CN106022954B - Multiple BP neural network load prediction method based on grey correlation degree - Google Patents
Multiple BP neural network load prediction method based on grey correlation degree Download PDFInfo
- Publication number
- CN106022954B CN106022954B CN201610323293.6A CN201610323293A CN106022954B CN 106022954 B CN106022954 B CN 106022954B CN 201610323293 A CN201610323293 A CN 201610323293A CN 106022954 B CN106022954 B CN 106022954B
- Authority
- CN
- China
- Prior art keywords
- load
- neural network
- layer
- output
- matrix
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Abstract
The invention discloses a multiple BP neural network load prediction method based on grey correlation degree, which comprises the following steps: firstly, load sequence correlation analysis based on grey correlation degree; secondly, determining a member set of the multiple BP neural networks based on the shortest distance method clustering; thirdly, determining the multiplicity of the multiple BP neural networks based on the effectiveness index; a momentum factor is introduced, and a mode of calculating and averaging for multiple times is adopted, so that the problem that the BP neural network is easy to fall into local convergence is solved, and the oscillation resistance of the BP neural network is improved; and fifthly, predicting the short-term power load by the established multi-BP neural network prediction model. The method of the invention improves the problem that the BP neural network is easy to fall into local convergence, improves the oscillation resistance of the BP neural network, and has better prediction effect compared with the traditional BP neural network prediction model.
Description
Technical Field
The invention relates to the technical field of short-term load prediction application of a power system, in particular to a multiple BP neural network load prediction method based on grey correlation degree.
Background
The power load prediction has very important significance in the aspects of ensuring the planning and the reliable and economic operation of a power system. With the continuous progress of modern technology and the deepening of smart power grids, load prediction theory and technology have been greatly developed. Over the years, power load prediction methods and theories are continuously emerging, and technologies such as a time series method, a fuzzy theory, a regression analysis method, a regression support vector machine, Bayes, a neural network and the like provide good technical support for power load prediction. However, the existing algorithm still has certain limitation. Time series method: the method has higher accuracy on historical data, is insensitive to weather factors during short-term load prediction, and is difficult to solve the problem of low short-term load prediction precision caused by weather factors. Regression analysis method: the quantitative relationship between the observed variables is quantitatively described from the point of view of statistical average significance, but is greatly limited by the size of the load data volume. Regression support vector machine: the method has good generalization capability, but the training time is excessively long due to the optimization of the penalty coefficient c, the loss function e and the gamma value parameter of the kernel function, and particularly, the method is more prominent when the training sample set is large in scale.
Considering that the BP neural network has strong nonlinear mapping capability, self-learning capability and fault-tolerant capability, when the BP neural network is applied to a load small data set, the BP neural network has the advantages of high prediction precision, high training speed and the like. However, when the conventional BP neural network is applied to load prediction, a key problem still exists, and as the number of load samples increases, the accuracy of network prediction may decrease, i.e. the so-called "overfitting" problem, and at the same time, the convergence rate of the neural network during training may become slow. The reason for this is that the load of the historical load data every day has peak-valley difference and large fluctuation, so that all the load data directly share one BP neural network. Obviously, in the training process, in order to emphasize the overall training error, the network will have an "overfitting" problem, which will result in a weak generalization capability in the later actual load prediction, and as the training samples increase, the load prediction speed will also decrease significantly.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a multiple BP neural network load prediction method based on gray correlation degree, aiming at the problem that the generalization capability is weak due to the fact that the traditional BP neural network is applied to load prediction and the overfitting exists, and defining an effectiveness index for representing cluster advantages and disadvantages based on the gray correlation degree and the shortest distance method so as to determine the reasonable weight of a prediction model.
In order to solve the technical problems, the invention adopts the technical scheme that:
a multiple BP neural network load prediction method based on grey correlation degree comprises the following steps: the method comprises the following steps: analyzing the relevance of the load sequence by adopting a grey relevance method; step two: determining a member set of the multiple BP neural networks by clustering according to a shortest distance method; step three: determining the multiplicity of the multiple BP neural networks according to the effectiveness index; step four: and (4) performing load prediction according to the analysis result in the first step and the BP neural network after determining the multiplicity in the second step and the third step.
