CN111898856A - Extreme learning machine-based physical-data fusion building analysis method - Google Patents
Extreme learning machine-based physical-data fusion building analysis method Download PDFInfo
- Publication number
- CN111898856A CN111898856A CN202010571025.2A CN202010571025A CN111898856A CN 111898856 A CN111898856 A CN 111898856A CN 202010571025 A CN202010571025 A CN 202010571025A CN 111898856 A CN111898856 A CN 111898856A
- Authority
- CN
- China
- Prior art keywords
- data
- building
- model
- temperature
- physical
- 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.)
- Granted
Links
- 230000004927 fusion Effects 0.000 title claims abstract description 58
- 238000004458 analytical method Methods 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 claims abstract description 50
- 238000012549 training Methods 0.000 claims abstract description 36
- 230000005611 electricity Effects 0.000 claims abstract description 32
- 230000006399 behavior Effects 0.000 claims abstract description 30
- 230000003068 static effect Effects 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000010438 heat treatment Methods 0.000 claims description 114
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 88
- 230000006870 function Effects 0.000 claims description 56
- 238000003860 storage Methods 0.000 claims description 53
- 238000010248 power generation Methods 0.000 claims description 44
- 238000013499 data model Methods 0.000 claims description 42
- 230000007613 environmental effect Effects 0.000 claims description 27
- 230000004913 activation Effects 0.000 claims description 23
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000005338 heat storage Methods 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000005457 optimization Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 10
- 238000010801 machine learning Methods 0.000 claims description 9
- 238000005259 measurement Methods 0.000 claims description 9
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims description 8
- 238000009423 ventilation Methods 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 6
- 238000011160 research Methods 0.000 claims description 5
- 230000005540 biological transmission Effects 0.000 claims description 4
- 238000011217 control strategy Methods 0.000 claims description 4
- 238000001816 cooling Methods 0.000 claims description 4
- 238000000354 decomposition reaction Methods 0.000 claims description 4
- 239000007789 gas Substances 0.000 claims description 4
- 230000014509 gene expression Effects 0.000 claims description 4
- 238000004519 manufacturing process Methods 0.000 claims description 4
- 239000000463 material Substances 0.000 claims description 4
- 238000013178 mathematical model Methods 0.000 claims description 4
- 239000003345 natural gas Substances 0.000 claims description 4
- 239000008400 supply water Substances 0.000 claims description 4
- 238000012546 transfer Methods 0.000 claims description 4
- 230000001052 transient effect Effects 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 claims description 2
- 238000009435 building construction Methods 0.000 claims description 2
- 238000004140 cleaning Methods 0.000 claims description 2
- 238000012850 discrimination method Methods 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 230000004044 response Effects 0.000 abstract description 4
- 238000005265 energy consumption Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000001364 causal effect Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000007500 overflow downdraw method Methods 0.000 description 2
- 238000004940 physical analysis method Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000004088 simulation Methods 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- 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 relates to the field of power systems, in particular to an analysis method for building modeling based on physics-data fusion of an extreme learning machine. The method comprises the following steps of data acquisition and pretreatment: building a physical model of the building based on a total measuring and distinguishing method through building data and electrical data which are collected and preprocessed; training a building physical model, the collected and preprocessed user data, environment data and actual power utilization data by using an extreme learning machine to obtain a physical-data fusion model; and inputting the static parameters of the electricity utilization behavior to be analyzed and the dynamic parameters into a physical-data fusion model through a physical model to obtain an analysis result. The system comprises a data acquisition and preprocessing module and a building physical module; and a physical-data fusion module. The invention provides an extreme learning machine-based physical-data fusion building modeling analysis method, which aims to solve the problems of low precision, slow response at a demand side and low load prediction precision of the existing building model.
Description
Technical Field
The invention relates to the field of power systems, in particular to an analysis method for building modeling based on physics-data fusion of an extreme learning machine.
Background
With the great promotion of the construction of the ubiquitous power internet of things, the arrangement of the mass sensing terminals enables the analysis and control of the power consumption and energy consumption of the digital house to become a hotspot of research and application, and the existing house energy efficiency analysis is usually based on an aggregation model of a distribution network platform area or a single building. Dynamic change of external environment, uneven heat distribution in buildings, change states of internal communication structures and even activity behaviors of personnel can cause parameter deviation of building electricity utilization models, and accurate analysis of house electricity utilization energy consumption is further influenced.
Therefore, the accurate house power utilization model is constructed, and the method is very important for achieving the purposes of accurate analysis of power utilization and energy consumption, further energy conservation management and control and the like. At present, building electric energy consumption models are established by three main methods, namely physical simulation modeling, building energy consumption analysis software combined with modeling methods such as finite element and space vector method; secondly, based on a statistic or data driving model, an energy consumption model is established through a large amount of measured data and an intelligent algorithm, and thirdly, a modeling mode based on a physical-data fusion idea is adopted. Among the three methods, the simulation method based on the physical model can effectively reflect the cause-effect relationship of building electricity-heat conversion, but the accuracy of the model is poor due to inaccurate measured data and high complexity of the physical model. The second method and the third method can train and calibrate the electricity consumption model of the building by utilizing the historical data and weather data of the building, the calculation time is short, and the method can be used for real-time operation decision, but the method is very sensitive to sample size, and the requirement of a training data set is far greater than that of the third method.
The existing analysis method based on the building electric energy consumption model cannot completely meet the requirement of multivariate data, and an analysis method with high space model precision, fast demand side response and high load prediction precision is urgently needed.
Disclosure of Invention
The purpose of the invention is as follows:
the invention provides an extreme learning machine-based physical-data fusion building modeling analysis method, which aims to overcome the defects of the traditional building modeling analysis method. The method aims to solve the problems that the existing building model is low in precision, slow in response of a demand side and low in load prediction precision.
The technical scheme is as follows:
the physical-data fusion building modeling analysis method based on the extreme learning machine comprises the following steps:
and inputting the static parameters of the electricity utilization behavior to be analyzed and the dynamic parameters into a physical-data fusion model through a physical model to obtain an analysis result.
The physical-data fusion building modeling analysis system based on the extreme learning machine comprises a data acquisition and preprocessing module and a building physical module; a physical-data fusion module; the acquisition and pretreatment module is used for acquiring and pretreating data; the building physical module is used for building a building physical model based on a general measurement and identification method through the collected and preprocessed building data and electrical data; the physical-data fusion module is used for training the building physical model, the collected and preprocessed user data, environment data and actual measured data by using an extreme learning machine to obtain a physical-data fusion model; and inputting the static parameters of the electricity utilization behavior to be analyzed and the dynamic parameters into a physical-data fusion model through a physical model to obtain an analysis result.
The advantages and effects are as follows:
the invention has the following advantages and beneficial effects:
1. the building model is modeled and analyzed by adopting a physical-data fusion method, the traditional building model is modeled from a single physical model, environmental factors and user behaviors are not considered, the physical model and the data model are fused based on a machine learning method, the building physical model and the data model are fused by the machine learning method, general main components are replaced by high-entropy causal data, and the reliability of output results of different physical simplified model methods is rapidly calculated by the machine learning method.
2. The building physics-data model and the actual load are trained by using the extreme learning machine, because the method can be used for mining knowledge from historical and empirical data of the building load and providing direct or auxiliary decision-making for services such as analysis, control and planning of the building load. Compared with other methods, the method has the following characteristics: firstly, hidden layer nodes/neurons do not need to be adjusted iteratively in the learning process; the feedforward network belongs to a general single hidden layer feedforward network and a multi-hidden layer feedforward network; the same structure can be used for various problems of feature learning, clustering, regression and classification; and fourthly, the weight parameter solving mode can ensure the global optimization of the solving result. And (3) repeatedly training the ELM network by changing the activation function of the hidden layer node of the ELM algorithm of the extreme learning machine, and calculating the training accuracy under different activation functions. The method has the advantages that the accuracy is used as an evaluation index, the activation function suitable for user electricity utilization behavior analysis is selected, the built physical-data model and the actual load are fitted to the maximum degree through the neural algorithm, the precision of the physical-data model is greatly improved, and the building model load prediction precision is greatly improved.
