CN105446139A - Construction energy consumption analysis method and construction energy consumption analysis system based on BP neural network - Google Patents

Abstract

The invention discloses a construction energy consumption analysis method and a construction energy consumption analysis system based on a BP neural network. The method is characterized in that input output vector quantity can be determined according to predetermined construction energy consumption type; a BP neural network model can be constructed according to the input output vector quantity; the established BP neutral network model can be trained by inputting a training sample; the energy consumption index data requiring the analysis can be input in the trained BP model for the processing and the analysis; and the data restoration of the output data of the neutral network can be carried out. The system comprises an energy saving potential quantification module, a data preprocessing module, a BP network analysis module, a data restoration module, a configuration management module, and a log recording module. The processing and the analysis of the construction energy consumption index data can be realized quickly and accurately, and therefore the unreasonable links during the construction energy use process can be acquired, and then the construction manager and even the common user lack of professional knowledge can be aware of the construction energy consumption current condition and the energy saving improvement measures.

Description

Based on analyzing methods of architectural energy saving and the system of BP neural network

Technical field

The present invention relates to the detection and classification of building energy consumption data, belong to Pattern classification techniques field category, particularly a kind of analyzing methods of architectural energy saving based on BP neural network and system.

Background technology

Building Energy Analysis is the focus that scholars pays close attention to and studies always, by correct effective energy consumption calculation and analysis, can reduce the energy consumption of each link in buildings life cycle.Current adoptable building energy consumption analytical approach is a lot, according to the mathematical model of institute's foundation, computing method can be divided into two large classes: a class is the static energy consumption analysis method be based upon in steady heat transfer theoretical foundation, another kind of is the dynamic energy consumption simulation be based upon in unsteady heat transfer theoretical foundation.

(1) static energy consumption analysis method: the ultimate principle of static energy consumption analysis method is calculated by steady state heat transfer theory by the heat consumption of each ten days in heating period or heating period, each moon, mainly degree of comprising BIN method, number of days method, equivalent peak value hourage method etc.The advantage of this method is fairly simple, be easy to hand computation, but precision is slightly poor, accuracy rate is lower, all analog computations are all that (meteorological condition, indoor temperature etc.) calculate under the ideal parameters of setting, can not reflect the status of energy consumption under building actual motion state;

(2) dynamic simulation method: theoretical based on unsteady heat transfer, computer technology is mainly utilized to carry out performance analysis and the dynamic similation of system, the mathematical model system that this method because require is set up and accurately, and be confined to linear and time invariant system.The method has higher requirement to professional domain knowledge in addition, mostly design for professional, and the actual user built or owner, estate management etc. are not possessed to the personnel of building energy system relevant professional knowledge, then cannot have a basic understanding to the house status of energy consumption of oneself in this way, therefore practical application is greatly limited.

Above-mentioned two kinds of traditional analyzing methods of architectural energy savings, due to the limitation of himself, do not make full use of existing energy consumption data, cannot accomplish the intellectual analysis to energy consumption, very limited to the support of energy-conservation decision-making.

Summary of the invention

The object of the invention is to overcome the shortcoming of prior art and deficiency, provide a kind of analyzing methods of architectural energy saving based on BP neural network, this method is on the basis analyzed building energy consumption achievement data and influence factor thereof and study, the Knowledge Discovery powerful by BP neural network learning technology and data analysis capabilities, realize carrying out Treatment Analysis to building energy consumption achievement data quickly and accurately, thus the unreasonable link obtained in energy for building process, building energy consumption present situation is understood by the domestic consumer helping building manager and even do not possess professional knowledge, specify energy-conservation Improving Measurements.

Another object of the present invention is to, a kind of Building Energy Analysis system based on BP neural network is provided.

In order to arrive above-mentioned first object, the present invention by the following technical solutions:

The present invention is based on the analyzing methods of architectural energy saving of BP neural network, it is characterized in that, comprise the following steps:

(1) according to appointment building energy consumption type determination input and output vector;

(2) according to input and output vector structure BP neural network model;

(3) input training sample to train the BP neural network model established;

(4) the energy consumption index data that actual needs carries out analyzing are input to trained BP model and carry out Treatment Analysis;

(5) data convert is carried out to the output data of neural network.

As preferred technical scheme, described step (1) comprises the steps: further

(1-1) building energy consumption type is divided into 3 classes, is respectively power consumption, water consumption and cold consumption adjusting system, the previous day day building energy consumption is measured in analysis, obtain corresponding measurement data as original building energy consumption data;

(1-2) input vector of building energy consumption achievement data value as model is calculated according to the computing formula of original building energy consumption data and green building assessment index,

(1-3) output vector of building energy consumption weak link as model of corresponding building energy consumption type is determined, the output variable of 3 energy consumption types;

(1-4) binary conversion treatment is carried out to the input vector determined;

(1-5) binary coding is carried out to the output vector determined, with energy consumption weak link number in building energy consumption type for binary coding length, thus output vector is adopted to the coded system of " getting 1 in n "; Wherein, n is the length of coding, i.e. the total number of energy consumption weak link of corresponding energy consumption type, and when certain building weak link is determined, the coding of its corresponding position is set to 1, and remaining n-1 position coding is all 0.

