CN107122569B - Power consumption prediction method for difference parallel HCMAC neural network bilateral variable flow ground source heat pump system - Google Patents
Power consumption prediction method for difference parallel HCMAC neural network bilateral variable flow ground source heat pump system Download PDFInfo
- Publication number
- CN107122569B CN107122569B CN201710380610.2A CN201710380610A CN107122569B CN 107122569 B CN107122569 B CN 107122569B CN 201710380610 A CN201710380610 A CN 201710380610A CN 107122569 B CN107122569 B CN 107122569B
- Authority
- CN
- China
- Prior art keywords
- neural network
- hcmac
- ground source
- heat pump
- power consumption
- 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.)
- Active
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 139
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000002146 bilateral effect Effects 0.000 title claims abstract description 20
- 230000006870 function Effects 0.000 claims description 33
- 239000013598 vector Substances 0.000 claims description 31
- 238000013139 quantization Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 125000004122 cyclic group Chemical group 0.000 claims 1
- 238000003062 neural network model Methods 0.000 abstract description 2
- 238000012360 testing method Methods 0.000 description 19
- 238000004422 calculation algorithm Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 5
- 238000013461 design Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000012530 fluid Substances 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000008400 supply water Substances 0.000 description 1
- 238000012956 testing procedure Methods 0.000 description 1
- 238000004148 unit process Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- 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
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Water Supply & Treatment (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Molecular Biology (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Air Conditioning Control Device (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a power consumption prediction method of a differential parallel HCMAC neural network bilateral variable flow ground source heat pump system, which comprises the steps of determining the input of a power consumption prediction model of the ground source heat pump system; dividing a sample data space into a plurality of subspaces according to the characteristics of sample data, determining a reference value output by each subspace to obtain a difference output by each subspace, and learning an output difference value by using an HCMAC neural network submodel; and inputting the sample to be predicted into the constructed differential parallel HCMAC neural network to obtain a power consumption prediction result of the ground source heat pump system. The prediction accuracy of the method is far higher than that of a conventional HCMAC neural network model, and the method can be used for predicting the operation power consumption of the bilateral variable flow ground source heat pump system.
Description
Technical Field
The invention relates to a power consumption prediction method for a differential parallel HCMAC neural network bilateral variable flow ground source heat pump system.
Background
The ground source heat pump technology utilizing shallow geothermal energy is rapidly developed in our country, and the project scale of the ground source heat pump is in the leading position in the world at present. However, the design department usually designs the ground source heat pump system according to the maximum load when designing the ground source heat pump system, which results in the ground source heat pump system working in the partial load state most of the time.
With the improvement of the ground source heat pump unit process technology, variable flow operation of the heat pump unit within a certain range becomes possible, and the flow of the evaporator and the condenser can be changed within 30-130% of the designed flow. The screw heat pump unit can realize stepless regulation of 10-100% of cold and heat load through the control of a slide valve of the screw heat pump unit. The development of the technologies enables variable-frequency variable-flow control of the circulating pump of the ground source heat pump system.
At present, a control method of a variable frequency pump mainly comprises constant temperature difference control and constant pressure difference control, but the control method cannot realize optimal control of a circulating pump under the current load. Generally, when the circulation pump frequency is reduced, the power consumption of the circulation pump is reduced, but at the same time, the power consumption of the heat pump unit is increased. To realize the optimal control of the circulating pump, it is very important to establish a power consumption prediction model of the ground source heat pump system. At present, a ground source heat pump system modeling method mainly adopts a physical mechanism modeling method, but some parameters are difficult to determine in the modeling process, so that difficulties are brought to modeling and model precision.
With the development of information technology, a large amount of data can be generated in the operation process of the ground source heat pump system, and no report is found on how to establish a power consumption model of the ground source heat pump system according to the collected effective data of the ground source heat pump system. At present, in the practical application of frequency conversion control of a ground source heat pump, most of the frequency conversion control of a load side is adopted, and a circulating pump at the ground source side usually works in a fixed frequency mode. From a large amount of measured engineering data, the temperature difference between the supply water and the return water at the ground source side is basically between 2 and 3 ℃, namely the ground source side works under the working condition of small temperature difference and large flow, which is unfavorable for energy saving. If the load side circulating pump and the ground source side circulating pump simultaneously carry out frequency conversion according to the change of the user load, namely the ground source heat pump system works in a bilateral variable flow state, the ground source heat pump system has larger energy-saving space.
Disclosure of Invention
The invention provides a method for predicting the power consumption of a differential parallel HCMAC (hybrid parallel double parallel hybrid HCMAC) neural network bilateral variable flow ground source heat pump system in order to solve the problems.
In order to achieve the purpose, the invention adopts the following technical scheme:
a power consumption prediction method for a differential parallel HCMAC neural network bilateral variable flow ground source heat pump system comprises the following steps:
(1) determining the input of a power consumption prediction model of a ground source heat pump system;
(2) dividing the sample data space into a plurality of subspaces according to the characteristics of the sample data, determining the reference value output by each subspace, obtaining the difference output by each subspace, and determining the quantization progression and neural network grid division of each HCMAC submodel.
(3) And learning the weight coefficient of each sub-model neural network of the DPHCMAC according to the sample learning data.
(4) And inputting the sample to be predicted into the constructed DPHCMAC neural network to obtain a power consumption prediction result of the ground source heat pump system.
Further, in the step (1), the power consumption of the ground source heat pump system is related to the frequency of the ground source side circulating pump, the frequency of the load side circulating pump and the current load of the user, and finally, the input of the power consumption prediction model of the ground source heat pump system is determined to be the frequency of the ground source side circulating pump, the frequency of the load side circulating pump and the current load of the user.
In the step (2), the subspaces are divided according to the input user load interval, and the frequency interval of the ground source side circulating pump and the frequency interval of the load side circulating pump of each subspace are kept consistent.
In the step (2), the DPHCMAC neural network structure is formed by connecting 8 sub-HCMAC models in parallel, and a learning sample data space is divided into 8 subspaces.
In the step (2), each input subspace is normalized, a reference value of each output subspace is determined according to the output space of each subspace, and an output difference value of each output subspace is calculated to form a new output difference subspace.
In the step (2), the neural network grids of each subspace are divided identically, and the frequency of the ground source side circulating pump, the frequency of the load side circulating pump and the current load of the user are quantized in the same number of stages.
In the step (2), the node vectors are formed by encoding the intersection points of the neural network grids, the node vectors of each neural network submodel are the same, and meanwhile, the weight vector of the neural network nodes is established, and under the initial condition, the weight of each node of the neural network is 0.
In the step (3), a neural network basis function is constructed, a Gaussian basis function is defined on the hypercomplex, and if the neural network nodes outside the hypercomplex are not selected, the basis function value is 0; the neural network nodes within the superobturator are selected and the closer to the center of the superobturator, the larger the value of the basis function.
In the step (3), a basis function value of each neural network node is calculated according to the current input to obtain a basis function vector, and each sub-model determines a basis function sub-matrix. The neural network grid division, the neural network node vector, the hyper-closed sphere radius and the standard deviation of each sub-model are the same, and the neural network weight vectors of different sub-models are obtained through the learning sample of each subspace.
In the step (3), the output of the sub-model HCMAC neural network is the algebraic sum of the products of all the activation node basis functions and the weight coefficient.
In the step (3), the C-L algorithm is adopted to carry out HCMAC submodel neural network learning.
In the step (4), the input subspace X is determined according to the value of the user load in the input datakSimultaneous determination of neural network submodels HCMACk。
In the step (4), input data is normalized, and HCMAC is carried out according to the established DPHCMAC neural network submodel HCMACkAnd calculating the output of the submodels.
And (4) carrying out differential inverse transformation according to the output of the sub-model HCMAC neural network to obtain the output of the DPHCMAC neural network.
Further, the method specifically comprises the following steps: determining the number of first-layer circulation times according to the number of subspace learning samples, determining the number of second-layer circulation times according to the vector length of neural network nodes, calculating the Euclidean distance between an input and each neural network node, determining the neural network nodes contained in a hypercomplex sphere, forming each input basis function vector, calculating each input basis function vector of a subspace, finally forming a basis function matrix, determining the number of neural network circulation learning times, and calculating the output of a neural network; learning a neural network submodel HCMAC from a deviation between an actual value and a neural network output valuekUntil the learning times are reached, the learning is finished.
Compared with the prior art, the invention has the beneficial effects that:
the DPHCMAC neural network divides a learning space into a plurality of subspaces according to the characteristics of power consumption learning data of a bilateral variable flow ground source heat pump system, and determines the output of each subspaceAnd obtaining the difference of each subspace output according to the obtained reference value. The learning of HCMAC neural network output is changed into differential learning, so that the domain interval and the value of learning data output are greatly reduced, compared with the conventional HCMAC model, the learning precision and the testing precision of the model are obviously improved, and the power consumption P of the bilateral variable-flow ground source heat pump system can be realizedzAccurate prediction of.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.
Fig. 1 is a power consumption structure diagram of a ground source heat pump system of the present invention;
FIG. 2 is a graph of sample learning data for the present invention;
FIG. 3 shows a load of 150kW P according to the present inventionzA variation graph;
FIG. 4 is P of the present inventionzA DPHCMAC prediction model structure diagram;
FIG. 5(a) shows P of the present inventionzLearning a prediction curve and an actual curve graph by HCMAC;
FIG. 5(b) is P of the present inventionzParallel HCMAC learning prediction curve and actual curve graph;
FIG. 5(c) is P of the present inventionzLearning a prediction curve and an actual curve graph by the DPHCMAC;
FIG. 5(d) is a schematic diagram comparing the learning errors of HCMAC and DPHCMAC of the present invention;
FIG. 6(a) shows P of the present inventionzTesting a prediction curve and an actual curve graph by the DPHCMAC;
FIG. 6(b) is P of the present inventionzDPHCMAC test prediction error graph.
The specific implementation mode is as follows:
the invention is further described with reference to the following figures and examples.
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
As introduced by the background art, in the prior art, parameters in the modeling process are difficult to determine, and difficulty is brought to modeling and model precision.
In a typical embodiment of the present application, the ground source heat pump system mainly includes a ground source heat pump unit, a load side circulating pump, a ground source side circulating pump, an air conditioner terminal, a ground heat exchanger, and the like. The power consumption of the ground source heat pump system is not considered, and comprises the power consumption of a ground source heat pump unit, the power consumption of a load side circulating pump and the power consumption of a ground source side circulating pump. Under similar working conditions, the power consumption of the circulating pump is in a cubic relation with the change of the frequency. The mathematical expression is:
in the formula, P0Power under power frequency of the circulating pump, kW;
p is circulating pump power under the current frequency, kW;
f-current circulation pump frequency, Hz.
The change of the circulating pump flow is in a linear relation with the change of the circulating pump frequency, namely:
in the formula, M0Power frequency down flow of circulating pump, m3/h；
M-circulation Pump flow at Current frequency, M3/h。
Along with the increase of the frequency of the ground source side circulating pump and the load side circulating pump, the flow rates of the evaporator and the condenser of the heat pump unit are increased linearly, but the operating power of the heat pump unit is gradually reduced. The evaporator and condenser flow rates can be reduced to 30% of the design flow rate. Generally, the minimum value of the frequency of the circulating pump can be set to be 20Hz, and according to the formula (2), the flow rate of the circulating pump is 40% of the design flow rate, so that the safe operation condition of the heat pump unit is met. Under the condition that the outlet temperature of the evaporator and the return water temperature of the condenser are not changed, the power consumption of the heat pump unit is related to the current user load, the flow of the evaporator and the flow of the condenser. The specific structural block diagram is shown in fig. 1.
The total power consumption of the ground source heat pump system can be expressed as:
Pz＝Pj(fl,fg，Q)+Pl(fl)+Pg(fg) (3)
in the formula (f)l-load side circulation pump frequency, Hz;
fg-ground source side circulation pump frequency, Hz;
q-current user load, kW;
Pl-load side circulation pump power consumption, kW;
Pg-power consumption of the ground source side circulating pump, kW;
Pjpower consumption of the ground source heat pump unit, kW;
Pz-total power consumption of the ground source heat pump system; kW.
The current load of the user is:
Q＝cρM(tg-th) (4)
in the formula, M is the evaporator flow rate, M3/h；
tg-evaporator leaving water temperature, deg.c;
th-evaporator return water temperature, deg.c;
rho-fluid Density, m3/kg；
c-specific heat capacity of fluid, kJ/(kg ℃).
From the above analysis, it can be seen that the power consumption of the ground source heat pump system is related to the frequency of the ground source side circulating pump, the frequency of the load side circulating pump and the current load of the user, and finally the input of the power consumption prediction model of the ground source heat pump system is determined to be fl、fgAnd Q. .
The learning and testing sample data come from a built bilateral variable flow TRNSYS simulation platform, the heat pump unit selects EXLSR250.1, the rated capacity is 225.7kW, the rated COP is 5, and the rated flow of the evaporator is 38.8m3Per hour, rated flow of condenser is 46.5m3The rated backwater temperature of the condenser is 26.667 ℃, and the rated outlet water temperature of the evaporator is 7 ℃.
The overall situation of the sample data of the simulation platform shows that the power consumption P of the ground source heat pump systemzThe power consumption P of the ground source heat pump system is different corresponding to different circulating pump frequencies on the ground source side and the load side under the same loadzBut its power consumption varies only to a small extent. As shown in FIG. 2, the power consumption of the global samples ranges from 15.57kW to 64.04kW when the user load varies from 50kW to 225 kW. When the user load is 150kW, the variation range of the power consumption of the ground source heat pump system is 32.45 kW-35.22 kW. FIG. 3 shows the frequency f of the circulation pump on the load side when the load of the user is 150kWlAnd the frequency f of the ground source side circulating pumpgP corresponding to the change from 20Hz to 50Hz respectivelyzA three-dimensional map is changed.
In order to improve the learning accuracy of the prediction model, a DPHCMAC neural network structure is proposed according to the characteristics of sample data, as shown in fig. 4.
In the figure, Pz01，…，Pz08The output reference values of the subspaces 1-8 are HCMAC1, HCMAC2, … and HCMAC8 which are HCMAC submodels. Delta PzIn order to output the differential value,
the prediction difference value is output for the neural network,
and outputting a predicted value for the neural network. The DPHCMAC neural network structure is formed by connecting 8 sub-HCMAC models in parallel, a learning sample data space is divided into 8 subspaces, and a specific algorithm is as follows.
DPHCMAC neural network initialization
(1) The input space X is 3-dimensional, X ═ Fl×FgXQ. Because the power frequency of the circulating pump is 50Hz, the rated capacity of the ground source heat pump unit is 225.7kW, and the input interval of each dimension is defined as Fl＝[0,50]，Hz；Fg＝[0,50]，Hz；Q＝[0,225.7]kW. According to the load interval of the user, dividing an input space X into s subspaces, wherein the subspaces are defined as Xk＝Flk×Fgk×QkK is 1,2, …, s. Interval of each dimension of the subspace: flk＝Fl＝[0,50]，Fgk＝Fg＝[0,50]All subspaces remain consistent. Q1＝[0,50],Q2＝[50,75]，Q3＝[75,100]，Q4＝[100,125]，Q5＝[125,150]，Q6＝[150,175]，Q7＝[175,200]，Q8＝[200,225.7]. Thereby dividing the entire input space into 8 subspaces.
(2) Normalizing each input subspace
In the formula, QkminAnd QkmaxRespectively, the lower and upper limit of the k-th subspace user load. Then arbitrarily input subspace XkArbitrary input x ofki＝[flki,fgki,Qki]Normalized to
(3) According to the output space P of each subspacezkDetermining a reference value P for each output subspacez0k. Satisfies the following conditions:
Pz0k<＝Pzkmin(6)
in the formula, PzkminIs a subspace PzkIs measured. General Pz0kTaking an integer.
(3) Computing each output subspace PzkForming a new output difference subspace.
ΔPzk＝Pzk-Pz0k(7)
(4) Since each subspace is normalized, the neural network meshing for each subspace is the same, here Flk、FgkAnd QkTaking the same quantization level number, wherein the quantization level number QL of each dimension is [0,0.2,0.4,0.6,0.8, 1.0%]. Coding the intersection points of the neural network grids to form node vectors PkThe number of the neural network nodes of each HCMAC neural network submodel is 63＝216，Pk＝[pk1,pk2,…,pkl]And l is 216, the node number is the same for the node vectors of 8 neural network submodels. Simultaneously establishing weight vectors q of neural network nodeskInitially, the weight of each node of the neural network is 0, q0k＝[0,0,…,0]l。
Neural network learning algorithm
(1) Basis function b of neural networkki(·)
Wherein j is 1,2, …, l.
At CiAbove defined Gaussian base function bki(·).
In the formula, σ represents a standard deviation. The neural network nodes outside the super-closed sphere are not selected, and the basis function value is 0; selecting the neural network node in the super-closed sphere and keeping the neural network node away from the center of the super-closed sphere
The more recent, the larger the value of the basis function. According to the current input
The basis function value of each neural network node can be calculated to obtain a basis function vector Bki，Bki＝[bki(p1),bki(p2),…,bki(pl)]. Finally, each submodel determines a submatrix of basis functions.
Because all input subspaces are normalized, each HCMAC submodel is consistent, and the neural network grid division, the neural network node vector and the radius R of the super-closed sphere are realizedbAnd the standard deviation σ are the same. Except that the neural network weight vectors obtained by learning are different.
(2) Compute submodel HCMACkOutput of neural network
Sub-model HCMACkThe output of the neural network is the algebraic sum of the products of the basis functions and the weight coefficients of all the activated nodes.
(3) HCMAC neural network learning algorithm
The HCMAC neural network learning algorithm adopts a C-L algorithm.
qk,m＝qk,m-1+Δqk,m-1(14)
Where m is the current number of learning times, ek,m-1Alpha and beta are constants, 0, for the difference between the actual value of the previous cycle and the predicted value of HCMAC<α<2，β>0。
(4) Inverse differential transform
And finishing learning each neural network submodel according to the method.
Neural network learning procedure
The DPHCMAC neural network is first initialized as described above. The k subspace neural network learning process is as follows, and the other subspace learning processes are the same.
Step 1: determining the number of first-layer circulation times N according to the number of subspace learning samplessk。
Step2: according to the neural network node vector PkThe length determines the number of second layer cycles/.
Step 3: calculating normalized input
And each neural network node pkjDetermining the neural network nodes contained in the superorbicular sphere, forming each input
Base function vector B ofki。
Step4: computing each input of the subspace
Base function vector B ofkiFinally, a basis function matrix is formed
Step5: and determining the number m of the neural network cycle learning.
Step 6: calculating a neural network output according to equation (11);
step 6: according to the deviation e between the actual value and the neural network output valuek,Learning neural network submodel HCMAC according to equations (12), (13) and (14)kWeighted coefficient vector q ofk。
Step 7: and the learning is finished after the learning times are reached.
Neural network testing procedure
Step1:StestFor the purpose of the sample testing the data space,
according to xiUser load QiIs determined as the input subspace XkSimultaneous determination of neural network submodels HCMACk。
Step2 according to the input subspace XkDiscourse interval of (2), according to formula (5) for xiCarrying out normalization treatment to obtain
Step 3: to be provided with
As a center, RbA hypercomplex sphere is drawn for the radius. Computing the basis function vector B at this inputki，Bki＝[bki,1,bki,2,…,bki,216]。
Step4 computing neural network outputs
Step5 calculation of Pzi
Completion of S according to the above methodtestThe test of (1).
Sample data come from a built bilateral variable flow TRNSYS simulation platform, and a conventional HCMAC neural network, a parallel HCMAC neural network and a DPHCMAC neural network provided by the specification are respectively adopted to carry out simulation experiments, wherein the quantization progression of each dimension of the HCMAC neural network is 6, the number of nodes of the neural network is 63216. The parallel HCMAC neural network and the DPHCMAC neural network respectively comprise 8 parallel HCMAC submodels, the quantization progression of each model is the same as that of the HCMAC neural network, and the number of the quantization progression is 63216, the total number of nodes of the parallel HCMAC neural network and the DPHCMAC neural network is 216 × 8 — 1728. The difference between the parallel HCMAC neural network and the DPHCMAC neural network is that the DPHCMAC neural network learns the difference of each subspace output, and the parallel HCMAC neural network is the same as the HCMAC neural network and learns the output. RbThe determination of the sum σ value needs to take into account the learning accuracy and generalization ability of the model. Empirically, take Rb0.3 and 0.15. The RMSE is used as an evaluation index,
in the formula, eiIs PzThe difference between the actual value and the predicted value.
Model learning
Firstly, an HCMAC neural network is adopted to learn sample data, and the learning error RMSE _ learning1 of the HCMAC model is obtained, namely 0.9816 kW. Is obviously PzThe difference value between the actual value and the predicted value is large, the prediction precision is not high, and the optimization of the power supply frequency of the variable flow circulation pump on both sides of the ground source heat pump is difficult to meet. Then, parallel HCMAC prediction model learning is adopted, and the learning error RMSE _ learning2 of the model is 0.6723 kW. The learning precision of the method is improved,but the number of nodes of the neural network increases. And finally, learning by adopting a DPHCMAC prediction model, wherein the learning error RMSE _ learning3 of the obtained model is 0.1513kW, the number of nodes of the neural network is the same as that of the parallel HCMAC neural network, and the prediction precision is obviously improved.
For clarity of illustration, only 256 learning data corresponding to the load Q of 125kW are plotted in fig. 5, and the learning effect is consistent with that of other loads. FIG. 5(a) is PzThe HCMAC model learns the predicted curve and the actual curve, and P is shown in FIG. 5(b)zLearning the predicted curve and the actual curve by the parallel HCMAC model, wherein P is shown in FIG. 5(c)zThe DPHCMAC model learns the prediction curve and the actual curve, and fig. 5(d) is a HCMAC model and DPHCMAC model learning error curve.
Model testing
And testing the established DPHCMAC neural network prediction model according to the test sample data, and obtaining a test error RMSE _ test of the DPHCMAC neural network model provided by the specification which is 0.2247kW through testing. For clarity of illustration, only 256 test data corresponding to the load Q of 112.5kW are plotted in fig. 6, and the test effect is consistent with that of other loads.
As can be seen from fig. 6, the proposed DPHCMAC neural network has better generalization capability and higher test accuracy, and can meet the requirement of power consumption prediction accuracy of a ground source heat pump system.
Power consumption of bilateral variable flow ground source heat pump system and frequency f of circulating pump at load sidelGround source side circulating pump frequency fgThe overall trend change of the power consumption of the ground source heat pump system is that the power consumption is gradually increased along with the increase of the load of a user; however, under the same load, the circulating pump frequency is different, and the power consumption of the ground source heat pump system fluctuates in a certain range. The method is characterized in that a sample data space is divided into a plurality of subspaces according to the characteristics of the sample data, a reference value output by each subspace is determined, and a difference output by each subspace is obtained.The HCMAC submodel learns the output differential value, and the learning precision of the model is obviously improved because the domain interval of the output differential is far smaller than the actually output domain interval. The learning error of the HCMAC neural network is RMSE _ learning 1-0.9816 kW with RMSE as an error evaluation index. The learning error of the proposed DPHCMAC neural network is RMSE _ learning 3-0.1513 kW, and the testing error of the DPHCMAC neural network is RMSE _ test-0.2247 kW. Obviously, the DPHCMAC neural network has better learning precision and testing precision, and can realize the power consumption P of the bilateral variable-flow ground source heat pump systemzAccurate prediction of.
The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (6)
1. A power consumption prediction method for a differential parallel HCMAC neural network bilateral variable flow ground source heat pump system is characterized by comprising the following steps: the method comprises the following steps:
(1) determining the input of a power consumption prediction model of a ground source heat pump system;
(2) dividing the sample data space into multiple subspaces according to the characteristics of the sample data, dividing the input space X into s subspaces according to the load interval of the user, wherein the subspaces are defined as Xk＝F1k×Fgk×Qk，k＝1,2,…,s，QkIs the current user load interval of the kth subspace, FgkIs the frequency interval of the circulating pump at the ground source side of the kth subspace F1kIs the kth subspace load side circulation pump frequency interval; seed of Japanese apricotInterval of each dimension of space: flk＝Fl＝[0,50]，Fgk＝Fg＝[0,50]In units of Hz, FlkAnd FgkMaintain consistency across subspaces, Q1＝[0,50]，Q2＝(50,75]，Q3＝(75,100]，Q4＝(100,125]，Q5＝(125,150]，Q6＝(150,175]，Q7＝(175,200]，Q8＝(200,225.7]Unit kW, so that the whole input space is divided into 8 subspaces, and normalization processing is performed on each subspace;
according to the output space P of each subspacezkDetermining a reference value P for each output subspacez0kAnd satisfies the following conditions:
Pz0k<＝Pzkmin(6)
in the formula, PzkminFor each subspace, an output space PzkMinimum value of (1), Pz0kTaking an integer;
computing the output space P for each subspacezkOutput differential value Δ P ofzkAnd a new output differential subspace is formed,
ΔPzk＝Pzk-Pz0k(7)；
the neural network meshing of each subspace is the same, here Flk、FgkAnd QkTaking the same quantization series, coding the intersection points of the neural network grids to form a neural network node vector Pk；
The node vectors of each neural network sub-model are the same, and a neural network node weight vector q is established simultaneouslykForming a DPHCMAC neural network, wherein the DPHCMAC is a differential parallel HCMAC neural network;
(3) learning the weight coefficient of each sub-model neural network of the DPHCMAC according to the sample learning data,
the neural network learning method of the DPHCMAC comprises the following steps:
step 1: determining the number of first-layer circulation times N according to the number of subspace learning samplessk；
Step2: according to the neural network node vector PkDetermining the cycle times l of the second layer by the length;
step 3: calculating normalized input
And each neural network node pkjDetermining the neural network nodes contained in the superorbicular sphere, forming each input
Base function vector B ofki；
Step4: computing each input of the subspace
Base function vector B ofkiFinally, a basis function matrix is formed
Step5: determining the number m of cyclic learning times of the neural network;
step 6: calculating the neural network output:
step 7: according to the deviation e between the actual value and the neural network output valuekLearning the neural network submodel HCMAC according to the equations (12), (13) and (14)kWeighted coefficient vector q ofk；
qk,m＝qk,m-1+Δqk,m-1(14)
Where m is the current cycle learning number, ek,m-1For last cycle actual value and HCMACkDifference between predicted values, α and β beingConstant value of 0<α<2，β>0；
Step 8: when the learning times are up, the learning is finished;
(4) and inputting the sample to be predicted into the constructed DPHCMAC neural network to obtain a power consumption prediction result of the ground source heat pump system.
2. The method for predicting the power consumption of the differentially parallel HCMAC neural network bilateral variable flow ground source heat pump system of claim 1, wherein the method comprises the following steps: in the step (1), the power consumption of the ground source heat pump system is related to the frequency of the ground source side circulating pump, the frequency of the load side circulating pump and the current load of the user, and finally the input of the power consumption prediction model of the ground source heat pump system is determined to be the frequency of the ground source side circulating pump, the frequency of the load side circulating pump and the current load of the user.
3. The method for predicting the power consumption of the differentially parallel HCMAC neural network bilateral variable flow ground source heat pump system of claim 1, wherein the method comprises the following steps: in the step (2), the node vectors are formed by encoding the intersection points of the neural network grids, the node vectors of each neural network submodel are the same, and meanwhile, the weight vector of the neural network nodes is established, and under the initial condition, the weight of each node of the neural network is 0.
4. The method for predicting the power consumption of the differentially parallel HCMAC neural network bilateral variable flow ground source heat pump system of claim 1, wherein the method comprises the following steps: in the step (3), a neural network basis function matrix B is constructedk(Nsk,l)Drawing a hyperclosure sphere by taking an input as a center, defining a Gaussian basis function, and if a neural network node outside the hyperclosure sphere is not selected, setting the basis function value to be 0; the neural network nodes within the superobturator are selected and the closer to the center of the superobturator, the larger the value of the basis function.
5. The method for predicting the power consumption of the differentially parallel HCMAC neural network bilateral variable flow ground source heat pump system of claim 1, wherein the method comprises the following steps: in the step (3), a basis function value of each neural network node is calculated according to the current input to obtain a basis function vector, and each nerveNetwork submodel HCMACkDetermining a basis function matrix
6. The method for predicting the power consumption of the differentially parallel HCMAC neural network bilateral variable flow ground source heat pump system of claim 1, wherein the method comprises the following steps: in the step (4), the input subspace X is determined according to the value of the user load in the sample data to be predictedkAnd simultaneously determining the sub-model HCMAC of the DPHCMAC neural networkk；
Normalizing the sample data to be predicted according to the established sub-model HCMAC of the DPHCMAC neural networkkCalculating the output of the submodel;
according to the sub-model HCMAC of the DPHCMAC neural networkkAnd performing differential inverse transformation on the output of the DPHCMAC neural network to obtain the output of the DPHCMAC neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710380610.2A CN107122569B (en) | 2017-05-25 | 2017-05-25 | Power consumption prediction method for difference parallel HCMAC neural network bilateral variable flow ground source heat pump system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710380610.2A CN107122569B (en) | 2017-05-25 | 2017-05-25 | Power consumption prediction method for difference parallel HCMAC neural network bilateral variable flow ground source heat pump system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107122569A CN107122569A (en) | 2017-09-01 |
| CN107122569B true CN107122569B (en) | 2020-10-30 |
Family
ID=59728947
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710380610.2A Active CN107122569B (en) | 2017-05-25 | 2017-05-25 | Power consumption prediction method for difference parallel HCMAC neural network bilateral variable flow ground source heat pump system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107122569B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816166B (en) * | 2019-01-17 | 2022-11-29 | 山东大学 | Ground source heat pump system performance prediction method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102589092A (en) * | 2012-03-12 | 2012-07-18 | 山东建筑大学 | Indoor-environment thermal comfort control method based on novel fuzzy controller |
| CN102854296A (en) * | 2012-08-30 | 2013-01-02 | 北京工业大学 | Sewage-disposal soft measurement method on basis of integrated neural network |
| JP5690650B2 (en) * | 2011-05-09 | 2015-03-25 | 新日鉄住金エンジニアリング株式会社 | Geothermal characteristics analysis method and apparatus, soil heat source heat pump system operation adjustment method and apparatus, and program |
| CN105302984A (en) * | 2015-11-12 | 2016-02-03 | 山东建筑大学 | Modeling and simulation method for ground source heat pump set |
-
2017
- 2017-05-25 CN CN201710380610.2A patent/CN107122569B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5690650B2 (en) * | 2011-05-09 | 2015-03-25 | 新日鉄住金エンジニアリング株式会社 | Geothermal characteristics analysis method and apparatus, soil heat source heat pump system operation adjustment method and apparatus, and program |
| CN102589092A (en) * | 2012-03-12 | 2012-07-18 | 山东建筑大学 | Indoor-environment thermal comfort control method based on novel fuzzy controller |
| CN102854296A (en) * | 2012-08-30 | 2013-01-02 | 北京工业大学 | Sewage-disposal soft measurement method on basis of integrated neural network |
| CN105302984A (en) * | 2015-11-12 | 2016-02-03 | 山东建筑大学 | Modeling and simulation method for ground source heat pump set |
Non-Patent Citations (2)
| Title |
|---|
| "基于多个并行CMAC神经网络的强化学习方法";李明爱 等;《***仿真学报》;20081231;第20卷(第24期);第6683-6685页 * |
| "超必球CMAC的性能分析及多CMAC结构";段培永，任化芝;《自动化学报》;20000731;第26卷(第4期);第563-567页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107122569A (en) | 2017-09-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106920006B (en) | Subway station air conditioning system energy consumption prediction method based on ISOA-LSSVM | |
| CN106600059B (en) | Intelligent power grid short-term load prediction method based on improved RBF neural network | |
| US20220027539A1 (en) | Method and system for manufacturing a heat exchanger for supercritical pressure fluid | |
| Zhou et al. | Comparison of different deep neural network architectures for isothermal indoor airflow prediction | |
| CN105760952A (en) | Load prediction method based on Kalman filtering and self-adaptive fuzzy neural network | |
| CN111069328A (en) | Isothermal extrusion process parameter optimization method based on particle swarm optimization | |
| CN107615186A (en) | The method and apparatus of Model Predictive Control | |
| CN116882708B (en) | Steel process flow control method and device based on digital twin and related equipment | |
| CN113343427B (en) | Structural topology configuration prediction method based on convolutional neural network | |
| Wu et al. | Application of GA-BPNN on estimating the flow rate of a centrifugal pump | |
| CN112800675A (en) | KPCA and ELM-based time-space separation distribution parameter system modeling method | |
| CN110826784A (en) | Energy use efficiency prediction method and device, storage medium and terminal equipment | |
| CN107122569B (en) | Power consumption prediction method for difference parallel HCMAC neural network bilateral variable flow ground source heat pump system | |
| Niu et al. | Model of turbine optimal initial pressure under off-design operation based on SVR and GA | |
| CN112966399B (en) | Pulse tube refrigerator working condition prediction method and system based on machine learning | |
| Ismail et al. | Optimizing the parameters of Sugeno based adaptive neuro fuzzy using artificial bee colony: A case study on predicting the wind speed | |
| Jin et al. | Artificial intelligence enabled energy-saving drive unit with speed and displacement variable pumps for electro-hydraulic systems | |
| CN117407637A (en) | High-index differential algebraic equation neural network solving method based on Radau method | |
| CN116717839A (en) | Heating control method, control device and heating system | |
| Yan-Gao et al. | Electricity price forecasting based on support vector machine trained by genetic algorithm | |
| CN111025898A (en) | Dimension reduction identification method for large-scale process control of process industry | |
| CN112989275B (en) | Multidirectional method for network large-scale control system | |
| CN112084596B (en) | Intelligent recommendation system and recommendation method for structural parameters of floor | |
| Setiyowati et al. | Water consumption prediction of semarang water utilities using support vector regression radial basic function kernel method | |
| Hui et al. | Neural network prediction methods of power consumption for GSHP system with bilateral variable flow |
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 |