In summary, the extreme learning machine and physical-data fusion method is used for building modeling analysis for the first time, and the method can fully take account of the heat-electricity conversion characteristics in the building and the dynamic changes of various environmental parameters and equipment electricity utilization behaviors, so that more accurate building electricity utilization model construction and refined electricity utilization energy consumption analysis are realized, and more accurate load prediction capability is realized.
Drawings
FIG. 1 is a block diagram of a physical-data fusion algorithm;
FIG. 2 is a block diagram of analyzing power usage behavior;
fig. 3 is a schematic diagram of an extreme learning machine.
Detailed Description
The following description of the embodiments of the present invention is provided in order to better understand the present invention for those skilled in the art with reference to the accompanying drawings.
The invention provides a physical-data fusion building modeling method based on an extreme learning machine, aiming at the problems that in the traditional building modeling analysis process, the structure of a built model is complex, the model precision is poor and the load prediction precision is low due to the fact that dynamic factors such as the sample demand is large, environmental data and user behaviors are not considered in the modeling process. The method is suitable for building models, improves the accuracy of the models on the basis of reducing the complexity of the models, and greatly increases the prediction precision of the load.
Based on the prior art, the invention provides a method for constructing a digital house electricity utilization model based on extreme learning machine physical and data fusion modeling by combining building, electrical, environmental and user behaviors and other metadata, and can realize refined load electricity utilization energy efficiency analysis. Compared with the traditional neural network model, the neural network model has the advantages of high learning speed, small training error and strong generalization capability. By combining data and a physical modeling method, the precision of a load model can be greatly improved, the demand side response of the load is realized, and the load prediction precision is greatly improved.
As shown in fig. 1 and fig. 2, the extreme learning machine-based physical-data fusion building modeling analysis method is characterized in that: the method comprises the following steps:
and inputting the static parameters of the electricity utilization behavior to be analyzed and the dynamic parameters into a physical-data fusion model through a physical model to obtain an analysis result.
The building physical model established by the total survey and discrimination method specifically comprises the following steps:
fusing a physical model and a data model through a machine learning method, describing the state of the building by using a physical simplified model reflecting a physical association relation, and generating data with high entropy characteristics, wherein the data are used as the input of the physical model and the data model;
the expressions of the physical model and the data model are:
where k represents a time status label, k +1 represents a future time, Xk+1A vector consisting of system state features predicted at the moment k + 1; x is the number ofk+1' is a state feature vector to be predicted preprocessed by a physical model at the moment k + 1; f and h respectively reflect the mapping relation between the measured data in the physical model and the data model and the characteristics of the data to be predicted; u is a random error vector in the calculation of the data model; xkAnd YkA vector of measured data of the power system, the difference being a vector X of measured datakProcessing the measured data vector Y by a physical modelkAnd (5) processing by a data model.
The physical model in the physical-data fusion model provides high-entropy input characteristics for the data model, so that the data model can more accurately express the characteristics of the problem to be solved, and further, an accurate data model is established.
The building data model and the physical model are fused through a machine learning method, high-entropy causal data are used for replacing conventional principal components, the reliability of output results of different physical simplified model methods is rapidly calculated through the machine learning method, the building physical-data fusion model is obtained, and then the building physical-data model is started for calculation.
The physical simplified model comprises the following specific steps:
firstly, building a physical model of a building; the building physical model comprises a building physical model and a heat pump physical model
In the heating model of the intelligent building, the following three ways of obtaining heat from the outside of the building are mainly adopted, namely heat conduction, heat convection and heat radiation. The heat storage capacity and the heat conduction capacity are the conceptual basis for building an RC network model of the intelligent building. Walls, ceilings, floors and room air can be provided as heat storage elements in the mould. The amount of heat they store is a function of their own mass and their specific heat capacity. Meanwhile, the heat is not only used for storage, but also can be exchanged with the outside world through the heat storage element in the intelligent building. The most commonly used RC network model at the present stage maintains the dynamic heating characteristic of the building while ensuring certain accuracy, so that the RC network model is widely applied.
TABLE 1 Unit analogy
As shown in Table 1, by analogy of Newton's law of cooling and ohm's law, heat is analogized to electric charge in a circuit, temperature difference is analogized to voltage, heat transfer rate is analogized to current, and thermal resistance is analogized to resistance; meanwhile, in order to further research the transient thermal behavior of the intelligent building model, the heat capacity, namely the heat storage capacity of an object, is introduced; in the RC network model, wall, indoor and outdoor air are the nodes in the circuit model.
The intelligent building model is as follows: assuming that the heating system is located in the technical room tr of the building; assuming that the heat loss from all parts of the heating system to the technical room is linear; these heat losses are expressed as heat pump losses hp, thermal resistances of the space heating storage tank sh and the domestic hot water storage tank dw; an air amb discharged into the surroundings by transmission and ventilation and discharged into the ground gnd, which is assumed to be thermostatted; since it is assumed that the floor and all walls are constructed using heavy materials, the groundThe thermal resistance of the plate and the outer wall is divided equally into several parts R on either side of the lumped thermal capacityw1、Rw2And Rf1、Rf2,Rw1、Rw2Is the equivalent thermal resistance of the outer wall, Rf1、Rf2Equivalent thermal resistance of the floor; modeling of ventilation heat loss as thermal resistance RveModeling solar energy and internal revenue as heat flow to all heat capacities;
so the building model can be written as:
wherein, Cin、Cwi、Cwe、Cf1、CtrRespectively representing indoor air, inner wall, outer wall, floor and technical room temperature equivalent capacitance, t represents time variable, QinIndicating the heating capacity of the chamber air, QwiDenotes the heating capacity of the interior wall, QweIndicating the heating capacity of the outer wall, Qf1Indicating floor heating capacity, Qhp,lossIndicating heat pump lost heat;
Tin、Tamb、Twi、Tf1、Tsh、Ttrrespectively showing indoor air, inflow indoor air, inner wall, floor, hot water pump and temperature of technical room;
in the optimal control problem, the heat loss of the heat pump to the machine room is neglected; the equation is discretized as:
wherein, TiAnd Ti+1Respectively representing the temperatures at time i and time i +1, and Δ t representing the heating time
The heat flow to the different components is calculated due to solar energy and internal gain as:
the operating temperature of the building is linearized:
wherein Q represents the heating capacity of the heat pump, P represents the required power input, and lt, mt and ht represent low temperature, medium temperature and high temperature respectively;
modeling the heating system of a heat pump heating system as follows:
the heat pump provides heat for space heating or domestic hot water production:
indicating the heating capacity of the heat pump at low temperature at time i,indicating the heating capacity of the heat pump at a low temperature of the heat pump at time i,the heating capacity of the water storage tank at low temperature at the moment i is shown, and the same is true;indicating the heating capacity of the heat pump at medium temperature at time i,indicating the heating capacity of the heater at time i when the heat pump is at medium temperature,represents the heating capacity of the storage tank at the medium temperature at time i,indicating the heating capacity of the heat pump at high temperature at time i,indicating the heating capacity of the heater when the heat pump is at a high temperature at time i,indicating the heating capacity of the storage tank at high temperature at time i;
the reasons for the variation of heat pump efficiency with ambient temperature, water temperature and compressor regulation are as follows:
phprepresenting the total power of the heat pump, php,htDenotes high temperature heating power, php,ltDenotes the low temperature heating power, php,mtDenotes medium heating power, ζlIndicates the degree of modulation, ζqIs a constraint coefficient;
the auxiliary heater efficiency is assumed to be 1:
the auxiliary heater power is for the time i,assisting the heating capacity of the heater at time i;
the space heat storage state is calculated by the following equation:
there are other limitations:
p represents the density of the water and,the volume of water lost to heat;represents the maximum supply water volume of the heat pump; nomRepresents the heating efficiency of the heat pump;