As preferred technical scheme, in step (1-2), the input variable of 3 class separate energy consumptions is as follows:

(1-2-1) power consumption part: comprise unit area air conditioner electric consumption, unit area illumination and socket electrisity consumption, unit area open air exhaust dynamo-electric consumption, unit area particular electrical consumption, also comprising inoperative period and working hour throws light on and the ratio of the ratio of the ratio of the ratio of socket power consumption, inoperative period and working hour air-conditioning power consumption, inoperative period and working hour room open air exhaust power consumption, inoperative period and working hour specific installation power consumption;

(1-2-2) water consumption part: the ratio of unit area water consumption, inoperative period and working hour water consumption;

(1-2-3) cold consumption adjusting system part: the cold consumption of unit area, comprise air-conditioning system Energy Efficiency Ratio, refrigeration system Energy Efficiency Ratio, handpiece Water Chilling Units operational efficiency, chilled water Transfer coefficient, chilled water Transfer coefficient, air conditioning terminal Energy Efficiency Ratio, cooling pump efficiency, efficiency of cooling tower, refrigerating water pump efficiency, water system supply backwater temperature difference and water system return water temperature consistance.

As preferred technical scheme, step (1-4) comprises the steps: further

(1-4-1) for each input pointer data of input vector, its index value is I _a, obtain the standard index value I that it is relevant _n;

(1-4-2) the energy-saving potential D of this input pointer is calculated according to formula (1);

$D=\frac{\mathrm{\Δ}I}{I_n}=\frac{I_a-I_n}{I_n}---\left(1\right)$

In formula: I _afor building the actual desired value calculated; I _nfor public building reference index value;

(1-4-3) D that step (1-4-2) calculates is carried out value by formula (2) rule, compare by energy-saving potential D threshold values T, if D>=T, binaryzation value f (x)=1 of this then input pointer; If D<T, then binaryzation value f (x)=0 of this input pointer;

$f\left(x_i\right)=\left\{\begin{array}{cc}0& D\left(x_i\right)\&le;T\\ 1& D\left(x_i\right)>T\end{array}\right.---\left(2\right)$

In formula: x _ifor the index parameter of i-th before binaryzation; F (x _i) be x _ivalue after binaryzation; D _ibe i-th index parameter x _ienergy-saving potential calculated value; T is energy-saving potential threshold values.

As preferred technical scheme, described step (2) comprises the steps: further

(2-1) according to the neuronal quantity of input vector, output vector determination input layer and output layer;

(2-2) determine BP neural network hidden layer neuron quantity according to the neuronal quantity of input layer and output layer, wherein, the excitation function of BP neural network is sigmoid function:

$f\left(x\right)=\frac{1}{1+e^{-x}}---\left(3\right)$

Described step (3), comprises the steps: further

The training parameter of setting BP neural network, and it is trained, wherein training parameter comprises: maximum frequency of training, anticipation error, momentum term numerical value and learning rate.

As preferred technical scheme, described step (5) comprises the steps: further

The output vector of BP neural network is traveled through, 1 is set to by maximum for numerical value one, remaining is set to 0, and the binary coding namely obtaining output vector exports, then according to coding and the corresponding relation analyzing conclusion content, binary coding is reduced to the understandable Word message of user; After obtaining the analysis conclusion fed back, user gets final product the place of clear and definite building energy consumption problem, and purposively can carry out emphasis investigation and safeguard, problem wherein comprises replacing or maintenance of equipment, the maintenance of reinforcement equipment and the operation reserve of adjustment equipment.

The present invention also provides a kind of Building Energy Analysis system based on BP neural network, and this system comprises energy-saving potential quantization modules, data preprocessing module, BP nework analysis module, data restoring module, Configuration Manager and logger module;

Described energy-saving potential quantization modules, for calculating the energy-saving potential of each building energy consumption achievement data;

Described data preprocessing module, for being responsible for the data basis providing analyzing methods of architectural energy saving;

Described BP nework analysis module, for being responsible for the core analysis flow elements realizing analyzing methods of architectural energy saving, this module is divided into knowledge acquisition submodule and energy consumption analysis submodule; Corresponding respectively BP network is used to carry out two stages analyzed: training stage and analysis phase;

Described data convert device module, for resolving the analysis conclusion through coding that BP neural network exports, convert the analysis conclusion through coding to user understandable Word message, output due to BP neural network is binary coding, so need to carry out content reduction to it, according to its scale-of-two output encoder, by the coding corresponding informance analyzed in database in conclusion table, the energy-conservation Improving Measurements analyzing conclusion and correspondence is combined into final analysis conclusion and feeds back to user;

Described Configuration Manager, for safeguarding the parameter configuration that responsible BP neural network builds, described parameter comprises the network number of plies, hidden layer neuron number, input and output item number, weight initialization scope, momentum arithmetic numerical value and learning rate initial value;

User operation behaviors all in described log pattern logging program operational process, the abnormal information of appearance and thread state, and be saved in relevant journal file.