Further, the method also comprises the following step five: and (3) improving the BP neural network, namely introducing momentum factors, and improving the BP neural network by adopting a mode of calculating and averaging for multiple times.
Furthermore, the BP neural network consists of two parts of forward transmission of signals and backward propagation of errors, the actual output is calculated according to the direction from input to output, and the correction process of weight and threshold values of each layer is carried out from the direction from output to input.
Further, the back propagation of the error comprises: and calculating the output error of each layer of neuron layer by layer from the output layer, and adjusting the weight and the threshold value of each layer according to an error gradient descent method to enable the final output of the adjusted network mapping to be close to the expected value.
Further, the first step specifically includes:
constructing a sequence matrix, and establishing an initial load sequence matrix L [ < L > ] based on a historical load data longitudinal 24-point load sequence1,L2,…Lm]Wherein the value of m is 24, N is the longitudinal dimension of the historical load data, corresponding to the number of days of load recording,
dimensionless, data processing is carried out by using an initial value method to obtain a dimensionless matrix which is recorded as L '═ L'1,L′2,…,L′m]，L′i(k)＝Li(k)/Li(1)i＝1,2,…,N；k＝1,2,…,m；
The calculation of the correlation coefficient is carried out,
wherein p and q are the serial numbers of the longitudinal 24-point load sequence,
for the resolution factor, k is the longitudinal length index, ξpq(k) The correlation coefficient of the load of the p column and the load of the q column at the k row;
calculation of degree of correlation, γpqThe correlation degree of the p-th column load and the q-th column load is calculated through the correlation coefficient matrix of the load sequence of the correlation degree,
further, the second step specifically includes:
quantitatively calculating distance vectors representing the similarity between the load incidence matrix sequences by adopting Euclidean distance;
acquiring a matrix containing clustering tree information by adopting a shortest distance method;
combined type
And determining a pedigree graph of the multiple BP neural network weights by adopting a shortest distance method according to the determined correlation coefficient matrix of the load sequence.
Further, the effectiveness index in the third step is
Wherein, lp (i), lq (i) respectively represent the loads at the time of the ith day p and the moment q in the same class, Lr (i) and Lt (i) respectively represent the loads at the time of the ith day r and the moment t in different classes, N is the longitudinal dimension of historical load data and corresponds to the number of days of load record.
Compared with the prior art, the invention has the beneficial effects that: a method for selecting the multiplicity of the multiple BP neural network based on gray correlation degree and shortest distance method clustering is provided to combine part of closely related load sequences and properly reduce the multiplicity of the multiple BP neural network. Meanwhile, a momentum factor is introduced, and a mode of calculating and averaging for multiple times is adopted, so that the problem that the BP neural network is easy to fall into local convergence is solved, and the oscillation resistance of the BP neural network is improved. Compared with a traditional BP neural network prediction model, the multi-BP neural network has a better prediction effect.
Drawings
Fig. 1 is a schematic diagram of a typical three-layer BP neural network structure according to the present invention.
FIG. 2 is a family chart of membership sets of the multiple BP neural network of the present invention.
FIG. 3 shows the short-term prediction effect of the multiple BP neural network model load in the present invention.
FIG. 4 is a graph of the load prediction results of the 6-fold BP neural network of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
1. Traditional BP neural network prediction principle analysis
1) Basic model of BP neural network
In 1986, scientists including Rumelhart and mccell proposed a BP neural network, which is a multi-layer feedforward neural network capable of learning and storing a large number of input-output pattern mappings without revealing a mathematical equation of such mappings beforehand, and is composed of an input layer, a hidden layer and an output layer. Fig. 1 is a structural diagram of a typical three-layer BP neural network, where layers are fully interconnected, there is no interconnection between the same layers, and a hidden layer may be one or more layers. In FIG. 1, xjRepresenting the input of the jth node of the input layer; w is aijRepresenting the weight from the ith node of the hidden layer to the jth node of the input layer; thetaiA threshold value of the ith node of the hidden layer; phi is the excitation function of the hidden layer; w is akiRepresenting the weight from the kth node of the output layer to the ith node of the hidden layer; alpha is alphakA threshold value of the kth node of the output layer; psi is the excitation function of the output layer; okRepresenting the output of the kth node.
2) BP neural network signaling and error correction
The basic BP neural network algorithm consists of two parts of forward transmission of signals and backward propagation of errors, namely, the actual output is calculated according to the direction from input to output, and the correction process of weight values and threshold values of each layer is carried out from the direction from output to input. And calculating and adjusting the output signal of the BP neural network, the weight values of each layer and the threshold according to the parameters shown in the figure 1.
(1) Forward propagation of input signals
According to the structure diagram of the BP neural network in FIG. 1, the input net of the ith node of the concerned hidden layer can be knowniAnd an output oiInput amount net of kth node of output layerkAnd an output okAre respectively as
(2) Counter-propagating process of error signal
And (3) the back propagation of the error, namely, calculating the output error of each layer of neuron layer by layer from the output layer, and then adjusting the weight and the threshold of each layer according to an error gradient descent method to enable the final output of the adjusted network mapping to be close to the expected value. According to the error gradient descent method, the weight correction quantity delta w from the hidden layer to the output layer can be corrected in sequencekiOutput layer threshold correction amount Δ αkAnd the weight correction quantity delta w from the input layer to the hidden layerijAnd hidden layer threshold correction amount delta thetaiAs shown in the formulas (5) to (8), η is the learning rate, and P is the total number of training samples.
2. Load sequence relevance analysis based on grey relevance
The correlation analysis is a method for analyzing the correlation degree of each factor in a system, which is proposed by a grey system theory, the basic idea is to judge the correlation degree according to the similarity degree between curves, and the calculation steps are as follows:
1) and constructing a sequence matrix. Based on the historical load data longitudinal 24-point load sequence, establishing an initial load sequence matrix L ═ L1,L2,…Lm]Wherein the value of m is 24, and N is the longitudinal dimension of the historical load data, corresponding to the number of days of load recording.
2) And (4) dimensionless. In order to eliminate the influence of dimension, data processing is carried out by using an initial value method. A dimensionless matrix can be obtained from equation (10) and is denoted as L '═ L'1,L′2,…,L′m]。
L′i(k)＝Li(k)/Li(1)i＝1,2,…,N；k＝1,2,…,m (10)
3) And calculating a correlation coefficient.
In the formula: p and q are both the serial numbers of the longitudinal 24-point load sequence;
the distinguishing coefficient has the effect of improving the significance of the difference between the correlation coefficients, and the value is generally 0.5; k is a longitudinal length index; xipq(k) The correlation coefficient of the load of the p column and the load of the q column at the k-th row is shown.
4) And calculating the relevance. In the formula (12) < gamma >, (Y)pqThe correlation degree of the p-th column load and the q-th column load is shown. And calculating a correlation coefficient matrix of the load sequence through the correlation degree.
3. Method for determining member set of multiple BP neural networks based on shortest distance method clustering
Clustering analysis algorithms are numerous, such as K-Means, grid and density based clustering algorithms, F clustering algorithms, and shortest distance methods. The shortest distance method is adopted for clustering, and compared with other algorithms, the method is simple and easy to operate. Therefore, the shortest distance method is selected for load sequence clustering, and an appropriate clustering distance amount is selected to determine the final multiple number of the multiple BP neural network. When clustering is carried out, firstly, Euclidean distance is adopted to quantitatively calculate distance vectors representing the similarity between load association matrix sequences, and then, a matrix containing clustering tree information is obtained by adopting a shortest distance method. And (3) determining a pedigree map of the multiple BP neural network weight by adopting a shortest distance method in combination with the correlation coefficient matrix of the load sequence determined by the formula (12).
4. Determining multiplicity of multiple BP neural networks based on effectiveness indicators
In order to measure the quality of the clustering result, an effectiveness index for representing the quality of the clustering effect is defined from the clustering essence. Considering that a better clustering result should have the characteristics that the smaller the inter-class distance is, the better the effectiveness index is defined as shown in formula (13).
In the formula: l isp(i)、Lq(i) Respectively representing the loads at p and q days in the same class; l isr(i)、Lt(i) Respectively representing the loads at r and t days of different classes; n is as defined in formula (9).
The process and advantageous effects of the present invention are described in more detail by the following specific examples.