calculating the state of the domestic hot water storage tank:
Cdw,ltindicating equivalent capacitance of water tank at low temperature, Cdw,mtEquivalent capacitance of water storage tank at medium temperature, Cdw,htIndicating equivalent capacitance of water tank at high temperature, Cdw,vhtThe equivalent capacitance of the water storage tank at ultrahigh temperature is shown,indicating the heating capacity at ultra high temperature at time i,the heating capacity of the auxiliary heater at the ultrahigh temperature at the moment i is represented;
other limitations are:
the temperature of the water storage tank at the ultrahigh temperature is shown,the temperature of the water storage tank at high temperature is shown,the temperature of the water storage tank at low temperature is shown,indicating the temperature of the water storage tank at the middle temperature; t ismainsIndicating the temperature of the thermostatic mixing valve after mixing the stored water with the cold water from the water pipe; t isdw,demRepresenting a required temperature of the water storage tank; c represents the specific heat capacity of water;
the multi-standard target is written as:
WEIEweight factor representing the environmental impact of the energy source, j being the power generation index, p, applicable to all power plantsimpact,jIn order to have a marginal impact on the power generation of the power plant,representing the power generated by a power plant at a power generation index over a period of time;
defining different control targets for a control strategy of a physical model of the heat pump building;
first, from an individual consumer's perspective, consumers want to minimize their own cost for providing thermal comfort and domestic hot water; this goal is achieved by minimizing the individual energy costs within the control range if thermal comfort is maintained; the cost of discomfort is added to the objective function from as a soft constraint:
by usingIndicating electricity prices at time steps i andandrespectively supplying power to the heat pump power supply and the auxiliary heater; quantifying the discomfort cost by multiplying the building operating temperature deviation outside the building comfort temperature boundary by the building discomfort cost p;
secondly, considering the viewpoint of environmental impact of energy use, the impact of the used energy on the environment is minimized; since the impact of each generation is different, optimization attempts to run the most environmentally friendly plant as much as possible; the objective function is then written as:
selecting the influence of power generation on different power plants as the single-stage influence of roof photovoltaic power generation, offshore wind power generation, nuclear power generation, combined cycle natural gas power generation and open cycle gas turbine power generation; finally, not only is the use of energy important in view of the environmental impact of power generation; for many power plants, large emissions are associated with the construction and decommissioning of the power plant;
indicating a threshold value for the heavy load of the heat pump,which is indicative of an over-loaded power,representing the remaining load power;
since this is an approximation, the installed capacity cost is included in the objective function, where the specified capacity price is pimpact; cap has little physical significance;
if an overload condition occurs, its impact is much greater than the impact of generating power at a power just below the overload threshold; the three viewpoints of consumer cost, energy influence and capacity influence are fused into a single multi-objective function; the energy influence is brought into the consumer cost function and the environmental influence of the weight factor energy introduction scale of the environmental influence of the consumer cost function; when the factors are small, the controller ignores the influence of power generation; the factors are large, and the capacity influence viewpoint can be simply added into the cost function; to change the behavior of the controller, the override threshold is also changed to minimize
lcapIt is indicated that the capacity limiting factor,representing the maximum remaining load power.
Training by using an extreme learning machine according to a building physical model, the collected and preprocessed user data, environmental data and actual measurement data, and specifically comprising the following steps:
using the optimal feature set as an input of the extreme learning machine network; attaching a classification label to data according to actual power consumption data to serve as output of the extreme learning machine network, so as to train the extreme learning machine network; simultaneously selecting an activation function suitable for analyzing the electricity utilization behavior of the user; after the activation function is determined, training the extreme learning machine network for multiple times by changing the number of nodes of the hidden layer, and calculating the accuracy rate of training results under different hidden layer nodes; meanwhile, the complexity of calculation is reduced, and a physical-data fusion model is obtained.
And selecting an optimal feature set suitable for the used user electricity utilization data according to a feature optimal strategy, performing feature optimal selection on a resident user load curve, wherein the optimal feature set comprises daily average load, valley electricity coefficient, flat section electricity consumption percentage and peak-hour electricity consumption rate, and performing normalization processing on the feature set.
The fusion of the physical model and the data model in the step 3 is to obtain the weight of the output layer based on multiple times of training of the extreme learning machineThe method comprises the following steps:
s1 for N arbitrarily different samples (x)i,ti) When the hidden layer unit of the extreme learning machine isWhen the activation function is g (x), the mathematical model can be represented by the following formula:
wherein j is 1,2 …, N; w is ai=[wi1,wi2,…,win]T is a weight vector connecting the input feature and the ith hidden layer unit; beta is ai=[βi1,βi2,…,βin]T is a weight vector connecting the ith hidden layer unit and the output result; biBiasing for the ith hidden layer unit: w is ai·xjRepresents the inner product of the two;
s2, selecting a function g (-) which meets infinitesimal conditions as an activation function of the hidden layer node, and randomly setting a weight w between the input layer and the hidden layer and a threshold b of the hidden layer node;
the number of hidden layer nodes determined at S3, further according to the following formula:
obtaining a hidden layer matrix; in the formula: w is aiAs hidden layer node OiA connection weight matrix with each node of the input layer; fjIs the input characteristic of the jth sample, one sample is composed of n points; n is the output value corresponding to the jth input sample, and the output of one sample consists of m points;
s4, calculating the Moor-Penrose generalized inverse of the output matrix H by using a singular value decomposition method, and then according to the following formula:
The method adopts a characteristic optimization strategy to extract an optimal characteristic set of the load curve, and takes the optimal characteristic set as the input of the ELM network. Attaching a classification label to the data according to the real measurement data, and using the classification label as the output of the ELM network so as to train the ELM network; and the trained network is used for realizing the classification of the power utilization behaviors of the users. And if the output result of the ELM network is the same as the original label of the group of data, the training is considered to be correct. And then, respectively calculating the accuracy of the classification of the training set and the test set, and comparing the influence of different parameters on the performance of the training result by changing the input parameters of the algorithm. And changing the activation function of the hidden layer node of the ELM algorithm, repeatedly training the ELM network, and calculating the training accuracy under different activation functions. Selecting an activation function suitable for analyzing the power utilization behavior of the user by taking the accuracy as an evaluation index; and after the activation function is determined, changing the number of the hidden layer nodes, repeating the step to train the ELM network, and calculating the accuracy of the training result under different hidden layer nodes. On the basis of ensuring the accuracy, the calculation complexity is reduced, and the number of nodes suitable for the model analysis hidden layer is selected.
Extreme learning machine-based physical-data fusion building modeling analysis system is characterized in that: the system comprises a data acquisition and preprocessing module and a building physical module; a physical-data fusion module;
the acquisition and pretreatment module is used for acquiring and pretreating data;
the building physical module is used for building a building physical model based on a general measurement and identification method through the collected and preprocessed building data and electrical data;
the physical-data fusion module is used for training the building physical model, the collected and preprocessed user data, environment data and actual measured data by using an extreme learning machine to obtain a physical-data fusion model;
and inputting the static parameters of the electricity utilization behavior to be analyzed and the dynamic parameters into a physical-data fusion model through a physical model to obtain an analysis result.
The acquisition and pretreatment module specifically comprises: the collected data comprises building construction data, electrical data, user data, environment data and actual measurement data; preprocessing the collected building data, electrical data, user data and environment data, cleaning abnormal data, and repairing error data based on Neville algorithm of Lagrange interpolation.
The building physical module specifically comprises:
fusing a physical model and a data model through a machine learning method, describing the state of the building by using a physical simplified model reflecting a physical association relation, and generating data with high entropy characteristics, wherein the data are used as the input of the physical model and the data model;
the expressions of the physical model and the data model are:
where k represents a time status label, k +1 represents a future time, Xk+1A vector consisting of system state features predicted at the moment k + 1; x is the number ofk+1' is a state feature vector to be predicted preprocessed by a physical model at the moment k + 1; f and h respectively reflect the mapping relation between the measured data in the physical model and the data model and the characteristics of the data to be predicted; u is a random error vector in the calculation of the data model; xkAnd YkA vector of measured data of the power system, the difference being a vector X of measured datakProcessing the measured data vector Y by a physical modelkProcessing by a data model;
the physical simplified model module specifically comprises:
firstly, building a physical model of a building; the building physical model comprises a building physical model and a heat pump physical model
By analogy of Newton's cooling law and ohm's law, heat is analogized to electric charge in a circuit, temperature difference is analogized to voltage, heat transfer rate is analogized to current, and thermal resistance is analogized to resistance; meanwhile, in order to further research the transient thermal behavior of the intelligent building model, the heat capacity, namely the heat storage capacity of an object, is introduced; in the RC network model, the wall, indoor and outdoor air act as nodes in the circuit model,
the intelligent building model is as follows: assuming that the heating system is located in the technical room tr of the building; assuming that the heat loss from all parts of the heating system to the technical room is linear; these heat losses are expressed as heat pump losses hp, thermal resistances of the space heating storage tank sh and the domestic hot water storage tank dw; an air amb discharged into the surroundings by transmission and ventilation and discharged in a manner assumed to be constantA warm ground gnd without considering temperature changes; since the floor and all walls are assumed to be constructed using heavy materials, the thermal resistance of the floor and the exterior walls are divided equally into several parts R on either side of the lumped thermal capacityw1、Rw2And Rf1、Rf2,Rw1、Rw2Is the equivalent thermal resistance of the outer wall, Rf1、Rf2Equivalent thermal resistance of the floor; modeling of ventilation heat loss as thermal resistance RveModeling solar energy and internal revenue as heat flow to all heat capacities;
so the building model can be written as:
wherein, Cin、Cwi、Cwe、Cf1、CtrRespectively representing indoor air, inner wall, outer wall, floor and technical room temperature equivalent capacitance, t represents time variable, QinIndicating the heating capacity of the chamber air, QwiDenotes the heating capacity of the interior wall, QweIndicating the heating capacity of the outer wall, Qf1Indicating floor heating capacity, Qhp,lossIndicating heat pump lost heat;
Tin、Tamb、Twi、Tf1、Tsh、Ttrrespectively showing indoor air, inflow indoor air, inner wall, floor, hot water pump and techniqueOperating room temperature;
in the optimal control problem, the heat loss of the heat pump to the machine room is neglected; the equation is discretized as:
wherein, TiAnd Ti+1Respectively representing the temperatures at time i and time i +1, and Δ t representing the heating time
The operating temperature of the building is linearized:
wherein Q represents the heating capacity of the heat pump, P represents the required power input, and lt, mt and ht represent low temperature, medium temperature and high temperature respectively;
modeling a heating system of a heat pump heating system is as follows
The heat pump provides heat for space heating or domestic hot water production:
indicating the heating capacity of the heat pump at low temperature at time i,indicating the heating capacity of the heat pump at a low temperature of the heat pump at time i,the heating capacity of the water storage tank at low temperature at the moment i is shown, and the same is true;indicating the heating capacity of the heat pump at medium temperature at time i,indicating the heating capacity of the heater at time i when the heat pump is at medium temperature,represents the heating capacity of the storage tank at the medium temperature at time i,indicating the heating capacity of the heat pump at high temperature at time i,indicating the heating capacity of the heater when the heat pump is at a high temperature at time i,indicating the heating capacity of the storage tank at high temperature at time i;
the reasons for the variation of heat pump efficiency with ambient temperature, water temperature and compressor regulation are as follows:
phprepresenting the total power of the heat pump, php,htDenotes high temperature heating power, php,ltDenotes the low temperature heating power, php,mtDenotes medium heating power, ζlIndicates the degree of modulation, ζqIs a constraint coefficient;
the auxiliary heater efficiency is assumed to be 1:
the auxiliary heater power is for the time i,assisting the heating capacity of the heater at time i;
the space heat storage state is calculated by the following equation:
there are other limitations:
p represents the density of the water and,the volume of water lost to heat;represents the maximum supply water volume of the heat pump; nomRepresents the heating efficiency of the heat pump;
calculating the state of the domestic hot water storage tank:
Cdw,ltindicating equivalent capacitance of water tank at low temperature, Cdw,mtEquivalent capacitance of water storage tank at medium temperature, Cdw,htIndicating equivalent capacitance of water tank at high temperature, Cdw,vhtThe equivalent capacitance of the water storage tank at ultrahigh temperature is shown,indicating the heating capacity at ultra high temperature at time i,the heating capacity of the auxiliary heater at the ultrahigh temperature at the moment i is represented;
other limitations are:
the temperature of the water storage tank at the ultrahigh temperature is shown,the temperature of the water storage tank at high temperature is shown,the temperature of the water storage tank at low temperature is shown,indicating the temperature of the water storage tank at the middle temperature; t ismainsIndicating the temperature of the thermostatic mixing valve after mixing the stored water with the cold water from the water pipe; t isdw,demRepresenting a required temperature of the water storage tank; c represents the specific heat capacity of water;
the multi-standard target is written as:
WEIEweight factor representing the environmental impact of the energy source, j being the power generation index, p, applicable to all power plantsimpact,jIn order to have a marginal impact on the power generation of the power plant,representing the power generated by a power plant at a power generation index over a period of time;
defining different control targets for a control strategy of a physical model of the heat pump building;
first, from an individual consumer's perspective, consumers want to minimize their own cost for providing thermal comfort and domestic hot water; this goal is achieved by minimizing the individual energy costs within the control range if thermal comfort is maintained; the cost of discomfort is added to the objective function from as a soft constraint:
by usingIndicating electricity prices at time steps i andandrespectively supplying power to the heat pump power supply and the auxiliary heater; quantifying the discomfort cost by multiplying the building operating temperature deviation outside the building comfort temperature boundary by the building discomfort cost p;
secondly, considering the viewpoint of environmental impact of energy use, the impact of the used energy on the environment is minimized; since the impact of each generation is different, optimization attempts to run the most environmentally friendly plant as much as possible; the objective function is then written as:
selecting the influence of power generation on different power plants as the single-stage influence of roof photovoltaic power generation, offshore wind power generation, nuclear power generation, combined cycle natural gas power generation and open cycle gas turbine power generation; finally, not only is the use of energy important in view of the environmental impact of power generation; for many power plants, large emissions are associated with the construction and decommissioning of the power plant;
indicating a threshold value for the heavy load of the heat pump,which is indicative of an over-loaded power,representing the remaining load power;
since this is an approximation, the installed capacity cost is included in the objective function, where the specified capacity price is pimpact; cap has little physical significance;
if an overload condition occurs, its impact is much greater than the impact of generating power at a power just below the overload threshold; the three viewpoints of consumer cost, energy influence and capacity influence are fused into a single multi-objective function; the energy influence is brought into the consumer cost function and the environmental influence of the weight factor energy introduction scale of the environmental influence of the consumer cost function; when the factors are small, the controller ignores the influence of power generation; the factors are large, and the capacity influence viewpoint can be simply added into the cost function; to change the behavior of the controller, the override threshold is also changed to minimize
lcapIt is indicated that the capacity limiting factor,representing the maximum remaining load power.
The physical-data fusion module comprises the following specific steps:
the fusion of the physical model and the data model is based on the multiple training of the extreme learning machine to obtain the weight of the output layerThe method comprises the following steps:
s1 for N arbitrarily different samples (x)i,ti) When the hidden layer unit of the extreme learning machine isWhen the activation function is g (x), the mathematical model can be represented by the following formula:
wherein j is 1,2 …, N; w is ai=[wi1,wi2,…,win]T is a weight vector connecting the input feature and the ith hidden layer unit; beta is ai=[βi1,βi2,…,βin]T is a weight vector connecting the ith hidden layer unit and the output result; biBiasing for the ith hidden layer unit: w is ai·xjRepresents the inner product of the two;
s2, selecting a function g (-) which meets infinitesimal conditions as an activation function of the hidden layer node, and randomly setting a weight w between the input layer and the hidden layer and a threshold b of the hidden layer node;
the number of hidden layer nodes determined at S3, further according to the following formula:
obtaining a hidden layer matrix; in the formula: w is aiAs hidden layer node OiA connection weight matrix with each node of the input layer; fjIs the input characteristic of the jth sample, one sample is composed of n points; n is the output value corresponding to the jth input sample, and the output of one sample consists of m points;
s4, calculating the Moor-Penrose generalized inverse of the output matrix H by using a singular value decomposition method, and then according to the following formula:
As shown in fig. 1 and 2, firstly, the power load sample data collected by a building is divided into building data, electrical data, environmental data and behavior data. By establishing a physical simple model and guiding and correcting the physical model by real power utilization data, the physical analysis method can provide high-entropy characteristic information for the data analysis method, and is beneficial to improving the efficiency of data model analysis, and the principle is as follows: the input characteristics comprise the building model to be predicted, and when the data model parameters are solved through the optimization process, the search space can be reduced, and the calculation complexity is reduced; and meanwhile, the method is helpful for establishing a better data model, namely: the high-entropy input characteristics enable the target for establishing the data model to be more definite, the model parameter optimization solution is more pertinent, and the situation that the data model is trapped in local optimization is avoided, so that the rationality of the data model is improved. The data driving method can make up for the problem of rule loss caused by model simplification and the like in a physical analysis method. And obtaining a final physical-data fusion model. And then inputting the electricity utilization behavior data to be predicted and analyzed into an electricity utilization analysis prediction model based on physical-data fusion to obtain a final physical-data fusion model.
As shown in fig. 3, an optimal feature set of the load curve is extracted by using a feature optimization strategy, and the optimal feature set is used as an input of the ELM network. Attaching a classification label to the data according to the real measurement data, and using the classification label as the output of the ELM network so as to train the ELM network; and the trained network is used for realizing the classification of the power utilization behaviors of the users. And if the output result of the ELM network is the same as the original label of the group of data, the training is considered to be correct. And then, respectively calculating the accuracy of the classification of the training set and the test set, and comparing the influence of different parameters on the performance of the training result by changing the input parameters of the algorithm. And changing the activation function of the hidden layer node of the ELM algorithm, repeatedly training the ELM network, and calculating the training accuracy under different activation functions. Selecting an activation function suitable for analyzing the power utilization behavior of the user by taking the accuracy as an evaluation index; and after the activation function is determined, changing the number of the hidden layer nodes, repeating the step to train the ELM network, and calculating the accuracy of the training result under different hidden layer nodes. On the basis of ensuring the accuracy, the calculation complexity is reduced, and the number of nodes suitable for the model analysis hidden layer is selected.
Claims (10)
1. The extreme learning machine-based physical-data fusion building modeling analysis method is characterized by comprising the following steps of: the method comprises the following steps:
step 1, data acquisition and pretreatment:
step 2, building a physical model of the building based on a total measuring and distinguishing method through the collected and preprocessed building data and electrical data;
step 3, training the building physical model, the collected and preprocessed user data, environment data and actual power utilization data by using an extreme learning machine to obtain a physical-data fusion model;
and inputting the static parameters of the electricity utilization behavior to be analyzed and the dynamic parameters into a physical-data fusion model through a physical model to obtain an analysis result.
2. The extreme learning machine-based physics-data fusion building modeling analysis method of claim 1, wherein: the building physical model established by the total survey and discrimination method specifically comprises the following steps:
fusing a physical model and a data model through a machine learning method, describing the state of the building by using a physical simplified model reflecting a physical association relation, and generating data with high entropy characteristics, wherein the data are used as the input of the physical model and the data model;
the expressions of the physical model and the data model are:
where k represents a time status label, k +1 represents a future time, Xk+1A vector consisting of system state features predicted at the moment k + 1; x is the number ofk+1' is a state feature vector to be predicted preprocessed by a physical model at the moment k + 1; f and h respectively reflect the physical model and the data modelMapping relation between measured data and data characteristics to be predicted in the model; u is a random error vector in the calculation of the data model; xkAnd YkA vector of measured data of the power system, the difference being a vector X of measured datakProcessing the measured data vector Y by a physical modelkAnd (5) processing by a data model.
3. The extreme learning machine-based physics-data fusion building modeling analysis method of claim 2, wherein: the physical simplified model comprises the following specific steps:
firstly, building a physical model of a building; the building physical model comprises a building physical model and a heat pump physical model
By analogy of Newton's cooling law and ohm's law, heat is analogized to electric charge in a circuit, temperature difference is analogized to voltage, heat transfer rate is analogized to current, and thermal resistance is analogized to resistance; meanwhile, in order to further research the transient thermal behavior of the intelligent building model, the heat capacity, namely the heat storage capacity of an object, is introduced; in the RC network model, the wall, indoor and outdoor air act as nodes in the circuit model,
the intelligent building model is as follows: assuming that the heating system is located in the technical room tr of the building; assuming that the heat loss from all parts of the heating system to the technical room is linear; these heat losses are expressed as heat pump losses hp, thermal resistances of the space heating storage tank sh and the domestic hot water storage tank dw; an air amb discharged into the surroundings by transmission and ventilation and discharged into the ground, which is assumed to be thermostatted; since the floor and all walls are assumed to be constructed using heavy materials, the thermal resistance of the floor and the exterior walls are divided equally into several parts R on either side of the lumped thermal capacityw1、Rw2And Rf1、Rf2,Rw1、Rw2Is the equivalent thermal resistance of the outer wall, Rf1、Rf2Equivalent thermal resistance of the floor; modeling of ventilation heat loss as thermal resistance RveModeling solar energy and internal revenue as heat flow to all heat capacities;
the building model is therefore written as:
wherein, Cin、Cwi、Cwe、Cf1、CtrRespectively representing indoor air, inner wall, outer wall, floor and technical room temperature equivalent capacitance, t represents time variable, QinIndicating the heating capacity of the chamber air, QwiDenotes the heating capacity of the interior wall, QweIndicating the heating capacity of the outer wall, Qf1Indicating floor heating capacity, Qhp,1ossIndicating heat pump lost heat;
Tin、Tamb、Twi、Tf1、Tsh、Ttrrespectively showing indoor air, inflow indoor air, inner wall, floor, hot water pump and temperature of technical room;
in the optimal control problem, the heat loss of the heat pump to the machine room is neglected; the equation is discretized as:
wherein, TiAnd Ti+1Respectively representing the temperatures at time i and time i +1, and Δ t representing the heating time
The operating temperature of the building is linearized:
wherein Q represents the heating capacity of the heat pump, P represents the required power input, and lt, mt and ht represent low temperature, medium temperature and high temperature respectively;
modeling a heating system of a heat pump heating system is as follows
The heat pump provides heat for space heating or domestic hot water production:
indicating the heating capacity of the heat pump at low temperature at time i,indicating the heating capacity of the heat pump at a low temperature of the heat pump at time i,the heating capacity of the water storage tank at low temperature at the moment i is shown, and the same is true;indicating the heating capacity of the heat pump at medium temperature at time i,indicating the heating capacity of the heater at time i when the heat pump is at medium temperature,represents the heating capacity of the storage tank at the medium temperature at time i,indicating the heating capacity of the heat pump at high temperature at time i,indicating the heating capacity of the heater when the heat pump is at a high temperature at time i,indicating the heating capacity of the storage tank at high temperature at time i;
the reasons for the variation of heat pump efficiency with ambient temperature, water temperature and compressor regulation are as follows:
phprepresenting the total power of the heat pump, php,htDenotes high temperature heating power, php,ltDenotes the low temperature heating power, php,mtDenotes medium heating power, ζlIndicates the degree of modulation, ζqIs a constraint coefficient;
The auxiliary heater efficiency is assumed to be 1:
the auxiliary heater power is for the time i,assisting the heating capacity of the heater at time i;
the space heat storage state is calculated by the following equation:
there are other limitations:
p represents the density of the water and,the volume of water lost to heat;represents the maximum supply water volume of the heat pump;nomrepresents the heating efficiency of the heat pump;
calculating the state of the domestic hot water storage tank:
Cdw,ltindicating equivalent capacitance of water tank at low temperature, Cdw,mtEquivalent capacitance of water storage tank at medium temperature, Cdw,htIndicating equivalent capacitance of water tank at high temperature, Cdw,vhtThe equivalent capacitance of the water storage tank at ultrahigh temperature is shown,indicating the heating capacity at ultra high temperature at time i,the heating capacity of the auxiliary heater at the ultrahigh temperature at the moment i is represented;
other limitations are:
the temperature of the water storage tank at the ultrahigh temperature is shown,the temperature of the water storage tank at high temperature is shown,the temperature of the water storage tank at low temperature is shown,indicating the temperature of the water storage tank at the middle temperature; t ismainsIndicating the temperature of the thermostatic mixing valve after mixing the stored water with the cold water from the water pipe; t isdw,demRepresenting a required temperature of the water storage tank; c represents the specific heat capacity of water;
the multi-standard target is written as:
WEIEweight factor representing the environmental impact of the energy source, j being the power generation index, p, applicable to all power plantsimpact,jIn order to have a marginal impact on the power generation of the power plant,representing the power generated by a power plant at a power generation index over a period of time;
defining different control targets for a control strategy of a physical model of the heat pump building;
first, from an individual consumer's perspective, consumers want to minimize their own cost for providing thermal comfort and domestic hot water; this goal is achieved by minimizing the individual energy costs within the control range if thermal comfort is maintained; the cost of discomfort is added to the objective function from as a soft constraint:
by usingIndicating electricity prices at time steps i andandrespectively supplying power to the heat pump power supply and the auxiliary heater; quantifying the discomfort cost by multiplying the building operating temperature deviation outside the building comfort temperature boundary by the building discomfort cost p;
secondly, considering the viewpoint of environmental impact of energy use, the impact of the used energy on the environment is minimized; since the impact of each generation is different, optimization attempts to run the most environmentally friendly plant as much as possible; the objective function is then written as:
selecting the influence of power generation on different power plants as the single-stage influence of roof photovoltaic power generation, offshore wind power generation, nuclear power generation, combined cycle natural gas power generation and open cycle gas turbine power generation; finally, not only is the use of energy important in view of the environmental impact of power generation; for many power plants, large emissions are associated with the construction and decommissioning of the power plant;
indicating a threshold value for the heavy load of the heat pump,which is indicative of an over-loaded power,representing the remaining load power;
since this is an approximation, the installed capacity cost is included in the objective function, where the specified capacity price is pimpact; cap has little physical significance;
if an overload condition occurs, its impact is much greater than the impact of generating power at a power just below the overload threshold; the three viewpoints of consumer cost, energy influence and capacity influence are fused into a single multi-objective function; the energy influence is brought into the consumer cost function and the environmental influence of the weight factor energy introduction scale of the environmental influence of the consumer cost function; when the factors are small, the controller ignores the influence of power generation; the factors are large, and the capacity impact viewpoint is simply added to the cost function; to change the behavior of the controller, the override threshold is also changed to minimize
4. The extreme learning machine-based physics-data fusion building modeling analysis method of claim 1, wherein: training by using an extreme learning machine according to a building physical model, the collected and preprocessed user data, environmental data and actual measurement data, and specifically comprising the following steps:
using the optimal feature set as an input of the extreme learning machine network; attaching a classification label to data according to actual power consumption data to serve as output of the extreme learning machine network, so as to train the extreme learning machine network; simultaneously selecting an activation function suitable for analyzing the electricity utilization behavior of the user; after the activation function is determined, training the extreme learning machine network for multiple times by changing the number of nodes of the hidden layer, and calculating the accuracy rate of training results under different hidden layer nodes; meanwhile, the complexity of calculation is reduced, and a physical-data fusion model is obtained.
5. The extreme learning machine-based physics-data fusion building modeling analysis method of claim 4, wherein: and selecting an optimal feature set suitable for the used user electricity utilization data according to a feature optimal strategy, performing feature optimal selection on a resident user load curve, wherein the optimal feature set comprises daily average load, valley electricity coefficient, flat section electricity consumption percentage and peak-hour electricity consumption rate, and performing normalization processing on the feature set.
6. The extreme learning machine-based physics-data fusion building modeling analysis method of claim 4, wherein: the fusion of the physical model and the data model in the step 3 is to obtain the weight of the output layer based on multiple times of training of the extreme learning machineThe method comprises the following steps:
s1 for N arbitrarily different samples (x)i,ti) When the hidden layer unit of the extreme learning machine isWhen the activation function is g (x), the mathematical model can be represented by the following formula:
wherein j is 1,2 …, N; w is ai=[wi1,wi2,…,win]T is a weight vector connecting the input feature and the ith hidden layer unit; beta is ai=[βi1,βi2,…,βin]T is a weight vector connecting the ith hidden layer unit and the output result; biBiasing for the ith hidden layer unit: w is ai·xjRepresents the inner product of the two;
s2, selecting a function g (-) which meets infinitesimal conditions as an activation function of the hidden layer node, and randomly setting a weight w between the input layer and the hidden layer and a threshold b of the hidden layer node;
the number of hidden layer nodes determined at S3, further according to the following formula:
obtaining a hidden layer matrix; in the formula: w is aiAs hidden layer node OiA connection weight matrix with each node of the input layer; fjIs the input characteristic of the jth sample, one sample is composed of n points; n is the output value corresponding to the jth input sample, and the output of one sample consists of m points;
s4, calculating the Moor-Penrose generalized inverse of the output matrix H by using a singular value decomposition method, and then according to the following formula:
7. Extreme learning machine-based physical-data fusion building modeling analysis system is characterized in that: the system comprises a data acquisition and preprocessing module and a building physical module; a physical-data fusion module;
the acquisition and pretreatment module is used for acquiring and pretreating data;
the building physical module is used for building a building physical model based on a general measurement and identification method through the collected and preprocessed building data and electrical data;
the physical-data fusion module is used for training the building physical model, the collected and preprocessed user data, environment data and actual measured data by using an extreme learning machine to obtain a physical-data fusion model;
and inputting the static parameters of the electricity utilization behavior to be analyzed and the dynamic parameters into a physical-data fusion model through a physical model to obtain an analysis result.
8. The extreme learning machine-based physics-data fusion building modeling analysis system of claim 7, wherein: the acquisition and pretreatment module specifically comprises: the collected data comprises building construction data, electrical data, user data, environment data and actual measurement data; preprocessing the collected building data, electrical data, user data and environment data, cleaning abnormal data, and repairing error data based on Neville algorithm of Lagrange interpolation.
9. The extreme learning machine-based physics-data fusion building modeling analysis system of claim 7, wherein: the building physical module specifically comprises:
fusing a physical model and a data model through a machine learning method, describing the state of the building by using a physical simplified model reflecting a physical association relation, and generating data with high entropy characteristics, wherein the data are used as the input of the physical model and the data model;
the expressions of the physical model and the data model are:
where k represents a time status label, k +1 represents a future time, Xk+1Systems for prediction of time k +1A vector composed of system state features; x is the number ofk+1' is a state feature vector to be predicted preprocessed by a physical model at the moment k + 1; f and h respectively reflect the mapping relation between the measured data in the physical model and the data model and the characteristics of the data to be predicted; u is a random error vector in the calculation of the data model; xkAnd YkA vector of measured data of the power system, the difference being a vector X of measured datakProcessing the measured data vector Y by a physical modelkProcessing by a data model;
the physical simplified model module specifically comprises:
firstly, building a physical model of a building; the building physical model comprises a building physical model and a heat pump physical model
By analogy of Newton's cooling law and ohm's law, heat is analogized to electric charge in a circuit, temperature difference is analogized to voltage, heat transfer rate is analogized to current, and thermal resistance is analogized to resistance; meanwhile, in order to further research the transient thermal behavior of the intelligent building model, the heat capacity, namely the heat storage capacity of an object, is introduced; in the RC network model, the wall, indoor and outdoor air act as nodes in the circuit model,
the intelligent building model is as follows: assuming that the heating system is located in the technical room tr of the building; assuming that the heat loss from all parts of the heating system to the technical room is linear; these heat losses are expressed as heat pump losses hp, thermal resistances of the space heating storage tank sh and the domestic hot water storage tank dw; an air amb discharged into the surroundings by transmission and ventilation and discharged into the ground gnd, which is assumed to be thermostatted; since the floor and all walls are assumed to be constructed using heavy materials, the thermal resistance of the floor and the exterior walls are divided equally into several parts R on either side of the lumped thermal capacityw1、Rw2And Rf1、Rf2,Rw1、Rw2Is the equivalent thermal resistance of the outer wall, Rf1、Rf2Equivalent thermal resistance of the floor; modeling of ventilation heat loss as thermal resistance RveModeling solar energy and internal revenue as heat flow to all heat capacities;
the building model is therefore written as:
wherein, Cin、Cwi、Cwe、Cf1、CtrRespectively representing indoor air, inner wall, outer wall, floor and technical room temperature equivalent capacitance, t represents time variable, QinIndicating the heating capacity of the chamber air, QwiDenotes the heating capacity of the interior wall, QweIndicating the heating capacity of the outer wall, Qf1Indicating floor heating capacity, Qhp,lossIndicating heat pump lost heat;
Tin、Tamb、Twi、Tf1、Tsh、Ttrrespectively showing indoor air, inflow indoor air, inner wall, floor, hot water pump and temperature of technical room;
in the optimal control problem, the heat loss of the heat pump to the machine room is neglected; the equation is discretized as:
wherein, TiAnd Ti+1Respectively representing the temperatures at time i and time i +1, and Δ t representing the heating time
The operating temperature of the building is linearized:
wherein Q represents the heating capacity of the heat pump, P represents the required power input, and lt, mt and ht represent low temperature, medium temperature and high temperature respectively;
modeling a heating system of a heat pump heating system is as follows
The heat pump provides heat for space heating or domestic hot water production:
indicating the heating capacity of the heat pump at low temperature at time i,indicating the heating capacity of the heat pump at a low temperature of the heat pump at time i,the heating capacity of the water storage tank at low temperature at the moment i is shown, and the same is true;indicating the heating capacity of the heat pump at medium temperature at time i,indicating the heating capacity of the heater at time i when the heat pump is at medium temperature,represents the heating capacity of the storage tank at the medium temperature at time i,indicating the heating capacity of the heat pump at high temperature at time i,indicating the heating capacity of the heater when the heat pump is at a high temperature at time i,indicating the heating capacity of the storage tank at high temperature at time i;
the reasons for the variation of heat pump efficiency with ambient temperature, water temperature and compressor regulation are as follows:
phprepresenting the total power of the heat pump, php,htDenotes high temperature heating power, php,ltDenotes the low temperature heating power, php,mtDenotes medium heating power, ζlIndicates the degree of modulation, ζqIs a constraint coefficient;
the auxiliary heater efficiency is assumed to be 1:
the auxiliary heater power is for the time i,assisting the heating capacity of the heater at time i;
the space heat storage state is calculated by the following equation:
there are other limitations:
p represents the density of the water and,the volume of water lost to heat;represents the maximum supply water volume of the heat pump;nomrepresents the heating efficiency of the heat pump;
calculating the state of the domestic hot water storage tank:
Cdw,ltindicating equivalent capacitance of water tank at low temperature, Cdw,mtEquivalent capacitance of water storage tank at medium temperature, Cdw,htIndicating equivalent capacitance of water tank at high temperature, Cdw,vhtThe equivalent capacitance of the water storage tank at ultrahigh temperature is shown,indicating the heating capacity at ultra high temperature at time i,the heating capacity of the auxiliary heater at the ultrahigh temperature at the moment i is represented;
other limitations are:
the temperature of the water storage tank at the ultrahigh temperature is shown,the temperature of the water storage tank at high temperature is shown,the temperature of the water storage tank at low temperature is shown,indicating the temperature of the water storage tank at the middle temperature; t ismainsIndicating the temperature of the thermostatic mixing valve after mixing the stored water with the cold water from the water pipe; t isdw,demRepresenting a required temperature of the water storage tank; c represents the specific heat capacity of water;
the multi-standard target is written as:
WEIEweight factor representing the environmental impact of the energy source, j being the power generation index, p, applicable to all power plantsimpact,jIn order to have a marginal impact on the power generation of the power plant,representing the power generated by a power plant at a power generation index over a period of time;
defining different control targets for a control strategy of a physical model of the heat pump building;
first, from an individual consumer's perspective, consumers want to minimize their own cost for providing thermal comfort and domestic hot water; this goal is achieved by minimizing the individual energy costs within the control range if thermal comfort is maintained; the cost of discomfort is added to the objective function from as a soft constraint:
by usingIndicating electricity prices at time steps i andandrespectively supplying power to the heat pump power supply and the auxiliary heater; quantifying the discomfort cost by multiplying the building operating temperature deviation outside the building comfort temperature boundary by the building discomfort cost p;
secondly, considering the viewpoint of environmental impact of energy use, the impact of the used energy on the environment is minimized; since the impact of each generation is different, optimization attempts to run the most environmentally friendly plant as much as possible; the objective function is then written as:
selecting the influence of power generation on different power plants as the single-stage influence of roof photovoltaic power generation, offshore wind power generation, nuclear power generation, combined cycle natural gas power generation and open cycle gas turbine power generation; finally, not only is the use of energy important in view of the environmental impact of power generation; for many power plants, large emissions are associated with the construction and decommissioning of the power plant;
indicating a threshold value for the heavy load of the heat pump,which is indicative of an over-loaded power,representing the remaining load power;
since this is an approximation, the installed capacity cost is included in the objective function, where the specified capacity price is pimpact; cap has little physical significance;
if an overload condition occurs, its impact is much greater than the impact of generating power at a power just below the overload threshold; the three viewpoints of consumer cost, energy influence and capacity influence are fused into a single multi-objective function; the energy influence is brought into the consumer cost function and the environmental influence of the weight factor energy introduction scale of the environmental influence of the consumer cost function; when the factors are small, the controller ignores the influence of power generation; the factors are large, and the capacity impact viewpoint is simply added to the cost function; to change the behavior of the controller, the override threshold is also changed to minimize
10. The extreme learning machine-based physics-data fusion building modeling analysis system of claim 7, wherein: the physical-data fusion module comprises the following specific steps:
physical model and data modelThe fusion of the method is based on the fact that the extreme learning machine trains for multiple times to obtain the weight of the output layerThe method comprises the following steps:
s1 for N arbitrarily different samples (x)i,ti) When the hidden layer unit of the extreme learning machine isWhen the activation function is g (x), the mathematical model can be represented by the following formula:
wherein j is 1,2 …, N; w is ai=[wi1,wi2,…,win]T is a weight vector connecting the input feature and the ith hidden layer unit; beta is ai=[βi1,βi2,…,βin]T is a weight vector connecting the ith hidden layer unit and the output result; biBiasing for the ith hidden layer unit: w is ai·xjRepresents the inner product of the two;
s2, selecting a function g (-) which meets infinitesimal conditions as an activation function of the hidden layer node, and randomly setting a weight w between the input layer and the hidden layer and a threshold b of the hidden layer node;
the number of hidden layer nodes determined at S3, further according to the following formula:
obtaining a hidden layer matrix; in the formula: w is aiAs hidden layer node OiA connection weight matrix with each node of the input layer; fjIs the input characteristic of the jth sample, one sample is composed of n points; n is the output value corresponding to the jth input sample, and the output of one sample consists of m points;
s4, calculating the Moor-Penrose generalized inverse of the output matrix H by using a singular value decomposition method, and then according to the following formula:
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2020102534371 | 2020-04-02 | ||
CN202010253437 | 2020-04-02 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111898856A true CN111898856A (en) | 2020-11-06 |
CN111898856B CN111898856B (en) | 2024-03-22 |
Family
ID=73206865
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010571025.2A Active CN111898856B (en) | 2020-04-02 | 2020-06-22 | Analysis method of physical-data fusion building based on extreme learning machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111898856B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112669446A (en) * | 2020-12-24 | 2021-04-16 | 联通(浙江)产业互联网有限公司 | Building scene modeling method and device |
CN113065190A (en) * | 2021-04-15 | 2021-07-02 | 天津大学 | Uncertainty-based residential building heat supply amount calculation method |
CN114237057A (en) * | 2021-12-20 | 2022-03-25 | 东南大学 | Dynamic modeling method, system, equipment and medium for intelligent building electricity analysis |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2015172560A1 (en) * | 2014-05-16 | 2015-11-19 | 华南理工大学 | Central air conditioner cooling load prediction method based on bp neural network |
CN105160437A (en) * | 2015-09-25 | 2015-12-16 | 国网浙江省电力公司 | Load model prediction method based on extreme learning machine |
CN107423839A (en) * | 2017-04-17 | 2017-12-01 | 湘潭大学 | A kind of method of the intelligent building microgrid load prediction based on deep learning |
CN108321793A (en) * | 2018-01-17 | 2018-07-24 | 东北电力大学 | The active distribution network of integrated intelligent building flexible load models and Optimization Scheduling |
CN109270841A (en) * | 2018-10-11 | 2019-01-25 | 东北电力大学 | A kind of energy flexibility regulation method of the intelligent building based on Model Predictive Control |
FR3077119A1 (en) * | 2018-01-23 | 2019-07-26 | Saint-Gobain Isover | METHOD AND DEVICE FOR PREDICTIVE DETERMINATION OF A THERMAL COMFORT INDICATOR OF A LOCAL |
-
2020
- 2020-06-22 CN CN202010571025.2A patent/CN111898856B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2015172560A1 (en) * | 2014-05-16 | 2015-11-19 | 华南理工大学 | Central air conditioner cooling load prediction method based on bp neural network |
CN105160437A (en) * | 2015-09-25 | 2015-12-16 | 国网浙江省电力公司 | Load model prediction method based on extreme learning machine |
CN107423839A (en) * | 2017-04-17 | 2017-12-01 | 湘潭大学 | A kind of method of the intelligent building microgrid load prediction based on deep learning |
CN108321793A (en) * | 2018-01-17 | 2018-07-24 | 东北电力大学 | The active distribution network of integrated intelligent building flexible load models and Optimization Scheduling |
FR3077119A1 (en) * | 2018-01-23 | 2019-07-26 | Saint-Gobain Isover | METHOD AND DEVICE FOR PREDICTIVE DETERMINATION OF A THERMAL COMFORT INDICATOR OF A LOCAL |
CN109270841A (en) * | 2018-10-11 | 2019-01-25 | 东北电力大学 | A kind of energy flexibility regulation method of the intelligent building based on Model Predictive Control |
Non-Patent Citations (2)
Title |
---|
汤奕 等: ""基于物理-数据融合的数字化楼宇用电模型构建方法"", 《供用电》, no. 10, pages 16 - 21 * |
陆俊 等: ""基于极限学习机的居民用电行为分类分析方法"", 《电力系统自动化》, vol. 43, no. 2, pages 97 - 104 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112669446A (en) * | 2020-12-24 | 2021-04-16 | 联通(浙江)产业互联网有限公司 | Building scene modeling method and device |
CN112669446B (en) * | 2020-12-24 | 2024-04-19 | 联通(浙江)产业互联网有限公司 | Building scene modeling method and device |
CN113065190A (en) * | 2021-04-15 | 2021-07-02 | 天津大学 | Uncertainty-based residential building heat supply amount calculation method |
CN113065190B (en) * | 2021-04-15 | 2022-05-27 | 天津大学 | Uncertainty-based residential building heat supply amount calculation method |
CN114237057A (en) * | 2021-12-20 | 2022-03-25 | 东南大学 | Dynamic modeling method, system, equipment and medium for intelligent building electricity analysis |
CN114237057B (en) * | 2021-12-20 | 2023-09-19 | 东南大学 | Dynamic modeling method, system, equipment and medium for intelligent building electricity analysis |
Also Published As
Publication number | Publication date |
---|---|
CN111898856B (en) | 2024-03-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Fu et al. | Applications of reinforcement learning for building energy efficiency control: A review | |
Harish et al. | A review on modeling and simulation of building energy systems | |
Zhang et al. | An optimal and learning-based demand response and home energy management system | |
CN111898856B (en) | Analysis method of physical-data fusion building based on extreme learning machine | |
US11859838B2 (en) | System and method for aligning HVAC consumption with renewable power production with the aid of a digital computer | |
Noye et al. | A review of advanced ground source heat pump control: Artificial intelligence for autonomous and adaptive control | |
Yan et al. | The performance prediction of ground source heat pump system based on monitoring data and data mining technology | |
CN113112077B (en) | HVAC control system based on multi-step prediction deep reinforcement learning algorithm | |
Olabi et al. | Application of artificial intelligence for prediction, optimization, and control of thermal energy storage systems | |
KR102194002B1 (en) | energy operation management system by using optimized physical learning model and machine learning methods | |
CN113572157A (en) | User real-time autonomous energy management optimization method based on near-end policy optimization | |
Deb et al. | A machine learning-based framework for cost-optimal building retrofit | |
Shann et al. | An active learning approach to home heating in the smart grid | |
Sun et al. | Research on control strategy integrated with characteristics of user's energy-saving behavior of district heating system | |
Abtahi et al. | Control-oriented thermal network models for predictive load management in Canadian houses with on-Site solar electricity generation: application to a research house | |
Han et al. | Physical-data fusion modeling method for energy consumption analysis of smart building | |
Ahmed et al. | Optimal design, operational controls, and data-driven machine learning in sustainable borehole heat exchanger coupled heat pumps: Key implementation challenges and advancement opportunities | |
CN108303898B (en) | Intelligent scheduling method of novel solar-air energy coupling cold and heat cogeneration system | |
Ferrarini et al. | Predictive control-oriented models of a domestic air-to-water heat pump under variable conditions | |
Danza et al. | Application of model predictive control for the optimization of thermo-hygrometric comfort and energy consumption of buildings. | |
Dominguez et al. | Distributed co-simulation for smart homes energy management in the presence of electrical thermal storage | |
Avotins et al. | IoT Solution Approach for Energy Consumption Reduction in Buildings: Part 3. Mathematical Model of Building and Experimental Results | |
Afram et al. | Data-driven modeling of thermal energy storage tank | |
Qin et al. | Exploring New Building Energy Saving Control Strategy Application under the Energy Internet of Things | |
Abraham-Reynolds | Quantifying the energy flexibility potential of commercial buildings with Thermally Active Building Systems |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |