CN112465256A - Building power consumption prediction method and system based on Stacking model fusion - Google Patents
Building power consumption prediction method and system based on Stacking model fusion Download PDFInfo
- Publication number
- CN112465256A CN112465256A CN202011443441.0A CN202011443441A CN112465256A CN 112465256 A CN112465256 A CN 112465256A CN 202011443441 A CN202011443441 A CN 202011443441A CN 112465256 A CN112465256 A CN 112465256A
- Authority
- CN
- China
- Prior art keywords
- power consumption
- model
- stacking
- building
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 230000004927 fusion Effects 0.000 title claims abstract description 29
- 230000010354 integration Effects 0.000 claims abstract description 29
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims description 28
- 230000005611 electricity Effects 0.000 claims description 26
- 238000007637 random forest analysis Methods 0.000 claims description 9
- 238000012706 support-vector machine Methods 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 5
- 230000002159 abnormal effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 3
- 230000008901 benefit Effects 0.000 abstract description 12
- 230000000694 effects Effects 0.000 abstract description 8
- 238000005265 energy consumption Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 8
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000002790 cross-validation Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 2
- 230000015654 memory Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 230000009193 crawling Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a building power consumption prediction method and system based on Stacking model fusion, and belongs to the field of building power consumption prediction. According to the method, a Stacking model fusion algorithm is adopted to integrate multiple regression models, a power consumption Stacking integration model is constructed, the advantages of multiple models are integrated, and prediction deviation is reduced; aiming at buildings with unstable power consumption, various influence factors such as historical power consumption, temperature, wind power, humidity, time information and the like are utilized, a power consumption Stacking integrated model is trained, the prediction accuracy is improved, the building manager can effectively control the building energy consumption, the situation that the power consumption is greatly different from the predicted power consumption is avoided, the building manager can reasonably predict and purchase when participating in electric power market transaction, the building manager can effectively control the power charge expenditure, the power selling arrangement of an electric power department or an electric power selling company is facilitated, the energy-saving and emission-reducing effects can be achieved, and good social benefits and economic benefits are achieved.
Description
Technical Field
The invention belongs to the field of building power consumption prediction, and particularly relates to a building power consumption prediction method and system based on Stacking model fusion.
Background
The monthly electricity consumption prediction of the building belongs to a time series prediction type. The time sequence is a group of random variables which are dependent on time, and the group of variables have dependency relations, and the correlation characteristics indicate the continuity of the development of the prediction object. The self-correlation characteristics contained in the time series are described by a mathematical model, so that the past value and the present value of the time series can be used for predicting the future value.
In the prior art of predicting monthly electricity consumption of buildings, a prediction means mostly adopts a multivariate regression means to predict, and multivariate modeling is carried out on local residency and living standard, so that the predicted residual sum is minimum to obtain a model. Or an SVM algorithm is used for carrying out certain approximation, and the predicted influence factors are projected onto different dimensions, so that a monthly power consumption system which is more unstable than multivariate regression can be predicted. There is also an autoregressive moving average model (ARMA) through time series, because the ARMA model only focuses on the influence of the time series, neglects many interference factors and limits the effectiveness in the prediction of the actual monthly power consumption, such as temperature, weather, whether it belongs to holidays, etc.
The methods have various advantages and various defects, and generally have higher requirements on the accuracy of historical data and better prediction effect in areas with stable power consumption, but have larger prediction errors under the action of exogenous factors (weather change, holidays).
Taking an office building as an example, an office building power system is a complex real-time dynamic system, and the stability and the safety of the office building power system in the operation process are greatly reduced due to complex and changeable influence factors. Meanwhile, the electricity consumption of an office building is different from that of common residents, and is influenced by various factors such as seasons, temperature, holidays, external environment and the like, for example, when the seasons alternate, the electricity consumption fluctuates greatly, which is mainly because of the heating demand in winter and the cooling demand in summer; for example, the size of the indoor space and the number of active personnel can influence the indoor temperature; office building equipment suffers loss over time, etc. Therefore, the office building power system is influenced by various factors such as seasons, holidays, the number of people in the office building, equipment loss and the like, and the monthly power consumption of the power system is not completely stable, so that the advantages of various models cannot be combined by using a traditional single model to predict the power consumption, the prediction result has great deviation with the actual power consumption, and the model cannot have strong generalization capability.