1. Example system and data processing
The data of the invention is derived from load data and weather data collected by a certain actual power grid, the sampling time interval period of each device is 1h, and the weather information is dry bulb temperature and dew point temperature. Although the data volume of the laboratory does not reach the scale of the big data, the experimental data can be used for carrying out algorithm correctness experiments, and a new method is provided for load prediction under the big data environment. The training range is from 1 month 1 day 2014 to 3 months 31 days 2014. The predicted day is the power load at different times of 4/1/2014, as shown in table 1. According to the characteristics and research results of a large number of documents, the sample attributes are determined to be load at the same moment in two weeks before the day, load at the same moment in one week before the day, load at the same moment in two days before the day, load at the same moment in one day before the day, dry bulb temperature at the same moment in the day before the day, dry bulb temperature at the same moment on the predicted day and dew point temperature at the same moment on the predicted day, the actual load at the same moment in the day is predicted, and the sample data are shown in Table 2.
Actual load data in table 12014 years, 4 months, 1 day
| Time of day/h | load/MW | Time of day/h | load/MW | Time of day/h | load/ |
| 1 | 11483 | 9 | 15870 | 17 | 14508 |
| 2 | 10924 | 10 | 15965 | 18 | 14332 |
| 3 | 10711 | 11 | 15978 | 19 | 14219 |
| 4 | 10728 | 12 | 15823 | 20 | 14702 |
| 5 | 11027 | 13 | 15556 | 21 | 15265 |
| 6 | 12128 | 14 | 15388 | 22 | 14557 |
| 7 | 14043 | 15 | 15060 | 23 | 13416 |
| 8 | 15413 | 16 | 14761 | 24 | 12135 |
TABLE 2 load training data sample set
| Properties | Value of |
| Load on day and two weeks simultaneously | 11600MW |
| Load at the same time of the week before the day | 11857MW |
| Load on day two days ahead at the same time | 11462MW |
| Load at the same time of day before day | 11203MW |
| Dry bulb temperature at the same time of day before | 46℃ |
| Dew point temperature at the same time of day before day | 43℃ |
| Predicting the dry bulb temperature at the same time of the day | 41℃ |
| Predicting the dew point temperature at the same time of the day | 18℃ |
| Predicting the actual load at the same moment of the day | 11483MW |
It is worth noting that the BP neural network is sensitive to the ratio of the numerical value between 0 and 1, so that before the original load sequence is input into the distributed BP neural network model, normalization processing needs to be performed on data, and after training is finished, inverse normalization processing needs to be performed to obtain an actual load predicted value.
2. Results and analysis of the experiments
The purpose of the experiment is to prove that compared with a multi-BP neural network model, the multi-BP neural network model directly shares one BP neural network model with all historical load data, and has better prediction accuracy. The combination formula (13) shows that it is relatively reasonable to establish a load prediction model of the 6-fold BP neural network, and the corresponding effectiveness index value is the minimum, which is 0.1422. The selection criterion of the heavy member can refer to fig. 2, namely when the selection criterion is set to 6, the selection criterion means that 6 BP load prediction models are created, load related sequences with time numbers of 2-8, 9, 1 and 20-24, 10 and 11 are respectively used as the input and output quantities of the BP load prediction models 1-5, and the load related sequences with the rest time are used as the input and output quantities of the BP load prediction model 6.
FIG. 3 is a graph of the predicted effect of averaging over multiple runs. The method can know that the influence of the weight of the BP neural network on the load prediction precision is large, a load short-term prediction model of the 6-weight BP neural network is established, and compared with the method that all historical load related sequences share the same BP neural network prediction model, the method has better prediction precision. The load prediction effect based on the 6-fold BP neural network is shown in FIG. 4, the average absolute relative error is 3.01%, the root mean square error is 1.63%, and the requirement of actual load prediction precision is met.
