CN106910144A - Based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method - Google Patents

Abstract

The invention discloses it is a kind of based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, propose timesharing and carve reality energy coefficient, for describing not descend building energy consumption characteristic in the same time, and the history data set using outdoor environment dry-bulb temperature, outdoor environment relative humidity, last moment building energy consumption value sets up building energy consumption forecast model, subsequent time building energy consumption is then obtained by online outdoor environment dry-bulb temperature, relative humidity, the building energy consumption value on-line prediction of obtaining.The present invention has following technique effect：The building energy consumption forecast model prediction data reliability that the method is set up is high, the building that can be used to predicting in single building or big regional extent by when energy consumption, the Energy Saving Control of building energy consumption, building energy consumption prediction and region in the occasion such as electric power peak clipping.

Description

Based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method

Technical field

It is more particularly to a kind of to be based on timesharing quarter actual use the present invention relates to the research field of heavy construction energy consumption prediction Several heavy constructions by when energy consumption on-line prediction method.

Background technology

As the continuous acceleration of the fast-developing and Development of China's Urbanization of China's economic society, heavy construction are continued to bring out, advise Mould also constantly increasing, comprehensive energy consumption also more and more higher, the proportion that its energy consumption accounts for social total energy consumption is also being continuously increased, large-scale to build Building energy-conservation turns into the important content for building green smart city.The rational short-term forecast of heavy construction energy consumption, for instructing summer The scheduling mechanism of power surges load, improves the theoretical system of constructing operation energy consumption research, adapts to future architecture operation energy consumption Growth requirement, realizes that the national emission reduction targets of China are significant.

Conventional building energy consumption short term prediction method includes univariate time series method, artificial neural network, supporting vector Various methods such as machine.Building on-line prediction algorithm needs to consider the factors such as real-time, engineering realizability.Univariate time series Predicted method is only with energy consumption time series as Prediction Parameters, it is impossible to effectively utilize the information with building energy consumption influence factor, very The precision of prediction of difficult further lifting operation energy consumption；Artificial neural network method output has unpredictability and inconsistency, holds Easily Local Minimum is absorbed in, it is difficult to find optimal models, the problems such as generalization ability is not strong；SVMs is more quick to missing data Sense, the pace of learning to extensive sample is slow；The above method is required for using tool box special, while modeling and predicted time It is long.Multiple nonlinear regression method advantage is fast predetermined speed, and to lacking, data are insensitive, there is detailed mathematic(al) representation solution Influence relation of each independent variable to dependent variable is released, but because some non-civil buildings such as megastore receive out door climatic parameter, people The influence of the factors such as current density, festivals or holidays, accurate forecast model is difficult to set up with Conventional polyol non-linear regression method.

The content of the invention

Shortcoming and deficiency it is a primary object of the present invention to overcome prior art, there is provided one kind carves actual use based on timesharing Can coefficient heavy construction by when energy consumption on-line prediction method, have effectively achieved the dynamic prediction of building energy consumption and accurate pre- Survey, can be not only used for single building energy consumption prediction, it can also be used to the energy consumption prediction of large building.

In order to achieve the above object, the present invention uses following technical scheme：

The present invention based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, including following steps Suddenly：

S1, data initialization,

S1.1, acquisition historical data, historical data include outdoor environment dry-bulb temperature T, outdoor environment phase in a period of time To humidity RH and building energy consumption Q by when data；

S1.2, historical data cleaning, sequence is carved by historical data by 24 timesharing are divided into constantly, each using 3 σ regulation analysis The abnormal data at moment, rejects the exceptional value at each moment respectively, and then the synchronization data with the previous day and one day after are average Value substitutes former data, and former data are substituted using extrapolated value method if abnormal data is MARG；

S1.3, timesharing quarter reality energy coefficient calculation of initial value, timesharing are carved reality energy coefficient and are defined as follows：

P (j) is jth moment energy consumption average value,It is P (j) in 1 to 24 moment Maximum；

S1.4, historical data conversion, using timesharing quarter reality energy coefficient ξ (j), j=1 ..., 24 carry out historical data turns Change, set ξ₀It is the limit value of reality energy coefficient, when actual energy coefficient ξ (j) ＞ ξ are carved in timesharing₀, construction history energy consumption conversion number According to forAs ξ (j)≤ξ₀, construction history energy consumption change data Q₁(i, j)=Q (i, j), that is, keep not Become；

S1.5, construction Multiple Non Linear Regression energy consumption forecast model input and output vector；

S1.6, historical data homogenization treatment, historical data include history energy consumption data and outdoor environment dry-bulb temperature and Outdoor environment relative humidity；

S1.7, historical data modeling, according to the taylor series expansion of the function of many variables, the Multiple Non Linear Regression mould of selection Type, tries to achieve multivariate nonlinear regression analysis model parameter；

S2, building energy consumption by when on-line prediction；

S2.1, the initialization of online acquisition moment：j₁=1；

S2.2, judge j₁Whether 24 are less than or equal to

If it is, being updated into next step S2.3 on-line data acquisitions and data set；

If it is not, then j₁=1, enter back into next step S2.3 on-line data acquisitions and updated with data set；

S2.3, data set update, i.e., terminate within D days to the since the 1st day, by the building energy at i+1 day whole day each moment Consumption value, outdoor environment dry-bulb temperature value and outdoor environment rh value are assigned to i-th day building energy consumption of synchronization, outdoor ring Border dry-bulb temperature and outdoor environment relative humidity；

I.e.：Data set updates current time for jth₁During the moment, first day jth is abandoned₁The data at moment, Q (1, j₁),T (1,j₁),RH(1,j₁),

On-line data acquisition：

Online acquisition and in the centrally stored current time data of data, the D+1 days jth₁Time data, including outdoor environment Dry-bulb temperature T (D+1, j₁), outdoor environment relative humidity RH (D+1, j₁) and building energy consumption Q (D+1, j₁)；

S2.4, data cleansing, energy consumption data Q (D+1, the j gathered using 3 σ regulation analysis current times₁) whether abnormal, If data exception, former data are substituted using extrapolated value method；

S2.5, timesharing carve actual energy coefficient online updating；Update method is as follows：

P(j₁) it is jth₁Moment energy consumption average value,It is P in 1 to 24 moment The maximum of (j)；

S2.6, online acquisition data conversion are actual with energy coefficient ξ (j using current time₁) carry out current time energy consumption number According to conversion, ξ is set₀It is the limit value of reality energy coefficient, when actual energy coefficient ξ (j are carved in timesharing₁) ＞ ξ₀, building energy consumption conversion number According to forAs ξ (j)≤ξ₀, building energy consumption change data Q₁(D+1, j₁)=Q (D+1, j₁), that is, keep constant；

S2.7, the treatment of online acquisition data normalization, using with standardized way and step S1.6 identical processing modes, Including online acquisition energy consumption data, outdoor environment dry-bulb temperature and outdoor environment relative humidity；

S2.8, subsequent time Multiple Non Linear Regression energy consumption on-line prediction；

S2.9, the treatment of energy consumption prediction data renormalization, using equation below：

It is the D+1 days j₁ + 1 moment energy consumption predicted value；

S2.10, energy consumption prediction data timesharing quarter energy data inverse transformation are simultaneously exported；

S2.11, forecast model online updating, are that subsequent time prediction is prepared；

Whether S2.12, determining program require is terminated.

As preferred technical scheme, in step S1.2, by historical data by be divided into constantly 24 timesharing carve sequence Q (i, J), T (i, j), RH (i, j), j=1,2 ... ... 23,24, i=1,2 ... ..., D, total number of days are D, then set：

P (j) is the energy consumption average value from the 1st day to the D days jth moment,

σ (j) be it is poor to the D days energy consumption standards at jth moment from the 1st day,

I.e.

If | Q (i, j)-P (j) | σ (j) of ＞ 3, and 1 ＜ i ＜ D, then Q (i, j)=(Q (i-1, j)+Q (i+1, j))/2

If | Q (i, j)-P (j) | ＞ 3 σ (j), i=1, then Q (i, j)=2Q (i+1, j)-Q (i+2, j)

If | Q (i, j)-P (j) | ＞ 3 σ (j), i=D, then Q (i, j)=2Q (i-1, j)-Q (i-2, j).

As preferred technical scheme, in step S1.5, specially：

Construction input energy consumption column vector Q ' (k), outdoor temperature column vector T ' (k) and outside relative humidity column vector RH ' K (), energy consumption gathers date column vector D (k), energy consumption gathers moment column vector H (k)；

Q ' (k)=[Q₁(1,1),Q₁(1,2),…,Q₁(1,24),Q₁(2,1) ..., Q₁(2,24) ...,

Q₁(D, 1) ..., Q₁(D,24)]^T

T ' (k)=[T (1,1), T (1,2) ..., T (1,24), T (2,1) ..., T (2,24) ...,

T (D, 1) ..., T (D, 24)]^T

RH ' (k)=[RH (1,1), RH (1,2) ..., RH (1,24), RH (2,1) ..., RH (2,24) ...,

RH (D, 1) ..., RH (D, 24)]^T

Construction input matrix INPUT (k)=[T ' (k-1), RH ' (k-1), Q ' (k-1)],

Output matrix Y (k)=[Q ' (k)], 2≤k≤L.

Used as preferred technical scheme, in step S1.6, the method for uniforming treatment is：

Wherein Q " (k), T " (k), RH " (k) is data after standardization；

Point Not Wei each time series minimum value,

The maximum of respectively each time series.

Used as preferred technical scheme, step S1.7 is specially：

According to the taylor series expansion of the function of many variables, multinomial highest number of times is taken for 2 times.The nonlinear multivariable of selection is returned Return model as follows：

Using least squares estimate, multivariate nonlinear regression analysis model parameter is asked for；

IfQ " (k) is the power consumption values at the k moment,It is the k moment Energy consumption predicted value；

Make function of many variables G (a₀,a₁,…,a₉) to a₀,a₁,…,a₉Partial derivative be zero, i.e.,

Training data is substituted into, the shaping such as form of AX=B is arranged, finally using X=A^-1B, you can try to achieve polynary non- Linear Regression Model Parameters a₀,a₁,…,a₉.Wherein, A is 10 × 10 matrix, and B is 10 × 1 matrixes.

Used as preferred technical scheme, step S2.3 is specially：

If

P(j₁) it is from the 2nd day to the D+1 days jth₁The energy consumption average value at moment,

σ(j₁) it is from the 2nd day to the D+1 days jth₁The energy consumption standard at moment is poor,

I.e.

If | Q (D+1, j₁)-P(j₁) | the σ (j of ＞ 3₁), then Q (D+1, j₁)=2Q (D, j₁)-Q(D-1,j₁)。

Used as preferred technical scheme, step S2.6 is specially：

(1) online acquisition energy consumption data standardization

(2) online acquisition outdoor environment dry-bulb temperature data normalization treatment

(3) online acquisition outdoor environment relative humidity data standardization

Used as preferred technical scheme, step S2.7 is specially：

Used as preferred technical scheme, step S2.7 is specially：

Construction input energy consumption column vector Q " (k), outdoor temperature column vector T " (k) and outside relative humidity column vector RH " (k)；

Q " (k)=[Q₁(1,j₁+1),Q₁(1,j₁+2),…,Q₁(1,24),Q₁(2,1)

..., Q₁(D, 1) ..., Q₁(D,24),Q₁(D+1,1) ..., Q₁(D+1,j₁)]^T

T " (k)=[T₁(1,j₁+1),T₁(1,j₁+2),…,T₁(1,24),T₁(2,1) ...,

T₁(D, 1) ..., T₁(D,24),T₁(D+1,1) ..., T₁(D+1,j₁)]^T

RH " (k)=[RH₁(1,j₁+1),RH₁(1,j₁+2),…,RH₁(1,24),RH₁(2,1) ...,

RH₁(D, 1) ..., RH₁(D,24),RH₁(D+1,1) ..., RH₁(D+1,j₁)]^T

Construction input matrix INPUT (k)=[T " (k-1), RH " (k-1), Q " (k-1)]

By in data substitution A, B matrix in Q " (k), Q " (k-1), T " (k-1), RH " (k-1), that is, substitute into

In AX=B, finally using X=A^-1B, you can try to achieve multivariate nonlinear regression analysis model parameter a₀,a₁,…,a₉。

Used as preferred technical scheme, step S2.11 is specially：

j₁=j₁+1

Judge whether to receive EP (end of program) instruction,

If it is, EP (end of program)；

If it is not, then going to S2.2 judges j₁Whether the step of less than or equal to 24, the flow of on-line prediction is performed again.

The present invention compared with prior art, has the following advantages that and beneficial effect：

1st, the present invention proposes that timesharing carves reality with energy coefficient to portray the regularity between building energy consumption and moment, to limit The flow of the people related to the moment and other influences of interference to energy consumption.Timesharing is carved reality energy coefficient and is defined as follows：

P (j) is jth moment energy consumption average value.Reality energy coefficient is carved to building using timesharing Building energy consumption carries out data prediction, and advantage is the influence for reducing the external interferences such as flow of the people to energy consumption, is predicted in simplification Precision of prediction is ensure that while journey.

2nd, the present invention takes multinomial highest number of times for 2 times according to the taylor series expansion of the function of many variables.That chooses is polynary Nonlinear regression model (NLRM) is as follows：

Multivariate nonlinear regression analysis model parameter is asked for using least squares estimate.

Order

Make function of many variables G (a₀,a₁,…,a₉) to a₀,a₁,…,a₉Partial derivative be zero, i.e.,

Training data is substituted into, the shaping such as form of AX=B is arranged, finally using X=A^-1B, you can try to achieve polynary non- Linear Regression Model Parameters a₀,a₁,…,a₉。

Compared with existing building energy consumption Forecasting Methodology such as neutral net, SVMs, time series, advantage is not Specific tool box special is needed, model is simple, and rapid modeling can obtain multivariate nonlinear regression analysis model in very short time Parameter a₀,a₁,…,a₉, shorten the modeling time.

3rd, the present invention is changed greatly according to building energy consumption in different periods, and the load variations rule of daily synchronization is similar Rule, energy consumption data time series is divided into 24 time series Q (*, j), using 3 σ regulation analysis historical datas by the hour With the abnormal data at online acquisition data each moment, the exceptional value at each moment is rejected respectively, then with the upper and lower data of synchronization Interpolation substitutes former data, and former data are substituted using extrapolated value method if abnormal data is MARG.

Advantage is the influence that can eliminate Outliers to model, improves precision of prediction.

4th, to modeling data using homogenization processing method, advantage is to eliminate outdoor environment dry-bulb temperature, room to the present invention External environment relative humidity, building energy consumption etc. difference variable between by physical significance or dimension it is inconsistent and cannot equality use institute The influence for bringing, it is to avoid the generation of un-reasonable phenomenon, improves precision of prediction.

Homogenization processing method is as follows：

Wherein x ' is data after standardization；It is the minimum value in x (k) sequences；It is the maximum in x (k) sequences.

5th, primary data length L of the present invention can set according to actual conditions.

6th, present invention employs Online MNR on-line training algorithms, model is constantly carried out with the renewal of online data Training, optimization, greatly improve precision of prediction.

7th, the present invention can be with the building energy consumption of on-line prediction subsequent time.

8th, the invention provides heavy construction by when energy consumption on-line prediction method, can be not only used for single building energy consumption it is pre- Survey, it can also be used to the energy consumption prediction of large building.

Brief description of the drawings

Fig. 1 is the flow chart of detection method.

Specific embodiment

With reference to embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited In this.

Embodiment

As shown in figure 1, the present embodiment be based on timesharing carve it is actual with can coefficient heavy construction energy consumption by when on-line prediction side Method, comprises the following steps：

1. data initialization；

1.1 historical datas are obtained；

Historical data is obtained, historical data includes outdoor environment dry-bulb temperature T, outdoor environment relative humidity in a period of time RH and building energy consumption Q by when data.Outdoor environment dry-bulb temperature and relative humidity can be obtained from weather bureau, building energy consumption By when data can gather remote transmission by building total ammeter and obtain.

If history data set data length is L=24 hours/day × number of days D, if not setting, default value is taken 6 months 180 My god, i.e. L=4320.

1.2 historical datas are cleaned

Due to causing data exception (without missing the problems such as external interference may be subject in ammeter data gatherer process Value), in order to reject the abnormal data in historical data, this patent by historical data by be divided into constantly 24 timesharing carve sequence Q (i, J), T (i, j), RH (i, j), j=1,2 ... ... 23,24, i=1,2 ... ..., D, total number of days be D, using 3 σ regulation analysis it is each when The abnormal data at quarter, rejects the exceptional value at each moment respectively, then the synchronization statistical average with the previous day and one day after Former data are substituted, former data are substituted using extrapolated value method if abnormal data is MARG.(acquiescence out door climatic parameter is obtained It is accurate.) set：

P (j) is the energy consumption average value from the 1st day to the D days jth moment,

σ (j) be it is poor to the D days energy consumption standards at jth moment from the 1st day,

I.e.

If | Q (i, j)-P (j) | σ (j) of ＞ 3, and 1 ＜ i ＜ D, then Q (i, j)=(Q (i-1, j)+Q (i+1, j))/2

If | Q (i, j)-P (j) | ＞ 3 σ (j), i=1, then Q (i, j)=2Q (i+1, j)-Q (i+2, j)

If | Q (i, j)-P (j) | ＞ 3 σ (j), i=D, then Q (i, j)=2Q (i-1, j)-Q (i-2, j)

Actual energy coefficient calculation of initial value is carved in 1.3 timesharing

The energy for building characteristic of synchronization is essentially identical, also similar by the influence degree of the interference such as flow of the people, light, The present invention propose timesharing carve it is actual with can coefficient portray this influence factor, with the reduction flow of the people related to the moment and other Influence of the disturbing factor to each moment energy consumption.Timesharing is carved reality energy coefficient and is defined as follows：

P (j) is jth moment energy consumption average value.It is P (j) in 1 to 24 moment Maximum.

1.4 historical datas are changed；

Carved using timesharing and actually use energy coefficient ξ (j), j=1 ..., 24 carry out historical data conversion.Setting ξ₀It is actual use The limit value of energy coefficient, ξ is taken in the present invention₀=0.2, when actual energy coefficient ξ (j) ＞ ξ are carved in timesharing₀, construction history energy consumption conversion number According to forAs ξ (j)≤ξ₀, construction history energy consumption change data Q₁(i, j)=Q (i, j), that is, keep not Become.

1.5 construction Multiple Non Linear Regression energy consumption forecast model input and output vectors；

Construction input energy consumption column vector Q ' (k), outdoor temperature column vector T ' (k) and outside relative humidity column vector RH ' K (), energy consumption gathers date column vector D (k), energy consumption gathers moment column vector H (k)；

Q ' (k)=[Q₁(1,1),Q₁(1,2),…,Q₁(1,24),Q₁(2,1) ..., Q₁(2,24) ...,

Q₁(D, 1) ..., Q₁(D,24)]^T

T ' (k)=[T (1,1), T (1,2) ..., T (1,24), T (2,1) ..., T (2,24) ...,

T (D, 1) ..., T (D, 24)]^T

RH ' (k)=[RH (1,1), RH (1,2) ..., RH (1,24), RH (2,1) ..., RH (2,24) ...,

RH (D, 1) ..., RH (D, 24)]^T

Construction input matrix INPUT (k)=[T ' (k-1), RH ' (k-1), Q ' (k-1)],

Output matrix Y (k)=[Q ' (k)], 2≤k≤L

1.6 historical data homogenizations are processed

In order to prevent different input/output variables due to physical significance or dimension it is inconsistent and cannot the problem that uses of equality, Need for each modeling input variable to carry out homogenization treatment, including history energy consumption data and outdoor environment dry-bulb temperature and outdoor ring Border relative humidity.

Standardization processing method can use normalization processing method：

Wherein Q " (k), T " (k), RH " (k) is data after standardization；

Point Not Wei each time series minimum value,

The maximum of respectively each time series.

The initialization modeling of 1.7 historical datas

According to the taylor series expansion of the function of many variables, multinomial highest number of times is taken for 2 times.The nonlinear multivariable of selection is returned Return model as follows：

Using least squares estimate, multivariate nonlinear regression analysis model parameter is asked for.

IfQ " (k) is the power consumption values at the k moment,It is the k moment Energy consumption predicted value.

Make function of many variables G (a₀,a₁,…,a₉) to a₀,a₁,…,a₉Partial derivative be zero, i.e.,

Training data is substituted into, the shaping such as form of AX=B is arranged, finally using X=A^-1B, you can try to achieve polynary non- Linear Regression Model Parameters a₀,a₁,…,a₉.Wherein, A is 10 × 10 matrix, and B is 10 × 1 matrixes.

Concrete principle is as follows：

ForTo a_nIt is given value in the partial derivative at k moment.

It is system of linear equations, can be write as matrix form

AX=B,

Wherein X=[a₀,a₁,…,a₉]^T,

Finally utilize X=A^-1B, you can try to achieve multivariate nonlinear regression analysis model parameter a₀,a₁,…,a₉。

2. building energy consumption by when on-line prediction

2.1 online acquisition moment initialized：j₁=1

2.2 judge j₁Whether 24 are less than or equal to

If it is, being updated into the on-line data acquisition of next step 2.3 and data set；

If it is not, then j₁=1, enter back into the on-line data acquisition of next step 2.3 and updated with data set；

2.3 on-line data acquisitions update with data set,

(default data gatherer process is continuous)

Current time is jth₁During the moment, first day jth is abandoned₁The data at moment, Q (1, j₁),T(1,j₁),RH(1,j₁), Online acquisition and in the centrally stored current time data of data, the D+1 days jth₁Time data, including outdoor environment dry-bulb temperature T(D+1,j₁), outdoor environment relative humidity RH (D+1, j₁) and building energy consumption Q (D+1, j₁).Outdoor environment dry-bulb temperature and relative Humidity can be obtained from weather bureau, it is also possible to by sensor Real-time Collection, and the real time data of architectural electric power consumption can be by building Total ammeter collection remote transmission is obtained.

2.4 data cleansings

Energy consumption data Q (D+1, the j gathered using 3 σ regulation analysis current times₁) whether abnormal, if data exception, adopt Former data are substituted with extrapolated value method.

If

P(j₁) it is from the 2nd day to the D+1 days jth₁The energy consumption average value at moment,

σ(j₁) it is from the 2nd day to the D+1 days jth₁The energy consumption standard at moment is poor,

I.e.

If | Q (D+1, j₁)-P(j₁) | the σ (j of ＞ 3₁), then Q (D+1, j₁)=2Q (D, j₁)-Q(D-1,j₁)

Actual energy coefficient online updating is carved in 2.5 timesharing

It is as follows that actual energy coefficient online updating method is carved in timesharing：

P(j₁) it is jth₁Moment energy consumption average value,It is P in 1 to 24 moment The maximum of (j).

2.6 online acquisition data conversions

It is actual with energy coefficient ξ (j using current time₁) carry out current time energy consumption data conversion.Setting ξ₀Actually to use energy The limit value of coefficient, ξ is taken in the present invention₀=0.2, when actual energy coefficient ξ (j are carved in timesharing₁) ＞ ξ₀, building energy consumption change data isAs ξ (j)≤ξ₀, building energy consumption change data Q₁(D+1, j₁)=Q (D+1, j₁), i.e., Keep constant.

The treatment of 2.7 online acquisition data normalizations

Standardization mode is identical with 1.6, including online acquisition energy consumption data, outdoor environment dry-bulb temperature and outdoor ring Border relative humidity.

(1) online acquisition energy consumption data standardization

(2) online acquisition outdoor environment dry-bulb temperature data normalization treatment

(3) online acquisition outdoor environment relative humidity data standardization

2.8 subsequent time Multiple Non Linear Regression energy consumption on-line predictions

2.9 consumption prediction data renormalization treatment：

2.10 consume prediction data timesharing quarters energy data inverse transformation and export

2.11 model online updatings, are that subsequent time prediction is prepared

Construction input energy consumption column vector Q " (k), outdoor temperature column vector T " (k) and outside relative humidity column vector RH " (k)；

Q " (k)=[Q₁(1,j₁+1),Q₁(1,j₁+2),…,Q₁(1,24),Q₁(2,1)

..., Q₁(D, 1) ..., Q₁(D,24),Q₁(D+1,1) ..., Q₁(D+1,j₁)]^T

T " (k)=[T₁(1,j₁+1),T₁(1,j₁+2),…,T₁(1,24),T₁(2,1) ...,

T₁(D, 1) ..., T₁(D,24),T₁(D+1,1) ..., T₁(D+1,j₁)]^T

RH " (k)=[RH₁(1,j₁+1),RH₁(1,j₁+2),…,RH₁(1,24),RH₁(2,1) ...,

RH₁(D, 1) ..., RH₁(D,24),RH₁(D+1,1) ..., RH₁(D+1,j₁)]^T

Construction input matrix INPUT (k)=[T " (k-1), RH " (k-1), Q " (k-1)]

During data substitution 1.7 in Q " (k), Q " (k-1), T " (k-1), RH " (k-1) is saved in A, B matrix, i.e., in following formula：

AX=B,

Wherein X=[a₀,a₁,…,a₉]^T；

ForTo a_nIt is given value in the partial derivative at k moment；

Finally utilize X=A^-1B, you can try to achieve multivariate nonlinear regression analysis model parameter a₀,a₁,…,a₉。

2.12 prepare for subsequent time data acquisition；

j₁=j₁+1

Whether 2.13 determining programs require is terminated,

If it is, EP (end of program)；

If it is not, then going to 2.2 judgement j₁Whether the step of less than or equal to 24, the stream of on-line prediction is then performed again Journey.

By the technical scheme of the present embodiment, the building energy consumption forecast model prediction data reliability of foundation is high, can be used for Building in single building of prediction or big regional extent by when energy consumption, the Energy Saving Control of building energy consumption, building energy consumption prediction and The occasions such as the electric power peak clipping in region.

Above-described embodiment is the present invention preferably implementation method, but embodiments of the present invention are not by above-described embodiment Limitation, it is other it is any without departing from Spirit Essence of the invention and the change, modification, replacement made under principle, combine, simplification, Equivalent substitute mode is should be, is included within protection scope of the present invention.

Claims (10)

1. based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, it is characterised in that including under State step：

S1, data initialization；

S1.1, acquisition historical data, historical data include that outdoor environment dry-bulb temperature T, outdoor environment are relatively wet in a period of time Degree RH and building energy consumption Q by when data；

S1.2, historical data cleaning, carve sequence, using 3 σ regulation analysis each moment by historical data by 24 timesharing are divided into constantly Abnormal data, the exceptional value at each moment is rejected respectively, then the synchronization statistical average with the previous day and one day after is replaced Generation former data, former data are substituted if abnormal data is MARG using extrapolated value method；

S1.3, timesharing quarter reality energy coefficient calculation of initial value, timesharing are carved reality energy coefficient and are defined as follows：

It is jth moment energy consumption average value；For P (j) is most in 1 to 24 moment Big value；

S1.4, historical data conversion, using timesharing quarter reality energy coefficient ξ (j), j=1 ..., 24 carry out historical data conversion, Setting ξ₀It is the limit value of reality energy coefficient, when actual energy coefficient ξ (j) ＞ ξ are carved in timesharing₀, construction history energy consumption change data isAs ξ (j)≤ξ₀, construction history energy consumption change data Q₁(i, j)=Q (i, j), that is, keep constant；

S1.5, construction Multiple Non Linear Regression energy consumption forecast model input and output vector；

S1.6, historical data homogenization treatment, historical data include history energy consumption data and outdoor environment dry-bulb temperature and outdoor Envionmental humidity；

S1.7, historical data modeling, according to the taylor series expansion of the function of many variables, the multivariate nonlinear regression analysis model of selection, Try to achieve multivariate nonlinear regression analysis model parameter；

S2, building energy consumption by when on-line prediction；

S2.1, the initialization of online acquisition moment：j₁=1；

S2.2, judge j₁Whether 24 are less than or equal to

If it is, being updated into next step S2.3 on-line data acquisitions and data set；

If it is not, then j₁=1, enter back into next step S2.3 on-line data acquisitions and updated with data set；

It is jth that S2.3, data set were updated with on-line data acquisition current time₁During the moment, first day jth is abandoned₁The number at moment According to Q (1, j₁),T(1,j₁),RH(1,j₁)

Data set updates, and online acquisition and in the centrally stored current time data of data, the D+1 days jth₁Time data, including Outdoor environment dry-bulb temperature T (D+1, j₁), outdoor environment relative humidity RH (D+1, j₁) and building energy consumption Q (D+1, j₁)；

S2.4, data cleansing, energy consumption data Q (D+1, the j gathered using 3 σ regulation analysis current times₁) whether abnormal, if data It is abnormal, then former data are substituted using extrapolated value method；

S2.5, timesharing carve actual energy coefficient online updating；Update method is as follows：

It is jth₁Moment energy consumption average value,It is P (j) in 1 to 24 moment Maximum；

S2.6, online acquisition data conversion are actual with energy coefficient ξ (j using current time₁) carry out current time energy consumption data turn Change, set ξ₀It is the limit value of reality energy coefficient, when actual energy coefficient ξ (j are carved in timesharing₁) ＞ ξ₀, building energy consumption change data isAs ξ (j)≤ξ₀, building energy consumption change data Q₁(D+1, j₁)=Q (D+1, j₁), i.e., Keep constant；

S2.7, the treatment of online acquisition data normalization, using with standardized way and step S1.6 identical processing modes, including Online acquisition energy consumption data, outdoor environment dry-bulb temperature and outdoor environment relative humidity；

S2.8, subsequent time Multiple Non Linear Regression energy consumption on-line prediction；

S2.9, the treatment of energy consumption prediction data renormalization, using equation below：

It is the D+1 days j₁When+1 Carve energy consumption predicted value；

S2.10, energy consumption prediction data timesharing quarter energy data inverse transformation are simultaneously exported；

$\hat{Q}(D+1,j_1+1)={\hat{Q}}_1(D+1,j_1+1)*\mathrm{\ξ}\left(j_1\right);$

S2.11, forecast model online updating, are that subsequent time prediction is prepared；

Whether S2.12, determining program require is terminated.

2. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, in step S1.2, historical data is carved into sequence Q (i, j) by being divided into 24 timesharing constantly, T (i, j), RH (i, J), j=1,2 ..., 23,24, i=1,2 ..., D, total number of days are D, then set：

P (j) is the energy consumption average value from the 1st day to the D days jth moment,

σ (j) be it is poor to the D days energy consumption standards at jth moment from the 1st day,I.e.

If | Q (i, j)-P (j) | σ (j) of ＞ 3, and 1 ＜ i ＜ D, then Q (i, j)=(Q (i-1, j)+Q (i+1, j))/2

If | Q (i, j)-P (j) | ＞ 3 σ (j), i=1, then Q (i, j)=2Q (i+1, j)-Q (i+2, j)

If | Q (i, j)-P (j) | ＞ 3 σ (j), i=D, then Q (i, j)=2Q (i-1, j)-Q (i-2, j).

3. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, in step S1.5, specially：

Construction input energy consumption column vector Q ' (k), outdoor temperature column vector T ' (k) and outside relative humidity column vector RH ' (k), energy Consumption collection date column vector D (k), energy consumption gathers moment column vector H (k)；

Q ' (k)=[Q₁(1,1),Q₁(1,2),…,Q₁(1,24),Q₁(2,1) ..., Q₁(2,24) ...,

Q₁(D, 1) ..., Q₁(D,24)]^T

T ' (k)=[T (1,1), T (1,2) ..., T (1,24), T (2,1) ..., T (2,24) ...,

T (D, 1) ..., T (D, 24)]^T

RH ' (k)=[RH (1,1), RH (1,2) ..., RH (1,24), RH (2,1) ..., RH (2,24) ...,

RH (D, 1) ..., RH (D, 24)]^T

Construction input matrix INPUT (k)=[T ' (k-1), RH ' (k-1), Q ' (k-1)],

Output matrix Y (k)=[Q ' (k)], 2≤k≤L.

4. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, in step S1.6, the method for uniforming treatment is：

$Q^{\′\′}\left(k\right)=\frac{Q^{\′}\left(k\right)-\min Q}{\max Q-\min Q}$

$T^{\′\′}\left(k\right)=\frac{T^{\′}\left(k\right)-\min T}{\max T-\min T}$

${\mathrm{RH}}^{\′\′}\left(k\right)=\frac{{\mathrm{RH}}^{\′}\left(k\right)-\min RH}{\max RH-\min RH}$

Wherein Q " (k), T " (k), RH " (k) is data after standardization；

Respectively The minimum value of each time series,

Respectively It is the maximum of each time series.

5. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, step S1.7 is specially：

According to the taylor series expansion of the function of many variables, it is 2 times, the Multiple Non Linear Regression mould of selection to take multinomial highest number of times Type is as follows：

$\begin{array}{c}{\hat{Q}}^{\′\′}\left(k\right)=f\[a_0,a_1,\mathrm{...},a_9,T^{\′\′}(k-1),{\mathrm{RH}}^{\′\′}(k-1),Q^{\′\′}(k-1)\]\\ =a_0+\left[\begin{array}{ccc}{a}_1& a_2& a_3\end{array}\right]{\left[\begin{array}{ccc}{T}^{\′\′}(k-1)& {\mathrm{RH}}^{\′\′}(k-1)& Q^{\′\′}(k-1)\end{array}\right]}^T\\ +\left[\begin{array}{ccc}{a}_4& a_5& a_6\end{array}\right]{\left[\begin{array}{ccc}{T}^{\′\′2}(k-1)& {\mathrm{RH}}^{\′\′2}(k-1)& Q^{\′\′2}(k-1)\end{array}\right]}^T\\ +\left[\begin{array}{ccc}{a}_7& a_8& a_9\end{array}\right]{\left[\begin{array}{ccc}{T}^{\′\′}(k-1)\·{\mathrm{RH}}^{\′\′}(k-1)& T^{\′\′}(k-1)\·Q^{\′\′}(k-1)& {\mathrm{RH}}^{\′\′}(k-1)\·Q^{\′\′}(k-1)\end{array}\right]}^T\end{array}$

Using least squares estimate, multivariate nonlinear regression analysis model parameter is asked for；

IfIt is the power consumption values at the k moment,It is the energy consumption at k moment Predicted value；

Make function of many variables G (a₀,a₁,…,a₉) to a₀,a₁,…,a₉Partial derivative be zero, i.e.,

$\frac{\∂G(a_0,a_1,...,a_9)}{\∂a_n}=0,n=0,1,...,9.$

Training data is substituted into, the shaping such as form of AX=B is arranged, finally using X=A^-1B, you can try to achieve nonlinear multivariable Parameters in Regression Model a₀,a₁,…,a₉, wherein, A is 10 × 10 matrix, and B is 10 × 1 matrixes.

6. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, the step S2.3 is specially：

Data set updates, i.e., terminate within D days to the since the 1st day, by building energy consumption value, the outdoor at i+1 day whole day each moment Environment dry-bulb temperature value and outdoor environment rh value are assigned to i-th day building energy consumption of synchronization, outdoor environment dry bulb temperature Degree and outdoor environment relative humidity；The step S2.4 is specially：

If

P(j₁) it is from the 2nd day to the D+1 days jth₁The energy consumption average value at moment,

σ(j₁) it is from the 2nd day to the D+1 days jth₁The energy consumption standard at moment is poor,

$\mathrm{\σ}\left(j_1\right)=\sqrt{\frac{1}{D}\underset{i=2}{\overset{D+1}{\Σ}}{\left(Q\right(i_1,j_1)-P(j_1\left)\right)}^2}$

I.e.

If | Q (D+1, j₁)-P(j₁) | the σ (j of ＞ 3₁), then Q (D+1, j₁)=2Q (D, j₁)-Q(D-1,j₁)。

7. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, step S2.7 is specially：

(1) online acquisition energy consumption data standardization

${Q_1}^{\′}(D+1,j_1)=\frac{Q_1(D+1,j_1)-\min Q}{\max Q-\min Q}$

(2) online acquisition outdoor environment dry-bulb temperature data normalization treatment

$T^{\′}(D+1,j_1)=\frac{T(D+1,j_1)-\min T}{\max T-\min T}$

(3) online acquisition outdoor environment relative humidity data standardization

${\mathrm{RH}}^{\′}(D+1,j_1)=\frac{RH(D+1,j_1)-\min RH}{\max RH-\min RH}.$

8. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, step S2.8 is specially：

$\begin{array}{c}{{\hat{Q}}_1}^{\′}(D+1,j_1+1)=f\[a_0,a_1,\mathrm{...},a_9,T^{\′}(D+1,j_1),{\mathrm{RH}}^{\′}(D+1,j_1),Q_1(D+1,j_1)\]\\ =a_0+\left[\begin{array}{ccc}{a}_1& a_2& a_3\end{array}\right]{\left[\begin{array}{ccc}{T}^{\′}(D+1,j_1)& {\mathrm{RH}}^{\′}(D+1,j_1)& Q^{\′}(D+1,j_1)\end{array}\right]}^T\\ +\left[\begin{array}{ccc}{a}_4& a_5& a_6\end{array}\right]{\left[\begin{array}{ccc}{T}^{\′2}(D+1,j_1)& {\mathrm{RH}}^{\′2}(D+1,j_1)& Q^{\′2}(D+1,j_1)\end{array}\right]}^T\\ +\left[\begin{array}{ccc}{a}_7& a_8& a_9\end{array}\right]{\left[\begin{array}{ccc}{T}^{\′}(D+1,j_1)\·{\mathrm{RH}}^{\′}(D+1,j_1)& T^{\′}(D+1,j_1)\·Q^{\′}(D+1,j_1)& {\mathrm{RH}}^{\′}(D+1,j_1)\·Q^{\′}(D+1,j_1)\end{array}\right]}^T.\end{array}$

9. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, step S2.11 is specially：

Construction input energy consumption column vector Q " (k), outdoor temperature column vector T " (k) and outside relative humidity column vector RH " (k)；

Q " (k)=[Q₁(1,j₁+1),Q₁(1,j₁+2),…,Q₁(1,24),Q₁(2,1)

..., Q₁(D, 1) ..., Q₁(D,24),Q₁(D+1,1) ..., Q₁(D+1,j₁)]^T

T " (k)=[T₁(1,j₁+1),T₁(1,j₁+2),…,T₁(1,24),T₁(2,1) ...,

T₁(D, 1) ..., T₁(D,24),T₁(D+1,1) ..., T₁(D+1,j₁)]^T

RH " (k)=[RH₁(1,j₁+1),RH₁(1,j₁+2),…,RH₁(1,24),RH₁(2,1) ...,

RH₁(D, 1) ..., RH₁(D,24),RH₁(D+1,1) ..., RH₁(D+1,j₁)]^T

Construction input matrix INPUT (k)=[T " (k-1), RH " (k-1), Q " (k-1)]

By in data substitution A, B matrix in Q " (k), Q " (k-1), T " (k-1), RH " (k-1), that is, substitute into

In AX=B, finally using X=A^-1B, you can try to achieve multivariate nonlinear regression analysis model parameter a₀,a₁,…,a₉。

10. according to claim 1 based on timesharing carve it is actual with can coefficient heavy construction by when energy consumption on-line prediction method, Characterized in that, step S2.12 is specially：

j₁=j₁+1

Judge whether to receive EP (end of program) instruction,

If it is, EP (end of program)；

If it is not, then going to S2.2 judges j₁Whether the step of less than or equal to 24, the flow of on-line prediction is then performed again.