CN110332647A - Subway underground station Load Prediction method and air-conditioning system - Google Patents
Subway underground station Load Prediction method and air-conditioning system Download PDFInfo
- Publication number
- CN110332647A CN110332647A CN201910624024.7A CN201910624024A CN110332647A CN 110332647 A CN110332647 A CN 110332647A CN 201910624024 A CN201910624024 A CN 201910624024A CN 110332647 A CN110332647 A CN 110332647A
- Authority
- CN
- China
- Prior art keywords
- neural network
- underground station
- load
- air
- energy consumption
- 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.)
- Granted
Links
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F24—HEATING; RANGES; VENTILATING
- F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
- F24F11/00—Control or safety arrangements
- F24F11/30—Control or safety arrangements for purposes related to the operation of the system, e.g. for safety or monitoring
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F24—HEATING; RANGES; VENTILATING
- F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
- F24F11/00—Control or safety arrangements
- F24F11/62—Control or safety arrangements characterised by the type of control or by internal processing, e.g. using fuzzy logic, adaptive control or estimation of values
- F24F11/63—Electronic processing
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F24—HEATING; RANGES; VENTILATING
- F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
- F24F2140/00—Control inputs relating to system states
- F24F2140/50—Load
Abstract
The invention patent relates to the predictions of air conditioner load, more particularly to a kind of subway underground station Load Prediction method, including following operating procedure: 1) environmental parameter of subway underground station being selected and inputted, variance analysis is carried out to environmental parameter, chooses parameter influential on air conditioning energy consumption using orthogonal experiment and method of analysis of variance；2) environmental parameter and air conditioning energy consumption mapping relations are established according to the environmental parameter of input；3) Parameters of Neural Network Structure and operating parameter are set, neural network is constructed；4) change environmental parameter, load is predicted using neural network, obtains prediction result；5) prediction result is evaluated with the ratio of total sum of squares of deviations, the root-mean-square error coefficient of variation using regression sum of square.Using this subway underground station Load Prediction method, so that cooling capacity and air quantity are supplied by public domain workload demand in standing, to realize that station air-conditioning system is energy saving by a relatively large margin.
Description
Technical field
The invention patent relates to the predictions of air conditioner load, and in particular to subway underground station Load Prediction method
And air-conditioning system.
Background technique
With the continuous development of urban track traffic, subway station quantity is also increasing.Year end 2017 of cut-off, urban rail are handed over
Logical 3234, station of putting into operation increases by 21.1% than last year.In City Rail Transit System station architecture, ventilation and air conditioning equipment
Energy consumption accounts for 65.71%, is most important large electricity consumer in rail traffic.As it can be seen that reducing ventilation and air conditioning system energy consumption to city
Rail Transit System station architecture energy conservation plays a crucial role.Subway underground station public domain ventilation and air conditioning system energy
It consumes with passenger flow, Train Interval, seasonal variations, the more apparent peak valley difference of presentation, and existing ventilation and air conditioning system control
Thought processed is mostly that systematic air flow, water are adjusted using feedback regulation.Due to the intrinsic lag characteristic of water system itself, lead
Cause air-conditioning system that cannot immediately, rapidly make a response to the load of each section environment, causing refrigeration unit is that station is public
The cooling capacity and the unmatched phenomenon of actually required cooling capacity that region provides.Therefore, how subsequent time is relatively accurately predicted
The air-conditioning refrigeration duty of underground station, being controlled in advance according to expected refrigeration duty ventilation and air conditioning system becomes rail vehicle
It stands the emphasis of ventilation and air conditioning system energy saving research.
Typical load forecasting method has exponential smoothing, gray prediction, linear regression and nerve net in engineer application at present
Network.Exponential smoothing is one kind of time series forecasting technology, is predicted according to the historical data of prediction object itself.Linearly
The Return Law is a kind of interpretation model based on regression analysis.Gray prediction theory is to implement to add up random sequence to weaken it
Randomness is regular to the reality for finding load development.
The Air-conditioning Load Prediction model structure that exponential smoothing is established is simple, and prediction cost is smaller, and system transplantation is preferable,
But the data for the correlative factor for having substantial connection with building loading cannot be efficiently used, it is more difficult to further promote precision of prediction.
Grey method is larger to the processing work difficulty of initial data before modeling, and precision of prediction is by original data processing result
It influences.Due to being non-linear relation between most influence factors and air conditioner load, thus multiple linear regression air-conditioning prediction in not
It is desirable.The precision of prediction highest of neural network, but it is not suitable for the air-conditioning system that newer, initial data lacks.
The research for carrying out thermal control to air-conditioning system with intelligent control methods such as artificial neural networks both at home and abroad at present is more
Concentrate on office building.Load prediction is carried out to certain university's administrative building, certain library.But the office building commuter time is constant,
Indoor airflow heat condition is basically unchanged, and air conditioner load variation has stronger regularity.
But be mostly based on big data correlation theory to the load prediction of subway underground station, Artificial Neural Network is used
It is less in the research of subway underground station Air-conditioning Load Prediction, and underground station public domain ventilation and air conditioning load is by passenger flow, column
The influence of vehicle interval, seasonal variations changes greatly, and is difficult to realize the Accurate Prediction to it using traditional load forecasting method.
The Air-condition system control strategy of existing subway station is the thought based on feedback control, according to room temperature or wet
The single influence factor such as degree in air-conditioning system unitary air handling unit, return/exhaust blower in air quantity of sending back to be adjusted, drop
Low fan energy consumption is to reach energy saving purpose.
Station public domain institute required airflow, cooling capacity determine that refrigeration duty is outdoor temperature humidity, passenger flow, sets by refrigeration duty in standing
Coupling output under the various factors effect such as standby power consumption, is adjusted air-conditioning system according to single factors, there is regulation
Inaccurate problem.In addition, feedback control has hysteresis quality, current control method not can be well solved Metro Air conditioner ventage
The volume hysteresis of system.
