CN111340305A - Building operation energy consumption prediction method - Google Patents
Building operation energy consumption prediction method Download PDFInfo
- Publication number
- CN111340305A CN111340305A CN202010162519.5A CN202010162519A CN111340305A CN 111340305 A CN111340305 A CN 111340305A CN 202010162519 A CN202010162519 A CN 202010162519A CN 111340305 A CN111340305 A CN 111340305A
- Authority
- CN
- China
- Prior art keywords
- energy consumption
- prediction
- office building
- operation energy
- svr
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005265 energy consumption Methods 0.000 title claims abstract description 140
- 238000000034 method Methods 0.000 title claims abstract description 75
- 239000011159 matrix material Substances 0.000 claims abstract description 29
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 230000000739 chaotic effect Effects 0.000 claims abstract description 19
- 238000012937 correction Methods 0.000 claims abstract description 16
- 238000012546 transfer Methods 0.000 claims abstract description 12
- 239000003245 coal Substances 0.000 claims abstract description 8
- 238000004458 analytical method Methods 0.000 claims abstract description 5
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 239000013598 vector Substances 0.000 claims description 37
- 230000007704 transition Effects 0.000 claims description 23
- 230000006870 function Effects 0.000 claims description 11
- 238000005291 chaos (dynamical) Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 7
- 238000012706 support-vector machine Methods 0.000 claims description 7
- 238000001816 cooling Methods 0.000 claims description 6
- 230000005611 electricity Effects 0.000 claims description 6
- 238000000926 separation method Methods 0.000 claims description 6
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000010438 heat treatment Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 101000802640 Homo sapiens Lactosylceramide 4-alpha-galactosyltransferase Proteins 0.000 claims description 3
- 102100035838 Lactosylceramide 4-alpha-galactosyltransferase Human genes 0.000 claims description 3
- 230000002068 genetic effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 230000001052 transient effect Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 239000007789 gas Substances 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 239000002131 composite material Substances 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000002737 fuel gas Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 230000001373 regressive effect Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 239000002699 waste material Substances 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
- G06N3/0418—Architecture, e.g. interconnection topology using chaos or fractal principles
-
- 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
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/08—Construction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a building operation energy consumption prediction method, which comprises the steps of preprocessing operation energy consumption data of an office building and performing characteristic analysis, unifying dimensions into a kilogram standard coal form, and obtaining a time sequence of the operation energy consumption of the office building; performing phase space reconstruction on the time sequence of the operation energy consumption of the office building by using a C-C correlation estimation method; judging whether the reconstructed office building operation energy consumption time sequence has the chaotic characteristic or not by using the maximum lyapunov index; performing short-term energy consumption prediction on the office building operation energy consumption time sequence with the Chaos characteristic by utilizing a Chaos-SVR neural network; dividing error intervals of the prediction result by using a mean-variance method, and constructing a Markov probability transfer matrix; and performing Markov chain error correction on the Markov probability transfer matrix, and predicting the operating energy consumption of the office building according to the predicted value. The invention obviously improves the prediction precision of the operation energy consumption of the office building and provides decision basis for the optimized operation and energy-saving management of the office building.
Description
Technical Field
The invention belongs to the technical field of building operation energy consumption prediction, and particularly relates to a building operation energy consumption prediction method.
Background
The problems of large energy consumption, low energy efficiency and the like of office buildings in the whole life cycle generally exist, and serious energy waste is caused. The energy-saving potential is huge, and the consumption reduction amplitude can reach 30-50%. Wherein, office building operation energy consumption accounts for the biggest. Therefore, the research on the operation energy consumption condition of the office building is of great significance, and the real-time and accurate prediction on the operation energy consumption condition can provide data decision for optimizing the operation efficiency, so that the energy-saving aim is achieved.
The prediction method of building energy consumption mainly comprises two categories: 1. a forward model; 2. the data drives the model. The machine learning prediction method is the most common short-term energy consumption prediction method and is widely applied to the energy fields of wind speed prediction, power demand prediction, building energy consumption prediction, building cold load prediction and the like.
At present, the method for predicting the energy consumption time series of buildings at home and abroad is widely applied as follows: artificial Neural Network (ANN), differential integrated Moving Average autoregressive (ARIMA), Support Vector Regression (SVR), Multiple Linear Regression (MLR), and the like. Except for the energy consumption of office building bodies such as outdoor meteorological parameters, building envelopes and the like, the energy consumption time sequence of the office building has nonlinear characteristics due to the operation energy consumption of electromechanical equipment, holidays, personnel occupancy and the like. In order to deeply excavate the nonlinear law in the time sequence and improve the prediction precision, researchers combine different algorithms with an artificial neural network and a support vector machine to establish various office building hybrid prediction models. The MLR method is utilized to carry out regression prediction on day-by-day cold load of the office building according to various factors such as weather, personnel and the like, and the average absolute percentage error is less than 8% compared with the actual load. The time-by-time energy consumption of the office building is respectively analyzed by using a feedback Neural Network (BPNN), a Radial Basis Function Neural Network (RBFNN), a Generalized Regression Neural Network (GRNN) and a Support Vector Machine (SVM) method, so that a better prediction effect is obtained, and the model is applied to a certain office building in Guangzhou China. A Wavelet-Partial Least Squares-Support vector machine (Wavelet-SVM) model is established to analyze the time-by-time energy consumption of the office building, and finally, prediction results of 1h, 2h, 3h and 24h in advance are obtained. In other application fields, the average speed, the average occupancy and the average traffic flow time sequence are integrated into one time sequence by using the Bayes theory, and the chaos theory and the SVR are combined for predicting the time sequence of the traffic flow, so that higher prediction precision is achieved. The research is based on multi-variable input, compared with the complex factors needing to be considered in a single-variable time sequence, the cold load of the office building at the historical moment is mapped to the same time dimension according to the Bayesian theory to be used as the single-variable input of a machine learning prediction model, and the building cold load time sequence is respectively predicted by using a Chaos-SVR (singular component-Support Vector Regression) and a WD-SVR (singular component-Support Vector Regression).
