CN115017799B - EM-DBN fusion-based solar radiation estimation method for quantitative region - Google Patents
EM-DBN fusion-based solar radiation estimation method for quantitative region Download PDFInfo
- Publication number
- CN115017799B CN115017799B CN202210520496.XA CN202210520496A CN115017799B CN 115017799 B CN115017799 B CN 115017799B CN 202210520496 A CN202210520496 A CN 202210520496A CN 115017799 B CN115017799 B CN 115017799B
- Authority
- CN
- China
- Prior art keywords
- model
- data
- dbn
- empirical
- solar radiation
- 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
- 230000005855 radiation Effects 0.000 title claims abstract description 75
- 230000004927 fusion Effects 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 53
- 230000002093 peripheral effect Effects 0.000 claims abstract description 42
- 230000006870 function Effects 0.000 claims abstract description 27
- 238000012360 testing method Methods 0.000 claims abstract description 23
- 230000014509 gene expression Effects 0.000 claims abstract description 14
- 238000009826 distribution Methods 0.000 claims description 16
- 238000005070 sampling Methods 0.000 claims description 16
- 238000010801 machine learning Methods 0.000 claims description 10
- 238000013526 transfer learning Methods 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 5
- 238000005259 measurement Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 230000006978 adaptation Effects 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000013508 migration Methods 0.000 abstract description 7
- 230000005012 migration Effects 0.000 abstract description 7
- 238000010248 power generation Methods 0.000 abstract description 5
- 238000011156 evaluation Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000001737 promoting effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012706 support-vector machine 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
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/04—Power grid distribution networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Photometry And Measurement Of Optical Pulse Characteristics (AREA)
Abstract
The invention discloses a method for estimating solar radiation in a non-measuring area based on EM-DBN fusion, which belongs to the technical field of power systems and comprises the following steps: based on the existing data set, selecting a solar total radiation estimation experience model; constructing a training set and a testing set by utilizing meteorological data and irradiance data of peripheral sites and target sites, and fitting a function expression of an empirical model based on training set data; the empirical model and the DBN model are fused in a characteristic dimension expansion mode, and an EM-DBN fusion model is constructed by utilizing peripheral site data; and combining with a migration learning algorithm, migrating the constructed EM-DBN fusion model to a non-measuring target site to obtain a solar radiation estimated value. The method solves the problem of radiation resource evaluation of the non-radiation measuring area, reduces the harm of uncertainty of solar power generation to a power system, and improves the utilization efficiency of solar energy.
Description
Technical Field
The invention belongs to the field of power systems, and particularly relates to a non-measuring area solar radiation estimation method based on EM-DBN fusion.
Background
Because solar radiation measuring equipment is high in manufacturing cost and high in maintenance cost, 2400 weather observation sites in China are provided, and only 98 solar radiation observation sites are provided. Therefore, the solar radiation in the non-radiation measuring area is estimated according to the existing meteorological parameter data, and the method has a certain research significance.
The total solar radiation estimate can be obtained by two methods, an empirical model and a machine learning model, wherein the empirical model describes the relationship between the total solar radiation and other meteorological parameters by using a specific formula. The irradiation intensity empirical model has the advantage that an explicit mechanism formula of solar radiation can be integrated into a calculation model, so that the estimation accuracy is high, but the calculation complexity is high because a plurality of regression coefficients are involved in calculation. The machine learning model, such as an artificial neural network, a support vector machine, a self-adaptive fuzzy neural reasoning system and the like, can directly input various types of data, has a simple modeling process, but the estimation result lacks theoretical basis and is difficult to explain by using a specific principle.
Disclosure of Invention
The invention aims to: the invention aims to provide a solar radiation estimation method based on EM-DBN fusion, which can obtain a solar radiation estimation value of a non-measuring area, improves the estimation precision of total solar radiation, solves the problem of radiation resource estimation of the non-measuring area, and has theoretical value and practical significance for reasonably utilizing solar resources, improving photovoltaic power generation capacity and promoting sustainable healthy development of economy.