Summary of the invention
It is a primary object of the present invention to be directed to subway underground station, provide it is a kind of predict it is more accurate, be more suitable
Load Prediction method and air-conditioning system for subway underground station.
To achieve the goals above, the application uses a kind of subway underground station Load Prediction method, packet
Include following operating procedure:
1) environmental parameter of subway underground station is selected and is inputted, wherein variance analysis is carried out to environmental parameter,
Parameter influential on air conditioning energy consumption is chosen using orthogonal experiment and method of analysis of variance；The parameter of selection should be to air-conditioning
The parameter that system energy consumption makes a significant impact；
2) environmental parameter and air conditioning energy consumption mapping relations are established according to the environmental parameter of input；
3) Parameters of Neural Network Structure and operating parameter are set, neural network is constructed；The neural network is feed forward neural
The Multi-layered Feedforward Networks of error back propagation in network；Feedback Neural Network is always deposited in actual neural fusion
In the transmission delay of information, these delays have an impact to the characteristic of neural network, and this Multi-layered Feedforward Networks is with extremely strong non-
Linear Mapping ability, and less calculation resources are occupied compared to feedback network；
4) change environmental parameter, load is predicted using neural network, obtains prediction result；
5) prediction result is evaluated；Wherein, evaluation when using regression sum of square and total sum of squares of deviations ratio
Value, the root-mean-square error coefficient of variation evaluate prediction result.
Subway underground station influences to fluctuate larger, this method compared to other public places, load factor affected by environment
Orthogonal experiment and method of analysis of variance are used when analyzing environmental parameter, orthogonal experiments analysis are commonly used very poor
Analytic approach or method of analysis of variance, extremum difference analysis calculation amount is smaller, is easily understood, but cannot distinguish between each horizontal institute under same factor
The difference of corresponding test result is that horizontal change causes or test error causes.Method of analysis of variance can be the change of factor level
The difference changed between test result caused by the fluctuation of the difference and error between caused test result distinguishes, and can provide
Reliable quantity survey.
Using this subway underground station Load Prediction method, use numerical analysis method to input parameter into
Row screening, by the utilization of neural network, can do the Public Areas Associated with Metro Station domain refrigeration duty under a variety of such environmental effects
It predicts out, and adjusts air-conditioner wind, water system accordingly, so that cooling capacity and air quantity are supplied by public domain workload demand in standing, with reality
Existing station air-conditioning system is energy saving by a relatively large margin.
It is further, establish environmental parameter and air conditioning energy consumption mapping relations the following steps are included:
1) according to subway underground station air conditioning energy consumption data, subway underground station air-Conditioning Load Calculation Method is determined；
2) according to air conditioning energy consumption and air conditioner load corresponding data, the environmental parameter for influencing air conditioner load is determined；
3) parameter influential on air conditioning energy consumption is chosen using orthogonal experiment and method of analysis of variance.
It is further, when establishing environmental parameter and air conditioning energy consumption mapping relations, according to track underground station energy
Consumption establishes energy consumption and environmental parameter mapping model；Wherein, subway underground station energy consumption includes: air-conditioning system cold source, power
Transmission & distribution, end-equipment energy consumption.
It is further, the environmental parameter include: passenger flow variation, equipment light power size, outdoor dry-bulb temperature,
Outside relative humidity and solar radiation quantity.
It is further, to environmental parameter carry out variance analysis the following steps are included:
1) according to orthogonal experiment contrived experiment scheme；The thought for utilizing fraction Factorial Design, using orthogonal experiment,
The representational point in part is picked out from comprehensive test according to orthogonality to be tested；
2) null hypothesis is proposed；I.e. according to test result, null hypothesis: H0- indifference is proposed；There were significant differences by H1-；It is proposed inspection
It tests hypothesis and is also known as null hypothesis, symbol is H0；The symbol of alternative hypothesis is H1；
3) selection check statistic；Select bulk testing data selection check statistic: what variance analysis generally used
Test statistics is F statistic, i.e., F value is examined；
4) observation and probability P value of test statistics are calculated；
5) it gives significance and makes decisions；Significance is given according to P value size, and is made decisions.
It is further, before carrying out variance analysis to environmental parameter, carry out range analysis；Compare range analysis knot
Fruit and the results of analysis of variance.The difference that difference analysis cannot distinguish between test result corresponding to each level under same factor is that level changes
Caused by change causes still test error, range analysis result can be used as the reference of the results of analysis of variance, have more analysis result
Objectivity.
It is further, when constructing neural network, determined by the size of significant property coefficient to air conditioning energy consumption shadow
The scale of size is rung, i.e., by being that the size of P value coefficient is determined on the scale of air conditioning energy consumption influence size.
It is further, Parameters of Neural Network Structure and operating parameter are set the following steps are included:
1) training data in neural network is set；I.e. in great number tested data, selection is suitable for neural network instruction
Experienced sample data；
2) screening and normalized are carried out to training data；Data screening, which refers to, pre-processes data, rejecting abnormalities
Data；Normalized, which refers to, to be mapped the data within the scope of 0~1；Sample data is screened in this way, replaces removing with interpolation
The wrong value, bad value gone, and overall data is normalized；
3) neural network node in hidden layer is set；According to existing empirical equation, hidden layer neuron number range is determined,
It according to the node in hidden layer that gathers when selecting training error minimum of dichotomy examination is neural network node in hidden layer in range；
Neural network node in hidden layer range is primarily determined with empirical equation, repeatedly chooses different node in hidden layer to nerve net
Network is trained, and determines suitable node in hidden layer；
4) all kinds of transmission function scope of applications are analyzed, choose transmission function in conjunction with prediction output valve；It analyzes all kinds of
With the transmission function scope of application, range selection input layer, hidden layer and output layer transmission function locating for prediction output valve are considered；
5) learning rate is chosen；I.e. under the premise of other parameters are fixed, when integrating output accuracy and neural network learning
Between, Rational choice learning rate
It is further, using regression sum of square and the ratio of total sum of squares of deviations, the root-mean-square error coefficient of variation pair
The prediction result evaluated the following steps are included:
1) change environmental parameter, subsequent time load is predicted using neural network；
2) R is used2Prediction result is evaluated with root-mean-square error coefficient of variation CVRMSE:
If R2>=0.8 and root-mean-square error coefficient of variation CVRMSE≤30% item think that prediction result is acceptable；That is the knot
Fruit is reliable, and the result is available.
If R2< 0.8 or root-mean-square error coefficient of variation CVRMSE > 30% then think that prediction result is unacceptable.
Above-mentioned R2It is statistically for measuring the statistic of the goodness of fit, the also referred to as coefficient of determination.
The present invention provides a kind of context aware systems for subway station air-conditioning system, comprising:
Environmental perception module, for monitoring and obtaining ambient data；
Neural network load prediction module is communicated to connect with environmental perception module, for receiving the environmental data, to reality
When load predicted, output prediction refrigeration duty value；
Main control module is communicated to connect with neural network load prediction module, for by receiving the prediction refrigeration duty
The outlet terminal of value control cooling capacity, air quantity.
The neural network load prediction module uses above-mentioned subway underground station Load Prediction method pair
Subway underground station air conditioner load is predicted.
To reach whole system control target, the realization of system purpose is based on ambient condition for the setting of above three module
Cognition technology, environmental perception module can be distributed multiple groups sensor and carry out information collection；Neural network load prediction module predicts
Input quantity of the load value come as regulation, main control module is to provide control instruction to all parts of whole system.
For environmental perception module after detecting, getting ambient data, neural network load prediction module receives ring
Output prediction refrigeration duty value after the data of border, main control module are carried out according to outlet terminal of the prediction refrigeration duty value to cooling capacity, air quantity
Control, air-conditioning system makes prediction to the Public Areas Associated with Metro Station domain refrigeration duty under a variety of such environmental effects, and adjusts accordingly
Cooling capacity outlet terminal is by workload demand supply in public domain in standing, to realize that station air-conditioning system is energy saving by a relatively large margin.
Further, the environmental perception module is provided with
Outdoor-monitoring sensing group, which is set to outside subway station and environmental perception module communicates to connect,
The outdoor-monitoring sensing group is for detecting subway station external environment；
Indoor monitoring sensing group, the indoor monitoring sensing group are set in subway station and communicate to connect with environmental perception module,
Indoor monitoring sensing group is for detecting environment in subway station.
Ambient condition include station external environment and stand in environment, refrigeration duty by indoor and outdoor surroundings factor coupling influence, here
Outdoor-monitoring sensing group, indoor monitoring sensing group is respectively set in environmental perception module, the limitation of single environment is avoided to lead to letter
Breath collects the inaccuracy insufficient, subsequent prediction program is caused to be predicted.
Further, the outdoor-monitoring sensing group includes:
Outdoor temperature sensor, for dry-bulb temperature outside sensing chamber；
Outside humidity sensor, for detecting outside relative humidity；
Solar radiation sensor, for solar radiation outside sensing chamber；
The outdoor temperature sensor, outside humidity sensor and solar radiation sensing are all set in ground.
Further, indoor monitoring sensing group includes:
Indoor temperature transmitter, for detecting indoor dry-bulb temperature；
Indoor humidity sensor, for detecting indoor relative humidity；
Passenger flow counter, for carrying out real-time statistics to occupancy.
Further, the passenger flow counter is the infrared passenger flow counter for being set to station passageway two sides.
Further, the outlet terminal includes water cooler, air-conditioning box and blower.
Further, the central processing module of above-mentioned environmental perception module is 8 CMOS microcontrollers, it specifically can be with
It is AT89S52 single-chip microcontroller.
Further, the neural network load prediction module includes BP neural network PID controller.
The present invention is described further with reference to the accompanying drawings and detailed description.The additional aspect of the present invention and excellent
Point will be set forth in part in the description, and partially will become apparent from the description below.Or practice through the invention
It solves.
Detailed description of the invention
The attached drawing for constituting a part of the invention is used to assist the understanding of the present invention, content provided in attached drawing and its
Related explanation can be used for explaining the present invention in the present invention, but not constitute an undue limitation on the present invention.In the accompanying drawings:
Fig. 1 is the schematic diagram for illustrating subway underground station public area load composition；
Fig. 2 is that different node in hidden layer correspond to MSE value schematic diagram；
Fig. 3 is transmission function when being ambipolar S function neural network prediction value and actual value fit solution line chart；
Fig. 4 is that different learning rates correspond to MSE value line chart；
Fig. 5 is the line chart of predicted value after change number；
Fig. 6 is the line chart of actual value after change number；
Fig. 7 is predicted value and actual value relative error line chart after change number；
Fig. 8 is that each burst error corresponds to CVRMSE value line chart when passenger flow changes；
Fig. 9 is predicted value and actual value fit solution figure after change Outdoor Air Parameters；
Each burst error corresponds to CVRMSE value line chart when Figure 10 is Changes in weather；
Figure 11 is the schematic diagram of the neural network used in present embodiment；
Figure 12 is the schematic diagram of this subway underground station Load Prediction method；
Figure 13 is a kind of schematic diagram of context aware systems for subway station air-conditioning system of the invention.
Specific embodiment
Clear, complete explanation is carried out to the present invention with reference to the accompanying drawing.Those of ordinary skill in the art are being based on these
The present invention will be realized in the case where explanation.Before in conjunction with attached drawing 1-13, the present invention will be described, need to particularly point out
It is:
The technical solution provided in each section including following the description and technical characteristic in the present invention are not rushing
In the case where prominent, these technical solutions and technical characteristic be can be combined with each other.
In addition, the embodiment of the present invention being related in following the description is generally only the embodiment of a branch of the invention, and
The embodiment being not all of.Therefore, based on the embodiments of the present invention, those of ordinary skill in the art are not making creativeness
Every other embodiment obtained, should fall within the scope of the present invention under the premise of labour.
About term in the present invention and unit.Term in description and claims of this specification and related part
" comprising " and its any deformation, it is intended that cover and non-exclusive include.
Related content of the invention is illustrated above.Those of ordinary skill in the art are in the feelings illustrated based on these
The present invention will be realized under condition.Based on above content of the invention, those of ordinary skill in the art are not making creativeness
Every other embodiment obtained, should fall within the scope of the present invention under the premise of labour.
Subway underground station Load Prediction method, including following operating procedure:
S1. the environmental parameter of subway underground station is selected and is inputted, wherein variance point is carried out to environmental parameter
Analysis chooses parameter influential on air conditioning energy consumption using orthogonal experiment and method of analysis of variance；
S2. environmental parameter and air conditioning energy consumption mapping relations are established according to the environmental parameter of input；
S3., Parameters of Neural Network Structure and operating parameter are set, neural network is constructed；The neural network is feed forward neural
The Multi-layered Feedforward Networks of error back propagation in network；Feedback Neural Network is always deposited in actual neural fusion
In the transmission delay of information, these delays have an impact to the characteristic of neural network, and this Multi-layered Feedforward Networks is with extremely strong non-
Linear Mapping ability, and less calculation resources are occupied compared to feedback network；
S4. change environmental parameter, load is predicted using neural network, obtains prediction result；
S5. the prediction is tied using regression sum of square and the ratio of total sum of squares of deviations, the root-mean-square error coefficient of variation
Fruit is evaluated.
Such as Figure 13, present embodiment uses a kind of context aware systems for subway station air-conditioning system, comprising:
Environmental perception module 1, for monitoring and obtaining ambient data；
Neural network load prediction module 2 is communicated to connect with environmental perception module 1, right for receiving the environmental data
Real-time load is predicted that refrigeration duty value is predicted in output；
Main control module 3 is communicated to connect with neural network load prediction module 2, for cold negative by receiving the prediction
Charge values control the outlet terminal of cooling capacity, air quantity.Here STC89C51 single-chip microcontroller specifically can be used in main control module 3, the master control
Molding block 3 is provided with communication module, can specifically use SIM800C communication module.
The neural network load prediction module 2 uses above-mentioned subway underground station Load Prediction method pair
Subway underground station air conditioner load is predicted.
The environmental perception module 1 is provided with
Outdoor-monitoring sensing group, which is set to outside subway station and environmental perception module 1 communicates to connect,
The outdoor-monitoring sensing group is for detecting subway station external environment；
Indoor monitoring sensing group, the indoor monitoring sensing group are set in subway station and communicate to connect with environmental perception module 1,
Indoor monitoring sensing group is for detecting environment in subway station.
The outdoor-monitoring sensing group includes:
Outdoor temperature sensor 101, for dry-bulb temperature outside sensing chamber；
Outside humidity sensor 102, for detecting outside relative humidity；
Solar radiation sensor 103, for solar radiation outside sensing chamber；
The outdoor temperature sensor 101, outside humidity sensor 102 and solar radiation sensor 103 are all set in ground
Face.
Indoor monitoring sensing group includes:
Indoor temperature transmitter 104, for detecting indoor dry-bulb temperature；
Indoor humidity sensor 105, for detecting indoor relative humidity；
Passenger flow counter 106, for carrying out real-time statistics to occupancy.
The concrete model of above-mentioned temperature sensor can use DS-18B20 digital temperature sensor.Above-mentioned humidity passes
Sensor can use HTU21D numeral output temperature humidity sensor module.
The passenger flow counter 106 is the infrared passenger flow counter 106 for being set to station passageway two sides.
The outlet terminal includes water cooler 301, air-conditioning box 302 and blower 303.
The environmental perception module 1 includes central processing module, signal acquisition module and A/D conversion module.Environment sensing
The central processing module of module 1 is 8 CMOS microcontrollers, the specific can be that AT89S52 single-chip microcontroller.
The neural network load prediction module 2 includes BP neural network PID controller.
When setting, the above-mentioned outdoor sensor of ground configuration AT STATION of environmental perception module 1 monitors outdoor dry bulb temperature respectively
Degree, outside relative humidity, outdoor solar radiation；In subway concourse placement sensor, the indoor dry-bulb temperature of monitoring, interior are relatively wet respectively
Degree, number.Each area monitoring system reads according to set time interval and calculates each area survey parameter value.
Neural network load prediction module 2 receives real-time parameter as input parameter, with neural network control theory pair
Real-time load is predicted that refrigeration duty value is predicted in output.
Main control module 3 receives prediction refrigeration duty value, by set control strategy, adjusts water cooler refrigerating capacity, combined type
Air-conditioning box returns/air draft fan delivery.
Concrete example explanation is carried out to local iron underground station Load Prediction method below, this method can be by
The following steps carry out:
1) subway underground station public domain air conditioner load composition and energy consumption factor are determined；
Such as Fig. 1, public domain air conditioner load is mainly made of eight parts: human-body radiating dissipates humidity load, and building enclosure dissipates
Heat dissipates humidity load, lighting load, new wind load, entrance air penetration load, station public area equipment heating load, shielding
Door leak out, thermally conductive load.Outdoor environment is closely bound up with hourly load, wherein outdoor dry-bulb temperature, outside relative humidity, wind-force
Size, Cloud amount and solar radiation etc. can all influence refrigeration duty.Passenger flow variation will cause fresh air volume and personnel's load in subway concourse
Variation also has larger impact to the variation of station total load.
2) parameter having a significant impact to air conditioning energy consumption is chosen using orthogonal experiment and method of analysis of variance, establishes ring
Border parameter and air conditioning energy consumption mapping relations；
A) test method design
Orthogonal experiment is used in the test of air conditioning energy consumption influence factor, it is assumed that 3 external environmental factor (outdoor airs
Temperature, outdoor air relative humidity, solar radiation), 3 Indoor Thermals disturb parameter (passenger flow, equipment heating amount, light calorific value) etc.
6 factors are mutually indepedent, are independent of each other.Choose 3 levels respectively in each influence factor, influence factor and horizontal value are shown in Table
1.After the completion of test, random error must be estimated when for statistical analysis to test result, therefore a reserved column blank column is as mistake
The results change of poor item, the column will be entirely from stochastic effects.
1 influence factor of table and horizontal value
Select L18 (37) type orthogonal arrage, carry out 18 groups of tests altogether.
Orthogonal experiments are shown in Table 2.
2 orthogonal experiments of table
A) test result analysis
Common extremum difference analysis or method of analysis of variance analyzed for orthogonal experiments, extremum difference analysis calculation amount is smaller,
The difference for being easily understood, but cannot distinguish between test result corresponding to each level under same factor is that horizontal change causes or try
Error is tested to cause.Method of analysis of variance can difference and error between test result caused by the variation factor level fluctuation institute
Difference between caused test result distinguishes, and can provide reliable quantity survey.Using method of analysis of variance to just in this example
Hand over test result analysis.Conclusion: outside air temperature, passenger flow variation influence all more significantly total load, but its influence degree
Size has larger difference.It sorts successively by the size of influence are as follows: passenger flow changes outside the room > (P=0.000713, P < 0.05)
Air themperature (P=0.000739, P < 0.05) > outdoor air relative humidity (P=0.00735, P < 0.05) > solar radiation
Intensity (P=0.04, P < 0.05).And equipment heating amount P=0.422, light calorific value P=0.087, it is all larger than 0.05, it is seen that
Equipment heating amount, light calorific value, which do not have total load, to be significantly affected.
The results of analysis of variance is shown in Table 3.
3 the results of analysis of variance of table
Dependent variable: total load
The side a.R=.996 (the adjustment side R=.980)
3) neural network relative parameters setting
A) training data is handled
Processing to train samples data includes data screening and normalized.Data screening refers to data
It is pre-processed, rejecting abnormalities data.When data are compared in the same time for certain data two groups of data adjacent thereto and front and back two days, phase
It is more than 150% to error and is considered as abnormal data.The point value after rejecting abnormalities data uses neighbouring authentic data interpolation generation
It replaces, to eliminate deviation.In addition, deleting corresponding data when night equipment is closed to improve neural metwork training speed.Data are returned
One change processing can be accelerated to train the convergence of network, in this method using the method for linear function conversion to training sample data into
Row normalization.Expression formula is as follows:
Y=(x-min value)/(max value-min value)
In formula: Y is the value after conversion, and x is the value before conversion, and max value and min value are respectively sample maximum
And minimum value.
B) node in hidden layer is arranged
The selection of hidden layer and its neuronal quantity has a major impact building neural network, the hidden layer foot of limited quantity
To solve non-linear and lag issues.The determination of hidden layer node number there is no ideal analytic expression or stringent mathematics side at present
Method generally takes following several determining methods:
(1) Kolmogorov be directed to BP neural network model, propose node in hidden layer generally can by 2n+1 (n be input
Node layer number) it chooses.This paper neural network model haves three layers altogether, and according to selected input parameter, input layer shares 16 sections
Point, hidden layer are set as 33 nodes.
(2) Jadid and Fairbairn proposes the upper bound algorithm of node in hidden layer: node in hidden layer is generally less than
Or it is equal to number of training/R* (input layer number+output layer number of nodes), R is generally in 5~10 section values.
(3) trial and error procedure, under identical neural network structure, by changing hidden layer neuron number to observe nerve net
The precision of prediction of network model, so that it is determined that optimal number.
According to existing empirical equation, this case primary election hidden layer neuron number is 30~50, n1=30, n2=50.It presses
Following steps determine hidden layer neuron number:
(1) n1=30 is taken, neural network when node in hidden layer is 30 is calculated, obtains the mean square error MSE1 of network,
It is the performance evaluation parameter for choosing number of nodes with mean square error MSE；
(2) n=(n1+n2)/2=40 is taken, training neural network obtains MSE2；
(3) if MSE1 < MSE2, n2=n is enabled, it is on the contrary then enable n1=n；
(4) if n1 < n2, return step (2), are recycled with this；Otherwise terminate examination to gather.
Above step is taken to carry out after trying to gather, the corresponding MSE variation of different number hidden layer neuron number is as shown in Figure 2.
Determine that hidden layer neuron number is 35.
C) selection of transmission function
Transmission function (excitation function) is to calculate output by input for neuron, multiple linear inputs can be converted to
Nonlinear relationship.The desirable arbitrary value of the input and output value of linear function, includes the following three types function:
(1) linear function: f (x)=k*x+c
(2) inclined-plane function:
(3) threshold function table:
Nonlinear function has:
(1) s type function: input value can use any value, and output valve codomain is (0,1)
Expression formula:
(2) bipolar s type function: input value can use any value, and output valve codomain is (- 1 ,+1)
Expression formula:
The main distinction of bipolar S type function and S type function is that the codomain of function, bipolar S type function codomain are (- 1,1),
S type function codomain is (0,1).BP algorithm requires transmission function that can lead, S type function and bipolar S type function be all it is guidable, therefore
Suitable in BP neural network.If the last layer of BP network is S type function, then the output of whole network is just limited in
In one lesser range (continuous quantity between 0~1)；If the last layer of BP network is purelin function, then whole
The output of a network can take arbitrary value.Since the air-conditioning refrigeration duty of prediction output may be 0, therefore output layer transmission function cannot
For s type function.First select training function tainlm, change different hidden layers, output layer transmission function, according to MSE situation and
Fit solution determines transmission function.
Fig. 3 hidden layer transmission function be tansig, output layer transmission function be purelin, from Fig. 3 analyze it is found that due to
There are 0 in true output, that is, needs to make Accurate Prediction to 0 value with trained neural network.When output layer transmission function
When for linear function, it is all unable to 0 value of Accurate Prediction.Therefore ambipolar S function is all selected to the transmission function of output layer, input layer.
D) selection of training function
Training function and learning function are to modify weight and threshold value based on error.Continue to change after completing primary training
Generation, until reaching the number of iterations or meeting precision.Levenberg-Marquardt (L-M) algorithm can in gauss-newton method and
It is smoothly reconciled between steepest descent method, the situation that iteration can not be calculated can be avoided the occurrence of, thus when reducing trained
Between.More various trained functions are it is found that when transmission function, node in hidden layer are identical, using Levenberg-Marquardt
When algorithm, MSE is minimum.
E) selection of learning rate
Learning rate is a hyper parameter in neural network setting.If learning rate is very low, training can become relatively reliable,
But optimization can expend longer time, because towards each step-length very little of loss function minimum value；If learning rate is very high,
Training result may never restrain, or even tend to diverging.It keeps other parameters constant, only changes learning rate, training network
MSE results change such as Fig. 4 difference learning rate corresponds to shown in MSE value, and combined training time and training precision consider, chooses lr=
0.01 is used as this neural network model learning rate.
4) prediction result is evaluated
A) passenger flow changes
It keeps light, machine utilization constant, changes station passenger flow, simulate the case where station passenger flow changes, use instruction
The neural network perfected is predicted (1~July 31 July, totally 26213 groups of data) prediction to station air-conditioning in July refrigeration duty
Value as shown in figure 5, actual value as shown in fig. 6, the relative error curve of predicted value and actual value as shown in fig. 7, by relative error
Curve is it is found that maximum relative error is no more than 9.8257%, and most predicted values and actual value relative error are within 5%.
When simulation passenger flow changes, the value of predicted value data bulk corresponding to the difference of actual value and its corresponding CVRMSE are as schemed
Shown in 8.The difference of 98% data actual value and actual value is between -5~10kw, and corresponding CVRMSE is smaller.
B) Changes in weather
Keep light, machine utilization, station passenger flow constant, change meteorologic parameter (is changed to In Chengdu weather number of files
According to), simulate the case where station outdoor weather conditions change.It is negative to station air-conditioning cold in July using trained neural network
Lotus predicted, whole month predicted value and actual value fitting effect and July 15 odd-numbered day fit solution it is as shown in Figure 9:
Known to the analysis of predicted load and actual value to 15 odd-numbered day of July: whole day fit solution is preferable, only in 17:00
When there is load peak, the difference of predicted value and actual value is 60.14988kW, relative error 7.847%, also engineering can
Receive in range.Whole month predicted value and actual value relative error are calculated it is found that maximum relative error is 11.675%, it is most
Predicted value and actual value relative error are within 8%.
Such as Figure 10, when Changes in weather, each burst error corresponds to CVRMSE value, although having only a few predicted value and practical value difference
Value is greater than 20kw, but the difference of a large amount of predicted values and actual value concentrates between -5~10kw, and CVRMSE value is respectively less than 30%.
When changing compared with passenger flow for corresponding prediction case, after outdoor environment changes, prediction of the neural network to refrigeration duty
Situation is slightly inadequate.Meet in variance analysis, outside air temperature and outdoor air relative humidity to air-conditioning cold loading effects compared with
Passenger flow variation influences significantly more feature on it.
Using this method, can make prediction to the Public Areas Associated with Metro Station domain refrigeration duty under a variety of such environmental effects,
And air-conditioner wind, water system are adjusted accordingly, so that cooling capacity and air quantity are supplied by public domain workload demand in standing, to realize station sky
Adjusting system is energy saving by a relatively large margin.
Related content of the invention is illustrated above.Those of ordinary skill in the art are in the feelings illustrated based on these
The present invention will be realized under condition.Based on above content of the invention, those of ordinary skill in the art are not making creativeness
Every other embodiment obtained, all should belong to protection of the present invention under the premise of labour.
Claims (10)
1. subway underground station Load Prediction method, which is characterized in that including following operating procedure:
1) environmental parameter of subway underground station is selected and is inputted, wherein variance analysis is carried out to environmental parameter, is utilized
Orthogonal experiment and method of analysis of variance choose parameter influential on air conditioning energy consumption；
2) environmental parameter and air conditioning energy consumption mapping relations are established according to the environmental parameter of input；
3) Parameters of Neural Network Structure and operating parameter are set, neural network is constructed；The neural network is feedforward neural network
In error back propagation Multi-layered Feedforward Networks；
4) change environmental parameter, load is predicted using neural network, obtains prediction result；
5) prediction result is evaluated；Wherein, evaluation when using regression sum of square and total sum of squares of deviations ratio,
The square error coefficient of variation evaluates prediction result.
2. subway underground station Load Prediction method as described in claim 1, which is characterized in that establish environment ginseng
It is several with air conditioning energy consumption mapping relations the following steps are included:
1) according to subway underground station air conditioning energy consumption data, subway underground station air-Conditioning Load Calculation Method is determined；
2) according to air conditioning energy consumption and air conditioner load corresponding data, the environmental parameter for influencing air conditioner load is determined；
3) parameter influential on air conditioning energy consumption is chosen using orthogonal experiment and method of analysis of variance.
3. subway underground station Load Prediction method as described in claim 1, which is characterized in that establish environment ginseng
When several and air conditioning energy consumption mapping relations, energy consumption and environmental parameter mapping model are established according to track underground station energy consumption；
Wherein, subway underground station energy consumption includes: air-conditioning system cold source, power transmission & distribution, end-equipment energy consumption.
4. subway underground station Load Prediction method as claimed in claim 3, which is characterized in that the environment ginseng
Number includes: passenger flow variation, equipment light power size, outdoor dry-bulb temperature, outside relative humidity and solar radiation quantity.
5. subway underground station Load Prediction method as described in claim 1, which is characterized in that environmental parameter
Carry out variance analysis the following steps are included:
1) according to orthogonal experiment contrived experiment scheme；
2) null hypothesis is proposed；
3) selection check statistic；
4) observation and probability P value of test statistics are calculated；
5) it gives significance and makes decisions.
6. subway underground station Load Prediction method as described in claim 1, which is characterized in that join to environment
Before number carries out variance analysis, range analysis is carried out；Compare range analysis result and the results of analysis of variance.
7. subway underground station Load Prediction method as described in claim 1, which is characterized in that building nerve net
When network, the scale that size is influenced on air conditioning energy consumption is determined by the size of significant property coefficient.
8. subway underground station Load Prediction method as claimed in claim 7, which is characterized in that setting nerve net
Network structural parameters and operating parameter the following steps are included:
1) training data in neural network is set；
2) screening and normalized are carried out to training data；
3) neural network node in hidden layer is set；
4) all kinds of transmission function scope of applications are analyzed, choose transmission function in conjunction with prediction output valve；
5) learning rate is chosen.
9. subway underground station Load Prediction method as described in claim 1, which is characterized in that flat using returning
Just and carrying out evaluation to the prediction result with the ratio of total sum of squares of deviations, the root-mean-square error coefficient of variation includes following step
It is rapid:
1) change environmental parameter, subsequent time load is predicted using neural network；
2) R is used2Prediction result is evaluated with root-mean-square error coefficient of variation CVRMSE:
If R2>=0.8 and root-mean-square error coefficient of variation CVRMSE≤30% item think that prediction result is acceptable；
If R2< 0.8 or root-mean-square error coefficient of variation CVRMSE > 30% then think that prediction result is unacceptable.
10. air-conditioning system, which is characterized in that including empty using subway underground station as claimed in any one of claims 1 to 9 wherein
The neural network load prediction module that adjusting system load forecasting method predicts subway underground station air conditioner load.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910624024.7A CN110332647B (en) | 2019-07-11 | 2019-07-11 | Load prediction method for air conditioning system of underground station of subway and air conditioning system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910624024.7A CN110332647B (en) | 2019-07-11 | 2019-07-11 | Load prediction method for air conditioning system of underground station of subway and air conditioning system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110332647A true CN110332647A (en) | 2019-10-15 |
| CN110332647B CN110332647B (en) | 2021-02-02 |
Family
ID=68146145
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910624024.7A Active CN110332647B (en) | 2019-07-11 | 2019-07-11 | Load prediction method for air conditioning system of underground station of subway and air conditioning system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110332647B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111237988A (en) * | 2020-01-15 | 2020-06-05 | 北京天泽智云科技有限公司 | Control method and system for subway vehicle-mounted air conditioning unit |
| CN112116070A (en) * | 2020-09-07 | 2020-12-22 | 北方工业大学 | Subway station environmental parameter monitoring method and device |
| CN112365103A (en) * | 2020-12-11 | 2021-02-12 | 同济大学建筑设计研究院(集团)有限公司 | Method, device, equipment and medium for predicting short-term load of building in operation stage |
| CN112556103A (en) * | 2020-12-07 | 2021-03-26 | 上海电科智能系统股份有限公司 | Large system energy-saving control mode and system based on load prediction |
| CN112686470A (en) * | 2021-01-18 | 2021-04-20 | 国网河北省电力有限公司经济技术研究院 | Power grid saturation load prediction method and device and terminal equipment |
| CN112747413A (en) * | 2019-10-31 | 2021-05-04 | 北京国双科技有限公司 | Air conditioning system load prediction method and device |
| CN112818286A (en) * | 2021-03-31 | 2021-05-18 | 中车青岛四方车辆研究所有限公司 | Method and system for calculating energy consumption of air conditioning system of metro vehicle in cooling season |
| CN113111419A (en) * | 2021-04-16 | 2021-07-13 | 西安建筑科技大学 | Method and system for establishing and predicting air-conditioning load prediction model in office building |
| CN113375304A (en) * | 2021-06-01 | 2021-09-10 | 深圳地铁建设集团有限公司 | Pre-adjusting method for air conditioning system of subway station |
| CN114151915A (en) * | 2021-11-22 | 2022-03-08 | 宁波奥克斯电气股份有限公司 | Compressor frequency control method and device and air conditioner |
| CN115186892A (en) * | 2022-07-08 | 2022-10-14 | 建德市安泰科技咨询有限公司 | Energy-saving assessment method, device and system based on big data |
| CN116558082A (en) * | 2023-06-16 | 2023-08-08 | 广东新拓展建筑工程有限公司 | Energy-saving ventilation system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102997374A (en) * | 2012-12-31 | 2013-03-27 | 深圳市奥宇控制系统有限公司 | Method and device for forecasting air-conditioning load and air-conditioner |
| CN104008427A (en) * | 2014-05-16 | 2014-08-27 | 华南理工大学 | Central air conditioner cooling load prediction method based on BP neural network |
| CN106874581A (en) * | 2016-12-30 | 2017-06-20 | 浙江大学 | A kind of energy consumption of air conditioning system in buildings Forecasting Methodology based on BP neural network model |
| CN111237988A (en) * | 2020-01-15 | 2020-06-05 | 北京天泽智云科技有限公司 | Control method and system for subway vehicle-mounted air conditioning unit |
-
2019
- 2019-07-11 CN CN201910624024.7A patent/CN110332647B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102997374A (en) * | 2012-12-31 | 2013-03-27 | 深圳市奥宇控制系统有限公司 | Method and device for forecasting air-conditioning load and air-conditioner |
| CN104008427A (en) * | 2014-05-16 | 2014-08-27 | 华南理工大学 | Central air conditioner cooling load prediction method based on BP neural network |
| CN106874581A (en) * | 2016-12-30 | 2017-06-20 | 浙江大学 | A kind of energy consumption of air conditioning system in buildings Forecasting Methodology based on BP neural network model |
| CN111237988A (en) * | 2020-01-15 | 2020-06-05 | 北京天泽智云科技有限公司 | Control method and system for subway vehicle-mounted air conditioning unit |
Non-Patent Citations (4)
| Title |
|---|
| 刘建周: "《化工基础实验》", 30 December 2006, 中国矿业大学出版社 * |
| 唐敏: "《基于负荷预测的地铁通风空调系统节能优化方案》", 《都市快轨交通》 * |
| 施彦: "《神经网络设计方法与实例分析》", 30 December 2009, 北京邮电大学出版社 * |
| 黎东升: "《统计学原理》", 31 August 2007, 中国农业出版社 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112747413A (en) * | 2019-10-31 | 2021-05-04 | 北京国双科技有限公司 | Air conditioning system load prediction method and device |
| CN112747413B (en) * | 2019-10-31 | 2022-06-21 | 北京国双科技有限公司 | Air conditioning system load prediction method and device |
| CN111237988A (en) * | 2020-01-15 | 2020-06-05 | 北京天泽智云科技有限公司 | Control method and system for subway vehicle-mounted air conditioning unit |
| CN112116070A (en) * | 2020-09-07 | 2020-12-22 | 北方工业大学 | Subway station environmental parameter monitoring method and device |
| CN112116070B (en) * | 2020-09-07 | 2024-04-05 | 北方工业大学 | Subway station environment parameter monitoring method and device |
| CN112556103A (en) * | 2020-12-07 | 2021-03-26 | 上海电科智能系统股份有限公司 | Large system energy-saving control mode and system based on load prediction |
| CN112365103B (en) * | 2020-12-11 | 2023-10-13 | 同济大学建筑设计研究院(集团)有限公司 | Method, device, equipment and medium for predicting short-term load of building in operation stage |
| CN112365103A (en) * | 2020-12-11 | 2021-02-12 | 同济大学建筑设计研究院(集团)有限公司 | Method, device, equipment and medium for predicting short-term load of building in operation stage |
| CN112686470A (en) * | 2021-01-18 | 2021-04-20 | 国网河北省电力有限公司经济技术研究院 | Power grid saturation load prediction method and device and terminal equipment |
| CN112818286A (en) * | 2021-03-31 | 2021-05-18 | 中车青岛四方车辆研究所有限公司 | Method and system for calculating energy consumption of air conditioning system of metro vehicle in cooling season |
| CN113111419A (en) * | 2021-04-16 | 2021-07-13 | 西安建筑科技大学 | Method and system for establishing and predicting air-conditioning load prediction model in office building |
| CN113375304A (en) * | 2021-06-01 | 2021-09-10 | 深圳地铁建设集团有限公司 | Pre-adjusting method for air conditioning system of subway station |
| CN114151915B (en) * | 2021-11-22 | 2023-11-07 | 宁波奥克斯电气股份有限公司 | Compressor frequency control method and device and air conditioner |
| CN114151915A (en) * | 2021-11-22 | 2022-03-08 | 宁波奥克斯电气股份有限公司 | Compressor frequency control method and device and air conditioner |
| CN115186892B (en) * | 2022-07-08 | 2023-09-15 | 建德市安泰科技咨询有限公司 | Big data-based energy-saving evaluation method, device and system |
| CN115186892A (en) * | 2022-07-08 | 2022-10-14 | 建德市安泰科技咨询有限公司 | Energy-saving assessment method, device and system based on big data |
| CN116558082A (en) * | 2023-06-16 | 2023-08-08 | 广东新拓展建筑工程有限公司 | Energy-saving ventilation system |
| CN116558082B (en) * | 2023-06-16 | 2023-11-17 | 广东新拓展建筑工程有限公司 | Energy-saving ventilation system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110332647B (en) | 2021-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110332647A (en) | Subway underground station Load Prediction method and air-conditioning system | |
| US5479358A (en) | Urban energy system for controlling an energy plant supplying energy to a community | |
| CN102779228B (en) | Method and system for online prediction on cooling load of central air conditioner in marketplace buildings | |
| CN103912966A (en) | Optimal control method for ground source heat pump refrigerating system | |
| CN106991504B (en) | Building energy consumption prediction method and system based on subentry measurement time sequence and building | |
| CN100501608C (en) | Posterior control and multi-sensor infusion method and central conditioner energy-saving control device | |
| CN105627506A (en) | Building cooling load prediction method and device of air conditioning refrigerator | |
| CN109932896A (en) | A kind of control method and system of building energy consumption | |
| CN109164707A (en) | A kind of indoor environment negative-feedback regu- lation system based on artificial neural network algorithm | |
| CN113240184B (en) | Building space unit cold load prediction method and system based on federal learning | |
| CN109828539A (en) | A kind of big data based on PDCA system is for heat energy control platform system | |
| CN115220351B (en) | Intelligent energy-saving optimization control method for building air conditioning system based on cloud side end | |
| CN107479388A (en) | A kind of classroom central air-conditioning and the energy-saving control system and control method of illumination | |
| CN114383299A (en) | Central air-conditioning system operation strategy optimization method based on big data and dynamic simulation | |
| CN113778215A (en) | Method for realizing data center PUE prediction and consumption reduction strategy based on big data | |
| CN111539580A (en) | Multi-scheme optimization method for urban greening ecological technology integration application | |
| CN111623498A (en) | Automatic air conditioner temperature determination method and system | |
| CN114565167A (en) | Novel dynamic load prediction and regulation method for heating power inlet | |
| CN113341792A (en) | Building data acquisition method and system based on BIM | |
| CN113947261A (en) | Optimization decision support method for building energy conservation transformation | |
| CN111428979A (en) | Public building operation energy efficiency comprehensive evaluation model based on input-output method | |
| Fang et al. | Optimization of Air Conditioning Energy Consumption Based on Indoor Comfort Degree | |
| CN114200839B (en) | Intelligent office building energy consumption control model for dynamic monitoring of coupling environment behaviors | |
| Ma et al. | Research on Accurate Prediction of Operating Energy Consumption of Green Buildings Based on Improved Machine Learning | |
| KR20200027079A (en) | Apparatus and method for optimum control of air conditioner using artificial neural 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |