reference model and the accuracy of the fault prediction result of the test set, the trained combined reference model is evaluated by using two evaluation indexes of Root Mean Square Error (RMSE) and average absolute percentage error (MAPE), and the calculation formulas of the two evaluation indexes are as follows:
wherein p is i 、a i Respectively obtaining a fault prediction result of the test set at the moment i and a fault actual result of the test set, wherein n is sample data in the test set;
if the evaluation index of the trained combined reference model meets the requirement, the parameters and the structure of the current combined reference model are stored, and the step S6 is carried out; if the requirement is not satisfied, the process returns to step S3.
Step S6: and inputting the processed current time sequence data into the trained combined reference model for fault diagnosis so as to output the current fault diagnosis result of the heat pump system.
In step S6, the trained combined reference model is the current combined reference model.
Preferably, the step of inputting the processed current time series data into the trained combined reference model to perform fault diagnosis so as to output a current fault diagnosis result of the heat pump system further includes:
batch processing is carried out on the time sequence data by utilizing a database MySQL, and the trained combined reference model is deployed in a cloud system or a local system;
and inputting the processed current time sequence data into the trained combined reference model to perform real-time fault diagnosis on the heat pump system, and performing alarm pushing on the diagnosed real-time fault information by using a cloud system or a local system and storing the information into a database.
According to the heat pump system fault diagnosis method based on the combined data driving model, the combined model is adopted to carry out fault diagnosis on the heat pump system, compared with a single model, the time series data characteristics can be better analyzed, the complex heat pump system has good composite characteristic capturing capability, and the fault diagnosis precision can be improved; the LSTM model can be used for solving the problems of gradient explosion and disappearance of a simple circulating neural network, and the PSO is utilized to optimize the LSTM, so that the hysteresis problem of the heat pump system can be well solved, and the problems of long training time and poor diagnosis effect caused by inaccurate parameters selected empirically can be solved.
The present invention is not limited to the above-mentioned embodiments, but is intended to be limited to the following embodiments, and any modifications, equivalents and modifications can be made to the above-mentioned embodiments without departing from the scope of the invention.
Claims (10)
1. The heat pump system fault diagnosis method based on the combined data driving model is characterized by comprising the following steps of:
step S1: acquiring time sequence data acquired by each sensor on the heat pump equipment, wherein the time sequence data of the heat pump equipment comprises historical time sequence data and current time sequence data;
step S2: processing the historical time sequence data and the current time sequence data respectively to correspondingly obtain processed historical time sequence data and processed current time sequence data, and dividing the processed historical time sequence data into a training set and a testing set;
step S3: respectively inputting the data in the training set into a Prophet model and an LSTM model optimized by a particle swarm algorithm, and simultaneously carrying out model training to obtain a trained Prophet model and a trained LSTM model, obtaining optimal weight coefficients according to failure prediction results in respective training, and obtaining a trained combined reference model according to the trained Prophet model and the corresponding optimal weight coefficients and the trained LSTM model and the corresponding optimal weight coefficients;
step S4: inputting the data in the test set into the trained combined reference model for fault prediction so as to output a fault prediction result of the test set;
step S5: judging whether the precision of the trained combined reference model meets the requirement according to the fault prediction result of the test set and the fault actual result of the test set, if so, entering a step S6, otherwise, returning to the step S3;
step S6: and inputting the processed current time sequence data into the trained combined reference model for fault diagnosis so as to output the current fault diagnosis result of the heat pump system.
2. The heat pump system fault diagnosis method based on the combined data driving model according to claim 1, wherein the time series data collected by each sensor comprises: compressor surface temperature, condensing temperature, evaporating temperature, discharge temperature, suction pressure, outlet water temperature and return water temperature.
3. The heat pump system fault diagnosis method based on a combined data driving model according to claim 2, wherein the processing the historical time series data and the current time series data respectively to correspond to the obtained processed historical time series data and the processed current time series data, and dividing the processed historical time series data into a training set and a testing set, further comprises:
preprocessing the historical time sequence data and the current time sequence data respectively to obtain preprocessed historical time sequence data and preprocessed current time sequence data;
respectively carrying out normalization processing on the preprocessed historical time sequence data and the preprocessed current time sequence data to correspondingly obtain normalized historical time sequence data and normalized current time sequence data;
and (3) the normalized historical time sequence data is processed according to 8:2 are divided into training and testing sets.
4. A heat pump system fault diagnosis method based on a combined data driving model according to claim 3, wherein the pre-processed historical time series data and the pre-processed current time series data are normalized, and the normalization processing formula is as follows:
wherein x is i X is the original data set min Is the minimum value in the original data set, x max X is the maximum in the original dataset * To normalize the processed data set, x * The value range of (1) is (0, 1).
5. The heat pump system fault diagnosis method based on a combined data driving model according to claim 1, wherein the data in the training set is respectively input into a propset model and an LSTM model optimized by a particle swarm algorithm for model training at the same time, so as to obtain a trained propset model and a trained LSTM model, and further comprising:
decomposing data y (t) in the training set into three parts using the propset model: the trend change function g (t), the holiday effect function h (t) and the periodic function s (t) have the following calculation formulas:
y(t)＝g(t)+s(t)+h(t)+ε(t)
wherein a trend change function g (t) processes the aperiodic changes of the data in the training set; a periodic function s (t) processes the periodic variation of the data in the training set; the holiday effect function h (t) is directed at the effect of holidays on the data in the training set; epsilon (t) is an error fluctuation function of the model;
and training the Prophet model by utilizing the data y (t) in the training set, obtaining model parameters by continuous evaluation and adjustment, and determining three decomposition functions to obtain a fault prediction result p (t) of the Prophet model.
6. The heat pump system fault diagnosis method based on the combined data driving model according to claim 5, further comprising:
(1) The calculation formula of the trend change function g (t) is as follows:
wherein C is the capacity of the Prophet model, k is the growth rate of the trend, n is the offset parameter, and t is the time used for trend change;
(2) Approximating the periodic function s (t) with a fourier series, the periodic function s (t) having a calculation formula:
wherein s is the period of the time sequence, 2N is the number of periods expected in the model, N is the order of Fourier transformation, a and b are estimated parameters;
(3) The calculation formula of h (t) of the holiday effect function is as follows:
wherein k is i I is holiday information, t is time required by a window period, and L is the length of training set data.
7. The method for diagnosing a heat pump system fault based on a combined data driven model according to claim 5, wherein the step of simultaneously performing model training on the data in the training set respectively input to the propset model and the LSTM model optimized by the particle swarm algorithm to obtain a trained propset model and a trained LSTM model, further comprises:
the LSTM model is composed of a plurality of unit structures, including three gates: input gate i t Forgetting door f t And an output gate o t The calculation formula is as follows:
wherein i is t 、f t 、o t Respectively an input door, a forget door and an output door; x is x t Input at time t; h is a t-1 Short term memory at time t; h is a t Short term memory at time t+1; sigma is a sigmoid function; tanh is a hyperbolic tangent function; c t-1 For the previous cell state, c after updating t ；Is a point-by-point product; w (w) i 、w f 、w o A weighting matrix for each threshold; b i 、b f 、b o A conversion deviation value for each threshold;
optimizing an LSTM model by using a particle swarm algorithm PSO, initializing various parameters of the particle swarm, and performing iterative computation to obtain the parameters of the optimized LSTM model, wherein the formula of the particle swarm algorithm PSO is as follows:
wherein w is an inertial weight for controlling weight distribution of the particles in local and global optima; c 1 、c 2 Adjusting the flight step length as an acceleration factor; r is (r) 1 、r 2 A random number between 0 and 1; the speed of the lower particle, the displacement of the lower particle, the local optimal solution of the position, the global optimal solution and the displacement of the optimal solution at the corresponding moment are respectively determined;the speed of the lower particle and the displacement of the lower particle at the next moment are respectively corresponding;
and training the optimized LSTM model by utilizing the data in the training set to obtain a trained LSTM model and a fault prediction result L (t) of the optimized LSTM model.
8. The method for diagnosing a heat pump system fault based on a combined data driven model according to claim 7, wherein the obtaining an optimal weight coefficient according to the fault prediction result during respective training, and then obtaining a trained combined reference model according to the trained propset model and the corresponding optimal weight coefficient thereof, and the trained LSTM model and the corresponding optimal weight coefficient thereof, further comprises:
assuming that at the time t, the failure prediction result of the Prophet model is p (t), and after optimizationThe failure prediction result of the LSTM model is L (t), and a dynamic weight value w is respectively given to the Prophet model and the optimized LSTM model 1 And w 2 Parameter optimization is carried out according to p (t) and L (t) to obtain an optimal weight coefficient w 1 And w 2 According to the trained Prophet model and the corresponding optimal weight coefficient w 1 And the trained LSTM model and the corresponding optimal weight coefficient w 2 Obtaining a trained combined reference model;
the method comprises the steps of carrying out linear weighting on a fault prediction result p (t) of the propset model and a fault prediction result L (t) of the optimized LSTM model according to a corresponding optimal weight coefficient to obtain an optimal fault diagnosis result R (t), wherein the calculation formula is as follows:
R(t)＝w 1 P(t)+w 2 L(t)
wherein w is 1 、w 2 For the optimal weight coefficient, the values satisfy the following conditions: w (w) 1 +w 2 =1, determined by the least squares method.