The technical scheme is as follows: the invention discloses a method for estimating solar radiation in a non-measuring area based on EM-DBN fusion, which comprises the following steps:
(1) Based on the existing data set, selecting a solar total radiation estimation experience model;
(2) Constructing a training set and a testing set by utilizing meteorological data and irradiance data of peripheral sites and target sites, and fitting a function expression of an empirical model based on training set data;
(3) The empirical model and the DBN model are fused in a characteristic dimension expansion mode, and an EM-DBN fusion model is constructed by utilizing peripheral site data;
(4) And combining with a migration learning algorithm, migrating the constructed EM-DBN fusion model to a non-measuring target site to obtain a solar radiation estimated value.
In the step (1), an empirical model with relatively high precision is selected as an empirical model for estimating total solar radiation, wherein the empirical model is selected as follows:
empirical model 1:
empirical model 2:
empirical model 3:
empirical model 4:
empirical model 5:
empirical formula 6:
wherein H is total daily radiation, H 0 Represents extraterrestrial radiation; s represents the actual sunshine duration; s is S 0 Representing the maximum possible sunshine duration; RH represents the average relative humidity; t (T) max and Tmin Respectively representing the highest daily temperature and the lowest daily temperature; t represents a daily average temperature; a. b, c, d, e and f are training parameters; wherein S is 0 Depending on the local latitude
The solar declination angle delta of the current day.
In the step (2), the training set and the testing set are constructed by using the meteorological data and irradiance data of the peripheral site and the target site, and the function expression of the empirical model is fitted, and the method specifically comprises the following steps:
(2.1) constructing a training set and a test set of weather data and irradiance data of peripheral sites and destination sites, wherein the training set comprises training samples of m days, and the test set comprises test samples of n days on the assumption that one sample is one day when modeling is performed on a target site; the input on the training set is X (X i ) I=1, 2, …, m, output is Y (Y i ) I=1, 2, …, m, the input on the test set is P (P j ) J=1, 2, …, n, output is Q (Q j ) J=1, 2, …, n; wherein x is i and pj For a row vector of 1×t, T represents the number of meteorological parameters required for building an empirical model, and the highest temperature T is obtained from the day of the day max Minimum temperature T min Average day temperature T, number of sunshine hours S, number of days n in one year day And relative humidity RH; y is i and qj The total solar radiation value of the current day;
(2.2) based on sample data on the training set, building an empirical model, fitting the data on the training set, and solving by least square.
In the step (3), the empirical model and the DBN model are fused in a characteristic dimension expansion mode, and an EM-DBN fusion model is constructed by utilizing peripheral site data, and the method specifically comprises the following steps:
(3.1) based on the input X (X) on the training set according to the empirical model function expression established in step (2) i ) Calculating a solar radiation estimated value of the empirical model on the training set; estimating the solar radiation estimated value of the empirical model and the highest temperature T of the day max Minimum temperature T min Average day temperature T, number of sunshine hours S, number of days n in one year day And average relative humidityRH together as input to the DBN model, the output is still Y (Y i )，i＝1，2，…，m；
(3.2) taking a DBN model as a basis of a fusion model, wherein the DBN model is formed by superposition and combination of a plurality of RBMs from bottom to top; after pre-training by RBM sampling, the output of the RBM of the upper layer is used as the input of the lower layer; training the DBN model layer by layer repeatedly; assume that
Is a set of training samples, and the specific calculation process in the RBM model is as follows:
in the formula ,lnLS Representing the partial derivatives of the respective parameters; n is n s Representing the number of training samples; η (eta)>0 represents a learning rate; θ= (W, a, b) represents parameters in RBM;
is a gradient of RBM; training +.A k-step contrast divergence algorithm is used>
For->
Take the initial value v (0) =v, then perform k steps Gibbs sampling; wherein the t-th step (t=1, 2, …, k) is first performed by using P (h|v (t-1) ) Sampling out h (t-1) And using P (v|h (t-1) ) Sampling v (t) Next, v obtained by sampling with the Gibbs in the k steps (k) To approximate the estimation formula:
wherein ,
and />
For the partial derivative in the gradient-rising iteration, its component is +.>
and />
And (3.3) fusing the experience model and the machine learning model by the characteristic dimension expansion mode in the step (3.1), selecting a peripheral site which is in the same weather zone with the destination site and is closer to the destination site, and constructing an EM-DBN fusion model by utilizing data of the peripheral site.
In the step (4), the constructed EM-DBN fusion model is migrated to a non-measuring target site to obtain a solar radiation estimated value by combining with a migration learning algorithm, specifically: on the basis of the original deep neural network, an adaptation layer is added in front of the last regression layer of the deep neural network; the maximum mean difference introduced into the transfer learning measures the distance of two different but related feature distributions, defining the distance of the two distributions as:
wherein X and T represent two different distributions of source and target domains, n and m the number of samples, X and T, X i and tj Representing the ith and jth samples in X and T, respectively, phi (·) represents the mapping function, and H represents the distance measured by the mapping of data into regenerated Hilbert space by phi (·).
The Gaussian kernel function k (·) is selected to map an infinite dimensional space, and the expression is as follows:
wherein u represents any point in space, v is the kernel function center, and sigma represents bandwidth;
then, the loss function is divided into two parts, and the difference degree between the estimation result and the actual data is expressed by using the two domain distances calculated by the MMD and the tag data estimation error of the peripheral station address, which is specifically as follows:
wherein θ represents the weight and bias of the network; n represents the number of samples of the peripheral site; x is x i An input feature representing an ith sample of the set of peripheral site samples; y is i An output radiation value representing an ith sample of the set of peripheral site samples; j (·) represents a loss function, λ represents a penalty coefficient, d 2 (D s ，D t ) MMD distance representing source and target domain; and utilizing the feature transformation capability of deep learning to transform the feature space until the feature distribution transformed by the peripheral site data set is matched with the feature distribution transformed by the non-measuring site data set, and successfully transferring the EM-DBN fusion model constructed by the peripheral site data to the non-measuring region, thereby obtaining the estimated solar radiation value of the non-measuring site.
The beneficial effects are that: compared with the prior art, the technical scheme of the invention has the beneficial effects that: the method comprises the steps of selecting peripheral stations of a non-measuring area to establish correlation between radiation data and other meteorological parameters, longitude and latitude and elevation, fitting a selected empirical model function expression, establishing an EM-DBN fusion model in a characteristic dimension expansion mode, effectively applying the fusion model to the non-measuring area in a transfer learning mode, improving estimation accuracy of total solar radiation, solving the problem of radiation resource estimation of the non-measuring area, reducing harm of uncertainty of solar power generation to a power system, and improving utilization efficiency of solar energy.
Drawings
FIG. 1 is a schematic flow chart of the present invention;
FIG. 2 is a schematic diagram of a DBN network structure according to the present invention;
FIG. 3 is a schematic diagram of an EM-DBN fusion model structure used in the present invention;
FIG. 4 is a schematic diagram of a deep neural network structure incorporating transfer learning according to the present invention;
FIG. 5 is a diagram showing the final prediction results using the proposed method in an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is described in detail below with reference to the detailed description and the attached drawings.
As shown in fig. 1, the method for estimating solar radiation without a measuring area based on EM-DBN fusion comprises the following steps:
(1) Based on the existing data set, selecting a solar total radiation estimation experience model;
(2) Constructing a training set and a testing set by utilizing meteorological data and irradiance data of peripheral sites and target sites, and fitting a function expression of an empirical model based on training set data;
(3) The empirical model and the DBN model are fused in a characteristic dimension expansion mode, and an EM-DBN fusion model is constructed by utilizing peripheral site data;
(4) And combining with a migration learning algorithm, migrating the constructed EM-DBN fusion model to a non-measuring target site to obtain a solar radiation estimated value.
The following describes in detail, with reference to specific examples, a specific implementation of the method of the present invention for non-survey solar radiation estimation. Taking 10 solar radiation sites randomly selected in China as an example, acquiring the solar radiation sites from 1 month in 1994 to 1 day to 2015 according to the sampling frequency of one sampling point in 1 dayMeteorological data and solar total radiation data of 22 years total of 12 months and 31 days, wherein the meteorological data comprises a daily maximum temperature T max Minimum temperature T min Average day temperature T, number of sunshine hours S, number of days n in one year day And relative humidity RH. The specific implementation steps are as follows:
the method comprises the following steps of (1) selecting a solar total radiation estimation empirical model with relatively high precision, wherein the selected empirical model is specifically as follows:
empirical model 1 (EM 1):
empirical model 2 (EM 2):
empirical model 3 (EM 3):
empirical model 4 (EM 4):
empirical model 5 (EM 5):
empirical formula 6 (EM 6):
wherein H represents total daily radiation; h 0 Represents extraterrestrial radiation; s represents the actual sunshine duration; s is S 0 Representing the maximum possible sunshine duration; RH is the average relative humidity; t (T) max and Tmin Respectively representing the highest daily temperature and the lowest daily temperature; t represents a daily average temperature; a. b, c, d, e and f represent training parameters. Wherein S is 0 Depending on the local latitude
Solar declination angle delta on the same day:
the step (2) is to construct a training set and a testing set by using meteorological data and irradiance data of peripheral sites and destination sites, and fit a function expression of an empirical model, and specifically comprises the following steps:
(2.1) first, a training set and a test set of peripheral site and destination site weather data and irradiance data are constructed. For each site, the data of 19 years from 1 month, 1 year, and 31 days, 1 year, 2012, and 31 years are selected as training sets, and the data of 3 years from 1 month, 1 year, 2013, 1 year, 12 months, 31 days, 2015 are selected as test sets. The training set contained m=7305 days of training samples and the test set contained n=1096 days of test samples. The input on the training set is X (X i ) I=1, 2, …, m, output is Y (Y i ) I=1, 2, …, m, the input on the test set is P (P j ) J=1, 2, …, n, output is Q (Q j ) J=1, 2, …, n. Wherein x is i and pj For a 1×6 row vector, the number of meteorological parameters selected by building an empirical model, namely the day maximum temperature T of the same day max Minimum temperature T min Average day temperature T, number of sunshine hours S, number of days n in one year day And relative humidity RH; y is i and qj Is the total solar radiation value of the current day.
(2.2) based on sample data on the training set, establishing 6 experience models in the step (1), fitting the data on the training set, solving by least square, and establishing the following models:
wherein ,X1 -X 6 Is the input of the empirical models EM1-EM 6; f (f) 1 (·)-f 6 (. Cndot.) are functional expressions of the fitted empirical models EM1-EM6, respectively; and H is the total solar radiation value calculated according to the fitted empirical model function expression.
The experimental model test results are shown in table 1, wherein the bolded values are the optimal values among the 6 experimental models. It can be seen from the table that the 6 empirical models selected have overall higher estimation accuracy. Comparing the estimation results of the plurality of sites, it can be found that the accuracy of the empirical model 5 (EM 5) is the worst among the 6 models, and the accuracy of the empirical model 3 (EM 3) is the highest.
Table 1 6 solar total radiation estimation results of empirical models
The step (3) fuses the experience model and the DBN model in a characteristic dimension expansion mode, and builds an EM-DBN fusion model by utilizing peripheral site data, and specifically comprises the following steps:
(3.1) based on the input X (X) on the training set according to the empirical model function expression established in step (2) i ) Calculating solar radiation estimated values H of 6 experience models on a training set train1 (h 1i )、 H train2 (h 2i )、H train3 (h 3i )、H train4 (h 4i )、H train5 (h 5i) and Htrain6 (h 6i ) Where i=1, 2, …, m. The solar radiation estimated value and the highest temperature T of the day of the 6 experience models max Minimum temperature T min Average day temperature T, number of sunshine hours S, number of days n in one year day Together with the relative humidity RH as input to the DBN model, the output is still Y (Y i )，i＝1，2，…，m。
And (3.2) taking the DBN model as the basis of the fusion model, wherein the DBN model is formed by superposition and combination of a plurality of RBMs from bottom to top, and the DBN network structure is shown in figure 2. After pre-training by RBM sampling, the output of the upper layer RBM will be the input of the lower layer. In order for each layer to learn more about the features of the previous layer, the DBN needs to be trained layer by layer repeatedly. Assume that
Is a set of training samples, and the specific calculation process in the RBM model is as follows:
wherein ,ns The number of training samples; η (eta)>0 is learning rate; θ= (W, a, b) represents parameters in RBM;
(lnL S partial derivatives with respect to the respective parameters) is the gradient of the RBM. Training +.>
For->
Take the initial value v (0) =v, then k steps of Gibbs sampling are performed. Wherein the t-th step (t=1, 2, …, k) is first performed by using P (h|v (t-1 ) Sampled h (t-1 ) And using P (v|h (t-1) ) Sampling v (t) Next, v obtained by sampling with the Gibbs in the k steps (k) To approximate the estimation formula:
wherein ,
and />
(its component is->
and />
) Is the partial derivative in the gradient-up iteration.
(3.3) fusing the empirical model and the machine learning model by means of feature dimension expansion in the step (3.1), wherein the structure of the EM-DBN fusion model is shown in fig. 3. And selecting the peripheral sites of the sites 2, 3 and 8 which are the same weather zone and are closer to each other as the site 7, and constructing an EM-DBN fusion model by using the data of the peripheral sites.
The method randomly selects the validity of the data test experience-machine learning fusion model of 4 sites. Likewise, for each site, 19 years of data from 1 month 1 day 1994 to 12 months 2012 31 were selected as training sets, 3 years of data from 1 month 1 day 2013 to 12 months 2015 were selected as test sets, and the test results are shown in table 2. In order to avoid redundancy, table 2 only shows the total solar radiation estimation results of the empirical model EM3 with the highest accuracy, and the rest of the empirical model results are not shown.
TABLE 2 estimation of total solar radiation on day for empirical-machine learning fusion model
From the results in the table, it can be seen that the deep learning model has a greater estimation accuracy improvement than the conventional empirical model, regardless of the site. Meanwhile, the experience-machine learning fusion model EM-DBN has certain precision improvement compared with the experience model EM3 and the deep learning model DBN, but the precision improvement compared with the experience model is obviously larger than that of the deep learning model.
And (4) combining with a transfer learning algorithm, and transferring the constructed EM-DBN fusion model to a non-measuring target site to obtain a radiation estimated value. On the basis of the original deep neural network, an adaptation layer is added in front of the last regression layer of the network, wherein the deep neural network structure combined with migration learning is shown in fig. 4. The maximum mean difference (maximum mean discrepancy, MMD) introduced into the transfer learning measures the distance of two different but related feature distributions, the distance of the two distributions being defined as:
wherein X and T represent two different distributions of source and target domains, n and m the number of samples, X and T, X i and tj Representing the ith and jth samples in X and T, respectively, phi (·) represents the mapping function, and H represents the distance measured by the mapping of data into the regenerated Hilbert space (RKHS) by phi (·). Since φ (-) is an infinite dimension and there is no way to directly solve, the Gaussian kernel function k (-) can map an infinite dimension space, so in most cases, the Gaussian kernel function k (-) is generally chosen:
wherein u Is an arbitrary point in the space of the device, v sigma represents bandwidth, being the kernel function center. Then the loss function is divided into two parts, using its peripheral siteThe difference degree between the estimation result and the actual data is expressed by the label data estimation error and the distance between two domains calculated by MMD, and the method is concretely as follows:
wherein θ represents the weight and deviation of the network, n represents the number of samples of the peripheral site, x i Input features representing the ith sample of the set of peripheral site samples, y i Output radiation value representing the ith sample of the set of peripheral site samples, J (·) representing the loss function, λ representing the penalty factor, d 2 (D s ，D t ) Representing MMD distance of the source domain and the target domain. And utilizing the feature transformation capability of deep learning to transform the feature space until the feature distribution transformed by the peripheral site data set is matched with the feature distribution transformed by the non-measuring site data set, and successfully transferring the EM-DBN fusion model constructed by the peripheral site data to the non-measuring region, thereby obtaining the estimated solar radiation value of the non-measuring site.
When calculating the characteristic distribution distance of the peripheral site and the non-measured site, the selected target domain (namely site 7) data and the source domain (peripheral site) data are all weather parameter data of 22 years from 1 month in 1994 to 12 months in 2015. Table 3 shows the total solar radiation estimation results for the day 7 of 2015 at the no-survey site. Fig. 5 shows a graph of the fit of the measured site-free solar radiation estimate to the true value, with a time span of from 1 month, 1 day, 2015 to 12 months, 31 days, 2015.
TABLE 3 estimation of total solar radiation at 2015, 7 month
As can be seen from Table 3, in most cases, the daily solar radiation estimation accuracy of the EM-DBN model adopting transfer learning on the non-radiation measurement site is higher, the error between the estimated value and the true value is smaller, and the effectiveness of combining the transfer learning is verified. The estimated value curve of the experience-machine learning model combined with the migration learning in fig. 5 is closer to the true value curve, which shows that the migration method has a certain effect on improving the solar radiation estimation accuracy of the no-quantity region.
In conclusion, the estimation method can be used for estimating solar radiation without a measuring area, and has theoretical value and practical significance for reasonably utilizing solar resources, improving photovoltaic power generation capacity and promoting sustainable healthy development of economy. Compared with other single solar radiation estimation models, the method provided by the invention combines the advantages of an experience model and a machine learning model to construct the model, the accuracy is obviously improved, and the fusion model is further transferred to a non-measurement site by utilizing a transfer learning mode, so that the method is favorable for evaluating radiation resources of the non-measurement region and provides valuable references for power generation of a photovoltaic power station.
Claims (2)
1. The method for estimating solar radiation in a non-measuring area based on EM-DBN fusion is characterized by comprising the following steps:
(1) Based on the existing data set, selecting a solar total radiation estimation experience model;
in the step (1), an empirical model with relatively high precision is selected as an empirical model for estimating total solar radiation, wherein the empirical model is selected as follows:
empirical model 1:
empirical model 2:
empirical model 3:
empirical model 4:
empirical model 5:
empirical formula 6:
wherein H represents total daily radiation; h 0 Represents extraterrestrial radiation; s represents the actual sunshine duration; s is S 0 Representing the maximum possible sunshine duration; RH represents the average relative humidity; t (T) max and Tmin Respectively representing the highest daily temperature and the lowest daily temperature; t represents a daily average temperature; a. b, c, d, e and f are training parameters; wherein S is 0 Depending on the local latitude
The declination angle delta of the sun on the same day;
(2) A training set and a testing set are constructed by utilizing meteorological data and irradiance data of peripheral sites and target sites, and a function expression of an empirical model is fitted based on training set data, and the method specifically comprises the following steps:
(2.1) constructing a training set and a test set of weather data and irradiance data of peripheral sites and destination sites, wherein the training set comprises training samples of m days, and the test set comprises test samples of n days on the assumption that one sample is one day when modeling is performed on a target site; the input on the training set is X (X i ) I=1, 2, …, m, output is Y (Y i ) I=1, 2, …, m, the input on the test set is P (P j ) J=1, 2, …, n, output is Q (Q j )，j＝1，2 ，…，n； wherein xi and pj For a row vector of 1×t, t represents the gas required to build an empirical modelThe number of image parameters, from the day maximum temperature T max Minimum temperature T min Average day temperature T, number of sunshine hours S, number of days n in one year day And relative humidity RH; y is i and qj The total solar radiation value of the current day;
(2.2) based on sample data on the training set, establishing an empirical model, fitting the data on the training set, and solving by adopting least square;
(3) The method comprises the steps of fusing an experience model and a DBN model in a characteristic dimension expansion mode, and constructing an EM-DBN fusion model by utilizing peripheral site data, and specifically comprises the following steps:
(3.1) based on the input X (X) on the training set according to the empirical model function expression established in step (2) i ) Calculating a solar radiation estimated value of the empirical model on the training set; estimating the solar radiation estimated value of the empirical model and the highest temperature T of the day max Minimum temperature T min Average day temperature T, number of sunshine hours S, number of days n in one year day Together with average humidity RH as input to the DBN model, the output is still Y (Y i )，i＝1，2，…，m；
(3.2) taking a DBN model as a basis of a fusion model, wherein the DBN model is formed by superposition and combination of a plurality of RBMs from bottom to top; after pre-training by RBM sampling, the output of the RBM of the upper layer is used as the input of the lower layer; training the DBN model layer by layer repeatedly; assume that
Is a set of training samples, and the specific calculation process in the RBM model is as follows:
in the formula ,lnLS Representing the partial derivatives of the respective parameters; n is n s Representing the number of training samples; η (eta)>0 represents a learning rate; θ= (W, a, b) represents parameters in RBM;
is a gradient of RBM; training +.A k-step contrast divergence algorithm is used>
For->
Take the initial value v (0) =v, then perform k steps Gibbs sampling; wherein the t-th step (t=1, 2, …, k) is first performed by using P (h|v (t-1) ) Sampling out h (t-1) And using P (v|h (t-1) ) Sampling v (t) Next, v obtained by sampling with the Gibbs in the k steps (k) To approximate the estimation formula:
wherein ,
and />
For the partial derivative in the gradient-rising iteration, its component is +.>
and />
(3.3) fusing the experience model and the machine learning model in the characteristic dimension expansion mode in the step (3.1), selecting a peripheral station address which is in the same weather zone with the destination station address and is closer to the destination station address, and constructing an EM-DBN fusion model by utilizing data of the peripheral station address;
(4) Transferring the constructed EM-DBN fusion model to a non-measuring target station site to obtain a solar radiation estimated value by combining a transfer learning algorithm; the method specifically comprises the following steps:
on the basis of the original deep neural network, an adaptation layer is added in front of the last regression layer of the deep neural network; the maximum mean difference introduced into the transfer learning measures the distance of two different but related feature distributions, defining the distance of the two distributions as:
wherein X and T represent two different distributions of source and target domains, n and m represent the number of samples of X and T, respectively, X i and tj Representing the ith and jth samples in X and T respectively,
represents a mapping function, H represents that this distance is defined by +.>
The data is mapped into the regenerated hilbert space for measurement.
2. The method for estimating solar radiation without measuring area based on EM-DBN fusion as claimed in claim 1, wherein Gaussian kernel function is selected
An infinite dimensional space is mapped, and the expression is:
wherein u represents any point in space, v is the kernel function center, and sigma represents bandwidth;
then, the loss function is divided into two parts, and the difference degree between the estimation result and the actual data is expressed by using the two domain distances calculated by the MMD and the tag data estimation error of the peripheral station address, which is specifically as follows:
wherein θ represents the weight and deviation of the network, n represents the number of samples of the peripheral site, x i Input features representing the ith sample of the set of peripheral site samples, y i Output radiation values representing the ith sample of the set of peripheral site samples,
represents a loss function, lambda represents a penalty coefficient, d 2 (D s ，D t ) MMD distance representing source and target domain; and utilizing the feature transformation capability of deep learning to transform the feature space until the feature distribution transformed by the peripheral site data set is matched with the feature distribution transformed by the non-measuring site data set, and successfully transferring the EM-DBN fusion model constructed by the peripheral site data to the non-measuring region, thereby obtaining the solar radiation estimated value of the non-measuring site. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210520496.XA CN115017799B (en) | 2022-05-13 | 2022-05-13 | EM-DBN fusion-based solar radiation estimation method for quantitative region |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210520496.XA CN115017799B (en) | 2022-05-13 | 2022-05-13 | EM-DBN fusion-based solar radiation estimation method for quantitative region |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115017799A CN115017799A (en) | 2022-09-06 |
| CN115017799B true CN115017799B (en) | 2023-05-23 |
Family
ID=83068600
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210520496.XA Active CN115017799B (en) | 2022-05-13 | 2022-05-13 | EM-DBN fusion-based solar radiation estimation method for quantitative region |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115017799B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116452055B (en) * | 2023-04-12 | 2023-10-17 | 深圳大学 | Urban photovoltaic potential calculation method integrating surface grid model and deep learning |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5571970B2 (en) * | 2010-02-10 | 2014-08-13 | 大阪瓦斯株式会社 | Solar radiation amount estimation device, photovoltaic power generation amount estimation device and system |
| CN104598743A (en) * | 2015-01-27 | 2015-05-06 | 国家电网公司 | Method for generating solar radiation data of immeasurable regions |
| CN110222411B (en) * | 2019-05-31 | 2020-04-03 | 河海大学 | Non-measurement-area solar radiation estimation method based on mRMR-DBN algorithm |
-
2022
- 2022-05-13 CN CN202210520496.XA patent/CN115017799B/en active Active
Non-Patent Citations (2)
| Title |
|---|
| 基于支路带节点功率估计的配网状态估计实用算法;杨成龙,高晓萍;电网技术(10);全文 * |
| 空间环境探测卫星用磁强计误差分析及在线标定;杨照华;余远金;祁振强;;宇航学报(08);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115017799A (en) | 2022-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20220214322A1 (en) | Air pollutants concentration forecasting method and apparatus and storage medium | |
| CN110070226B (en) | Photovoltaic power prediction method and system based on convolutional neural network and meta-learning | |
| CN111008728B (en) | Prediction method for short-term output of distributed photovoltaic power generation system | |
| Mohandes | Modeling global solar radiation using Particle Swarm Optimization (PSO) | |
| CN106022521B (en) | Short-term load prediction method of distributed BP neural network based on Hadoop architecture | |
| Chaabene et al. | Neuro-fuzzy dynamic model with Kalman filter to forecast irradiance and temperature for solar energy systems | |
| CN110807554B (en) | Generation method and system based on wind power/photovoltaic classical scene set | |
| Dong et al. | A general radial basis function neural network assisted hybrid modeling method for photovoltaic cell operating temperature prediction | |
| Abdel-Nasser et al. | A novel smart grid state estimation method based on neural networks | |
| CN107562992B (en) | Photovoltaic array maximum power tracking method based on SVM and particle swarm algorithm | |
| CN110222411B (en) | Non-measurement-area solar radiation estimation method based on mRMR-DBN algorithm | |
| CN114004163B (en) | PM2.5 inversion method based on MODIS and long-short-term memory network model | |
| CN109948920B (en) | Electric power market settlement data risk processing method based on evidence theory | |
| CN112418346B (en) | Numerical weather forecast total radiation system error classification calculation method | |
| CN115017799B (en) | EM-DBN fusion-based solar radiation estimation method for quantitative region | |
| Chauhan et al. | Approximating parameters of photovoltaic models using an amended reptile search algorithm | |
| CN113592144A (en) | Medium-and-long-term runoff probability forecasting method and system | |
| Bilgiç et al. | Comparison of different techniques for estimation of incoming longwave radiation | |
| CN112101626A (en) | Distributed photovoltaic power generation power prediction method and system | |
| CN114004152B (en) | Multi-wind-field wind speed space-time prediction method based on graph convolution and recurrent neural network | |
| CN111815051B (en) | GRNN photovoltaic power generation prediction method considering weather influence factors | |
| Zhang et al. | All‐factor short‐term photovoltaic output power forecast | |
| Mellit et al. | Neural network adaptive wavelets for sizing of stand-alone photovoltaic systems | |
| CN111461297B (en) | Solar irradiation quantity optimization prediction algorithm based on MPC and ELM neural network | |
| CN111859303B (en) | Soil humidity fusion method and system based on dynamic Bayesian average |
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 |