Claims (3)
1. A multiple BP neural network load prediction method based on grey correlation degree is characterized by comprising the following steps:
the method comprises the following steps: analyzing the relevance of the load sequence by adopting a grey relevance method;
the first step specifically comprises:
constructing a sequence matrix, and establishing an initial load sequence matrix L [ < L > ] based on a historical load data longitudinal 24-point load sequence1,L2,…Lm]Wherein the value of m is 24, N is the longitudinal dimension of the historical load data, corresponding to the number of days of load recording,
dimensionless method using initial valueProcessing the data to obtain a dimensionless matrix which is recorded as L ═ L'1,L′2,…,L′m]，L′i(k)＝Li(k)/Li(1)i＝1,2,…,N；k＝1,2,…,m；
The calculation of the correlation coefficient is carried out,
in the formula, p and q are the serial numbers of the longitudinal 24-point load sequence, theta is a resolution coefficient, k is a longitudinal length index, and xipq(k) The correlation coefficient of the load of the p column and the load of the q column at the k row;
calculation of degree of correlation, γpqThe correlation degree of the p-th column load and the q-th column load is calculated through the correlation coefficient matrix of the load sequence of the correlation degree,
step two: determining a member set of the multiple BP neural networks by clustering according to a shortest distance method;
the second step specifically comprises:
quantitatively calculating distance vectors representing the similarity between the load incidence matrix sequences by adopting Euclidean distance;
acquiring a matrix containing clustering tree information by adopting a shortest distance method;
combined type
Determining a pedigree graph of the multiple BP neural network weight number by adopting a shortest distance method according to the determined correlation coefficient matrix of the load sequence;
the member set of the multiple BP neural networks is from a pedigree graph of multiple BP neural networks;
step three: determining the multiplicity of the multiple BP neural networks according to the effectiveness index;
the effectiveness index in the third step is
Wherein, lp (i), lq (i) respectively represent the loads at the time of the ith day p and the moment q in the same class, Lr (i) and Lt (i) respectively represent the loads at the time of the ith day r and the moment t in different classes, N is the longitudinal dimension of historical load data and corresponds to the number of days of load recording;
determining the clustering result with the minimum effectiveness index J value in the pedigree graph of the multiple BP neural networks as the optimal BP neural network weight;
step four: load prediction is carried out according to the analysis result of the step one and the BP neural network after the duplication is determined in the step two and the step three;
step five: and (3) improving the BP neural network, namely introducing momentum factors, and improving the BP neural network by adopting a mode of calculating and averaging for multiple times.
2. The method as claimed in claim 1, wherein the BP neural network consists of forward transmission of signals and backward propagation of errors, the calculation of actual output is performed in the direction from input to output, and the correction of weights and thresholds of each layer is performed in the direction from output to input.
3. The multiple BP neural network load prediction method based on gray correlation according to claim 2, wherein the back propagation of the error comprises: and calculating the output error of each layer of neuron layer by layer from the output layer, and adjusting the weight and the threshold value of each layer according to an error gradient descent method to enable the final output of the adjusted network mapping to be close to the expected value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610323293.6A CN106022954B (en) | 2016-05-16 | 2016-05-16 | Multiple BP neural network load prediction method based on grey correlation degree |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610323293.6A CN106022954B (en) | 2016-05-16 | 2016-05-16 | Multiple BP neural network load prediction method based on grey correlation degree |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106022954A CN106022954A (en) | 2016-10-12 |
| CN106022954B true CN106022954B (en) | 2022-04-15 |
Family
ID=57097390
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610323293.6A Active CN106022954B (en) | 2016-05-16 | 2016-05-16 | Multiple BP neural network load prediction method based on grey correlation degree |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106022954B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106529724A (en) * | 2016-11-14 | 2017-03-22 | 吉林大学 | Wind power prediction method based on grey-combined weight |
| CN108427867A (en) * | 2018-01-22 | 2018-08-21 | 中国科学院合肥物质科学研究院 | One kind being based on Grey BP Neural Network interactions between protein Relationship Prediction method |
| CN108629401A (en) * | 2018-04-28 | 2018-10-09 | 河海大学 | Character level language model prediction method based on local sensing recurrent neural network |
| CN108469783B (en) * | 2018-05-14 | 2021-02-02 | 西北工业大学 | Deep hole roundness error prediction method based on Bayesian network |
| CN111091217B (en) * | 2018-10-23 | 2022-07-08 | 中国电力科学研究院有限公司 | Building short-term load prediction method and system |
| CN109255505B (en) * | 2018-11-20 | 2021-09-24 | 国网辽宁省电力有限公司经济技术研究院 | Short-term load prediction method of multi-model fusion neural network |
| CN109508835B (en) * | 2019-01-01 | 2020-11-24 | 中南大学 | Smart power grid short-term power load prediction method integrating environmental feedback |
| CN110009140B (en) * | 2019-03-20 | 2021-10-08 | 华中科技大学 | Daily power load prediction method and prediction device |
| CN112990597B (en) * | 2021-03-31 | 2024-02-27 | 国家电网有限公司 | Ultra-short-term prediction method for industrial park power consumption load |
| CN113222216A (en) * | 2021-04-14 | 2021-08-06 | 国网江苏省电力有限公司营销服务中心 | Method, device and system for predicting cooling, heating and power loads |
| CN114881382B (en) * | 2022-07-13 | 2022-10-11 | 天津能源物联网科技股份有限公司 | Optimized adjusting method for autonomous prediction of heat supply demand load |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101383023A (en) * | 2008-10-22 | 2009-03-11 | 西安交通大学 | Neural network short-term electric load prediction based on sample dynamic organization and temperature compensation |
| CN103390199A (en) * | 2013-07-18 | 2013-11-13 | 国家电网公司 | Photovoltaic power generation capacity/power prediction device |
-
2016
- 2016-05-16 CN CN201610323293.6A patent/CN106022954B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101383023A (en) * | 2008-10-22 | 2009-03-11 | 西安交通大学 | Neural network short-term electric load prediction based on sample dynamic organization and temperature compensation |
| CN103390199A (en) * | 2013-07-18 | 2013-11-13 | 国家电网公司 | Photovoltaic power generation capacity/power prediction device |
Non-Patent Citations (3)
| Title |
|---|
| "Prediction technique for transformer oil breakdown voltage via multi-parameter correlation based on grey theory and BP neural network";Zhi Li 等;《2010 International Conference on Information, Networking and Automation (ICINA)》;20101019;全文 * |
| "基于模糊聚类的灰色关联分析结合的神经网络符合预测";康丽峰等;《第八届中国青年运筹信息管理学者大会论文集》;20060831;第91-95页 * |
| "基于灰色关联理论和BP神经网络的智慧城市发展潜力评价";邹凯等;《科技进步与对策》;20150910;124页右边栏-127页左边栏 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106022954A (en) | 2016-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106022954B (en) | Multiple BP neural network load prediction method based on grey correlation degree | |
| CN109492822B (en) | Air pollutant concentration time-space domain correlation prediction method | |
| CN109816095B (en) | Network flow prediction method based on improved gated cyclic neural network | |
| CN106022521B (en) | Short-term load prediction method of distributed BP neural network based on Hadoop architecture | |
| CN108596327B (en) | Seismic velocity spectrum artificial intelligence picking method based on deep learning | |
| CN108710974B (en) | Water ammonia nitrogen prediction method and device based on deep belief network | |
| CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
| CN105488563A (en) | Deep learning oriented sparse self-adaptive neural network, algorithm and implementation device | |
| CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
| CN110020712B (en) | Optimized particle swarm BP network prediction method and system based on clustering | |
| CN109308544B (en) | Blue algae bloom prediction method based on contrast divergence-long and short term memory network | |
| CN109523021A (en) | A kind of dynamic network Structure Prediction Methods based on long memory network in short-term | |
| CN110348608A (en) | A kind of prediction technique for improving LSTM based on fuzzy clustering algorithm | |
| CN106067034B (en) | Power distribution network load curve clustering method based on high-dimensional matrix characteristic root | |
| CN111047078B (en) | Traffic characteristic prediction method, system and storage medium | |
| CN113868938A (en) | Short-term load probability density prediction method, device and system based on quantile regression | |
| CN110289987B (en) | Multi-agent system network anti-attack capability assessment method based on characterization learning | |
| CN109408896B (en) | Multi-element intelligent real-time monitoring method for anaerobic sewage treatment gas production | |
| CN112819087B (en) | Method for detecting abnormality of BOD sensor of outlet water based on modularized neural network | |
| CN114596726A (en) | Parking position prediction method based on interpretable space-time attention mechanism | |
| CN111863153A (en) | Method for predicting total amount of suspended solids in wastewater based on data mining | |
| CN112232570A (en) | Forward active total electric quantity prediction method and device and readable storage medium | |
| CN113191689B (en) | Land suitability evaluation method for coupling principal component analysis and BP neural network | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device | |
| CN114971090A (en) | Electric heating load prediction method, system, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |