CN109961177A - A kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network - Google Patents
A kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network Download PDFInfo
- Publication number
- CN109961177A CN109961177A CN201910178747.9A CN201910178747A CN109961177A CN 109961177 A CN109961177 A CN 109961177A CN 201910178747 A CN201910178747 A CN 201910178747A CN 109961177 A CN109961177 A CN 109961177A
- Authority
- CN
- China
- Prior art keywords
- data
- shot
- energy consumption
- long term
- neural network
- 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
- 238000005265 energy consumption Methods 0.000 title claims abstract description 40
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 23
- 230000000306 recurrent effect Effects 0.000 title claims abstract description 23
- 230000007787 long-term memory Effects 0.000 title claims abstract description 18
- 238000004378 air conditioning Methods 0.000 claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 29
- 238000012360 testing method Methods 0.000 claims abstract description 23
- 230000007613 environmental effect Effects 0.000 claims abstract description 8
- 230000007774 longterm Effects 0.000 claims abstract description 8
- 230000002349 favourable effect Effects 0.000 claims abstract description 5
- 230000008569 process Effects 0.000 claims abstract description 4
- 238000001816 cooling Methods 0.000 claims description 20
- 238000007710 freezing Methods 0.000 claims description 12
- 230000008014 freezing Effects 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 11
- 239000008400 supply water Substances 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 8
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 238000013480 data collection Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 241001269238 Data Species 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 4
- 238000003646 Spearman's rank correlation coefficient Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims 1
- 230000000875 corresponding effect Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000015654 memory Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000003466 anti-cipated effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013501 data transformation Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/045—Combinations of 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
- G06Q10/04—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
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Strategic Management (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Air Conditioning Control Device (AREA)
Abstract
A kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network, the following steps are included: step 1, the water cooled central air conditioner data for multiple water cooled central air conditioner projects normal operation that acquisition favourable opposition energy company provides, corresponding environmental data;Step 2, data prediction is carried out to the water cooled central air conditioner data that are operated normally, corresponding environmental data, and multiple data sets is integrated, obtains integrated data set;Step 3 is trained data set, realize Energy consumption forecast for air conditioning, Recognition with Recurrent Neural Network is remembered using LSTM-RNN shot and long term, using pretreated data set and corresponding power consumption as the input of LSTM-RNN shot and long term memory Recognition with Recurrent Neural Network, after carrying out network training, final prediction model is obtained;Step 4, by test data input prediction model, obtain the power consumption values under air-conditioning current working.This invention simplifies model training processes, improve predictablity rate.
Description
Technical field
The present invention relates to a kind of general water cooled central air conditioner energy consumption predictions based on shot and long term memory Recognition with Recurrent Neural Network
Method.
Background technique
Building automation system (BAS) is to be integrated with technology of Internet of things, the system of the technologies such as control technology, network technology.
It by the monitoring of various equipment Comprehensive Automations to building (group) and management, for owner and user provide safety, comfortably,
The work and living environment of convenient and efficient, and whole system and one of the various equipment is made to be in optimal working condition.With this
Meanwhile in BAS system, a large amount of air-conditioning data such as temperature, humidity, flow, power etc. are all recorded in the database.Pass through
Mass data is analyzed air-conditioning system, is modeled, and air conditioning energy consumption can be better anticipated, and reflects air conditioning condition in building and carries out
Automatic management in real time, realizes automatic management and energy conservation, while improving the comfort of personnel in building.And present analysis modeling
Technology can only be mostly both for a project, and the part modeling in a building or even a region is predicted.This makes pre-
The difficulty of survey increases, and needs repeatedly modeling, then predicted to guarantee estimating for whole energy consumption.Use the data of multiple projects
To same type and the little central air-conditioning of gap carries out versatility comprehensive modeling, then carries out forecast assessment, energy to air conditioning energy consumption
A large amount of data are preferably utilized, modeling and forecasting job simplification is made.
Summary of the invention
In order to overcome, air-conditioned energy consumption modeling and forecasting step is many and diverse and the modeling limited deficiency in region, the present invention mention
It is higher for a kind of accuracy and have with certain versatility based on shot and long term memory Recognition with Recurrent Neural Network general water cooling center
Energy consumption forecast for air conditioning method.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network, including it is following
Step:
Step 1 obtains data set, and data set is provided by favourable opposition energy company, the water cooling center comprising multiple normal operations
The data of air-conditioning project, corresponding environmental data;
Step 2, data prediction, multiple project datas progress data prediction that step 1 is obtained, including data are clear
It washes, data transformation, feature selecting and data normalization;
Step 3 obtains final data collection, merges to the multiple data sets pre-processed, and upset and be split as training
Collection and test set obtain the data set eventually for training and test;
Step 4 establishes Energy consumption forecast for air conditioning model, and prediction model remembers Recognition with Recurrent Neural Network using LSTM-RNN shot and long term
Training, the training set that the final data that step 3 is obtained is concentrated are defeated as LSTM-RNN shot and long term memory Recognition with Recurrent Neural Network
Enter, carries out network training;Air-conditioning prediction model is assessed using the test set that final data is concentrated, obtains final air-conditioning
Energy consumption prediction model;
Step 5, by test data input prediction model, obtain the power consumption values under air-conditioning current working.
Further, in the step 2, data prediction step are as follows:
2.1) data cleansing: there are attribute missing or exception for data, but close on data and do not lack, use mean value interpolation method
To the missing or abnormal filling;It, can be direct if data have attribute to lack or abnormal, and close on data also and have more missing or exception
Batch deletes missing data;
2.2) data convert: constructing new attribute into property set, help Accurate Prediction air conditioning energy consumption.Time, date
Have larger impact to flow of the people, the cooling needs in groups of building also with flow of the people variation and change, affect air-conditioning indirectly
Total energy consumption, therefore tag along sort is stamped to month, working time, the addition as characteristic;It cools down supply water temperature and cools back
The difference of coolant-temperature gage can intuitively embody cooling tower heat-sinking capability, and related to total energy consumption;Backwater temperature difference and freezing are freezed for water temperature
The difference of difference can intuitively embody air conditioner refrigerating ability, and related to total energy consumption, therefore the cooling supply backwater temperature difference of addition, cooling supply back
Two characteristic items of water temperature difference;
2.3) feature selecting: if feature corresponding eigenvalue is continuous variable, pearson product is calculated away from related coefficient, sieve
Select the feature that related coefficient is greater than 10%;For ordinal data or be unsatisfactory for normal distribution hypothesis data at equal intervals, calculate
Spearman rank correlation coefficient filters out the feature that related coefficient is greater than 10%;
The Pearson correlation coefficient of described two variables calculates as follows:
The Spearman related coefficient is defined as the Pearson correlation coefficient between grade variables.Initial data
The descending position average in conceptual data according to it, is assigned a corresponding grade.
2.4) data normalization: merging data, and normalizes, and obtains multiple available project data collection;
Further, the step 2, in 3, obtain final training set and test set are as follows:
(1) input variable: wherein the input quantity includes at least outdoor temperature, outside humidity, cooling supply water temperature, cooling
Return water temperature, freezing supply water temperature, freezing return water temperature, general pipeline flow, month label and time tag;
(2) predictive variable: predictive variable is the air conditioning energy consumption under current working;
(3) data normalization: being normalized using mean variance, formula are as follows:
Wherein x' is the data after normalization, and x is input data, and μ is the mean value of data, and σ is the standard deviation of data;
(4) data merge: due to feature having the same in similar Water cooled air conditioners project data, and attribute in step 2
Most of useless attribute has been rejected when screening, can directly have been merged the attribute of same characteristic features;
(5) data are upset and are split: directly using the train_test_split method in sklearn packet, will merge
Data set input, random factor is set, and primary contract is set, is obtained eventually for trained training set and for test
Test set.
Further, in the step 3, LSTM-RNN shot and long term remembers Recognition with Recurrent Neural Network prediction model network structure such as
Under:
(1) input layer: first converting a two-dimensional array for the air-conditioning data of input, and abscissa is to have number of data n altogether,
Ordinate is that the data attribute attribute input_size of input enters back into activation primitive by a weight and bigoted conversion
Tanh, then a three-dimensional array is converted by data, as the input for entering hidden layer, x coordinate is to train batch into this
Number batch_size, y-coordinate are the number of data of a batch, that is, the primary data bulk into rnn training, z coordinate
For cell_size, that is, hide layer unit number;
(2) hidden layer: the data that will be handled well, and the hiding layer state input lstm unit that last time rnn is obtained, if
Training for the first time, then hidden layer state initialization is 0, by input gate, forgets door, out gate, obtains hidden layer output data
And hiding layer state at this time;
(3) output layer: the output that hidden layer is obtained carries out redeformation, and passes through output layer weight and bigoted conversion,
The air-conditioning power consumption finally entered;
(4) other: in the backpropagation of period, weight, it is bigoted, hide layer state adjustment all by loss loss function Lai
It determines, pass through construction loss function and carries out gradient decline processing, parameters are corrected in backpropagation.
Further, in the step 3, it is as follows that LSTM-RNN shot and long term remembers Recognition with Recurrent Neural Network prediction model parameters:
(1) Timesteps: its value represents the length of time series that RNN can be utilized, and is typically set to a cycle or half
A period since data set is random ordering, and needs the model with more versatility, and directlying adopt timesteps is 1;
(2) activation primitive: selecting tanh as activation primitive in RNN, it can the continuous real value of input " compressed " to-
Between 1 and 1, if it is very big negative, then output is exactly -1;If it is very big positive number, output is exactly 1;
(3) neuron selects: shot and long term memory (long-short term memory, LSTM) network is one in RNN
Kind specific neuron in hidden layer, by input gate, forgets door, and the structure of out gate controls retaining for information, can sieve
Important for result and unessential information in information flow is selected, keeps result more accurate;
(4) loss function: being modified mean square error (mean squared error, MSE), when calculating predicted value
predictedtWith actual value obsernertDifference square after, then divided by actual value observert, obtain square-error for reality
The ratio of actual value, then be averaged;So that model suitably focuses on the adjustment of big error rate data, but can be very good to exclude from
The interference of group's point, some particular values can also be looked after, and obtained result is average, can more preferably accomplish universality;
(5) other parameters: the hidden layer that model uses 10 layers of LSTM node to constitute, every layer has 50 nodes, adaptive ladder
Descent method is spent, exercise wheel number is 5000 wheels, batch_size 30, learning rate 0.00005.
Technical concept of the invention are as follows: in multiple air-conditioning project datas that favourable opposition energy science and technology company provides, and the external world
On the basis of environmental data, certain data prediction is carried out, Feature Selection obtains more higher with the air conditioning energy consumption degree of association
Characteristic, then multiple project datas are integrated, characteristic and energy consumption data are instructed by specific algorithm
Practice, generates energy consumption prediction model, predicted, worked as further according to air-conditioning data and environmental data data, and using the model
Air conditioning energy consumption under preceding operating condition.
Beneficial effects of the present invention are mainly manifested in: when handling air-conditioning data, in statistics
The methods of related coefficient excludes some extraneous features, and increases some correlated characteristics according to information such as times, and to data set into
Row integration obtains Water cooled air conditioners integrated data set;On this basis, carry out training pattern using LSTM-RNN, it is simple to a certain extent
Change model training process, improves predictablity rate.
Detailed description of the invention
Fig. 1 is the general water cooled central air conditioner energy consumption prediction side of the present invention that Recognition with Recurrent Neural Network is remembered based on shot and long term
The flow chart of method;
Fig. 2 is LSTM-RNN basic block diagram belonging to the present invention;
Fig. 3 is LSTM unit basic block diagram.
Specific embodiment
The invention will be further described below in conjunction with the accompanying drawings.
Referring to Fig.1~Fig. 3, a kind of general water cooled central air conditioner energy consumption prediction based on shot and long term memory Recognition with Recurrent Neural Network
Method the described method comprises the following steps:
Step 1, data set is obtained, data set is provided by favourable opposition energy company, the water cooling center comprising multiple normal operations
The data of air-conditioning project, corresponding environmental data;
Table 1 is the description to data set, and table 2 is description of the project 7 to air-conditioning data and environmental data:
Table 1
Table 2
Step 2, data prediction, multiple project datas that step 1 is obtained carry out data prediction;
The realization process of the data prediction are as follows:
2.1) data cleansing: there are attribute missing or exception for data, but close on data and do not lack, use mean value interpolation method
To the missing or abnormal filling;It, can be direct if data have attribute to lack or abnormal, and close on data also and have more missing or exception
Batch deletes missing data;As shown in table 3,14 days 18: 40 data of August in 2017 are whole missing, at this time upper and lower two numbers
According to complete, and gap is little, and mean value interpolation method can be used, and can also directly be deleted, and table 3 is 8 data of project
Table 3
2.2) data convert: constructing new attribute into property set, help Accurate Prediction air conditioning energy consumption.Time, date
Have larger impact to flow of the people, the cooling needs in groups of building also with flow of the people variation and change, affect air-conditioning indirectly
Total energy consumption, therefore tag along sort is stamped to month, working time, the addition as characteristic;It cools down supply water temperature and cools back
The difference of coolant-temperature gage can intuitively embody cooling tower heat-sinking capability, and related to total energy consumption;Backwater temperature difference and freezing are freezed for water temperature
The difference of difference can intuitively embody air conditioner refrigerating ability, and related to total energy consumption, therefore the cooling supply backwater temperature difference of addition, cooling supply back
Two characteristic items of water temperature difference;The increased attribute of project 8 is as shown in table 4:
Table 4
(3) feature selecting: if feature corresponding eigenvalue is continuous variable, pearson product is calculated away from related coefficient, screening
Related coefficient is greater than 10% feature out;For ordinal data or be unsatisfactory for normal distribution hypothesis data at equal intervals, calculate
Spearman rank correlation coefficient filters out the feature that related coefficient is greater than 10%;
The Pearson correlation coefficient of described two variables calculates as follows:
The Spearman related coefficient is defined as the Pearson correlation coefficient between grade variables.Initial data
The descending position average in conceptual data according to it, is assigned a corresponding grade.The each feature of table 5 and air-conditioning power consumption
Pearson correlation coefficient.
Table 5
(4) data normalization: merging data, and normalizes, and obtains multiple available project data collection;Due to
Freezing supply water temperature, freezing return water temperature, freezing supply backwater temperature difference, and cooling supply water temperature, cooling backwater temperature, cooling confession
Three features of backwater temperature difference can select maximum two features of Pearson correlation coefficient two-by-two in conjunction with obtaining third feature.
Step 3 obtains final data collection, merges to the multiple data sets pre-processed, and upset and be split as training
Collection and test set obtain the data set eventually for training and test;
(1) input variable: wherein the input quantity includes at least outdoor temperature, outside humidity, cooling supply water temperature, cooling
Return water temperature, freezing supply water temperature, freezing return water temperature, general pipeline flow, month label and time tag;
(2) predictive variable: predictive variable is the air conditioning energy consumption under current working;
(3) data normalization: being normalized using mean variance, formula are as follows:
Wherein x' is the data after normalization, and x is input data, and μ is the mean value of data, and σ is the standard deviation of data;
(4) data merge: due to feature having the same in similar Water cooled air conditioners project data, and attribute in step 2
Most of useless attribute has been rejected when screening, can directly have been merged the attribute of same characteristic features.
(5) data are upset and are split: directly using the train_test_split method in sklearn packet, will merge
Data set input, setting random factor is 1, carries out out-of-order arrangement to data, it is ensured that the diversity per a batch of data;And
Setting primary contract is 7:3, is obtained eventually for trained training set and for the test set of test.
Step 4 establishes Energy consumption forecast for air conditioning model, and prediction model remembers Recognition with Recurrent Neural Network using LSTM-RNN shot and long term
Training, the training set that the final data that step 3 is obtained is concentrated are defeated as LSTM-RNN shot and long term memory Recognition with Recurrent Neural Network
Enter, carries out network training;Air-conditioning prediction model is assessed using the test set that final data is concentrated, obtains final air-conditioning
Energy consumption prediction model;
According to Fig. 1, LSTM-RNNj structure and training step are as follows:
(1) input layer: first converting a two-dimensional array for the air-conditioning data of input, and abscissa is to have number of data n altogether,
Ordinate is that the data attribute attribute input_size of input enters back into activation primitive by a weight and bigoted conversion
Tanh, then a three-dimensional array is converted by data, as the input for entering hidden layer, x coordinate is to train batch into this
Number batch_size, y-coordinate are the number of data of a batch, that is, the primary data bulk into rnn training, z coordinate
For cell_size, that is, hide layer unit number;
(2) hidden layer: the data that will be handled well, and hiding layer state that last time rnn is obtained (if training for the first time,
Then hidden layer state initialization is 0) to input lstm unit, by input gate, forgets door, out gate, obtains hidden layer output number
Hiding layer state accordingly and at this time;
(3) output layer: the output that hidden layer is obtained carries out redeformation, and passes through output layer weight and bigoted conversion,
The air-conditioning power consumption finally entered;
(4) other: in the backpropagation of period, weight, it is bigoted, hide layer state adjustment all by loss loss function Lai
It determines, pass through construction loss function and carries out gradient decline processing, parameters are corrected in backpropagation.
The parameter of LSTM-RMM:
(1) Timesteps: its value represents the length of time series that RNN can be utilized, and is typically set to a cycle or half
A period since data set is random ordering, and needs the model with more versatility, and directlying adopt timesteps is 1;
(2) activation primitive: selecting tanh as activation primitive in RNN, it can the continuous real value of input " compressed " to-
Between 1 and 1, particularly, if it is very big negative, then output is exactly -1;If it is very big positive number, output is exactly
1;
(3) neuron selects: shot and long term memory (long-short term memory, LSTM) network is one in RNN
The specific neuron of kind, controls retaining for information in hidden layer by the structure of 3 doors, can filter out weight in information flow
To keep result more accurate with unessential information.As shown in figure 3, LSTM cellular construction is as follows:
It in time t, inputs as Xt, the preceding input of hidden layer is ht-1 generation, and indicating may be not too important in last data
Information.The preceding input of unit is that Ct-1 represents possible important information in last data.These three inputs pass through input
Door it, the processing for forgeing door ft and out gate ot, form unit output state Ct, and hidden layer exports ht and final output Yt.
Input gate:
Forget door:
Out gate:
Unit input:
Unit output:
Hidden layer output:
ht=ot*tanh(Ct)
WhereinBy xtIt is connected to the weight matrix of three doors and unit input,It is by ht-1It is connected to the weight matrix of three doors and the input of unit unit, bi, bf, bi, bCIt is three
The shift term of door and unit input, σ represent sigmoid functionTanh is exactly hyperbolic tangent function
(4) loss function: being modified mean square error (mean squared error, MSE), when calculating predicted value
predictedtWith actual value observertDifference square after, then divided by actual value observert, obtain square-error for reality
The ratio of actual value, then be averaged so that model suitably focuses on the adjustment of big error rate data, and can be very good to exclude from
The interference of group's point, some particular values can also be looked after, and obtained result is average, can more preferably accomplish universality;
(5) other parameters: the hidden layer that model uses 10 layers of LSTM node to constitute, every layer has 50 nodes, adaptive ladder
Descent method is spent, exercise wheel number is 5000 wheels, batch_size 30, learning rate 0.00005.
Step 5, by test data input prediction model, obtain the power consumption values under air-conditioning current working.Table 6 is of the invention
Trained and test result:
Data set | Training set error rate | Test set error rate |
Project 7 | 0.00852 | 0.01103 |
Project 8 | 0.00892 | 0.01633 |
Project 9 | 0.00824 | 0.01386 |
Integrated data set | 0.00974 | 0.01965 |
Table 6
For the final evaluation criterion that the present invention uses for mean error ME (Mean Error), formula is as follows:
Mean error refers in equal precision measurement, the arithmetic mean of instantaneous value of the random error of measured all measured values.
Claims (4)
1. a kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network, feature exist
In the described method comprises the following steps:
Step 1 obtains data set, and data set is provided by favourable opposition energy company, the water cooled central air conditioner comprising multiple normal operations
The data of project, corresponding environmental data;
Step 2, data prediction, multiple project datas that step 1 is obtained carry out data prediction, including data cleansing, number
According to transformation, feature selecting, data normalization;
Step 3, obtain final data collection, multiple data sets pre-process are merged, and upset be split as training set with
Test set obtains the data set eventually for training and test;
Step 4 establishes Energy consumption forecast for air conditioning model, and prediction model is using LSTM-RNN shot and long term memory Recognition with Recurrent Neural Network instruction
To practice, the training set that the final data that step 3 is obtained is concentrated remembers the input of Recognition with Recurrent Neural Network as LSTM-RNN shot and long term,
Carry out network training;Air-conditioning prediction model is assessed using the test set that final data is concentrated, obtains final air-conditioning energy
Consume prediction model;
Step 5, by test data input prediction model, obtain the power consumption values under air-conditioning current working.
2. a kind of general water cooled central air conditioner energy consumption based on shot and long term memory Recognition with Recurrent Neural Network according to claim 1
Prediction technique, it is characterised in that: in the step 2, data prediction step are as follows:
2.1) data cleansing: there are attribute missing or exception for data, but close on data and do not lack, using mean value interpolation method to this
Missing or abnormal filling;It, can direct batch if data have attribute to lack or abnormal, and close on data also and have more missing or exception
Delete missing data;
2.2) data convert: constructing new attribute into property set, help Accurate Prediction air conditioning energy consumption.Time, date are to people
Flow has larger impact, the cooling needs in groups of building also with flow of the people variation and change, affect air-conditioning total energy indirectly
Consumption, therefore tag along sort is stamped to month, working time, the addition as characteristic;It cools down supply water temperature and cools back water temperature
The difference of degree can intuitively embody cooling tower heat-sinking capability, and related to total energy consumption;Backwater temperature difference and freezing are freezed for water temperature difference
Difference can intuitively embody air conditioner refrigerating ability, and related to total energy consumption, therefore cooling supply backwater temperature difference, cooling is added for return water temperature
Poor two characteristic items;
2.3) feature selecting: if feature corresponding eigenvalue is continuous variable, pearson product is calculated away from related coefficient, is filtered out
Related coefficient is greater than 10% feature;For ordinal data or be unsatisfactory for normal distribution hypothesis data at equal intervals, calculate
Spearman rank correlation coefficient filters out the feature that related coefficient is greater than 10%;
2.4) data normalization: merging data, and normalizes, and obtains multiple available project data collection.
3. a kind of general water cooled central air conditioner based on shot and long term memory Recognition with Recurrent Neural Network according to claim 1 or 2
Energy consumption prediction technique, it is characterised in that: in the step 3, the step of obtaining final training set and test set is as follows:
3.1) input variable: wherein the input quantity includes at least outdoor temperature, outside humidity, cooling supply water temperature, cools back
Coolant-temperature gage, freezing supply water temperature, freezing return water temperature, general pipeline flow, month label and time tag;
3.2) predictive variable: predictive variable is the air conditioning energy consumption under current working;
3.3) data normalization: being normalized using mean variance, formula are as follows:
Wherein x' is the data after normalization, and x is input data, and μ is the mean value of data, and σ is the standard deviation of data;
3.4) data merge: due to feature having the same in similar Water cooled air conditioners project data, and attribute selection in step 2
When rejected most of useless attribute, directly the attribute of same characteristic features can be merged;
3.5) data are upset and are split: directly using the train_test_split method in sklearn packet, the number that will merge
It is inputted according to collection, random factor is set, and primary contract is arranged, is obtained eventually for trained training set and for the test of test
Collection.
4. a kind of general water cooled central air conditioner based on shot and long term memory Recognition with Recurrent Neural Network according to claim 1 or 2
Energy consumption prediction technique, it is characterised in that: in the step 4, LSTM-RNN shot and long term remembers Recognition with Recurrent Neural Network prediction model net
Network structure is as follows:
(1) Timesteps: its value represents the length of time series that RNN can be utilized, and is typically set to a cycle or half week
Phase since data set is random ordering, and needs the model with more versatility, and directlying adopt timesteps is 1;
(2) activation primitive: selecting tanh as activation primitive in RNN, it can be the continuous real value of input " compressed " to -1 and 1
Between, if it is very big negative, then output is exactly -1;If it is very big positive number, output is exactly 1;
(3) neuron selects: shot and long term memory LSTM network is the neuron of one of RNN, passes through 3 doors in hidden layer
Structure control retaining for information, can filter out important with unessential information in information flow;
(4) loss function: mean square error MSE is modified.As calculating predicted value predictedtAnd actual value
observertDifference square after, then divided by actual value observert, show that square-error for the ratio of actual value, then carries out
It is average;
(5) other parameters: the hidden layer that model uses 10 layers of LSTM node to constitute, every layer has 50 nodes, under self-adaption gradient
Drop method, exercise wheel number are 5000 wheels, batch_size 30, learning rate 0.00005.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910178747.9A CN109961177A (en) | 2019-03-11 | 2019-03-11 | A kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910178747.9A CN109961177A (en) | 2019-03-11 | 2019-03-11 | A kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109961177A true CN109961177A (en) | 2019-07-02 |
Family
ID=67024198
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910178747.9A Pending CN109961177A (en) | 2019-03-11 | 2019-03-11 | A kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109961177A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110866592A (en) * | 2019-10-28 | 2020-03-06 | 腾讯科技(深圳)有限公司 | Model training method and device, energy efficiency prediction method and device and storage medium |
CN111442476A (en) * | 2020-03-06 | 2020-07-24 | 财拓云计算(上海)有限公司 | Method for realizing energy-saving temperature control of data center by using deep migration learning |
CN111507752A (en) * | 2020-03-26 | 2020-08-07 | 杭州电子科技大学 | Abnormal user behavior identification method based on bidirectional long-short term memory network |
CN111797980A (en) * | 2020-07-20 | 2020-10-20 | 房健 | Self-adaptive learning method for personalized floor heating use habits |
CN111859625A (en) * | 2020-06-28 | 2020-10-30 | 五邑大学 | Energy-saving control method and device based on big data and storage medium |
CN112183826A (en) * | 2020-09-15 | 2021-01-05 | 湖北大学 | Building energy consumption prediction method based on deep cascade generation countermeasure network and related product |
CN112963946A (en) * | 2021-02-26 | 2021-06-15 | 南京邮电大学 | Heating, ventilating and air conditioning system control method and device for shared office area |
CN113432247A (en) * | 2021-05-20 | 2021-09-24 | 中南大学 | Water chilling unit energy consumption prediction method and system based on graph neural network and storage medium |
CN113486433A (en) * | 2020-12-31 | 2021-10-08 | 上海东方低碳科技产业股份有限公司 | Method for calculating energy consumption shortage number of net zero energy consumption building and filling system |
CN113835341A (en) * | 2021-09-18 | 2021-12-24 | 中邮科通信技术股份有限公司 | Energy consumption analysis and diagnosis method based on intelligent building AI |
WO2022052547A1 (en) * | 2020-09-14 | 2022-03-17 | 青岛海信日立空调系统有限公司 | Method for predicting energy efficiency of air-conditioning system, and air-conditioning system |
WO2022112895A1 (en) * | 2020-11-30 | 2022-06-02 | International Business Machines Corporation | Automated deep learning architecture selection for time series prediction with user interaction |
CN116757534A (en) * | 2023-06-15 | 2023-09-15 | 中国标准化研究院 | Intelligent refrigerator reliability analysis method based on neural training network |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160146493A1 (en) * | 2014-11-26 | 2016-05-26 | International Business Machines Corporation | Building thermal control techniques |
CN108954680A (en) * | 2018-07-13 | 2018-12-07 | 电子科技大学 | A kind of air-conditioning energy consumption prediction technique based on operation data |
CN109102103A (en) * | 2018-06-26 | 2018-12-28 | 上海鲁班软件股份有限公司 | A kind of multi-class energy consumption prediction technique based on Recognition with Recurrent Neural Network |
-
2019
- 2019-03-11 CN CN201910178747.9A patent/CN109961177A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160146493A1 (en) * | 2014-11-26 | 2016-05-26 | International Business Machines Corporation | Building thermal control techniques |
CN109102103A (en) * | 2018-06-26 | 2018-12-28 | 上海鲁班软件股份有限公司 | A kind of multi-class energy consumption prediction technique based on Recognition with Recurrent Neural Network |
CN108954680A (en) * | 2018-07-13 | 2018-12-07 | 电子科技大学 | A kind of air-conditioning energy consumption prediction technique based on operation data |
Non-Patent Citations (2)
Title |
---|
廖文强,王江宇 等: "基于长短期记忆神经网络的暖通空调系统能耗预测", 《制冷技术》 * |
易丽蓉,王绍宇 等: "基于多变量 LSTM 的工业传感器时序数据预测", 《智能计算机与应用》 * |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110866592A (en) * | 2019-10-28 | 2020-03-06 | 腾讯科技(深圳)有限公司 | Model training method and device, energy efficiency prediction method and device and storage medium |
CN110866592B (en) * | 2019-10-28 | 2023-12-29 | 腾讯科技(深圳)有限公司 | Model training method, device, energy efficiency prediction method, device and storage medium |
CN111442476A (en) * | 2020-03-06 | 2020-07-24 | 财拓云计算(上海)有限公司 | Method for realizing energy-saving temperature control of data center by using deep migration learning |
CN111507752A (en) * | 2020-03-26 | 2020-08-07 | 杭州电子科技大学 | Abnormal user behavior identification method based on bidirectional long-short term memory network |
CN111859625A (en) * | 2020-06-28 | 2020-10-30 | 五邑大学 | Energy-saving control method and device based on big data and storage medium |
CN111797980A (en) * | 2020-07-20 | 2020-10-20 | 房健 | Self-adaptive learning method for personalized floor heating use habits |
WO2022052547A1 (en) * | 2020-09-14 | 2022-03-17 | 青岛海信日立空调系统有限公司 | Method for predicting energy efficiency of air-conditioning system, and air-conditioning system |
CN112183826A (en) * | 2020-09-15 | 2021-01-05 | 湖北大学 | Building energy consumption prediction method based on deep cascade generation countermeasure network and related product |
CN112183826B (en) * | 2020-09-15 | 2023-08-01 | 湖北大学 | Building energy consumption prediction method based on deep cascade generation countermeasure network and related products |
WO2022112895A1 (en) * | 2020-11-30 | 2022-06-02 | International Business Machines Corporation | Automated deep learning architecture selection for time series prediction with user interaction |
GB2616770A (en) * | 2020-11-30 | 2023-09-20 | Ibm | Automated deep learning architecture selection for time series prediction with user interaction |
CN113486433A (en) * | 2020-12-31 | 2021-10-08 | 上海东方低碳科技产业股份有限公司 | Method for calculating energy consumption shortage number of net zero energy consumption building and filling system |
CN112963946B (en) * | 2021-02-26 | 2022-06-17 | 南京邮电大学 | Heating, ventilating and air conditioning system control method and device for shared office area |
CN112963946A (en) * | 2021-02-26 | 2021-06-15 | 南京邮电大学 | Heating, ventilating and air conditioning system control method and device for shared office area |
CN113432247A (en) * | 2021-05-20 | 2021-09-24 | 中南大学 | Water chilling unit energy consumption prediction method and system based on graph neural network and storage medium |
CN113432247B (en) * | 2021-05-20 | 2022-04-26 | 中南大学 | Water chilling unit energy consumption prediction method and system based on graph neural network and storage medium |
CN113835341A (en) * | 2021-09-18 | 2021-12-24 | 中邮科通信技术股份有限公司 | Energy consumption analysis and diagnosis method based on intelligent building AI |
CN113835341B (en) * | 2021-09-18 | 2024-05-17 | 中邮科通信技术股份有限公司 | Intelligent building AI-based energy consumption analysis and diagnosis method |
CN116757534A (en) * | 2023-06-15 | 2023-09-15 | 中国标准化研究院 | Intelligent refrigerator reliability analysis method based on neural training network |
CN116757534B (en) * | 2023-06-15 | 2024-03-15 | 中国标准化研究院 | Intelligent refrigerator reliability analysis method based on neural training network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109961177A (en) | A kind of general water cooled central air conditioner energy consumption prediction technique based on shot and long term memory Recognition with Recurrent Neural Network | |
CN109959122A (en) | A kind of Energy consumption forecast for air conditioning method based on shot and long term memory Recognition with Recurrent Neural Network | |
CN109959123A (en) | A kind of energy-saving method for air conditioner based on genetic algorithm and shot and long term memory Recognition with Recurrent Neural Network | |
Cai et al. | Predicting the energy consumption of residential buildings for regional electricity supply-side and demand-side management | |
Sha et al. | Overview of computational intelligence for building energy system design | |
US11755976B2 (en) | Method for predicting oxygen load in iron and steel enterprises based on production plan | |
CN106920006B (en) | Subway station air conditioning system energy consumption prediction method based on ISOA-LSSVM | |
Yi et al. | A neural network model forecasting for prediction of daily maximum ozone concentration in an industrialized urban area | |
Alcalá et al. | Fuzzy control of HVAC systems optimized by genetic algorithms | |
CN105320184B (en) | Building Indoor Environment intelligent monitor system | |
CN108053052B (en) | A kind of oil truck oil and gas leakage speed intelligent monitor system | |
CN114322199B (en) | Digital twinning-based ventilation system autonomous optimization operation regulation and control platform and method | |
CN113326651A (en) | Refrigerating station load and energy efficiency ratio dynamic modeling method based on T-S fuzzy model | |
CN112524751B (en) | Dynamic air conditioning system energy consumption prediction model construction and prediction method and device | |
CN104933629A (en) | Power user equipment evaluation method based on interval level analysis and interval entropy combination | |
CN115249094A (en) | Building energy efficiency management and optimization method based on big data | |
Song et al. | An indoor temperature prediction framework based on hierarchical attention gated recurrent unit model for energy efficient buildings | |
CN114611201B (en) | Multi-objective energy-saving optimization method and system for future climate buildings | |
CN113240184A (en) | Building space unit cold load prediction method and system based on federal learning | |
CN114200839A (en) | Office building energy consumption intelligent control model for dynamic monitoring of coupled environmental behaviors | |
CN107818340A (en) | Two-stage Air-conditioning Load Prediction method based on K value wavelet neural networks | |
CN107256439A (en) | Joint EEMD and neutral net short-term load forecasting method and system | |
Yu et al. | An active construction dynamic schedule management model: using the fuzzy earned value management and BP neural network | |
CN114418177A (en) | New product material distribution prediction method for digital twin workshop based on generation of countermeasure network | |
CN106403162A (en) | Local heat comfort control method, local heat comfort controller and control system |
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 |
Application publication date: 20190702 |
|
RJ01 | Rejection of invention patent application after publication |