Disclosure of Invention
The technical problem to be solved by the invention is to provide a building operation energy consumption prediction method aiming at the defects in the prior art, and the energy consumption short-term prediction is carried out by utilizing a time series method in a data driving model according to the acquired historical energy consumption data characteristics of the office building.
The invention adopts the following technical scheme:
a building operation energy consumption prediction method comprises the following steps:
s1, collecting and preprocessing operation energy consumption data of office buildings;
s2, performing characteristic analysis on the preprocessed office building operation energy consumption data, unifying the form of kilogram standard coal in dimension, and obtaining a time sequence of the office building operation energy consumption;
s3, performing phase space reconstruction on the time sequence of the operation energy consumption of the office building by using a C-C correlation estimation method;
s4, judging whether the office building operation energy consumption time sequence reconstructed in the step S3 has the chaos characteristic or not by using the maximum lyapunov index;
s5, performing short-term energy consumption prediction on the office building operation energy consumption time sequence with the chaotic characteristic in the step S4 by utilizing a Chaos-SVR neural network;
s6, dividing an error interval of the prediction result in the step S5 by utilizing a Chaos-SVR neural network, and constructing a Markov probability transfer matrix;
and S7, carrying out Markov chain error correction on the Markov probability transfer matrix in the step S6 to obtain a prediction value, and predicting the operation energy consumption of the office building.
Specifically, in step S1, the preprocessing specifically includes: the operation energy consumption data of the office building is divided into water consumption, electricity consumption, gas consumption, centralized cooling/heating and other energy consumption.
Specifically, in step S3, statistics are formed by the correlation integral of the office building operation energy consumption time-by-time series, and the delay time τ and the optimum window width τ are calculated from the relationship between the statistics and the delay timewAccording to τωThe embedding dimension is obtained as (m-1) τ, and the reconstructed phase space is obtained.
Specifically, in step S4, the maximum Lyapunov exponent is obtained by using a small data volume method, which specifically includes: firstly, calculating an average period, delay time and embedding dimension, then carrying out phase space reconstruction, finding out a closest point, limiting transient separation, carrying out straight line fitting by using a least square method after measuring the average separation, and when the maximum Lyapunov index value is a positive value, obtaining the total energy consumption time sequence of the office building and having chaotic characteristics.
Specifically, in step S5, a chaos theory is used to construct a nonlinear mapping, and a phase space reconstruction technique is used to reconstruct information implied by the original time series data, which is used as an input of the SVR model for nonlinear training and prediction.
Specifically, the method specifically comprises the following steps:
s501, constructing a prediction model structure, including input vectors, nonlinear change of support vectors and output;
s502, obtaining an epsilon-SVR model by utilizing an epsilon insensitive loss function in a support vector machine, and adopting a Gaussian radial basis function f (x)i,x)=exp(-γ||xi-x||2) As a kernel of the epsilon-SVR, optimizing a penalty factor C and a kernel function parameter gamma by adopting a genetic algorithm;
s503, performing network loop learning, correcting the output and weight of the SVR until the error is controlled in an allowable range or the iteration number reaches an upper limit, and finishing training;
s504, taking continuous Q data points after the first n data points are sampled as prediction data, predicting the model, wherein the output value of the network is a prediction value, and performing corresponding inverse normalization processing on the prediction value to obtain a predicted actual value.
Specifically, step S6 specifically includes:
s601, use E respectively1,E2,...,EmIndicating an error state interval, with state transitions only at t1,t2,...,tmCan occur at several moments and define the mean value of errorsAnd standard deviation S, the error state intervals are divided into 5 groups according to the central limit principle, i.e.
S602, Markov chain is formed by state EiTransitioning to State E through k StepsjFor transition probability ofRepresenting, calculating a state transition probability matrix of the k steps according to the state transition probability matrix of the one step;
s603, setting the initial vector as P(0)Judging the state space of the k steps by the state vector after the k steps of transfer;
and S604, obtaining a value after error correction according to the probability transition matrix obtained in the step S602 and the state vector obtained in the step S603.
wherein the content of the first and second substances,and (d) a predicted value matrix obtained by the Chaos-SVR prediction model, and A (t) is a true value matrix.