As preferred technical scheme, described data preprocessing module comprises input data binaryzation submodule and exports data encoding submodule;

Described input data binaryzation submodule, for being responsible for, binary conversion treatment is carried out to the input item of BP neural network, each sample that this Module cycle traversal input amendment is concentrated, its energy-saving potential is calculated to each input item of sample, and this energy-saving potential and threshold values T are compared, if >T, then the binaryzation value of input item is 1; Otherwise get 0;

Described output data encoding submodule, for carrying out binary coding to output item, if there is this coding in the analysis conclusion table in database, directly taking out and carrying out using; If there is not this coding in database, then use after needing that unified binary coding is carried out to this alanysis conclusion.

As preferred technical scheme, described knowledge acquisition submodule was specially in the training acquisition stage:

According to the analysis rule in knowledge base building energy consumption desired value and Building Energy Analysis conclusion combined as training sample set pair BP neural network and train, allow its automatic acquisition knowledge and rule, each sample in described knowledge acquisition submodule circuit training sample set, input item is the vector of building index binaryzation, output item is the binary coding of energy consumption analysis conclusion, repeat this operation until training error is less than anticipation error or reaches maximum frequency of training, after training terminates, namely this BP neural network gets the rule knowledge required for the analysis phase, its form of expression is that the weights of each neuron node in BP network are with threshold values,

Described energy consumption analysis submodule was specially in the energy consumption analysis stage:

Being combined into vector after carrying out binaryzation by needing the achievement data carrying out analyzing and processing is input in trained network, the energy consumption weak link of output to building according to network is analyzed, BP neural network after training can be analyzed input vector rapidly, and exports corresponding binary coding analysis conclusion.

As preferred technical scheme, the information of described logger module record comprises:

User operation log, to the Operation Log of database, comprising: increase, revise, delete and inquiry;

System cloud gray model abnormal information, comprising: the operation exception of database manipulation exception, file resource operation exception and other resources;

The Parameter transfer that foreground is called background service, comprising: the title of method, the parameter of method and numerical value.

Compared with prior art, tool has the following advantages and beneficial effect in the present invention:

(1) the present invention carries out computing by the achievement data of the power consumption part to building, water consumption part and cold consumption adjusting system part, analyze the energy consumption weak link that may exist in building, and obtain corresponding energy-conservation Improving advice in this, as basis, user can be effectively helped to understand the situation of energy for building, help user to carry out purposive investigation and maintenance to energy consumption link and equipment thereof, provide decision support to the daily administration of energy conservation of building and energy-conservation improvement.

(2) the present invention adopts BP neural network to analyze, quantitative information and qualitative information can be processed, each node adopts distributed variable-frequencypump, process data capability is strong, its analysis strategy adopting the knowledge representation mode of implicit expression to solve problem, not only can be avoided the conflict of analysis rule, and also improve the speed of analysis;

(3) the present invention passes through training study, be associated network between building energy consumption achievement data and energy consumption analysis conclusion, can well the nonlinear relationship of processing load and correlative factor, automatically the noise in building energy consumption achievement data can not only be eliminated, and the work draw correct conclusion of can remaining valid in noise circumstance;

(4) the present invention utilizes error back propagation constantly to adjust the weights and bias of neural network, on this basis by carrying out binary conversion treatment to input data, can not only effective departure scope, and analysis precision can be made to remain on higher level.

Accompanying drawing explanation

Fig. 1 is originally based on the analysis process figure of the analyzing methods of architectural energy saving of BP neural network.

Fig. 2 is originally based on the input data binaryzation process flow diagram of the analyzing methods of architectural energy saving of BP neural network.

Fig. 3 is originally based on the network structure of the Building Energy Analysis system of BP neural network.

Fig. 4 is originally based on the training process flow diagram of the analyzing methods of architectural energy saving of BP neural network.

Embodiment

Below in conjunction with embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited thereto.

Embodiment

See Fig. 1, the process flow diagram of the Building Energy Analysis based on BP neural network of the present invention, next elaborates to step described in the method.

The electrisity consumption of this example to building is analyzed, and obtains the energy consumption weak link that this architectural electric power consumption part exists.

Step 1., according to appointment building energy consumption type determination input and output vector, can adopt following steps to realize:

(1-1) in analysis end time the previous day day (24:00), electric power is respectively itemized electric quantity data collection, obtain corresponding electric quantity data as original energy consumption data, comprise: unit area illumination in this year and socket power consumption, this year unit area air-conditioning power consumption, this year, unit area open air exhaust electromechanics consumed, this year unit area specific installation power consumption and the inoperative period yesterday illumination with socket electrisity consumption, the illumination of working hour yesterday and socket power consumption value, inoperative yesterday period air conditioner electric consumption, working hour yesterday air-conditioning power consumption value, the dynamo-electric consumption of inoperative yesterday period open air exhaust, working hour yesterday air-conditioning power consumption value, the dynamo-electric consumption of inoperative yesterday period open air exhaust, the dynamo-electric consumption value of working hour yesterday open air exhaust, inoperative yesterday period specific installation electrisity consumption, working hour yesterday specific installation power consumption value,

(1-2) calculate according to following computing formula, obtain the energy consumption index data of electrisity consumption part, comprise: unit area illumination and socket electrisity consumption, unit area air conditioner electric consumption, the dynamo-electric consumption of unit area open air exhaust, unit area particular electrical consumption, inoperative period and working hour throw light on and the ratio of socket power consumption, the ratio of inoperative period and working hour air-conditioning power consumption, the inoperative period with working hour room open air exhaust power consumption ratio and the ratio of inoperative period and working hour specific installation power consumption, therefore here using the input variable of the energy consumption index data of electrisity consumption part as BP neural network, namely the input variable of BP neural network is exactly the vector of one 8 dimension.

(1-3) output vector of building energy consumption weak link as model of corresponding building energy consumption type is determined, the energy consumption weak link of electrisity consumption part is as shown in table 1, object vector is 8 energy consumption weak links of electrisity consumption part, and namely output vector is the vector of one 8 dimension.

Table 1 power consumption part building energy consumption index calculating method

(1-4) binary conversion treatment is carried out to the input vector of the energy consumption index data of electrisity consumption part;

(1-4-1) the energy-saving potential D of this input pointer is calculated according to following formula;

$D=\frac{\mathrm{\&Delta;}I}{I_n}=\frac{I_a-I_n}{I_n}---\left(4\right)$

In formula: I _afor building the actual desired value calculated; I _nfor public building reference index value.

(1-4-2) D that step (1-4-1) calculates is carried out value by following formula rule.

$f\left(x_i\right)=\left\{\begin{array}{cc}0& D\left(x_i\right)\&le;T\\ 1& D\left(x_i\right)>T\end{array}\right.---\left(5\right)$

In formula: x _ifor the index parameter of i-th before binaryzation; F (x _i) be x _ivalue after binaryzation; D _ibe i-th index parameter x _ienergy-saving potential calculated value; T is energy-saving potential threshold values.

(1-5) binary coding is carried out to the output vector of the energy consumption weak link of electrisity consumption part, as shown in table 2;

Table 2 electrisity consumption part energy consumption weak link is encoded

Step 2., according to input and output vector structure BP neural network model, can adopt following steps to realize:

(2-1) according to the amount of element determination input layer of input vector, output vector and the neuronal quantity of output layer.In this example, input vector and output vector have 8 elements, so the neuron number of input layer and output layer is all 8.

(2-2) determine BP neural network hidden layer neuron quantity according to the neuronal quantity of input layer and output layer, tentatively draft the number of hidden nodes by following experimental formula:

$p=\sqrt{nm}---\left(6\right)$

Wherein, n is input number of nodes, and m is output node number, and P is the number of hidden nodes.

And the excitation function of BP neural network is sigmoid function:

$f\left(x\right)=\frac{1}{1+e^{-x}}---\left(7\right)$

Step 3. inputs training sample and trains the BP neural network model established;

(3-1) according to the neuronal quantity setting training parameter of constructed BP neural network model, network training is carried out to BP neural network, after BP neural metwork training, just may be used for the practical application of Building Energy Analysis.Described training parameter comprises: maximum frequency of training, anticipation error, momentum term numerical value and learning rate.When BP neural network structure more complicated, when neuron number is many, can suitably increase frequency of training and learning rate.Optimum configurations is as shown in table 3 in this example:

Table 3 architectural electric power consumption amount partial nerve network parameter

As shown in Fig. 4 process flow diagram, perform following steps:

(3-2) method of random assignment is adopted to carry out the weights W of input layer to hidden layer _ij, hidden layer is to the weights W of output layer _hoinitialization, the span of random value is [-1,1];

(3-3) training sample in sample set is input in neural network one by one;

(3-4) calculate each neuron of hidden layer according to the following formula to export:

ho _h(k)＝f(hi _h(k))h＝1,2,...,p

(3-5) calculate each neuron of output layer according to the following formula to export:

yo _o(k)＝f(yi _o(k))o＝1,2,...,q

(3-6) the actual error with desired output that exports is calculated

$e=\frac{1}{2}\underset{o=1}{\overset{q}{\&Sigma;}}{\left(d_o\right(k)-y_o(k\left)\right)}^2---\left(8\right)$

Wherein, d _ofor the value of desired output, y _ofor the actual output of neural network.

(3-7) circulation step (3-3) to (3-6) is until all samples that training sample is concentrated all have been processed;

(3-8) calculation training sample set global error and judge the end condition of BP learning algorithm, if error is less than anticipation error 0.001 or has reached maximum frequency of training 50000 times, then terminate training process; Otherwise forward step (3-9) to;

(3-9) from output layer, carry out anti-pass to error, error of calculation function is to each neuronic partial derivative of output layer according to the following formula:

$\frac{\&part;e}{\&part;w_{ho}}=\frac{\&part;e}{\&part;{\mathrm{yi}}_o}\frac{\&part;{\mathrm{yi}}_o}{\&part;w_{ho}}=-{\mathrm{\&delta;}}_o\left(h\right){\mathrm{ho}}_h\left(k\right)$

(3-10) according to the following formula error of calculation function to each neuronic partial derivative of hidden layer:

$\frac{\&part;e}{\&part;w_{ih}}=\frac{\&part;e}{\&part;{\mathrm{hi}}_h\left(k\right)}\frac{\&part;{\mathrm{hi}}_h\left(k\right)}{\&part;w_{ih}}=-{\mathrm{\&delta;}}_h\left(k\right)x_i\left(k\right)$

(3-11) utilize each neuronic partial derivative of output layer and each neuronic output of hidden layer to revise connection weight w _ho(k):

${\mathrm{\&Delta;w}}_{ho}\left(k\right)=-\mathrm{\&eta;}\frac{\&part;e}{\&part;w_{ho}}={\mathrm{\&eta;\&delta;}}_o\left(k\right){\mathrm{ho}}_h\left(k\right)$

w _ho(k+1)＝w _ho(k)+ηδ _o(k)ho _h(k)

(3-12) each neuronic partial derivative of hidden layer is utilized to be connected weight w with each neuronic Introduced Malaria of input layer _ih(k), and forward step (3-3) to;

${\mathrm{\&Delta;w}}_{ih}\left(k\right)=-\mathrm{\&eta;}\frac{\&part;e}{\&part;w_{ih}}=-\mathrm{\&eta;}\frac{\&part;e}{\&part;{\mathrm{hi}}_b\left(k\right)}\frac{\&part;{\mathrm{hi}}_h\left(k\right)}{\&part;w_{ih}}={\mathrm{\&eta;\&delta;}}_h\left(k\right)x_i\left(k\right)$

w _ih(k+1)＝w _ih(k)+ηδ _h(k)x _i(k)

Step 4. is input to trained BP model the energy consumption index data that actual needs carries out analyzing and carries out Treatment Analysis.After carrying out network training with training sample to BP neural network, to the electrisity consumption energy consumption index vector of BP neural network input yesterday, output valve is assay value, and the indicator of power consumption data that building needs carry out analyzing are as shown in table 4:

The indicator of power consumption data needing to carry out analyzing built by table 4

Use binaryzation operation to obtain after needing the power consumption part index number vector analyzed to carry out pre-service, pretreated indicator of power consumption data are as shown in table 5:

The pretreated indicator of power consumption data of table 5

After the calculating of BP neural network, obtain the output of network, operation result is: out=0.002316; 0.861639;

5.053419E-9；0.031438；9.345239E-6；1.611918E-4；5.146818E-7；0.364439。

The output data of step 5. pair neural network carry out data convert.Following steps can be adopted to realize:

(5-1) travel through the output vector of BP neural network, be set to 1 by maximum for numerical value one, remaining is set to 0, and the binary coding namely obtaining output vector exports, and obtaining output encoder is: 01000000;

(5-2) according to coding and the corresponding relation analyzing conclusion content, binary coding is reduced to the understandable Word message of user again.The scale-of-two calculated is exported and converts the understandable conclusion content of user to, the conclusion information of output encoder 01000000 is: inoperative period/evening, light fixture or office equipment (computer screen, host computer, printer) do not close or there is standby situation;

(5-3) after obtaining the analysis conclusion of feedback, user gets final product the place of clear and definite building energy consumption problem, and purposively can carry out emphasis investigation and safeguard.Visit investigation through evening, determine really to exist in building After Hours office equipment (computer screen, host computer, printer) the no situation such as pass, and corridor, stair light fixture be held open state the whole night.Therefore two corresponding corresponding Improving Measurements are taked to be: one is carried out arranging according to the automatic control of illumination count value to illumination in energy source monitoring system, it is as follows that this arranges rule: if detect that room brilliancy is lower than timing 15Lx from illumination meter first time, if room brilliancy continues five minutes all lower than 15Lx, then automatically open lighting; If illumination meter detects that room brilliancy is higher than 30Lx, and this brightness continues more than five minutes, then automatically close lighting, wherein Lx is unit of illuminance.Another Improving Measurements is that a requirement building managerial personnel strengthen inspection, inoperative period/close the office equipment and lighting that do not need to use, and strengthen the energy-conservation knowledge of publicity in the evening in time, improves the awareness of saving energy of user of service.

The present embodiment is based on the Building Energy Analysis system of BP neural network.Its network structure is as follows:

1. bottom hardware data acquisition unit.Data acquisition unit can be metering ammeter, water meter, cooling metering table etc., and they are installed in each controlled building, the responsible original energy consumption data building end and the service data using energy equipment of gathering.

2. data acquisition server.Data acquisition program primary responsibility in data acquisition server to the equipment manager of administering in building carry out data extraction, and the data collected are carried out valid cache, to support the data access of upper layer data storehouse server.

3. database server.Mainly the energy consumption data of each building gathered in database server and carry out effectively storing and backup.The capture program wherein disposed regularly can call the interface acquisition energy consumption data that each data acquisition server is issued, and then carries out the data got into library storage, is supplied to upper layer application and uses.

4. application server.Application server major deployments, based on the Building Energy Analysis system of BP neural network, carries out analyzing and processing to the energy consumption data of building.Whole system program is divided into six large modules, is energy-saving potential quantization modules, data preprocessing module, BP nework analysis module, data restoring module, Configuration Manager and logger module respectively.

(1) energy-saving potential quantization modules

This module primary responsibility calculates the energy-saving potential of each building energy consumption achievement data, because need the energy-saving potential of first parameter when carrying out pre-service to index input data, then is carried out the binary conversion treatment of input item by the energy-saving potential of index.This operating process is first obtain correction factor corresponding to this index by calling correction factor computing module, then calculates according to the quantitative formula of energy-saving potential, obtains the energy-saving potential numerical value of this evaluation index;

(2) data preprocessing module

Data preprocessing module is a module relatively important in this system, because this module in charge provides the data basis of analyzing methods of architectural energy saving, plays conclusive effect to the subsequent operation of energy consumption analysis system.Data preprocessing module primary responsibility processes the data of the energy consumption index that building energy consumption metrics evaluation system-computed obtains further, which includes input data binaryzation and exports data encoding two submodules.

(2-1) input item binaryzation: this submodule primary responsibility carries out binary conversion treatment to the input item of BP neural network.Each sample that this Module cycle traversal input amendment is concentrated, its energy-saving potential is calculated to each input item of sample, and this energy-saving potential and threshold values (T=5%) are compared, if >5%, then the binaryzation value of input item is 1; Otherwise get 0.

(2-2) output item coding: this submodule carries out binary coding to output item, if there is this coding in the analysis conclusion table in database, directly takes out and carries out using; If there is not this coding in database, then use after needing that unified binary coding is carried out to this alanysis conclusion.

(3) BP nework analysis module

This module is most important part in program, be responsible for the core analysis flow elements realizing analyzing methods of architectural energy saving, this module is divided into two submodules, be knowledge acquisition and energy consumption analysis two parts respectively, this is also corresponding uses BP network to carry out two stages analyzed: training stage and analysis phase.Core algorithm module realizes from building energy consumption index to the transfer process of energy consumption analysis conclusion, is connecting building index phenomenon and the tie analyzing conclusion.

(3-1) the knowledge acquisition stage: in the training stage, according to the analysis rule in knowledge base building energy consumption desired value and Building Energy Analysis conclusion combined as training sample set pair BP neural network and train, allow its automatic acquisition knowledge and rule.Each sample in this submodule circuit training sample set, input item is the vector of building index binaryzation, and output item is the binary coding of energy consumption analysis conclusion, repeats this operation until training error is less than anticipation error or reaches maximum frequency of training.After training terminates, namely this BP neural network gets the rule knowledge required for the analysis phase, and its form of expression is that the weights of each neuron node in BP network are with threshold values.

(3-2) in the energy consumption analysis stage: in the analysis phase of method, be combined into vector be input in trained network by needing the achievement data carrying out analyzing and processing after carrying out binaryzation, the energy consumption weak link of output to building according to network is analyzed.BP neural network after training can be analyzed input vector rapidly, and exports corresponding binary coding analysis conclusion.

(4) data convert device module

Data convert device module primary responsibility resolves the analysis conclusion through coding that BP neural network exports, and converts the analysis conclusion through coding to user understandable Word message.Output due to BP neural network is binary coding, so need to carry out content reduction to it, according to its scale-of-two output encoder, by the coding corresponding informance analyzed in database in conclusion table, the energy-conservation Improving Measurements analyzing conclusion and correspondence is combined into final analysis conclusion and feeds back to user.

(5) Configuration Manager

This module major maintenance is responsible for the parameter configuration that BP neural network builds, and comprises the network number of plies, hidden layer neuron number, input and output item number, weight initialization scope, momentum arithmetic numerical value and learning rate initial value etc.

(6) logger module

User operation behaviors all in this module in charge logging program operational process, the abnormal information of appearance and thread state, and be saved in relevant journal file.The information of record comprises:

(6-1) User operation log, to the Operation Log of database, comprising: increase, revise, delete and inquiry;

(6-2) system cloud gray model abnormal information, comprising: the operation exception of database manipulation exception, file resource operation exception and other resources;

(6-3) Parameter transfer that calls background service of foreground, comprising: the title of method, the parameter of method and numerical value.

5. terminal device.Terminal device can be computing machine, server, mobile phone or other internet devices, terminal device runs the WWW application of B/S pattern.By terminal device, user can browse the energy consumption analysis conclusion of history, can also arrange simultaneously, and carry out the knowledge maintenance of energy consumption analysis database to the parameter of BP neural network.

Above-described embodiment is the present invention's preferably embodiment; but embodiments of the present invention are not restricted to the described embodiments; change, the modification done under other any does not deviate from Spirit Essence of the present invention and principle, substitute, combine, simplify; all should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (10)

1. based on the analyzing methods of architectural energy saving of BP neural network, it is characterized in that, comprise the following steps:

(1) according to appointment building energy consumption type determination input and output vector;

(2) according to input and output vector structure BP neural network model;

(3) input training sample to train the BP neural network model established;

(4) the energy consumption index data that actual needs carries out analyzing are input to trained BP model and carry out Treatment Analysis;

(5) data convert is carried out to the output data of neural network.

2. the analyzing methods of architectural energy saving based on BP neural network according to claim 1, is characterized in that, described step (1) comprises the steps: further

(1-1) building energy consumption type is divided into 3 classes, is respectively power consumption, water consumption and cold consumption adjusting system, the previous day day building energy consumption is measured in analysis, obtain corresponding measurement data as original building energy consumption data;

(1-2) input vector of building energy consumption achievement data value as model is calculated according to the computing formula of original building energy consumption data and green building assessment index,

(1-3) output vector of building energy consumption weak link as model of corresponding building energy consumption type is determined, the output variable of 3 energy consumption types;

(1-4) binary conversion treatment is carried out to the input vector determined;

(1-5) binary coding is carried out to the output vector determined, with energy consumption weak link number in building energy consumption type for binary coding length, thus output vector is adopted to the coded system of " getting 1 in n "; Wherein, n is the length of coding, i.e. the total number of energy consumption weak link of corresponding energy consumption type, and when certain building weak link is determined, the coding of its corresponding position is set to 1, and remaining n-1 position coding is all 0.

3. the analyzing methods of architectural energy saving based on BP neural network according to claim 2, is characterized in that, in step (1-2), the input variable of 3 class separate energy consumptions is as follows:

(1-2-1) power consumption part: comprise unit area air conditioner electric consumption, unit area illumination and socket electrisity consumption, unit area open air exhaust dynamo-electric consumption, unit area particular electrical consumption, also comprising inoperative period and working hour throws light on and the ratio of the ratio of the ratio of the ratio of socket power consumption, inoperative period and working hour air-conditioning power consumption, inoperative period and working hour room open air exhaust power consumption, inoperative period and working hour specific installation power consumption;

(1-2-2) water consumption part: the ratio of unit area water consumption, inoperative period and working hour water consumption;

(1-2-3) cold consumption adjusting system part: the cold consumption of unit area, comprise air-conditioning system Energy Efficiency Ratio, refrigeration system Energy Efficiency Ratio, handpiece Water Chilling Units operational efficiency, chilled water Transfer coefficient, chilled water Transfer coefficient, air conditioning terminal Energy Efficiency Ratio, cooling pump efficiency, efficiency of cooling tower, refrigerating water pump efficiency, water system supply backwater temperature difference and water system return water temperature consistance.

4. the analyzing methods of architectural energy saving based on BP neural network according to claim 2, is characterized in that, step (1-4) comprises the steps: further

(1-4-1) for each input pointer data of input vector, its index value is I _a, obtain the standard index value I that it is relevant _n;

(1-4-2) the energy-saving potential D of this input pointer is calculated according to formula (1);

$D=\frac{\mathrm{\&Delta;}I}{I_n}=\frac{I_a-I_n}{I_n}---\left(1\right)$

In formula: I _afor building the actual desired value calculated; I _nfor public building reference index value;

(1-4-3) D that step (1-4-2) calculates is carried out value by formula (2) rule, compare by energy-saving potential D threshold values T, if D>=T, binaryzation value f (x)=1 of this then input pointer; If D<T, then binaryzation value f (x)=0 of this input pointer;

$f\left(x_i\right)=\left\{\begin{array}{cc}1& D\left(x_i\right)\&le;T\\ 0& D\left(x_i\right)>T\end{array}\right.---\left(2\right)$

In formula: x _ifor before binaryzation _iindividual index parameter; F (x _i) be x _ivalue after binaryzation; D _ibe i-th index parameter x _ienergy-saving potential calculated value; T is energy-saving potential threshold values.

5. the analyzing methods of architectural energy saving based on BP neural network according to claim 1, is characterized in that, described step (2) comprises the steps: further

(2-1) according to the neuronal quantity of input vector, output vector determination input layer and output layer;

(2-2) determine BP neural network hidden layer neuron quantity according to the neuronal quantity of input layer and output layer, wherein, the excitation function of BP neural network is sigmoid function:

$f\left(x\right)=\frac{1}{1+e^{-x}}---\left(3\right)$

Described step (3), comprises the steps: further

The training parameter of setting BP neural network, and it is trained, wherein training parameter comprises: maximum frequency of training, anticipation error, momentum term numerical value and learning rate.

6. the analyzing methods of architectural energy saving based on BP neural network according to claim 1, is characterized in that, described step (5) comprises the steps: further

The output vector of BP neural network is traveled through, 1 is set to by maximum for numerical value one, remaining is set to 0, and the binary coding namely obtaining output vector exports, then according to coding and the corresponding relation analyzing conclusion content, binary coding is reduced to the understandable Word message of user; After obtaining the analysis conclusion fed back, user gets final product the place of clear and definite building energy consumption problem, and purposively can carry out emphasis investigation and safeguard, problem wherein comprises replacing or maintenance of equipment, the maintenance of reinforcement equipment and the operation reserve of adjustment equipment.

7. the Building Energy Analysis system based on BP neural network according to claim 1, it is characterized in that, comprise energy-saving potential quantization modules, data preprocessing module, BP nework analysis module, data restoring module, Configuration Manager and logger module;

Described energy-saving potential quantization modules, for calculating the energy-saving potential of each building energy consumption achievement data;

Described data preprocessing module, for being responsible for the data basis providing analyzing methods of architectural energy saving;

Described BP nework analysis module, for being responsible for the core analysis flow elements realizing analyzing methods of architectural energy saving, this module is divided into knowledge acquisition submodule and energy consumption analysis submodule; Corresponding respectively BP network is used to carry out two stages analyzed: training stage and analysis phase;

Described data convert device module, for resolving the analysis conclusion through coding that BP neural network exports, convert the analysis conclusion through coding to user understandable Word message, output due to BP neural network is binary coding, so need to carry out content reduction to it, according to its scale-of-two output encoder, by the coding corresponding informance analyzed in database in conclusion table, the energy-conservation Improving Measurements analyzing conclusion and correspondence is combined into final analysis conclusion and feeds back to user;

Described Configuration Manager, for safeguarding the parameter configuration that responsible BP neural network builds, described parameter comprises the network number of plies, hidden layer neuron number, input and output item number, weight initialization scope, momentum arithmetic numerical value and learning rate initial value;

User operation behaviors all in described log pattern logging program operational process, the abnormal information of appearance and thread state, and be saved in relevant journal file.

8. the Building Energy Analysis system based on BP neural network according to claim 7, is characterized in that, described data preprocessing module comprises input data binaryzation submodule and exports data encoding submodule;

Described input data binaryzation submodule, for being responsible for, binary conversion treatment is carried out to the input item of BP neural network, each sample that this Module cycle traversal input amendment is concentrated, its energy-saving potential is calculated to each input item of sample, and this energy-saving potential and threshold values T are compared, if >T, then the binaryzation value of input item is 1; Otherwise get 0;

Described output data encoding submodule, for carrying out binary coding to output item, if there is this coding in the analysis conclusion table in database, directly taking out and carrying out using; If there is not this coding in database, then use after needing that unified binary coding is carried out to this alanysis conclusion.

9. the Building Energy Analysis system based on BP neural network according to claim 7, is characterized in that, described knowledge acquisition submodule was specially in the training acquisition stage:

According to the analysis rule in knowledge base building energy consumption desired value and Building Energy Analysis conclusion combined as training sample set pair BP neural network and train, allow its automatic acquisition knowledge and rule, each sample in described knowledge acquisition submodule circuit training sample set, input item is the vector of building index binaryzation, output item is the binary coding of energy consumption analysis conclusion, repeat this operation until training error is less than anticipation error or reaches maximum frequency of training, after training terminates, namely this BP neural network gets the rule knowledge required for the analysis phase, its form of expression is that the weights of each neuron node in BP network are with threshold values,

Described energy consumption analysis submodule was specially in the energy consumption analysis stage:

Being combined into vector after carrying out binaryzation by needing the achievement data carrying out analyzing and processing is input in trained network, the energy consumption weak link of output to building according to network is analyzed, BP neural network after training can be analyzed input vector rapidly, and exports corresponding binary coding analysis conclusion.

10. the Building Energy Analysis system based on BP neural network according to claim 7, is characterized in that, the information of described logger module record comprises:

User operation log, to the Operation Log of database, comprising: increase, revise, delete and inquiry;

System cloud gray model abnormal information, comprising: the operation exception of database manipulation exception, file resource operation exception and other resources;

The Parameter transfer that foreground is called background service, comprising: the title of method, the parameter of method and numerical value.