9. The method for diagnosing a heat pump system fault based on a combined data driving model according to claim 1, wherein the method for determining whether the accuracy of the trained combined reference model meets the requirement according to the fault prediction result of the test set and the fault actual result of the test set, if yes, step S6 is entered, otherwise, step S3 is returned to, and further comprising:
and evaluating the trained combined reference model by using two evaluation indexes of Root Mean Square Error (RMSE) and average absolute percentage error (MAPE), wherein the calculation formulas of the two evaluation indexes are as follows:
wherein p is i 、a i Respectively obtaining a fault prediction result of the test set at the moment i and a fault actual result of the test set, wherein n is sample data in the test set;
if the evaluation index of the trained combined reference model meets the requirement, the parameters and the structure of the current combined reference model are stored, and the step S6 is carried out; if the requirement is not satisfied, the process returns to step S3.
10. The heat pump system fault diagnosis method based on a combined data driving model according to claim 1, wherein the step of inputting the processed current time series data into the trained combined reference model for fault diagnosis to output the current fault diagnosis result of the heat pump system further comprises the steps of:
deploying the trained combined reference model on a cloud system or a local system;
and inputting the processed current time sequence data into the trained combined reference model to perform real-time fault diagnosis on the heat pump system, and performing alarm pushing on the diagnosed real-time fault information by using a cloud system or a local system and storing the information into a database.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310984152.9A CN117093896A (en) | 2023-08-07 | 2023-08-07 | Heat pump system fault diagnosis method based on combined data driving model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310984152.9A CN117093896A (en) | 2023-08-07 | 2023-08-07 | Heat pump system fault diagnosis method based on combined data driving model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117093896A true CN117093896A (en) | 2023-11-21 |
Family
ID=88774250
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310984152.9A Pending CN117093896A (en) | 2023-08-07 | 2023-08-07 | Heat pump system fault diagnosis method based on combined data driving model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117093896A (en) |
-
2023
- 2023-08-07 CN CN202310984152.9A patent/CN117093896A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110738360B (en) | Method and system for predicting residual life of equipment | |
| CN109766583B (en) | Aircraft engine life prediction method based on unlabeled, unbalanced and initial value uncertain data | |
| KR101098037B1 (en) | System and method for automatic virtual metrology | |
| CN108320016B (en) | Short-term prediction method for building energy consumption | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
| CN110288046B (en) | Fault prediction method based on wavelet neural network and hidden Markov model | |
| CN110555230B (en) | Rotary machine residual life prediction method based on integrated GMDH framework | |
| CN111814956B (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
| CN112116162B (en) | Power transmission line icing thickness prediction method based on CEEMDAN-QFAOA-LSTM | |
| CN114218872B (en) | DBN-LSTM semi-supervised joint model-based residual service life prediction method | |
| CN111898867B (en) | Airplane final assembly production line productivity prediction method based on deep neural network | |
| CN107644297B (en) | Energy-saving calculation and verification method for motor system | |
| CN110757510A (en) | Method and system for predicting remaining life of robot | |
| CN113947182A (en) | Traffic flow prediction model construction method based on double-stage stack graph convolution network | |
| CN112884236A (en) | Short-term load prediction method and system based on VDM decomposition and LSTM improvement | |
| CN113325721A (en) | Model-free adaptive control method and system for industrial system | |
| CN114548591A (en) | Time sequence data prediction method and system based on hybrid deep learning model and Stacking | |
| CN116992759A (en) | Freeze-thawing concrete strength evaluation method and device based on interpretable neural network | |
| CN115128978A (en) | Internet of things environment big data detection and intelligent monitoring system | |
| CN115096357A (en) | Indoor environment quality prediction method based on CEEMDAN-PCA-LSTM | |
| CN116303786B (en) | Block chain financial big data management system based on multidimensional data fusion algorithm | |
| CN117093896A (en) | Heat pump system fault diagnosis method based on combined data driving model | |
| CN115062764B (en) | Intelligent illuminance adjustment and environmental parameter Internet of things big data system | |
| Wenqiang et al. | Remaining useful life prediction for mechanical equipment based on temporal convolutional network |
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 |