CN109308519A - A kind of refrigeration equipment failure prediction method neural network based - Google Patents
A kind of refrigeration equipment failure prediction method neural network based Download PDFInfo
- Publication number
- CN109308519A CN109308519A CN201811149566.5A CN201811149566A CN109308519A CN 109308519 A CN109308519 A CN 109308519A CN 201811149566 A CN201811149566 A CN 201811149566A CN 109308519 A CN109308519 A CN 109308519A
- Authority
- CN
- China
- Prior art keywords
- sensor
- refrigeration equipment
- neural network
- fault
- time series
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- G06N3/047—Probabilistic or stochastic networks
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Educational Administration (AREA)
- Biophysics (AREA)
- Quality & Reliability (AREA)
- General Business, Economics & Management (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Tourism & Hospitality (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Air Conditioning Control Device (AREA)
Abstract
The invention proposes a kind of refrigeration equipment failure prediction methods neural network based, comprising the following steps: obtains the collected time series data collection of N number of sensor institute in refrigeration equipment to Fault-Sensitive；Shot and long term neural network model is constructed according to time series data collection, and the actual prediction result of time series data collection is obtained according to shot and long term neural network model；Fault distinguishing model f (x) is determined according to time series data collection；The real data set of the fault condition of refrigeration equipment known to fetching portion, and fault threshold is determined according to fault distinguishing model f (x) and real data set；The probability density that actual prediction result is determined according to fault distinguishing model compares the health condition to judge refrigeration equipment by the probability density and fault threshold of actual prediction result.The present invention carries out failure predication to refrigeration equipment in such a way that shot and long term neural network model and fault distinguishing model f (x) combine, and the health status of refrigeration equipment can be effectively predicted.
Description
Technical field
The present invention relates to technical field of refrigeration equipment, and in particular to a kind of refrigeration equipment failure predication neural network based
Method.
Background technique
With the improvement of people's living standards, people are higher and higher to the dependency degree of chilled food, freezing, refrigeration food
The fast-developing impetus is presented in product industry, and freezer is also built on a large scale.However, the management level of freezer equipment is universal at present
Lowly, skilled worker's level of skill is not high, and refrigeration system is not maintained reasonably, and frequent unskilled labourer does big work, is operating only haltingly,
Cause equipment annual test in bad repair, aging accelerates, so that phenomena such as refrigeration system ammonia refrigerant emits, reveals occur, seriously affects freezer
Cooling quality.Therefore, it is badly in need of providing a kind of failure prediction method for refrigeration equipment, the failure of refrigeration equipment is avoided to occur.
Summary of the invention
In view of the deficiencies of the prior art, the present invention proposes a kind of refrigeration equipment failure prediction method neural network based,
Effectively the failure of refrigeration equipment can be predicted.
The technical scheme of the present invention is realized as follows: a kind of refrigeration equipment failure prediction method neural network based,
The following steps are included:
Step 1, the collected time series data collection of N number of sensor institute in refrigeration equipment to Fault-Sensitive is obtained；
Step 2, shot and long term neural network model is constructed according to the time series data collection, and according to the shot and long term nerve net
Network model obtains the actual prediction result of the time series data collection；
Step 3, fault distinguishing model f (x) is determined according to the time series data collection；
Step 4, the real data set of the fault condition of refrigeration equipment known to fetching portion, and according to the fault distinguishing mould
Type f (x) and the real data set determine fault threshold；
Step 5, the probability density that the actual prediction result is determined according to the fault distinguishing model, passes through the reality
The probability density of prediction result and the fault threshold compare the health condition to judge refrigeration equipment.
Optionally, the sensor includes temperature sensor, humidity sensor, pressure sensor, vibrating sensor, liquid level
Sensor and concentration sensor.
Optionally, further comprising the steps of:
Time series data collection collected is pre-processed, the pretreatment is filled up including abnormality value removing, missing data
And normalized.
Optionally, in step 2, the actual prediction of the time series data collection is obtained according to the shot and long term neural network model
As a result the step of are as follows:
Data Tendency Forecast Based is carried out to the time series data collection by shot and long term neural network model, obtains prediction result；
Anti-normalization processing is carried out to the prediction result, obtains actual prediction result.
Optionally, in step 2, include: according to the step of time series data collection building shot and long term neural network model
The time series data collection is divided into training set train and checksum set val, then according to preset time step-length by institute
It states the training set train and checksum set val and is respectively divided into input train_x, val_x and output train_y, val_y, and
It is formatted as three-dimensional array train_X, val_X, train_Y, val_Y respectively；
Nonlinear shot and long term memory nerve is constructed by described three-dimensional array train_X, val_X, train_Y, val_Y
Network Prediction Model simultaneously determines the optimal number of nodes of hidden layer using Fibonacci method, and then determines neural network optimum prediction model
Structure.
Optionally, described train_x, val_x, train_y, val_y are respectively using sample number as X-axis, preset time step
A length of Y-axis, the number of parameters of N number of sensor are three-dimensional array train_X, val_X, train_Y, val_Y as Z axis format.
Optionally, in step 3, the expression formula of the fault distinguishing model f (x) are as follows:
WhereinFor the variance of the time series data collection, μ is the mean value of the time series data collection, and x is the time series data collection
Respective value.
Optionally, in step 4, fault threshold is determined according to the fault distinguishing model f (x) and the real data set
Step includes:
Step 41, the truthful data is calculated by the fault distinguishing model f (x) to concentrate caused by each sensor
The probability density of data；
Step 42, probability density maximum value and probability density minimum value are selected from the probability density of each sensor, and
Probability density maximum value is defined to total step-length number of probability density minimum value, wherein one step-length of the every advance of probability density minimum value
For a probability threshold value；
Step 43, the F1 score of all probability threshold values of each sensor is calculated, the wherein highest probability threshold of F1 score is selected
Value is the fault threshold of respective sensor.
Optionally, the expression formula of the F1 score are as follows:
Wherein, precision is the precision ratio that truthful data concentrates some sensor, and recall is the real data set
In some sensor recall ratio, FN indicates that the truthful data concentration of some sensor is judged as negative sample, but is in fact
The data bulk of positive sample；FP indicates that the truthful data concentration of some sensor is judged as positive sample, but is in fact negative sample
This data bulk；TP indicates that the truthful data concentration of some sensor is judged as positive sample, in fact and positive sample
Data bulk.
Optionally, in step 5, by the probability density of the actual prediction result and the fault threshold compare come
The step of judging the health condition of refrigeration equipment include:
Step 51, by the probability density of the actual prediction result of each sensor compared with corresponding failure threshold value carries out；
Step 52, if its probability density is greater than its corresponding failure threshold value, then it represents that its respective sensor position detected
There are failure,
Step 53, if its probability density is less than its corresponding failure threshold value, then it represents that its respective sensor position detected
There is no failures.
Compared with prior art, the invention has the following advantages that the present invention uses shot and long term neural network model and failure
The mode that discrimination model f (x) is combined carries out failure predication to refrigeration equipment, and the health status of refrigeration equipment can be effectively predicted,
Improve the failure predication rate to refrigeration equipment.In addition, due to the time series data collection that the present invention is acquired using multiple sensors, and root
According to the unbalanced situation of refrigeration equipment data distribution of different freezers, using LSTM shot and long term Memory Neural Networks algorithm to history
Data are trained modeling, so that fault prediction model has the characteristics that dynamic self-adapting, can adapt to working condition and environment becomes
Change etc., nonlinear failure prediction is realized, failure predication rate is improved.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention without any creative labor, may be used also for those of ordinary skill in the art
To obtain other drawings based on these drawings.
Fig. 1 is that the present invention is based on the flow charts of the refrigeration equipment failure prediction method embodiment one of neural network；
Fig. 2 is that the present invention is based on the flow charts of the refrigeration equipment failure prediction method embodiment two of neural network.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is that the present invention is based on the flow chart of the refrigeration equipment failure prediction method embodiment one of neural network, the implementations
Example specifically includes the following steps:
Step 101, the collected time series data collection of N number of sensor institute in refrigeration equipment to Fault-Sensitive is obtained；
In the embodiment of the present invention, N number of sensor to Fault-Sensitive may include temperature sensor, humidity sensor, pressure
Sensor, vibrating sensor, liquid level sensor and concentration sensor (gas concentration lwevel, Funing tablet etc.) etc. are to refrigeration equipment
The influential sensor of failure.
Wherein, temperature sensor is the temperature conditions for detecting refrigeration equipment, and humidity sensor is for detecting refrigeration
The humidity condition of equipment, pressure sensor are the pressure conditions for detecting refrigeration equipment, and vibrating sensor is for detecting system
The Vibration Condition of cool equipment, liquid level sensor are used to detect the specific level condition of refrigeration equipment, and concentration sensor is for examining
Survey gas concentration lwevel, the Funing tablet etc. of refrigeration equipment.It is handled and is analyzed by the data to these sensors, to sentence
The health condition of disconnected refrigeration equipment.
Step 102, time series data collection collected is pre-processed, the pretreatment includes abnormality value removing, missing
Data filling and normalized；
In the embodiment of the present invention, after first being filled up to time series data collection collected progress abnormality value removing and missing data,
Normalizing post-processing is carried out again.Wherein, averaging method can be used and carry out abnormality value removing, missing data is carried out using interpolation method and is filled out
It mends.0-1 is mainly narrowed down to by all calculative data to the purpose that data set collected is normalized
Between, effectively simplify and calculate, saves computing resource.
Step 103, shot and long term neural network model is constructed according to the time series data collection, and according to the shot and long term nerve
Network model obtains the actual prediction result of the time series data collection；
Wherein, after shot and long term neural network model is built, refrigeration equipment can be carried out to predict its fault condition.For example,
When needing to predict the fault condition of refrigeration equipment, by acquiring N number of sensor (temperature sensor, humidity biography to Fault-Sensitive
Sensor, pressure sensor, vibrating sensor, liquid level sensor and concentration sensor etc.) data, then pass through established length
Short-term neural network model sorts out data, then predicts the data of acquisition back, finally will be according to being predicted
Result carry out breakdown judge.
The present invention, can be before refrigeration equipment prepares to break down to refrigeration by using shot and long term neural network model
Equipment running status is predicted, to repair in advance to refrigeration equipment, reduces loss.
Step 104, fault distinguishing model f (x) is determined according to the time series data collection；
In the embodiment of the present invention, fault distinguishing model f (x) is gauss of distribution functionWherein, the mean value of gauss of distribution function f (x)With variance μ by being acquired in step 101
Time series data collection determine.
Wherein, in principle, each sensor time series data collection collected determines a fault distinguishing model f (x)i, i.e.,
Gauss of distribution function f (x)i；If wherein several sensors are associated, this several sensor time series data collection collected
Determine a fault distinguishing model f (x)i, i.e. gauss of distribution function f (x)i。
Step 105, the real data set of the fault condition of refrigeration equipment known to fetching portion, and according to the fault distinguishing
Model f (x) and the real data set determine fault threshold；
In the embodiment of the present invention, acquired real data set includes the fault data and non-faulting data of refrigeration equipment.
For example, a refrigeration equipment, the collected data of N number of sensor institute are non-faulting data when working normally, and there are failures
When N number of sensor institute collected data be fault data.
Specifically, determining that fault threshold includes following son by the fault distinguishing model f (x) and the real data set
Step:
Step 1051, the truthful data is calculated by corresponding failure discrimination model f (x) to concentrate produced by each sensor
Data probability density；
Step 1052, probability density maximum value and probability density minimum value are selected from the probability density of each sensor,
And probability density maximum value is defined to total step-length number of probability density minimum value, wherein every advance one of probability density minimum value walks
An a length of probability threshold value；
For example, total step-length of the definition wherein between the probability density maximum value and probability density minimum value of some sensor is
When 1000, probability density maximum value is subtracted into probability density minimum value later again divided by total step-length, threshold steps can be obtained.It will be general
Rate density minimum value is advanced by threshold steps, wherein a probability threshold value can be obtained in step-length of advancing every time, due to total step-length
Number is 1000, and 1000 probability threshold values can be obtained at this time.
Wherein, total step-length number can be set according to actual needs, and total step-length number is bigger in principle, and precision is higher,
It calculates also more complicated.
In the embodiment of the present invention, the fault threshold of all the sensors is calculated.
Step 1053, the F1 score of all probability threshold values of each sensor is calculated, the wherein highest probability of F1 score is selected
Threshold value is the fault threshold of respective sensor.
Wherein, the expression formula of F1 score are as follows:
In expression formula, precision is the precision ratio that truthful data concentrates some sensor, i.e., the knot returned after retrieval
In fruit, truly correct number accounts for the ratio of entire result；Recall is that the truthful data concentrates looking into for some sensor complete
Rate, i.e., truly correct number accounts in entire data set (retrieving and not retrieving) truly correct in search result
Several ratios；FN (False Negative) indicates that the truthful data concentration of some sensor is judged as negative sample, but true
On be positive sample data bulk；FP (False Positive) indicates that the truthful data concentration of some sensor is judged as just
Sample, but be in fact the data bulk of negative sample；TN (True Negative) indicates that the truthful data of some sensor is concentrated
It is judged as negative sample, in fact and the data bulk of negative sample；TP (True Positive) indicates the true of some sensor
Real data concentration is judged as positive sample, in fact and the data bulk of positive sample.
Step 106, the probability density that the actual prediction result is determined according to the fault distinguishing model, passes through the reality
The probability density of border prediction result and the fault threshold compare the health condition to judge refrigeration equipment.
Specifically, being compared by the probability density of the actual prediction result with the fault threshold to judge to freeze
The health condition of equipment includes following sub-step:
Step 1061, the probability density of the actual prediction result of each sensor and corresponding failure threshold value ∈ i are subjected to phase
Than；
Step 1062, if its probability density is greater than its corresponding failure threshold value ∈ i, then it represents that its respective sensor is detected
Position there are failure,
Step 1063, if its probability density is less than its corresponding failure threshold value ∈ i, then it represents that its respective sensor is detected
Failure is not present in position.
If the probability density of all the sensors data collected is respectively less than its corresponding failure threshold value ∈, the refrigeration is proved
Failure is not present in equipment.
The present invention is in such a way that shot and long term neural network model and fault distinguishing model f (x) combine to refrigeration equipment
Failure predication is carried out, the health status of refrigeration equipment can be effectively predicted, improves the failure predication rate to refrigeration equipment.
The present invention predicts N number of sensor data collected by constructing shot and long term neural network model, obtains
Its practical prediction result；Then its fault threshold is obtained further through building fault distinguishing model, finally by by this N number of sensor
The probability density of actual prediction result the health condition to judge refrigeration equipment is compared with corresponding fault threshold respectively,
If it is which position of refrigeration equipment is broken down that refrigeration equipment there are failure, may also be aware of according to judging result, so as to
It quickly repairs, avoids loss.
Fig. 2 is that the present invention is based on the flow chart of the refrigeration equipment failure prediction method embodiment two of neural network, the implementations
Example specifically includes the following steps:
Step 201, the collected time series data collection { x1 (t- of N number of sensor institute in refrigeration equipment to Fault-Sensitive is obtained
1),x2(t-1),x3(t-1),x4(t-1),...,x1(t),x2(t),x3(t),x4(t)}；
As above-described embodiment, N number of sensor to Fault-Sensitive includes temperature sensor, humidity sensor, pressure
Sensor and concentration sensor (gas concentration lwevel, Funing tablet etc.) etc. are on the influential sensor of the failure of refrigeration equipment.
Step 202, using averaging method and interpolation method to the time series data collection of acquisition carry out respectively abnormality value removing and
Missing data is filled up, then to treated time series data collection { x1 (t-1), x2 (t-1), x3 (t-1), x4 (t-1) ..., x1
(t), x2 (t), x3 (t), x4 (t) } it is normalized, time series data collection { X1 (t-1), X2 (t-1), X3 after being normalized
(t-1),X4(t-1),...,X1(t),X2(t),X3(t),X4(t)}；
In the embodiment of the present invention, data set collected is normalized, is in order to will be all calculative
Data all narrow down between 0-1, effectively simplify and calculate, and save computing resource.
Step 203, shot and long term neural network model is constructed according to the time series data collection, and according to the shot and long term nerve
Network model obtains the actual prediction result of the time series data collection；
Specifically, including following sub-step:
Step 2031, the time series data collection is divided into training set train and checksum set val；
For example, the time series data collection is divided into training set train and checksum set val in the ratio of 8:2, or press it
He divides ratio.
Step 2032, the training set train and the checksum set val are respectively divided into according to preset time step-length defeated
Enter train_x, val_x and output train_y, val_y；
For example, the time step of training set train and verifying collection val each record is 10 minutes, then just temporally
Step-length 10/2 divide equally respectively training set train and verifying collection val, must can input train_x, val_x and export train_y,
Val_y, wherein train_x is preceding 5 minutes data, and train_y is rear 5 minutes numbers, and val_x is preceding 5 minutes data,
Val_y is rear 5 minutes numbers.
Step 2033, train_x, val_x, train_y, val_y are formatted as respectively three-dimensional array train_X,
val_X,train_Y,val_Y；
For example, train_x has 100 samples, each sample 5-minute data, each minute data are made of 4 parameters
(x1, x2, x3, x4), then being used as y-axis within 5 minutes, 4 parameters constitute three-dimensional as z-axis just using 100 samples as x-axis
Array [sample number, duration, number of parameters], other are similarly.
Step 2034, nonlinear length is constructed by described three-dimensional array train_X, val_X, train_Y, val_Y
Phase Memory Neural Networks prediction model, and the optimal number of nodes of hidden layer is determined using Fibonacci method, and then determine neural network
Optimum prediction model structure；
Step 2035, operation data trend prediction is carried out by the optimum prediction model, obtains prediction result { y1 (t-
1),y2(t-1),y3(t-1),y4(t-1),...,y1(t),y2(t),y3(t),y4(t)}；
Step 2036, anti-normalization processing is carried out to the prediction result, obtains actual prediction result { Y1 (t-1), Y2
(t-1),Y3(t-1),Y4(t-1),...,Y1(t),Y2(t),Y3(t),Y4(t)}。
In the embodiment of the present invention, mould is predicted by the shot and long term Memory Neural Networks that building is in the nature non-linear topological structure
Type can adapt to working condition and environmental change etc. so that prediction model has the characteristics that dynamic self-adapting, realize non-linear event
Barrier prediction.
Wherein, a upper embodiment can also construct shot and long term Memory Neural Networks prediction model using this method, and carry out
Predict its practical prediction result.
Step 204, fault distinguishing model f (x) is determined according to the time series data collection after normalization；
In the embodiment of the present invention, fault distinguishing model f (x) is gauss of distribution functionWherein, the mean value of gauss of distribution function f (x)With variance μ by it is collected when ordinal number
It is determined according to collection.
Likewise, each sensor time series data collection collected determines a fault distinguishing model f (x)i, i.e. Gauss point
Cloth function f (x)i；If wherein several sensors are associated, this several sensor time series data collection collected determines one
A fault distinguishing model f (x)i, i.e. gauss of distribution function f (x)i。
Step 205, the real data set of the fault condition of refrigeration equipment known to fetching portion, and according to the fault distinguishing
Model f (x) and the real data set determine fault threshold；
In the embodiment of the present invention, the determination method of fault threshold can be consistent with a upper embodiment, equally to calculate all
The fault threshold of sensor.
Step 206, the probability density that the actual prediction result is determined according to the fault distinguishing model, passes through the reality
The probability density of border prediction result and the fault threshold compare the health condition to judge refrigeration equipment.
Specifically, being compared by the probability density of the actual prediction result with the fault threshold to judge to freeze
The step of health condition of equipment includes:
Step 1061, the probability density of the actual prediction result of each sensor and corresponding failure threshold value ∈ i are subjected to phase
Than；
Step 1062, if its probability density is greater than its corresponding failure threshold value ∈ i, then it represents that its respective sensor is detected
Position there are failure,
Step 1063, if its probability density is less than its corresponding failure threshold value ∈ i, then it represents that its respective sensor is detected
Failure is not present in position.
Likewise, being demonstrate,proved if the probability density of all the sensors data collected is respectively less than its corresponding failure threshold value ∈
Failure is not present in the bright refrigeration equipment.
The present invention applies the method for big data analysis in terms of the failure predication of refrigeration equipment, effectively improves refrigeration
The failure predication rate of equipment；And the shot and long term neural network model that the present invention constructs is prediction model, has dynamic self-adapting
Feature can adapt to working condition and environmental change etc., realize nonlinear failure prediction.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of refrigeration equipment failure prediction method neural network based, which comprises the following steps:
Step 1, the collected time series data collection of N number of sensor institute in refrigeration equipment to Fault-Sensitive is obtained；
Step 2, shot and long term neural network model is constructed according to the time series data collection, and according to the shot and long term neural network mould
Type obtains the actual prediction result of the time series data collection；
Step 3, fault distinguishing model f (x) is determined according to the time series data collection；
Step 4, the real data set of the fault condition of refrigeration equipment known to fetching portion, and according to the fault distinguishing model f
(x) and the real data set determines fault threshold；
Step 5, the probability density that the actual prediction result is determined according to the fault distinguishing model, passes through the actual prediction
As a result probability density and the fault threshold compares the health condition to judge refrigeration equipment.
2. refrigeration equipment failure prediction method neural network based as described in claim 1, which is characterized in that the sensor
Including temperature sensor, humidity sensor, pressure sensor, vibrating sensor, liquid level sensor and concentration sensor.
3. refrigeration equipment failure prediction method neural network based as described in claim 1, which is characterized in that further include following
Step:
Time series data collection collected is pre-processed, the pretreatment is filled up and returned including abnormality value removing, missing data
One change processing.
4. refrigeration equipment failure prediction method neural network based as claimed in claim 3, which is characterized in that in step 2, root
The step of obtaining the actual prediction result of the time series data collection according to the shot and long term neural network model are as follows:
Data Tendency Forecast Based is carried out to the time series data collection by shot and long term neural network model, obtains prediction result；
Anti-normalization processing is carried out to the prediction result, obtains actual prediction result.
5. the refrigeration equipment failure prediction method neural network based as described in Claims 1 to 4 is any, which is characterized in that step
In rapid 2, include: according to the step of time series data collection building shot and long term neural network model
The time series data collection is divided into training set train and checksum set val, then according to preset time step-length by the instruction
Practice the collection train and checksum set val and be respectively divided into input train_x, val_x and output train_y, val_y, and respectively
It is formatted as three-dimensional array train_X, val_X, train_Y, val_Y；
Nonlinear shot and long term Memory Neural Networks are constructed by described three-dimensional array train_X, val_X, train_Y, val_Y
Prediction model simultaneously determines the optimal number of nodes of hidden layer using Fibonacci method, and then determines neural network optimum prediction model knot
Structure.
6. refrigeration equipment failure prediction method neural network based as claimed in claim 5, which is characterized in that the train_
X, respectively using sample number as X-axis, preset time step-length is Y-axis, the parameter number of N number of sensor by val_x, train_y, val_y
Amount is three-dimensional array train_X, val_X, train_Y, val_Y as Z axis format.
7. refrigeration equipment failure prediction method neural network based as claimed in claim 6, which is characterized in that in step 3, institute
State the expression formula of fault distinguishing model f (x) are as follows:
WhereinFor the variance of the time series data collection, μ is the mean value of the time series data collection, and x is the phase of the time series data collection
Answer numerical value.
8. refrigeration equipment failure prediction method neural network based as claimed in claim 7, which is characterized in that in step 4, root
The step of determining fault threshold according to the fault distinguishing model f (x) and the real data set include:
Step 41, the truthful data is calculated by the fault distinguishing model f (x) and concentrates data caused by each sensor
Probability density；
Step 42, probability density maximum value and probability density minimum value are selected from the probability density of each sensor, and are defined
Probability density maximum value is to total step-length number of probability density minimum value, and wherein the every step-length of advancing of probability density minimum value is one
A probability threshold value；
Step 43, the F1 score for calculating all probability threshold values of each sensor, select wherein the highest probability threshold value of F1 score for
The fault threshold of respective sensor.
9. refrigeration equipment failure prediction method neural network based as claimed in claim 8, which is characterized in that the F1 score
Expression formula are as follows:
Wherein, precision is the precision ratio that truthful data concentrates some sensor, and recall is that the truthful data concentrates certain
The recall ratio of a sensor, FN indicate that the truthful data concentration of some sensor is judged as negative sample, but are in fact positive samples
This data bulk；FP indicates that the truthful data concentration of some sensor is judged as positive sample, but is in fact negative sample
Data bulk；TP indicates that the truthful data concentration of some sensor is judged as positive sample, in fact and the data of positive sample
Quantity.
10. refrigeration equipment failure prediction method neural network based as claimed in claim 9, which is characterized in that in step 5,
The health condition to judge refrigeration equipment is compared by the probability density and the fault threshold of the actual prediction result
The step of include:
Step 51, by the probability density of the actual prediction result of each sensor compared with corresponding failure threshold value carries out；
Step 52, if its probability density is greater than its corresponding failure threshold value, then it represents that its respective sensor position detected exists
Failure,
Step 53, if its probability density is less than its corresponding failure threshold value, then it represents that do not deposit at its respective sensor position detected
In failure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811149566.5A CN109308519A (en) | 2018-09-29 | 2018-09-29 | A kind of refrigeration equipment failure prediction method neural network based |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811149566.5A CN109308519A (en) | 2018-09-29 | 2018-09-29 | A kind of refrigeration equipment failure prediction method neural network based |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109308519A true CN109308519A (en) | 2019-02-05 |
Family
ID=65225360
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811149566.5A Pending CN109308519A (en) | 2018-09-29 | 2018-09-29 | A kind of refrigeration equipment failure prediction method neural network based |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109308519A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109992916A (en) * | 2019-04-09 | 2019-07-09 | 山东超越数控电子股份有限公司 | A kind of engineering outfield aircraft engine failure prediction method based on flight data, terminal and readable storage medium storing program for executing |
| CN110097134A (en) * | 2019-05-08 | 2019-08-06 | 合肥工业大学 | A kind of mechanical breakdown method of early diagnosis based on time series |
| CN110490266A (en) * | 2019-08-23 | 2019-11-22 | 北京邮电大学 | A kind of sensing data uploads, Transducer-fault Detecting Method and device |
| CN110516941A (en) * | 2019-08-16 | 2019-11-29 | 北京慧辰资道资讯股份有限公司 | One kind producing hidden danger method and device based on big data analysis enterprise |
| CN110704508A (en) * | 2019-09-30 | 2020-01-17 | 佛山科学技术学院 | Intelligent production line abnormal data processing method and device |
| CN111439267A (en) * | 2020-03-30 | 2020-07-24 | 上海商汤临港智能科技有限公司 | Method and device for adjusting cabin environment |
| CN111815056A (en) * | 2020-07-10 | 2020-10-23 | 中国人民解放军空军工程大学 | Aircraft external field aircraft fuel system fault prediction method based on flight parameter data |
| CN112183830A (en) * | 2020-09-16 | 2021-01-05 | 新奥数能科技有限公司 | Method and device for predicting temperature of chilled water |
| CN113536989A (en) * | 2021-06-29 | 2021-10-22 | 广州博通信息技术有限公司 | Refrigerator frosting monitoring method and system based on camera video frame-by-frame analysis |
| JP2021184132A (en) * | 2020-05-21 | 2021-12-02 | 東芝ライフスタイル株式会社 | Information processing system |
| EP4071571A4 (en) * | 2020-01-21 | 2023-01-25 | Mitsubishi Heavy Industries Engine & Turbocharger, Ltd. | Prediction apparatus, prediction method, and program |
| CN115879054A (en) * | 2023-03-03 | 2023-03-31 | 泰安市特种设备检验研究院 | Method and device for determining liquid ammonia refrigeration state based on image processing |
| CN116579762A (en) * | 2023-04-14 | 2023-08-11 | 广州林旺空调工程有限公司 | Intelligent operation and maintenance platform for cooling tower |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107341462A (en) * | 2017-06-28 | 2017-11-10 | 电子科技大学 | A kind of video classification methods based on notice mechanism |
| CN107769972A (en) * | 2017-10-25 | 2018-03-06 | 武汉大学 | A kind of power telecom network equipment fault Forecasting Methodology based on improved LSTM |
| US20180075348A1 (en) * | 2016-09-09 | 2018-03-15 | Cylance Inc. | Machine learning model for analysis of instruction sequences |
-
2018
- 2018-09-29 CN CN201811149566.5A patent/CN109308519A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180075348A1 (en) * | 2016-09-09 | 2018-03-15 | Cylance Inc. | Machine learning model for analysis of instruction sequences |
| CN107341462A (en) * | 2017-06-28 | 2017-11-10 | 电子科技大学 | A kind of video classification methods based on notice mechanism |
| CN107769972A (en) * | 2017-10-25 | 2018-03-06 | 武汉大学 | A kind of power telecom network equipment fault Forecasting Methodology based on improved LSTM |
Non-Patent Citations (1)
| Title |
|---|
| PANKAJ MALHOTRA等: "Long Short Term Memory Networks for Anomaly Detection in time series", 《ESANN 2015 PROCEEDINGS, EUROPEAN SYMPOSIUM ON ARTIFICIAL NEURAL NETWORKS, COMPUTATIONAL INTELLIGENCE AND MACHINE LEARNING》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109992916A (en) * | 2019-04-09 | 2019-07-09 | 山东超越数控电子股份有限公司 | A kind of engineering outfield aircraft engine failure prediction method based on flight data, terminal and readable storage medium storing program for executing |
| CN110097134A (en) * | 2019-05-08 | 2019-08-06 | 合肥工业大学 | A kind of mechanical breakdown method of early diagnosis based on time series |
| CN110516941A (en) * | 2019-08-16 | 2019-11-29 | 北京慧辰资道资讯股份有限公司 | One kind producing hidden danger method and device based on big data analysis enterprise |
| CN110490266A (en) * | 2019-08-23 | 2019-11-22 | 北京邮电大学 | A kind of sensing data uploads, Transducer-fault Detecting Method and device |
| CN110704508A (en) * | 2019-09-30 | 2020-01-17 | 佛山科学技术学院 | Intelligent production line abnormal data processing method and device |
| CN110704508B (en) * | 2019-09-30 | 2023-04-07 | 佛山科学技术学院 | Intelligent production line abnormal data processing method and device |
| EP4071571A4 (en) * | 2020-01-21 | 2023-01-25 | Mitsubishi Heavy Industries Engine & Turbocharger, Ltd. | Prediction apparatus, prediction method, and program |
| WO2021196721A1 (en) * | 2020-03-30 | 2021-10-07 | 上海商汤临港智能科技有限公司 | Cabin interior environment adjustment method and apparatus |
| CN111439267B (en) * | 2020-03-30 | 2021-12-07 | 上海商汤临港智能科技有限公司 | Method and device for adjusting cabin environment |
| CN111439267A (en) * | 2020-03-30 | 2020-07-24 | 上海商汤临港智能科技有限公司 | Method and device for adjusting cabin environment |
| JP2021184132A (en) * | 2020-05-21 | 2021-12-02 | 東芝ライフスタイル株式会社 | Information processing system |
| CN111815056A (en) * | 2020-07-10 | 2020-10-23 | 中国人民解放军空军工程大学 | Aircraft external field aircraft fuel system fault prediction method based on flight parameter data |
| CN112183830A (en) * | 2020-09-16 | 2021-01-05 | 新奥数能科技有限公司 | Method and device for predicting temperature of chilled water |
| CN113536989A (en) * | 2021-06-29 | 2021-10-22 | 广州博通信息技术有限公司 | Refrigerator frosting monitoring method and system based on camera video frame-by-frame analysis |
| CN115879054A (en) * | 2023-03-03 | 2023-03-31 | 泰安市特种设备检验研究院 | Method and device for determining liquid ammonia refrigeration state based on image processing |
| CN115879054B (en) * | 2023-03-03 | 2023-05-30 | 泰安市特种设备检验研究院 | Method and device for determining liquid ammonia refrigeration state based on image processing |
| CN116579762A (en) * | 2023-04-14 | 2023-08-11 | 广州林旺空调工程有限公司 | Intelligent operation and maintenance platform for cooling tower |
| CN116579762B (en) * | 2023-04-14 | 2023-10-20 | 广州林旺空调工程有限公司 | Intelligent operation and maintenance platform for cooling tower |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109308519A (en) | A kind of refrigeration equipment failure prediction method neural network based | |
| CN109858244B (en) | Method and system for detecting abnormal behaviors of processes in container | |
| CN110362068A (en) | A kind of mechanical equipment fault method for early warning, system and readable storage medium storing program for executing based on industrial Internet of Things | |
| AU2016286280A1 (en) | Combined method for detecting anomalies in a water distribution system | |
| CN110647539A (en) | Prediction method and system for vehicle faults | |
| CN110414155A (en) | A kind of detection of fan part temperature anomaly and alarm method with single measuring point | |
| CN107797537A (en) | A kind of prognostic and health management method applied to automatic production line | |
| CN106104496A (en) | The abnormality detection not being subjected to supervision for arbitrary sequence | |
| CN107796609B (en) | Water chilling unit fault diagnosis method based on DBN model | |
| NO343417B1 (en) | System and method for condition assessment of downhole tools | |
| CN108027611B (en) | Decision assistance system and method for machine maintenance using expert opinion supervised decision mode learning | |
| CN106973039A (en) | A kind of network security situation awareness model training method and device based on information fusion technology | |
| Qin et al. | Remaining useful life prediction for rotating machinery based on optimal degradation indicator | |
| CN116380496B (en) | Automobile door fatigue endurance test method, system and medium | |
| CN115962797B (en) | Sensor reliability test method and system based on temperature stress | |
| CN115392782A (en) | Method and system for monitoring and diagnosing health state of process system of nuclear power plant | |
| CN113361939B (en) | Dynamic association method and system for quality and equipment management of wrapping machine type | |
| CN117743120A (en) | Network card abnormality detection method based on big data technology | |
| CN117891234A (en) | Method and device for detecting running state of machine room, storage medium and electronic equipment | |
| EP4273520A1 (en) | Method and system for comprehensively diagnosing defect in rotating machine | |
| CN108664696A (en) | A kind of assessment method and device of handpiece Water Chilling Units operating status | |
| CN111259494A (en) | Health monitoring and analyzing method for heavy machine equipment | |
| CN117555716B (en) | Abnormal mode self-adaptive multidimensional root cause positioning device and root cause positioning method | |
| US11047833B2 (en) | Method for automatic determination of trend in graphic analysis of turbomachines | |
| CN111160454B (en) | Quick change signal detection method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190205 |