Further, in step S602, k stages of the state transition probability matrix P(k)Comprises the following steps:
P(k)=(P(1))k
where m is the number of rows.
Further, in step S603, the state vector after k steps of transfer is:
P(k)=P(0)×P(k)=P(0)×(P(1))k。
compared with the prior art, the invention has at least the following beneficial effects:
the invention relates to a building operation energy consumption prediction method, which is characterized in that on the basis of a support vector regression neural network, a decisive rule hidden behind a chaos phenomenon is explored to enable the chaos theory to be applied to the possibility of the office building operation energy consumption prediction field, whether the office building operation energy consumption has chaos characteristics or not is judged by utilizing a maximum lyapunov index, firstly, dimension unification is carried out on the office building operation energy consumption data, and the office building operation energy consumption data is converted into a kilogram standard coal (kgce) form; calculating the maximum embedding dimension and the optimal window width according to a C-C correlation estimation method, reconstructing a phase space of an office building operation energy consumption time sequence, and then calculating a maximum lyapunov index by using a small data volume method, wherein the maximum lyapunov index is a positive value, and the office building operation energy consumption has a chaotic characteristic; and finally, forecasting the operation energy consumption of the office building by utilizing a chaos theory and a support vector regression combination model, and considering that some relative errors still exist in the forecasting result part of the combination model, further correcting the forecasting result by adopting a Markov chain to improve the precision, fully reflecting the invalidity characteristic of the Markov chain in the error correction of the combination model, only processing the final forecasting result without influencing the operation process of the traditional forecasting model, remarkably improving the forecasting precision after the Markov chain is corrected, more conforming to the change rule of the operation energy consumption of the actual office building, and providing a sufficient decision basis for the optimized operation and energy-saving management of the office building.
Further, the operation energy consumption of the office building comprises water consumption, electricity consumption, gas consumption, centralized cooling/heating and other energy consumption. In order to analyze the chaos characteristics of the operating energy consumption time sequence of the office building, because the composition structure is complex, a plurality of items of data need to be unified in dimension. Therefore, all the operation energy consumption dimension is converted into the form of kilogram standard coal (kgce), so as to obtain the time-by-time sequence of the operation energy consumption of the office building.
Further, phase spaceThe method is a tool for describing a dynamic system of univariate or multivariate time series reconstruction generated by a deterministic chaotic system for determining the degree of freedom. The main principle of phase space reconstruction is to recover the dynamic characteristics of a chaotic attractor from a high-dimensional space in a topological equivalent sense, form statistic through the correlation integral of a time-by-time sequence of the operating energy consumption of an office building, and calculate the delay time tau and the optimal window width tau through a relational graph of the statistic and the delay timewAccording to τωThe embedding dimension is obtained as (m-1) τ, and the reconstructed phase space is obtained. The reasonable selection of the embedding dimension and the delay time determines the quality of the phase-space reconstruction of the energy consumption time series of the office building, and simultaneously ensures that the prediction precision is high enough in the actual prediction.
Furthermore, the small data volume method adopted in step S4 can fully utilize all available data, is relatively reliable for small data sets, has small calculation amount, is relatively easy to operate, and has high calculation result precision.
Further, the prediction method of the Chaos-SVR of the office building mainly uses the Chaos theory to construct nonlinear mapping, utilizes the phase space reconstruction technology to restore and reconstruct the information implied by the original time series data, and uses the information as the input of the SVR model to carry out nonlinear training and prediction. The possibility of applying the chaos theory to the field of prediction of the operation energy consumption of office buildings is provided by exploring the decisive rule hidden behind the chaos phenomenon.
Further, the Markov chain is a random process with no after effect, and is suitable for correcting and describing a prediction problem with volatility. It can obtain the probability distribution of the next time by deducing according to the time condition of a certain known condition, and obtain the state of the next time without any relation with other times. Therefore, the accumulated error is corrected by using a Markov chain on the basis of the traditional Chaos-SVR prediction model, so that the accurate prediction of the energy consumption time sequence of the office building is realized.
Furthermore, accumulated errors existing in parameter transmission in the combined model are effectively improved through a Markov chain, and the correction process of the system is suitable for a nonlinear system of office building operation energy consumption. The invalidity characteristic of the Markov chain is fully reflected in error correction of the combined model, only the final prediction result is processed without influencing the operation process of the traditional prediction model, and the change rule of the operation energy consumption of the actual office building is better met.
In conclusion, the invention constructs nonlinear mapping according to the Chaos theory to perform phase space reconstruction on the office building operation energy consumption time sequence, judges the Chaos characteristic of the office building energy consumption time sequence, provides a Chaos-SVR combined prediction model, and utilizes a Markov chain to correct accumulated errors in the combined model, thereby obviously improving the office building operation energy consumption prediction precision and providing a sufficient decision basis for the optimized operation and energy-saving management of the office building.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
FIG. 1 is a block flow diagram of the present invention;
FIG. 2 is a graph of statistical quantity of total energy consumption time series for reconstructing an office building by a C-C correlation estimation method according to the present invention;
FIG. 3 is a least squares fit line graph of the maximum lyapunov exponent obtained by the present invention;
FIG. 4 is a diagram of a structure of the Chaos-SVR model of the present invention;
FIG. 5 is a graph of the fitting effect of the model of the present invention;
FIG. 6 is a flow chart of a small data volume method.
Detailed Description
The invention provides a building operation energy consumption prediction method, which is used for carrying out phase space reconstruction on a time sequence of a research object, judging that the time sequence has chaotic characteristics, establishing a chaotic theory and a support vector regression combined model for training, eliminating an accumulated error of the combined model due to parameter transmission by adopting a Markov chain, and obtaining a final prediction result to predict the operation energy consumption of a building.
Referring to fig. 1, the method for predicting building operation energy consumption of the present invention includes the following steps:
s1, processing the operation energy consumption data of the office building;
the operating energy consumption of the office building is processed and classified into water consumption, electricity consumption, gas consumption, centralized cooling/heating and other energy consumption.
S2, performing characteristic analysis on the operation energy consumption of the office building, including water consumption, electricity consumption, fuel gas quantity, centralized cooling/heat supply and other energy consumption, and unifying the dimension into a kilogram standard coal (kgce) form;
in order to analyze the chaos characteristics of the operating energy consumption time sequence of the office building, because the operating energy consumption of the office building has a complex structure, a plurality of items of data need to be unified firstly. Therefore, all the operation energy consumption dimension is converted into the form of kilogram standard coal (kgce), so as to obtain the time-by-time sequence of the operation energy consumption of the office building.
Through a building energy consumption monitoring platform in the city of Xian, various classified energy consumption data from 0 point in 8 month and 1 day to 23 points in 9 month and 30 days in 2019 of a certain office building are collected, and are shown in table 1.
TABLE 1 energy consumption statistics for certain office building in Xian
As can be seen from table 1, the office building operation energy consumption includes water consumption, electricity consumption, gas consumption, central cooling/heating and other energy consumption. In order to analyze the chaos characteristics of the operating energy consumption time sequence of the office building, because the composition structure is complex, a plurality of items of data need to be unified in dimension. Therefore, all the operation energy consumption dimension is converted into the form of kilogram standard coal (kgce), so as to obtain the time-by-time sequence of the operation energy consumption of the office building.
S3, performing phase space reconstruction on the time sequence of the operation energy consumption of the office building by using a C-C correlation estimation method (a method for estimating delay time by using correlation integration);
the phase space is a tool for describing the certainty of determining the degree of freedomAnd a dynamic system for single variable or multivariable time sequence reconstruction generated by the chaotic system. The main principle of phase space reconstruction is to recover the dynamic characteristics of a chaotic attractor from a high-dimensional space in the topological equivalence sense, and for a time-by-time sequence x of total energy consumption1,x2,x3,...,xn-1,xnAnd properly selecting the embedding dimension m and the delay time tau to obtain a reconstructed phase space.
The C-C correlation estimation method forms statistic by the correlation integral of the time-by-time sequence of the operation energy consumption of the office building, and simultaneously calculates the delay time tau and the optimal window width tau by the relation graph of the statistic and the delay timewThen according to τωThe embedding dimension is obtained as (m-1) τ, and the reconstructed phase space is obtained.
S4, judging that the office building operation energy consumption time sequence has chaotic characteristics by using the maximum lyapunov index;
and (4) solving the maximum Lyapunov index by using a small data volume method, thereby judging whether the operating energy consumption time sequence of the office building has a chaotic characteristic.
Referring to fig. 6, an average period, a delay time, and an embedding dimension are calculated first, then phase space reconstruction is performed to find a closest point, transient separation is limited, and after average separation is measured, a least square method is used to perform straight line fitting to complete a fractional data quantity method; the method can fully utilize all available data, is relatively reliable for small data groups, has small calculation amount, is relatively easy to operate, and has high calculation result precision. Therefore, a small data method is adopted to calculate the maximum Lyapunov exponent.
One or more positive values in the Lyapunov exponent spectra of a system can confirm that the chaotic characteristics exist.
According to the optimal delay time tau which is obtained by the phase space reconstruction C-C method and the embedding dimension m which is 2, a least square method fitting straight line of the office building energy consumption time sequence shown in figure 3 is calculated by a small data quantity method.
The maximum Lyapunov index value is 0.0077 which is a positive value according to the slope of the straight line, so that the total energy consumption time series of the office building has chaotic characteristics.
After the phase space is reconstructed, two adjacent trajectories in the phase space gradually diverge or converge along time, the Lyapunov index is the convergence or divergence rate of the trajectories, a positive maximum Lyapunov index reflects that a time sequence has a purity characteristic, and a negative maximum Lyapunov index indicates that the time sequence has randomness or periodicity; the larger the maximum Lyapunov exponent, the stronger the time series nonlinearity and the more sensitive it is to the initial value.
S5, performing short-term energy consumption prediction by utilizing a Chaos-SVR neural network;
the prediction method of the Chaos-SVR of the office building utilizes the Chaos theory to construct nonlinear mapping, utilizes the phase space reconstruction technology to restore and reconstruct the information implied by the original time series data, and uses the information as the input of the SVR model to carry out nonlinear training and prediction.
S501, constructing a prediction model structure, which mainly comprises three parts: an input vector, a nonlinear variation of a support vector, and an output;
the phase space of the reconstructed time sequence of the operation energy consumption of the certain office building in the Western Ann is Y2(i)=[x(i),x(i+2)]And the number of phase points is N-N-2, each phase point comprises the main characteristic of each sequence, and the real state of the building energy consumption can be approximated. Vector xi=(xiL) is the input vector for the model, β ═ β1,β2,...βl) Is the weight vector of the output, yi=f(xi) Is the output of the model. The structure of the Chaos-SVR model is shown in FIG. 4;
s502, obtaining an epsilon-SVR model by utilizing an epsilon insensitive loss function in a support vector machine, and adopting a Gaussian radial basis function f (x)i,x)=exp(-γ||xi-x||2) As a kernel for ε -SVR, the ε -SVR estimation function has the form:
α thereini,αi *Is a lagrange multiplier; for ε -SVR, an important issue is the selection of model parameters, which are optimized using Genetic Algorithm (GA)A penalizing factor C and a kernel function parameter gamma;
s503, performing network loop learning, correcting the output and weight of the SVR until the error is controlled in an allowable range or the iteration number reaches an upper limit, and finishing training;
s504, taking continuous Q data points after the first n data points are sampled as prediction data, predicting the model, wherein the output value of the network is a prediction value, and performing corresponding inverse normalization processing on the prediction value to obtain a predicted actual value.
S6, dividing error intervals by using the prediction result of the Chaos-SVR neural network, and constructing a Markov probability transfer matrix;
and dividing error intervals for the prediction result by using a mean-variance method, classifying the errors according to the divided intervals, constructing a Markov probability transition matrix according to the classification result, and determining the state of the prediction time period from the state of the initial vector.
S601, dividing error state interval
The error state interval represents the error state of the predicted value at the future time, and is represented by E1,E2,...,EmAre respectively expressed, and the state transition is only at t1,t2,...,tmAnd so on may occur at several times. Defining mean of errorAnd the standard deviation S are respectively:
wherein the content of the first and second substances,a matrix of predicted values determined for the Chaos-SVR prediction model, A (t) a matrix of true values, the error state intervals being divided into 5 groups according to the central limit principle, i.e.
S602, state transition probability matrix
Markov chain from state EiTransitioning to State E through k StepsjFor transition probability ofThe meaning of this is which error state interval will be in at the future time, and is calculated as follows:
wherein N isiRepresents state EiThe total number of occurrences;represents state EiThe error is transferred to the state E through k stepsjThe number of times of (c); n is the number of divided state intervals, the one-step state transition probability matrix is as follows:
the transition probability matrix for K steps is calculated using the C-K equation (Chepman-Kolmogorov equation) as:
P(k)=(P(1))k
s603, prediction model
Let the initial vector be P(0)And the state vector after k steps of transfer is as follows:
P(k)=P(0)×P(k)=P(0)×(P(1))k
therefore, the state space of the k steps can be judged.
And S604, obtaining a value after error correction according to the probability transition matrix obtained in the step S602 and the state vector obtained in the step S603.
And S7, carrying out Markov chain error correction to obtain a final prediction value to predict the operation energy consumption of the building.
And determining the state vector of each time period according to the error interval determined in the step S6, and calculating the predicted value of the Chaos-SVR neural network prediction model after the Markov chain correction according to the state transition vector and the probability transition matrix.
And the error correction is to obtain an eigenvalue and an eigenvector by using the divided error interval and the Markov probability transfer matrix, and obtain a corrected result according to the eigenvalue and the eigenvector.
The Markov chain parameters are set by taking the forecast data of 9 months and 25 days of the office building as an example.
Firstly, the average value of errors of a predicted energy consumption value and an actual energy consumption value is obtainedAnd standard deviation S-0.088 the error state space is determined as shown in table 3.
TABLE 3 error State intervals for Markov chains
And carrying out state division on the error value of the predicted value of the Chaos-SVR model according to the divided state interval, wherein the division result is shown in a table 4.
TABLE 4Chaos-SVR model prediction error classification results
According to the state transition situation of the Markov chain of the Chaos-SVR prediction result obtained by determining the error state interval, further calculating a state transition probability matrix P as follows:
and obtaining an energy consumption prediction state vector of the office building in 9 months and 25 days according to the Markov chain correction model, and correcting the accumulated error existing in the Chao-SVR prediction model to obtain a final prediction value of the energy consumption of the office building.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
According to the building operation energy consumption prediction method, various classified energy consumption data from 0 point in 8 month and 1 day to 23 points in 9 months and 30 days in 2019 of an office building are collected through a building energy consumption monitoring platform in the city of Xian to be experimentally verified, so that the model can accurately predict the change condition of the operation energy consumption of the office building in practical application, and plays a very positive role in energy conservation and emission reduction.
480 groups of data totaling 20 days from 9/month and 4 to 9/month and 24 days in 2019 are selected as training data, and 48 groups of data totaling 25 days in 9/month and 25 days in 26 days are selected as prediction data to perform model verification. Compared with the predicted value and the actual value of two machine learning methods which are widely applied in the field of time sequence prediction at present, namely a Nonlinear autoregressive neural network (Nonlinear Auto Regressive neural network) method and a Support Vector Regression (SVR) method, the method utilizes the Mean square error RMSE (root Mean Squared error) average Absolute percentage error MAPE (Mean Absolute percentage error) as an evaluation index.
The trained model combining the Chaos-SVR and the Markov chain is used for analyzing on a test data set, the fitting effect of the test data is shown in figure 5, the abscissa is a sampling time point, the ordinate is an energy consumption value for operating the office building, and as can be seen from figure 5, the Narnet prediction curve has the largest fluctuation, the SVR prediction curve is relatively gentle, the Chaos-SVR prediction curve is closest to a real value curve, and partial values are almost completely fitted.
According to the office building operation energy consumption composite prediction method based on the combination of the Chaos-SVR and the Markov chain, the predicted specific load value is different from the real value, but the whole rising and falling change trend of the predicted value curve can accurately reflect the real load curve.
Referring to fig. 2, the analysis results,first zero orThe minimum value obtained for the first time is the optimal delay time τ of 2 corresponding to the first local maximum value independent in time series. Scor(t) taking the minimum value, the time-series independent first overall maximum time window t being the optimum window width tau w2, according to the embedding time window width formula τwThe embedding dimension m 2 can be determined by (m-1) τ. Therefore, the total energy consumption time series reconstruction phase space of the office building is Y2(i)=[x(i),x(i+2)]。
Referring to fig. 3, the maximum Lyapunov index value of 0.0077, which is a positive value, can be obtained according to the slope of the straight line, so that the total energy consumption time series of the office building has a chaotic characteristic.
Please refer to fig. 4, which is a structure of a prediction model, mainly including three parts: an input vector, a non-linear variation of a support vector, and an output. The phase space of the reconstructed time sequence of the operation energy consumption of the certain office building in the Western Ann is Y2(i)=[x(i),x(i+2)]The number of phase points is N-N-2, each phase point comprises the main characteristic of each sequence, and the energy consumption of the building can be approximatedThe true state. Vector xi=(xiL) is the input vector for the model, β ═ β1,β2,...βl) Is the weight vector of the output, yi=f(xi) Is the output of the model.
According to experimental data, short-term prediction of office building operation energy consumption is carried out by adopting a nonlinear autoregressive neural network (Narnet), Support Vector Regression (SVR), Chao-SVR and Chaos-SVR-Markov method, and a prediction result is compared with an actual value to obtain a prediction result error comparison shown in a table 5.
TABLE 5 prediction error comparison
As can be seen from Table 1, the RMSE of the Chaos-SVR prediction model is 6.3261, and the MAPE is 0.2875, which is the smallest compared to the other two methods. The experiment is carried out on the basis of the same environment and the same group of data, and compared with a Narnet method and an SVR method, the Chaos-SVR prediction method has a better prediction effect. The RMSE and MAPE after the Markov correction are reduced, which shows that the accumulated error caused by parameter transmission in the combined model is effectively improved through the Markov chain, and the correction process is suitable for a nonlinear system of the operating energy consumption of office buildings. The invalidity characteristic of the Markov chain is fully reflected in error correction of the combined model, only the final prediction result is processed without influencing the operation process of the traditional prediction model, and the change rule of the operation energy consumption of the actual office building is better met. Therefore, the Chaos-SVR time series prediction model based on the Markov chain modification is superior to the Chaos-SVR time series prediction model.
In conclusion, the office building operation energy consumption composite prediction method based on the combination of the Chaos-SVR neural network and the Markov chain analyzes the operation energy consumption characteristics of a certain practical office building in Xian, discovers that the time sequence data has chaotic characteristics, uses the chaotic characteristics as the input variables of the SVR model, corrects errors of the combined model, reduces the redundancy rate and the complexity of the characteristic model, and improves the operation efficiency of the algorithm. Experiments prove that the short-term energy consumption condition of the office building can be timely and accurately reflected in actual application, and the method plays a positive role in predicting the operation energy consumption of the office building.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.
Claims (10)
1. A building operation energy consumption prediction method is characterized by comprising the following steps:
s1, collecting and preprocessing operation energy consumption data of office buildings;
s2, performing characteristic analysis on the preprocessed office building operation energy consumption data, unifying the form of kilogram standard coal in dimension, and obtaining a time sequence of the office building operation energy consumption;
s3, performing phase space reconstruction on the time sequence of the operation energy consumption of the office building by using a C-C correlation estimation method;
s4, judging whether the office building operation energy consumption time sequence reconstructed in the step S3 has the chaos characteristic or not by using the maximum lyapunov index;
s5, performing short-term energy consumption prediction on the office building operation energy consumption time sequence with the chaotic characteristic in the step S4 by utilizing a Chaos-SVR neural network;
s6, dividing an error interval of the prediction result in the step S5 by utilizing a Chaos-SVR neural network, and constructing a Markov probability transfer matrix;
and S7, carrying out Markov chain error correction on the Markov probability transfer matrix in the step S6 to obtain a prediction value, and predicting the operation energy consumption of the office building.
2. The method for predicting the energy consumption for building operation according to claim 1, wherein in step S1, the preprocessing specifically comprises: the operation energy consumption data of the office building is divided into water consumption, electricity consumption, gas consumption, centralized cooling/heating and other energy consumption.
3. The method for predicting the operational energy consumption of buildings according to claim 1, wherein in step S3, the statistics are formed by the correlation integral of the office building operational energy consumption time-by-time series, and the delay time τ and the optimal window width τ are calculated from the relationship between the statistics and the delay timewAccording to τωThe embedding dimension is obtained as (m-1) τ, and the reconstructed phase space is obtained.
4. The building operation energy consumption prediction method according to claim 1, wherein in step S4, a minimum data quantity method is used to obtain a maximum Lyapunov index, specifically: firstly, calculating an average period, delay time and embedding dimension, then carrying out phase space reconstruction, finding out a closest point, limiting transient separation, carrying out straight line fitting by using a least square method after measuring the average separation, and when the maximum Lyapunov index value is a positive value, obtaining the total energy consumption time sequence of the office building and having chaotic characteristics.
5. The building operation energy consumption prediction method according to claim 1, wherein in step S5, a chaos theory is used to construct a nonlinear mapping, and a phase space reconstruction technique is used to reconstruct information implied by the original time series data, which is used as an input of the SVR model for nonlinear training and prediction.
6. The building operation energy consumption prediction method according to claim 5, characterized by specifically comprising:
s501, constructing a prediction model structure, including input vectors, nonlinear change of support vectors and output;
s502, obtaining an epsilon-SVR model by utilizing an epsilon insensitive loss function in a support vector machine, and adopting a Gaussian radial basis function f (x)i,x)=exp(-γ||xi-x||2) As a kernel of the epsilon-SVR, optimizing a penalty factor C and a kernel function parameter gamma by adopting a genetic algorithm;
s503, performing network loop learning, correcting the output and weight of the SVR until the error is controlled in an allowable range or the iteration number reaches an upper limit, and finishing training;
s504, taking continuous Q data points after the first n data points are sampled as prediction data, predicting the model, wherein the output value of the network is a prediction value, and performing corresponding inverse normalization processing on the prediction value to obtain a predicted actual value.
7. The method for predicting the energy consumption for building operation according to claim 1, wherein the step S6 is specifically as follows:
s601, use E respectively1,E2,...,EmIndicating an error state interval, with state transitions only at t1,t2,...,tmCan take place at any moment, defines the error mean value X and standard deviation S, and divides the error state interval into 5 groups according to the central limit principle, namely
S602, Markov chain is formed by state EiTransitioning to State E through k StepsjFor transition probability ofRepresenting, calculating a state transition probability matrix of the k steps according to the state transition probability matrix of the one step;
s603, setting the initial vector as P(0)Judging the state space of the k steps by the state vector after the k steps of transfer;
and S604, obtaining a value after error correction according to the probability transition matrix obtained in the step S602 and the state vector obtained in the step S603.
8. The method for predicting the energy consumption for building operation according to claim 7, wherein in step S601, the mean error valueAnd the standard deviation S are respectively:
10. The method for predicting building operation energy consumption according to claim 7, wherein in step S603, the state vector after k steps of transferring is:
P(k)=P(0)×P(k)=P(0)×(P(1))k。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010162519.5A CN111340305A (en) | 2020-03-10 | 2020-03-10 | Building operation energy consumption prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010162519.5A CN111340305A (en) | 2020-03-10 | 2020-03-10 | Building operation energy consumption prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111340305A true CN111340305A (en) | 2020-06-26 |
Family
ID=71186579
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010162519.5A Pending CN111340305A (en) | 2020-03-10 | 2020-03-10 | Building operation energy consumption prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111340305A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111830350A (en) * | 2020-07-23 | 2020-10-27 | 珠海格力电器股份有限公司 | Energy consumption metering method and device and electric appliance |
CN112180733A (en) * | 2020-10-14 | 2021-01-05 | 西安建筑科技大学 | Fuzzy logic-based building energy consumption system prediction control parameter setting method |
CN112288139A (en) * | 2020-10-10 | 2021-01-29 | 浙江中烟工业有限责任公司 | Air conditioner energy consumption prediction method and system based on chaotic time sequence and storage medium |
CN115081763A (en) * | 2022-08-24 | 2022-09-20 | 山东鲁晟精工机械有限公司 | Energy consumption prediction system for turning process of numerical control lathe |
CN117294019A (en) * | 2023-10-11 | 2023-12-26 | 中铁十四局集团建筑工程有限公司 | Environment-friendly building energy consumption monitoring method and system based on Internet of things |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110264004A (en) * | 2019-06-20 | 2019-09-20 | 西安建筑科技大学 | A kind of air-conditioning refrigeration duty dynamic prediction method combined based on PSO-BP with Markov chain |
CN110260470A (en) * | 2019-06-20 | 2019-09-20 | 西安建筑科技大学 | Central air-conditioning parallel connection cold load optimal distribution method based on colony intelligence framework |
CN110621031A (en) * | 2019-07-31 | 2019-12-27 | 中南大学 | Multi-user multi-task mobile edge computing energy-saving method based on heterogeneous energy acquisition |
-
2020
- 2020-03-10 CN CN202010162519.5A patent/CN111340305A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110264004A (en) * | 2019-06-20 | 2019-09-20 | 西安建筑科技大学 | A kind of air-conditioning refrigeration duty dynamic prediction method combined based on PSO-BP with Markov chain |
CN110260470A (en) * | 2019-06-20 | 2019-09-20 | 西安建筑科技大学 | Central air-conditioning parallel connection cold load optimal distribution method based on colony intelligence framework |
CN110621031A (en) * | 2019-07-31 | 2019-12-27 | 中南大学 | Multi-user multi-task mobile edge computing energy-saving method based on heterogeneous energy acquisition |
Non-Patent Citations (3)
Title |
---|
ZHOUXUAN ET AL.: "Forecasting performance comparison of two hybrid machine learning models for cooling load of a large-scale commercial building", 《JOURNAL OF BUILDING ENGINEERING》 * |
王富强 等: "基于混沌相空间重构与支持向量机的风速预测", 《太阳能学报》 * |
郭杰昊: "基于混沌时间序列的短期负荷预测研究", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111830350A (en) * | 2020-07-23 | 2020-10-27 | 珠海格力电器股份有限公司 | Energy consumption metering method and device and electric appliance |
CN112288139A (en) * | 2020-10-10 | 2021-01-29 | 浙江中烟工业有限责任公司 | Air conditioner energy consumption prediction method and system based on chaotic time sequence and storage medium |
CN112180733A (en) * | 2020-10-14 | 2021-01-05 | 西安建筑科技大学 | Fuzzy logic-based building energy consumption system prediction control parameter setting method |
CN115081763A (en) * | 2022-08-24 | 2022-09-20 | 山东鲁晟精工机械有限公司 | Energy consumption prediction system for turning process of numerical control lathe |
CN115081763B (en) * | 2022-08-24 | 2022-11-11 | 山东鲁晟精工机械有限公司 | Energy consumption prediction system for turning process of numerical control lathe |
CN117294019A (en) * | 2023-10-11 | 2023-12-26 | 中铁十四局集团建筑工程有限公司 | Environment-friendly building energy consumption monitoring method and system based on Internet of things |
CN117294019B (en) * | 2023-10-11 | 2024-03-22 | 中铁十四局集团建筑工程有限公司 | Environment-friendly building energy consumption monitoring method and system based on Internet of things |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111340305A (en) | Building operation energy consumption prediction method | |
CN111626506B (en) | Regional photovoltaic power probability prediction method based on federal learning and cooperative regulation and control system thereof | |
Gao et al. | Day-ahead power forecasting in a large-scale photovoltaic plant based on weather classification using LSTM | |
CN108846517B (en) | Integration method for predicating quantile probabilistic short-term power load | |
CN110705743B (en) | New energy consumption electric quantity prediction method based on long-term and short-term memory neural network | |
CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
Wang et al. | Design of a combined system based on two-stage data preprocessing and multi-objective optimization for wind speed prediction | |
CN110729764B (en) | Optimal scheduling method for photovoltaic power generation system | |
CN113205207A (en) | XGboost algorithm-based short-term power consumption load fluctuation prediction method and system | |
CN111353656A (en) | Steel enterprise oxygen load prediction method based on production plan | |
CN113107626B (en) | Load prediction method of combined cycle generator set based on multivariable LSTM | |
CN110880789A (en) | Economic dispatching method for wind power and photovoltaic combined power generation system | |
CN113344192B (en) | Enterprise-level motor system energy-saving optimization automatic control method and system | |
CN116167531A (en) | Photovoltaic power generation prediction method based on digital twin | |
CN115965132A (en) | Distributed photovoltaic digital twin system power prediction method based on GA-BP neural network | |
CN109190786A (en) | A kind of combination forecasting method of photovoltaic efficiency | |
Kuan et al. | Short-term CHP heat load forecast method based on concatenated LSTMs | |
Wang et al. | Load forecasting method of integrated energy system based on CNN-BiLSTM with attention mechanism | |
Qu et al. | A trend forecast of import and export trade total volume based on LSTM | |
Ren et al. | Calculation Method of the Line Loss Rate in Transformer District Based on Neural Network with Optimized Input Variables | |
CN112669169B (en) | Short-term photovoltaic power prediction device and method | |
Yao et al. | Short term power load forecasting method based on particle swarm optimization algorithm and recurrent neural network | |
Li et al. | Short-Term Load Forecasting of Energy Internet Based on Energy Consumption Structure | |
Sheng et al. | Research on Application of Neural Network Based on Cloud Computing in Power Load Forecasting | |
Chen et al. | Short-Term Photovoltaic Power Prediction Algorithm Based on Improved BP Neural Network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200626 |
|
RJ01 | Rejection of invention patent application after publication |