Disclosure of Invention
Aiming at the defects or improvement requirements in the prior art, the invention provides a building power consumption prediction method and system based on Stacking model fusion, and aims to solve the technical problem that the existing method has a large prediction error in building power consumption prediction under the condition of unstable power use state.
In order to achieve the above object, according to an aspect of the present invention, there is provided a building power consumption prediction method based on Stacking model fusion, including:
s1, collecting temperature, wind power, humidity and time information of a building to be predicted in a historical time period and power consumption data of a corresponding time period as a training set;
s2, constructing a power consumption Stacking integration model; the power consumption Stacking integration model is obtained by integrating a plurality of regression models by adopting a Stacking model fusion algorithm, and comprises a first layer and a second layer; each model in the first layer obtains an initial power consumption predicted value according to input temperature, wind power, humidity and time information; the second layer corrects the initial power consumption predicted value output by the first layer to obtain a final power consumption predicted value;
s3, taking temperature, wind power, humidity and time information of a building historical period to be predicted as input, taking power consumption data of a corresponding period as expected output, and training a power consumption Stacking integration model;
and S4, inputting the temperature, wind power, humidity and time information corresponding to the future monthly degree to be predicted into the trained power consumption Stacking integrated model to obtain a power consumption prediction result corresponding to the monthly degree.
Further, before training the power consumption Stacking integration model, the method further comprises the following preprocessing of the training set: filling missing data; rejecting abnormal and repeated data; extracting at the lowest temperature and the highest temperature; extracting time characteristics; and carrying out dimensionless processing on various types of data.
Further, the time characteristics include week, hour, season, and holiday information.
Further, the first layer of the power consumption Stacking integration model comprises a random forest, KNN, LSTM and LightGBM.
Furthermore, a support vector machine regression model is adopted in the second layer of the power consumption Stacking integration model.
According to another aspect of the invention, a building power consumption prediction system based on Stacking model fusion is provided, which comprises:
the data set collection module is used for collecting temperature, wind power, humidity and time information of a building to be predicted in a historical time period and power consumption data of a corresponding time period as a training set;
the model building module is used for building a power consumption Stacking integration model; the power consumption Stacking integration model is obtained by integrating a plurality of regression models by adopting a Stacking model fusion algorithm, and comprises a first layer and a second layer; each model in the first layer obtains an initial power consumption predicted value according to input temperature, wind power, humidity and time information; the second layer corrects the initial power consumption predicted value output by the first layer to obtain a final power consumption predicted value;
the training module is used for training the electricity consumption Stacking integrated model by taking temperature, wind power, humidity and time information of a building historical period to be predicted as input and taking electricity consumption data of a corresponding period as expected output;
and the prediction module is used for inputting the temperature, wind power, humidity and time information corresponding to the future monthly degrees to be predicted into the trained power consumption Stacking integrated model to obtain the power consumption prediction result corresponding to the monthly degrees.
Further, the first layer of the power consumption Stacking integration model comprises a random forest, KNN, LSTM and LightGBM.
Furthermore, a support vector machine regression model is adopted in the second layer of the power consumption Stacking integration model.
In general, the above technical solutions contemplated by the present invention can achieve the following advantageous effects compared to the prior art.
(1) According to the method, a Stacking model fusion algorithm is adopted to integrate multiple regression models, a power consumption Stacking integration model is constructed, the advantages of multiple models are integrated, and prediction deviation is reduced; and aiming at buildings with unstable power consumption (such as large commercial power utilization facilities of office buildings, markets, schools and the like), various influence factors such as historical power consumption, temperature, holidays, season information and the like are utilized, a power consumption Stacking integrated model is trained, monthly power consumption is obtained by predicting daily power consumption, the prediction period is shortened, and the prediction accuracy is improved.
(2) The method can accurately predict the monthly electricity consumption and the predicted electricity consumption of the building, is more beneficial to a building manager to effectively control the energy consumption of the building, avoids the condition that the difference between the electricity consumption and the predicted electricity consumption is too large, reasonably predicts and purchases the electricity when participating in the electric power market transaction, enables the building manager to effectively control the electricity expense, is convenient for the electricity selling arrangement of an electric power department or an electricity selling company, can achieve the effects of energy conservation and emission reduction, and has good social benefit and economic benefit.
Drawings
FIG. 1 is a schematic diagram of a web crawler according to the present invention;
FIG. 2 is a diagram of a data preprocessing process provided by the present invention;
FIG. 3 is a schematic diagram of an electric power consumption Stacking integration model provided by the present invention;
FIG. 4 is a specific flowchart of the power consumption Stacking integration model training provided by the present invention;
FIG. 5 is a schematic diagram of a random forest model provided by the present invention;
FIG. 6 is a schematic diagram of a KNN model provided by the present invention;
FIG. 7 is a block diagram of an LSTM loop module provided by the present invention;
FIG. 8 is a schematic diagram of a GBDT model provided by the present invention;
FIG. 9 is a process diagram for power usage prediction provided by the method of the present invention;
FIG. 10 is a graph comparing the predicted monthly electricity consumption of an office building with real data;
FIG. 11 is a histogram comparing the monthly electricity consumption prediction model indexes of an office building.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The invention provides a building power consumption prediction method based on Stacking model fusion, which comprises the following steps of:
s1, collecting temperature, wind power, humidity and time information of a building to be predicted in a historical time period and power consumption data of a corresponding time period;
because the temperature is known to have great influence on the electricity consumption, the local weather data of the building needs to be crawled by using the web crawler and the current electricity consumption data together as a data set for subsequent model training, the weather data needing to be crawled generally has the highest temperature, the lowest temperature, wind power and humidity, and the web crawler principle is shown in fig. 1.
The method comprises the steps of combining power consumption data and weather data crawled by a crawler, dividing time characteristics, filling missing data, detecting and eliminating abnormal values and dimensionless. Referring to fig. 2, the operation process of each part is as follows: merging the power consumption data and the weather data, and merging the power consumption data and the weather data according to time; the time division characteristics divide the common date and time into the characteristics of the week number, the month, the season, the holidays and the like. The season information comprises spring, summer, autumn and winter, and the holiday information comprises weekends, holidays of 3 days or more, holidays of 7 days or more and spring holidays; filling missing data, and filling missing values by adopting random forest regression; detecting and eliminating abnormal values, drawing the noise of a boxed graph detection data set and directly eliminating the noise; dimensionless, the features were normalized using minmaxscale by sklern.
The following are partial data before and after preprocessing.
TABLE 1 partial data before pretreatment
| Time | Maximum temperature | Lowest temperature | Wind power | Humidity | Electric quantity |
| 2019-6-1 00:00:00 | 23 | 19 | 8 | 43 | 55.7 |
| 2019-6-1 01:00:00 | 23 | 19 | 8 | 43 | 57.0 |
| 2019-6-1 02:00:00 | 23 | 19 | 8 | 43 | 54.9 |
| 2019-6-1 03:00:00 | 23 | 19 | 8 | 43 | 58.6 |
| 2019-6-1 04:00:00 | 23 | 19 | 8 | 43 | 59.3 |
TABLE 2 partial data after pretreatment
Table 1 shows that the combined raw data is collected by the crawler in this experiment, the time is not subjected to characteristic division, and the characteristics of the highest temperature, the lowest temperature, the wind power and the humidity are not subjected to dimensionless method; and table 2 is training data after the experiment preprocessing, time is divided into hours, weeks, seasons, holidays, and features of the highest temperature, the lowest temperature, wind power and humidity are dimensionless, so that a training set has more available features, and the influence of different feature units on feature importance is reduced.
S2, constructing a power consumption Stacking integration model; the power consumption Stacking integration model is obtained by integrating a plurality of regression models by adopting a Stacking model fusion algorithm, and comprises a first layer and a second layer; each model in the first layer obtains an initial power consumption prediction value according to input information such as temperature, wind power, humidity, week, month, season, holidays and the like; the second layer corrects the initial power consumption predicted value output by the first layer to obtain a final power consumption predicted value;
the power consumption Stacking integration model is obtained by integrating a plurality of regression models through a Stacking model fusion algorithm; as shown in fig. 3, the model is divided into two layers, the first layer uses various regression models to learn and predict input information such as temperature, holidays, seasons and the like, the prediction result is used as input data of the second layer, the second layer uses a relatively simple regression model to reduce the risk of overfitting, and the prediction result is corrected according to actual power consumption to obtain a final prediction result; in the embodiment of the present invention, the first layer uses a random forest, KNN, LSTM, LightGBM; since the first layer already uses a large number of models with higher complexity, in order to avoid overfitting of the final model, the second layer model should try to select a single model with larger difference from the first layer. In the first layer, both the random forest and the LightGBM are tree models, and the LSTM belongs to a recurrent neural network model, so that a support vector machine regression model different from all models in the first layer is selected in the second layer, and the support vector machine adopts a structural risk minimization criterion to design a learning machine, so that the empirical risk and the confidence range are considered in a compromise manner, and the method has better popularization capability; the support vector machine is specially used for the limited sample case, the aim is to obtain the optimal solution under the existing information, not only the optimal solution when the number of samples tends to be infinite, but also the number of samples and the number of models used in the first layer in the invention are limited.
The following briefly introduces each base learner selected by the first layer in the Stacking integration model of the present invention:
1) a random forest is a supervised learning algorithm, which in short builds multiple decision trees and combines them together to obtain a more accurate and stable prediction, which can be used both for classification and for regression problems, the principle of which is shown in fig. 5.
2) KNN (K-Nearest Neighbor) is also a machine learning algorithm that can be used for both classification and regression. The KNN regression algorithm is to establish a vector space model, find K training samples and use the average value of the characteristics of the training samples for the characteristics to be predicted by using an averaging method. Referring to fig. 6, the basic flow is that, first, the training set and the test set are divided, then, the euclidean distance between the sample data and the prediction sample is calculated, and finally, the euclidean distances are listed from small to large, and the training data in the first K ranks are removed, so that the average value of the training data is calculated, that is, the final prediction value. The KNN algorithm has the advantages of simplicity and good generalization capability.
3) The LSTM (Long Short Term memory) refers to a long Short Term memory network, is a special RNN (recurrent neural network) algorithm, makes up for the defects of the traditional RNN, and introduces a three-layer door mechanism, namely a forgetting door, an input door and an output door. The input gate mainly plays a role in how information of an input layer is transmitted to the cell unit, the forgetting gate plays a role in selectively memorizing historical information, and the output gate plays a role in manipulating output data. In dealing with the time series problem, the LSTM has a stronger generalization ability, performs selective memory and forgetting, and has stronger coordination and stability than the conventional RNN, and the structure of the LSTM loop module is shown in fig. 7.
4) LightGBM is a distributed gradient spanning tree algorithm open by Microsoft corporation, and when facing massive data and data with good characteristic dimensions, the LightGBM has quicker training, less memory usage and excellent accuracy than XGBoost, and the model can be used for processing classification and regression problems. The LightGBM algorithm is an improvement over XGboost from GBDT, the principle of GBDT is shown in fig. 8.
S3, taking temperature, wind power, humidity, week, month, season and holiday information of the historical time period of the building to be predicted as input, taking power consumption data of the corresponding time period as expected output, and training a power consumption Stacking integration model;
as shown in fig. 4, the power consumption Stacking integration model training mode is as follows:
for a data set:
S＝{(yn,xn),n＝1,…,N} (1)
(1) in the formula xnIs the feature vector of the nth sample, ynIs the predicted value of the nth sample, and N is the number of included features.
The original data is divided into K subsets using K-fold cross-validation, the subsets being equal in size:
S＝{S1,S2,…,Sk} (2)
dividing K times from the subsets divided in (2) and taking each subset as a test set ScThe other subset being a training set
Inputting a training set to obtain trained base models for K base learners in the first layer:
M＝{Mk,k＝1,…,K} (3)
for the K-fold test set S in the K-fold cross validationcEach sample x innRadical learning machine MkFor which the predicted result is rkn. After the cross validation is completed, the output data of the K base learners are formed into a new data set:
S′＝{(yn,r1n,…,rkn),n＝1,…,N} (4)
the new data set is used as a second layer data set of the Stacking integration model, induction learning is carried out by using a support vector regression model of the second layer, finally, the advantages of all models of the first layer can be fully exerted, and the prediction error of each model of the first layer is reduced.
S4, crawling the weather data of one month from the weather website, and inputting the temperature, wind power, humidity, week, month, season and holiday information corresponding to the future time period to be predicted into the trained power consumption Stacking integrated model to obtain the power consumption prediction result of the corresponding time period.
The method of the invention is completely implemented, and refer to fig. 9.
The prediction and evaluation index of the power consumption Stacking integrated model adopts the average absolute percentage error (MAPE), and the expression is as follows:
(5) in the formula XiFor the actual daily electricity consumption, YiTo predict the daily electricity consumption; and finally, counting the accuracy of the predicted total power consumption and the actual total power consumption of the predicted month as a final evaluation index, wherein the corresponding mathematical expression is as follows:
wherein, YsTo predict the actual total electricity consumption of a month, PsA total power usage is predicted for the current month of the predicted month.
To verify the effectiveness of the method of the present invention, the monthly electricity consumption of the power system of the office building 2020 from 8/month 1 to 8/month 31 is predicted, and the comparison result between the electricity consumption prediction result and the real data is shown in fig. 10, wherein the abscissa represents the time series from 8/month 1 to 2020/month 31 in the year in the hour unit, the ordinate is the electricity consumption in the hour unit, the absolute average percentage error is calculated to be 10.5% in the hour unit, and the absolute average percentage error is calculated to be 1.1% in the month unit. Therefore, the method is more suitable for monthly electric quantity prediction, and the effectiveness of the method is proved.
In order to further verify the effectiveness of the method of the present invention, the present embodiment also performs a lateral comparison with the prediction efficiency of other various single models. The models used for comparison are GBDT, LSTM and ARIMA, wherein GBDT is a tree model, LSTM is a recurrent neural network model, ARIMA is a time sequence model, and the used test set is the monthly power consumption of a certain office building 2020 from 8 months 1 to 8 months 31. The model index comparison histogram of each model is shown in fig. 11, the evaluation index in the histogram is 1-MAPE, and it can be seen from the graph that the generalization capability of the Stacking integration model used in the method is optimal, then the tree model represented by GBDT, then the LSTM neural network model, and the least effective is the ARIMA model based on time series. The reason is mainly that ARIMA is a model based on time series prediction, and the effect is the worst because the influence of meteorological factors on power consumption is not considered, but the prediction effect and the stability of LSTM and GBDT are not as good as model fusion.
The method is different from other methods mainly in the following two aspects. Firstly, the effect of the Stacking model fusion is better than that of single model modeling prediction, because the Stacking model fusion is combined with various algorithms, the advantages of the algorithms are fully exerted, data space and structure are observed from different angles, and the condition that a single model is locally optimal is avoided. Secondly, the Stacking model fusion method used in the method abandons the conventional method of fusion by adopting a plurality of algorithms with higher similarity, and through experimental comparison, the algorithm fusion with high difference and strong learning capability is adopted for optimization, so that the prediction effect of the Stacking model fusion can reach the optimum.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (8)
1. A building power consumption prediction method based on Stacking model fusion is characterized by comprising the following steps:
s1, collecting temperature, wind power, humidity and time information of a building to be predicted in a historical time period and power consumption data of a corresponding time period as a training set;
s2, constructing a power consumption Stacking integration model; the power consumption Stacking integration model is obtained by integrating a plurality of regression models by adopting a Stacking model fusion algorithm, and comprises a first layer and a second layer; each model in the first layer obtains an initial power consumption predicted value according to input temperature, wind power, humidity and time information; the second layer corrects the initial power consumption predicted value output by the first layer to obtain a final power consumption predicted value;
s3, taking temperature, wind power, humidity and time information of a building historical period to be predicted as input, taking power consumption data of a corresponding period as expected output, and training a power consumption Stacking integration model;
and S4, inputting the temperature, wind power, humidity and time information corresponding to the future monthly degree to be predicted into the trained power consumption Stacking integrated model to obtain a power consumption prediction result corresponding to the monthly degree.
2. The building power consumption prediction method based on the Stacking model fusion as claimed in claim 1, wherein before the training of the power consumption Stacking integration model, the method further comprises the following pre-processing of the training set: filling missing data; rejecting abnormal and repeated data; extracting at the lowest temperature and the highest temperature; extracting time characteristics; and carrying out dimensionless processing on various types of data.
3. The building power consumption prediction method based on Stacking model fusion as claimed in claim 2, wherein the time characteristics include week, hour, season and holiday information.
4. The building power consumption prediction method based on Stacking model fusion as claimed in claim 1, wherein the first layer of the power consumption Stacking integration model comprises random forest, KNN, LSTM, LightGBM.
5. The building power consumption prediction method based on the Stacking model fusion as claimed in claim 4, characterized in that the Stacking integrated model second layer adopts support vector machine regression model.
6. A building power consumption prediction system based on Stacking model fusion is characterized by comprising:
the data set collection module is used for collecting temperature, wind power, humidity and time information of a building to be predicted in a historical time period and power consumption data of a corresponding time period as a training set;
the model building module is used for building a power consumption Stacking integration model; the power consumption Stacking integration model is obtained by integrating a plurality of regression models by adopting a Stacking model fusion algorithm, and comprises a first layer and a second layer; each model in the first layer obtains an initial power consumption predicted value according to input temperature, wind power, humidity and time information; the second layer corrects the initial power consumption predicted value output by the first layer to obtain a final power consumption predicted value;
the training module is used for training the electricity consumption Stacking integrated model by taking temperature, wind power, humidity and time information of a building historical period to be predicted as input and taking electricity consumption data of a corresponding period as expected output;
and the prediction module is used for inputting the temperature, wind power, humidity and time information corresponding to the future monthly degrees to be predicted into the trained power consumption Stacking integrated model to obtain the power consumption prediction result corresponding to the monthly degrees.
7. The system of claim 6, wherein the first layer of the Stacking model comprises random forest, KNN, LSTM, LightGBM.
8. The system for predicting power consumption of buildings based on Stacking model fusion as claimed in claim 7, wherein the Stacking integrated model second layer adopts support vector machine regression model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011443441.0A CN112465256A (en) | 2020-12-08 | 2020-12-08 | Building power consumption prediction method and system based on Stacking model fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011443441.0A CN112465256A (en) | 2020-12-08 | 2020-12-08 | Building power consumption prediction method and system based on Stacking model fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112465256A true CN112465256A (en) | 2021-03-09 |
Family
ID=74801791
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011443441.0A Pending CN112465256A (en) | 2020-12-08 | 2020-12-08 | Building power consumption prediction method and system based on Stacking model fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112465256A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113327029A (en) * | 2021-05-27 | 2021-08-31 | 天津安捷物联科技股份有限公司 | Office building energy strategy generation method and system |
| CN113344254A (en) * | 2021-05-20 | 2021-09-03 | 山西省交通新技术发展有限公司 | Method for predicting traffic flow of expressway service area based on LSTM-LightGBM-KNN |
| CN113505537A (en) * | 2021-07-27 | 2021-10-15 | 上海上实龙创智能科技股份有限公司 | Building energy consumption detection method and device, computer equipment and storage medium |
| CN114004425A (en) * | 2021-12-29 | 2022-02-01 | 北京京东振世信息技术有限公司 | Article circulation information prediction model generation method, information generation method and device |
| CN114035629A (en) * | 2021-10-20 | 2022-02-11 | 科大讯飞股份有限公司 | Control method of drying equipment and related equipment |
| CN114529071A (en) * | 2022-02-11 | 2022-05-24 | 杭州致成电子科技有限公司 | Method for predicting power consumption of transformer area |
| CN115222150A (en) * | 2022-08-17 | 2022-10-21 | 南京航空航天大学 | Stacking-based flight delay prediction method and system |
| CN116526479A (en) * | 2023-07-03 | 2023-08-01 | 国网北京市电力公司 | Method, device, equipment and medium for predicting power supply quantity |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245798A (en) * | 2019-06-18 | 2019-09-17 | 天津安捷物联科技股份有限公司 | A kind of monthly electricity demand forecasting method and system of office building electric system |
| CN110503251A (en) * | 2019-08-12 | 2019-11-26 | 江苏方天电力技术有限公司 | A kind of non-festivals or holidays load forecasting method based on Stacking algorithm |
-
2020
- 2020-12-08 CN CN202011443441.0A patent/CN112465256A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245798A (en) * | 2019-06-18 | 2019-09-17 | 天津安捷物联科技股份有限公司 | A kind of monthly electricity demand forecasting method and system of office building electric system |
| CN110503251A (en) * | 2019-08-12 | 2019-11-26 | 江苏方天电力技术有限公司 | A kind of non-festivals or holidays load forecasting method based on Stacking algorithm |
Non-Patent Citations (4)
| Title |
|---|
| 史佳琪等: "基于多模型融合Stacking集成学习方式的负荷预测方法", 《中国电机工程学报》, vol. 39, no. 14, 20 July 2019 (2019-07-20), pages 4032 - 4041 * |
| 李昆明等: "基于利用BP神经网络进行Stacking模型融合算法的电力非节假日负荷预测研究", 《软件》 * |
| 李昆明等: "基于利用BP神经网络进行Stacking模型融合算法的电力非节假日负荷预测研究", 《软件》, vol. 40, no. 9, 30 September 2019 (2019-09-30), pages 176 - 181 * |
| 陆泉等: "基于大数据挖掘的医疗健康公共服务", 武汉大学出版社, pages: 450 - 456 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113344254A (en) * | 2021-05-20 | 2021-09-03 | 山西省交通新技术发展有限公司 | Method for predicting traffic flow of expressway service area based on LSTM-LightGBM-KNN |
| CN113327029A (en) * | 2021-05-27 | 2021-08-31 | 天津安捷物联科技股份有限公司 | Office building energy strategy generation method and system |
| WO2023005120A1 (en) * | 2021-07-27 | 2023-02-02 | 上海上实龙创智能科技股份有限公司 | Energy consumption prediction method and apparatus for building, and computer device and storage medium |
| CN113505537A (en) * | 2021-07-27 | 2021-10-15 | 上海上实龙创智能科技股份有限公司 | Building energy consumption detection method and device, computer equipment and storage medium |
| CN114035629A (en) * | 2021-10-20 | 2022-02-11 | 科大讯飞股份有限公司 | Control method of drying equipment and related equipment |
| CN114035629B (en) * | 2021-10-20 | 2023-04-07 | 科大讯飞股份有限公司 | Drying equipment control method and related equipment |
| CN114004425A (en) * | 2021-12-29 | 2022-02-01 | 北京京东振世信息技术有限公司 | Article circulation information prediction model generation method, information generation method and device |
| CN114004425B (en) * | 2021-12-29 | 2022-06-07 | 北京京东振世信息技术有限公司 | Article circulation information prediction model generation method, information generation method and device |
| CN114529071A (en) * | 2022-02-11 | 2022-05-24 | 杭州致成电子科技有限公司 | Method for predicting power consumption of transformer area |
| CN115222150A (en) * | 2022-08-17 | 2022-10-21 | 南京航空航天大学 | Stacking-based flight delay prediction method and system |
| CN115222150B (en) * | 2022-08-17 | 2024-02-20 | 南京航空航天大学 | Flight delay prediction method and system based on Stacking |
| CN116526479A (en) * | 2023-07-03 | 2023-08-01 | 国网北京市电力公司 | Method, device, equipment and medium for predicting power supply quantity |
| CN116526479B (en) * | 2023-07-03 | 2023-10-13 | 国网北京市电力公司 | Method, device, equipment and medium for predicting power supply quantity |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112465256A (en) | Building power consumption prediction method and system based on Stacking model fusion | |
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Almalaq et al. | A review of deep learning methods applied on load forecasting | |
| Gonzalez-Briones et al. | Machine learning models for electricity consumption forecasting: a review | |
| Kalogirou et al. | Artificial intelligence techniques in solar energy applications | |
| Peng et al. | Multi-source transfer learning guided ensemble LSTM for building multi-load forecasting | |
| Sengar et al. | Ensemble approach for short term load forecasting in wind energy system using hybrid algorithm | |
| Kaboli et al. | An expression-driven approach for long-term electric power consumption forecasting | |
| CN113240153A (en) | Photovoltaic power generation data prediction method and device, computing equipment and storage medium | |
| Chen et al. | MultiCycleNet: multiple cycles self-boosted neural network for short-term electric household load forecasting | |
| Čurčić et al. | Gaining insights into dwelling characteristics using machine learning for policy making on nearly zero-energy buildings with the use of smart meter and weather data | |
| CN114595861A (en) | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method | |
| Wang et al. | Short-term wind power probabilistic forecasting using a new neural computing approach: GMC-DeepNN-PF | |
| CN114240687A (en) | Energy hosting efficiency analysis method suitable for comprehensive energy system | |
| CN117113086A (en) | Energy storage unit load prediction method, system, electronic equipment and medium | |
| Prema et al. | LSTM based Deep Learning model for accurate wind speed prediction | |
| CN115481788B (en) | Phase change energy storage system load prediction method and system | |
| Guanoluisa-Pineda et al. | Short-Term forecasting of photovoltaic power in an isolated area of Ecuador using deep learning techniques | |
| Ye et al. | A comparative study of ensemble support vector regression methods for short-term load forecasting | |
| Sala et al. | Smart multi-model approach based on adaptive Neuro-Fuzzy Inference Systems and Genetic Algorithms | |
| CN113283638A (en) | Load extreme curve prediction method and system based on fusion model | |
| CN111680818A (en) | Short-term reactive load prediction method and system | |
| CN114049143B (en) | Node-holiday power load-oriented derivative cluster model prediction method and system | |
| Tian et al. | Wind Speed Prediction based on Long Short-Term Memory network and Policy Suggestions for Wind Power Industry Development | |
| Haque et al. | Short-Term Electrical Load Prediction for Future Generation Using Hybrid Deep Learning Model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210309 |
|
| RJ01 | Rejection of invention patent application after publication |