CN110135623A - Load prediction input feature vector screening technique based on hierarchical clustering and neural network - Google Patents
Load prediction input feature vector screening technique based on hierarchical clustering and neural network Download PDFInfo
- Publication number
- CN110135623A CN110135623A CN201910294713.6A CN201910294713A CN110135623A CN 110135623 A CN110135623 A CN 110135623A CN 201910294713 A CN201910294713 A CN 201910294713A CN 110135623 A CN110135623 A CN 110135623A
- Authority
- CN
- China
- Prior art keywords
- cluster
- clustering
- load prediction
- neural network
- distance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a kind of load prediction input feature vector screening technique based on hierarchical clustering and neural network, this method is clustered the raw data set including load prediction using Hierarchical clustering analysis, data set is divided into different classes according to its feature, feature in every one kind is analyzed, it is analyzed using prediction effect of the neural network to the different characteristic in every one kind, it is final to reject the poor feature of prediction effect, obtain the input feature vector collection of the building load prediction of final optimization pass.The present invention can screen lengthy and jumbled load prediction input data set, reduce the complexity of load prediction, improve the calculating speed of load prediction, while retain the validity feature of data concentration, guarantee the accuracy of load prediction.
Description
Technical field
The invention belongs to load prediction domains, and in particular to a kind of load prediction input based on hierarchical clustering and neural network
Feature Selection method and parameter prediction effect judgment method.
Background technique
Load prediction generally includes the cooling and heating load prediction in load forecast and building, is primarily referred to as according to system
Feature and relevant parameter and it is some influence load correlative factor determine future under conditions of meeting certain required precision
The load data of certain particular moment.Load prediction is either in the economic adjusting and control of electric system still in the optimization of building system
Control aspect has important role.For electric system, accurate load prediction can be with the arrangement power grid of economical rationality inside
The start and stop of generating set keep the security and stability of operation of power networks, effectively reduce cost of electricity-generating, increase economic efficiency and social
Benefit.For building system, accurate Building Cooling load prediction can instruct the operation for building interior heating ventilation air-conditioning system, to the greatest extent
Amount keeps building system for the matching of energy and building load, while controlling Architectural Equipment and running in the state of efficiently, improves
Its economy and reliability of system.
The more accurate machine learning method that data-driven is mainly based upon with efficient method in load prediction at present, this
For kind method compared to other based on physical modeling and based on the prediction technique of cinder box, the simple accuracy simultaneously of modeling is higher.However,
Obtaining accurate load prediction results to the method by machine learning needs a large amount of data to be trained model, and mistake
More input datas will lead to the problems such as model calculation time is long, and generalization ability caused by model over-fitting is poor.It would therefore be desirable to have
The data characteristics screening technique of effect carries out dimensionality reduction to the data of input model, while this method needs reducing input data dimension
Guarantee that the critical data feature of load prediction is complete, avoids causing the accuracy of load prediction excessive while spending as far as possible
It influences.The processing method of the input data of load prediction at present is mainly the statistical feature according to data, such as correlation analysis
Feature Selection and dimensionality reduction are carried out Deng to data.However, the relationship of statistical feature and load prediction accuracy is still not clear, pass through
The key feature that this method may cause prediction is removed and the result accuracy of load prediction is caused to decline.
Therefore the dimension of input data how is reduced during data processing, while guaranteeing the negative of machine learning as far as possible
The technical issues of key characteristics in lotus prediction are not removed, and are urgent solutions at present
Summary of the invention
To overcome above-mentioned the deficiencies in the prior art, the present invention provides a kind of load based on clustering and neural network
Predict the screening technique of input feature vector, the method carries out Hierarchical clustering analysis to the initial data of load prediction；Utilize nerve
Network judges the prediction effect of different parameters in same class, rejects the bad parameter of prediction effect, and what is optimized is negative
Lotus predicts input feature vector collection.Compared to other existing load prediction input data processing methods, obtained feature set more adapts to be based on
The load forecasting method of machine learning, and can preferably guarantee the accuracy of the result of load prediction.
To achieve the above object, a kind of technical solution of the present invention: load prediction based on hierarchical clustering and neural network
Input feature vector screening technique, comprising the following steps:
Hierarchical clustering analysis is carried out to the initial data of load prediction, it is embedding to obtain having levels for load prediction input data
Cover clustering tree；
Building load prediction input Parameter Clustering cluster number is determined according to distance between the cluster in hierarchical clustering；
Judged using prediction effect of the neural network to different parameters in the data of every one kind, it is poor to reject prediction
Parameter, the load prediction input feature vector collection optimized.
Further, the hierarchical clustering is seen each sample in data set using the aggregation strategy of " bottom-up "
Make an initial clustering cluster, every one-step polymerization merges two nearest clustering clusters of distance, constantly repeats until being polymerized to
One clustering cluster.
Further, the distance function uses Euclidean distance function, to sample xi=(xi1；xi2；…；xi3) and xj=
(xj1；xj2；…；xj3), distance function dist () are as follows:
Further, distance is calculated using average distance between the cluster, for giving clustering cluster CiAnd Cj, cluster spacing
From are as follows:
Size relation between cluster used by using polymerizeing every time in cluster process between distance is as the final cluster of selection
The foundation of cluster number, the similitude according to the clustering cluster of the small formation of distance between cluster is big, the similitude of the class of the big formation of between class distance
Small principle will if distance is big between cluster used by this time polymerize between cluster when the pre-polymerization of the obvious ratio of value added of distance
Cluster number of clusters before this time polymerization is the cluster number of clusters for finally dividing cluster.
Further, parameter prediction effect judgment method neural network based are as follows:
(1) test set is established for the different parameters in each clustering cluster, includes clustering cluster where the parameter in test set
Whole parameters of other outer clustering clusters, not comprising the other parameters in the clustering cluster in addition to the parameter；
(2) the identical neural network model of structure is established for the same prediction target to each test set；
(3) using the prediction result of neural network as the foundation for the prediction effect for evaluating corresponding parameter；
(4) root-mean-square error (RMSE), mean absolute error (MAE), deterministic coefficient (R are utilized2) and root-mean-square error
Evaluation index of the coefficient of variation (CV-RMSE) as the prediction effect to different parameters, calculation method are defined as follows:
Wherein: Fi--- actual measured value；
Fi' --- model predication value.
Further, the corresponding parameter of neural network model that prediction effect is most bad in each clustering cluster is rejected, is obtained
The load prediction input data set of final optimization.
Beneficial effect
(1) the present invention is based on clusterings classifies to load prediction input data, the data characteristics tool in same class
There is certain similitude, reject partial data therein, remaining data can still retain part of it feature, therefore in cluster
On the basis of on data carry out screening reject the diversity of data characteristics is influenced it is smaller, on the accuracy of load prediction influence compared with
It is small.
(2) the present invention is based on prediction effect of the neural network to different parameters to evaluate, and evaluation result can fit well
Should be based on the load forecasting model of machine learning, this method is compared to the data characteristics screening technique based on data statistics parameter
The data characteristics conducive to load forecasting model can be preferably remained with, guarantees the accuracy of load prediction.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow charts of hierarchical clustering and the load prediction input feature vector screening technique of neural network；
Fig. 2 is the corresponding diagram of data and its title in the embodiment of the present invention；
Fig. 3 is the Hierarchical clustering analysis result in the embodiment of the present invention；
Fig. 4 is the state of aggregation tabular drawing in the middle-level cluster of the embodiment of the present invention；
Fig. 5 is that neural network evaluates tabular drawing to load prediction input data prediction effect in the embodiment of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference
Attached drawing, the present invention is described in further detail.
As shown in Figure 1, being sieved the present embodiment provides a kind of based on the load prediction input feature vector of clustering and neural network
Choosing method, comprising the following steps:
Step 1: using the hierarchy clustering method of " bottom-up ", the input Parameter Clustering formation of load prediction being had levels
Nested clustering tree.
Be selected as case is the Feature Selection fission for the input data of prediction model for being directed to building load, to building
The input data of load prediction generally includes to disturb parameter in building (such as: opening of device situation in the interior personnel amount of building, building
Deng), building is outer to disturb parameter (such as: outdoor temperature, outside humidity, sun load value), parameters history value.Due to building enclosure knot
The history value of the thermal inertia of structure, feature of the building load with delay, relevant parameter and load data is to the pre- of building load
Survey plays a very important role.Included input data in the present embodiment is specifically shown in Fig. 2.
Using each input data as an initial clustering cluster, the Euclidean distance between all clustering clusters is counted
It calculates, calculation formula is as follows:
Wherein, xiAnd xjFor initial clustering cluster；dist(xi,xj) be initial clustering cluster Euclidean distance.
The smallest two data of Euclidean distance are merged into a clustering cluster, record the Euclidean distance.For what is newly formed
Euclidean distance is recalculated between clustering cluster, using average distance to cluster when calculating Euclidean distance for the clustering cluster that polymerization is formed
Between distance calculated, calculation formula is as follows:
Wherein, Ci,CjRespectively different clustering clusters, x, z are respectively Ci,CjIn data.
The smallest clustering cluster polymerization of distance between cluster is become into a cluster again, constantly repeatedly above step, until all
Data aggregate becomes a clustering cluster, and finally obtained cluster result such as Fig. 3 is indicated, distance between the data and cluster of every step polymerization
As shown in Figure 4.Abscissa is sample number in Fig. 3, and ordinate is the clustering cluster distance by scale again.
Step 2: the number of final clustering cluster is determined according to distance between cluster.
Clustering cluster is determined as final according to distance between the cluster between two clustering clusters in one-step polymerization every in cluster
Several foundations.
Distance is as shown in figure 3, know cluster process in total between each step clusters in embodiment clustering cluster number and cluster
Undergo 11 steps that all initial clustering clusters are polymerized to one kind.By the statistics between distance cluster it is found that not with clustering cluster
Disconnected cohesion, the number of clustering cluster are constantly reduced, and the distance between cluster is being gradually increased, before cluster proceeds to step 9, cluster spacing
Smaller from the amplitude of increase, the similitude according to the clustering cluster of the small formation of distance between cluster is big, the phase of the class of the big formation of between class distance
Like the small principle of property, distance increases amplitude increase between the cluster compared to previous step polymerization of distance between cluster when finding step 9 polymerization
Obviously, it is big also to compare former steps for the increase of distance between the cluster of the polymerization process and after step 9, therefore step 9 is polymerize it
For the number of the clustering cluster of preceding formation as final clustering cluster number, the position for clustering stopping is as shown in phantom in Figure 3, i.e., finally will
Input data is divided into 4 classes.Specifically, the first kind includes: Toutp、ΔTp、SRp, the second class includes: Twsp、Twrp、CLp、Gp、
Tinp, third class includes: RHop、RHip、ΔTwp, the 4th class includes: Occp
Step 3: being tested and evaluated using prediction effect of the neural network to the parameter in every cluster, it is pre- to reject load
Survey the bad parameter of effect, the load prediction data feature set optimized.
According to cluster as a result, the data in embodiment are divided into 4 clustering clusters, the data in each clustering cluster have one
Fixed similitude, therefore it also has certain similitude in the effect of load prediction, using neural network model to different numbers
According to prediction effect evaluated.The input layer data of neural network is constructed for some data, including different from the data
All data of clustering cluster, do not include with the other parameters in its same clustering cluster, output layer be building load value.In embodiment
In, it include the outdoor temp angle value Tout of previous hour in one of clustering clusterp, the indoor and outdoor temperature difference Δ of previous hour
TpAnd the intensity of solar radiation SR of previous hourp, for ToutpNeural network prediction model is established, inputs in layer data and wraps
Include ToutpAnd the data in other clustering clusters, without including Δ TpAnd SRp.It equally include SRpNeural network prediction model
It does not include ToutpWith Δ Tp, including Δ TpNeural network prediction model in do not include SRpAnd Toutp.For all parameters according to
Same method establishes neural network prediction model, divides training set and test set, each neural network of training obtain with each
The corresponding neural network prediction model of data.All neural network models are tested using same test set, finally
Result root-mean-square error (RMSE), mean absolute error (MAE), deterministic coefficient (R2) and the root-mean-square error coefficient of variation
(CV-RMSE) it is compared as evaluation index, calculation method is defined as follows:
Wherein: Fi--- actual measured value；
Fi' --- model predication value.
Final result is analyzed, the parameter that neural network model effect is worst in same clustering cluster is rejected, for
Result such as Fig. 5, Δ T of the forecasting accuracy index of the neural network prediction model of a certain clustering cluster in embodimentpEvaluation refer to
Be marked on it is worst in all data of this clustering cluster, therefore reject Δ Tp, SRpAnd ToutpIt is not including Δ TpNerve net
It is available than including Δ T in network prediction modelpThe better prediction effect of neural network prediction model, illustrate SRpAnd Toutp
It can be used as the building load prediction input feature vector of final optimization pass, while excessive data characteristics will not be lost.It is poly- to each
Class cluster all carries out same screening process, finally obtains the input feature vector collection of building load prediction, comprising: the first kind includes:
Toutp、SRp、Twrp、CLp、Gp、Tinp、RHip、ΔTwp、Occp。
Particular embodiments described above has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects
Describe in detail bright, it should be understood that the above is only a specific embodiment of the present invention, is not intended to restrict the invention, it is all
Within the spirit and principles in the present invention, any modification, equivalent substitution, improvement and etc. done should be included in protection of the invention
Within the scope of.
Claims (7)
1. the load prediction input feature vector screening technique based on hierarchical clustering and neural network, which is characterized in that including following step
It is rapid:
1) Hierarchical clustering analysis is carried out to the initial data of load prediction, obtains the nesting of load prediction input data having levels
Clustering tree；
2) building load prediction input Parameter Clustering cluster number is determined according to distance between the cluster in hierarchical clustering；
3) judged using prediction effect of the neural network to different parameters in the data of every one kind, reject and predict poor ginseng
Number, the load prediction input feature vector collection optimized.
2. the load prediction input feature vector screening technique according to right 1 based on hierarchical clustering and neural network, feature
It is, the hierarchical clustering is regarded each sample in data set as one and initially gathered using the aggregation strategy of " bottom-up "
Class cluster, every one-step polymerization merge two nearest clustering clusters of distance, constantly repeat until being polymerized to a clustering cluster.
3. the load prediction input feature vector screening technique according to right 2 based on hierarchical clustering and neural network, feature
It is, the distance function uses Euclidean distance function, and expression formula is as follows:
To sample xi=(xi1；xi2；…；xi3) and xj=(xj1；xj2；…；xj3), distance function dist () are as follows:
4. the load prediction input feature vector screening technique according to right 2 based on hierarchical clustering and neural network, feature
It is, distance is calculated using average distance between the cluster, and expression formula is as follows:
For giving clustering cluster CiAnd Cj, distance between cluster are as follows:
5. the load prediction input feature vector screening technique according to right 1 based on hierarchical clustering and neural network, feature
Be, using polymerizeing every time in cluster process used by size relation between cluster between distance it is a as final clustering cluster is chosen
Several foundations；Similitude according to the clustering cluster of the small formation of distance between cluster is big, and the similitude of the class of the big formation of between class distance is small
Principle: if distance is big between cluster used by this time polymerize between cluster when the pre-polymerization of the obvious ratio of value added of distance, by this time
Cluster number of clusters before polymerization is the cluster number of clusters for finally dividing cluster.
6. the load prediction input feature vector screening technique according to right 1 based on hierarchical clustering and neural network, feature
It is, the constructive method of the load prediction input data set of the optimization is that reject prediction effect in each clustering cluster most bad
The corresponding parameter of neural network model, the data set constituted is the load prediction input feature vector collection of final optimization.
7. the load prediction input feature vector screening technique according to right 1 based on hierarchical clustering and neural network, feature
It is, the prediction effect judgment step are as follows:
(1) test set is established for the different parameters in each clustering cluster, includes outside the clustering cluster of parameter place in test set
Whole parameters of other clustering clusters, not comprising the other parameters in the clustering cluster in addition to the parameter；
(2) the identical neural network model of structure is established for the same prediction target to each test set；
(3) using the prediction result of neural network as the foundation for the prediction effect for evaluating corresponding parameter；
(4) root-mean-square error (RMSE), mean absolute error (MAE), deterministic coefficient (R are utilized2) and root-mean-square error variation lines
Evaluation index of the number (CV-RMSE) as the prediction effect to different parameters, calculation method are defined as follows:
Wherein: Fi--- actual measured value；
Fi' --- model predication value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910294713.6A CN110135623A (en) | 2019-04-12 | 2019-04-12 | Load prediction input feature vector screening technique based on hierarchical clustering and neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910294713.6A CN110135623A (en) | 2019-04-12 | 2019-04-12 | Load prediction input feature vector screening technique based on hierarchical clustering and neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110135623A true CN110135623A (en) | 2019-08-16 |
Family
ID=67569938
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910294713.6A Pending CN110135623A (en) | 2019-04-12 | 2019-04-12 | Load prediction input feature vector screening technique based on hierarchical clustering and neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110135623A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110851762A (en) * | 2019-08-30 | 2020-02-28 | 北京中环合创环保能源科技有限公司 | Building dynamic simulation model building method of solar centralized heating system |
| CN111476455A (en) * | 2020-03-03 | 2020-07-31 | 中国南方电网有限责任公司 | Power grid operation section feature selection and online generation method based on two-stage structure |
| CN113610152A (en) * | 2021-08-06 | 2021-11-05 | 天津大学 | Air conditioning system flexible operation strategy making method based on load mode |
| CN114004513A (en) * | 2021-11-04 | 2022-02-01 | 浙江大学 | Demand prediction method, system and storage medium |
| US11628848B2 (en) | 2020-03-31 | 2023-04-18 | Toyota Research Institute, Inc. | Systems and methods for training a neural network for estimating a trajectory of a vehicle |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007299059A (en) * | 2006-04-27 | 2007-11-15 | Matsushita Electric Works Ltd | Signal discrimination method |
| CN108874959A (en) * | 2018-06-06 | 2018-11-23 | 电子科技大学 | A kind of user's dynamic interest model method for building up based on big data technology |
| CN109086928A (en) * | 2018-07-27 | 2018-12-25 | 福州大学 | Photovoltaic plant realtime power prediction technique based on SAGA-FCM-LSSVM model |
| CN109309382A (en) * | 2018-09-13 | 2019-02-05 | 广东工业大学 | A kind of Short-Term Load Forecasting Method |
| CN109595742A (en) * | 2018-11-29 | 2019-04-09 | 天津大学 | A kind of inexpensive adapting method of existing large public building air-conditioning system |
-
2019
- 2019-04-12 CN CN201910294713.6A patent/CN110135623A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007299059A (en) * | 2006-04-27 | 2007-11-15 | Matsushita Electric Works Ltd | Signal discrimination method |
| CN108874959A (en) * | 2018-06-06 | 2018-11-23 | 电子科技大学 | A kind of user's dynamic interest model method for building up based on big data technology |
| CN109086928A (en) * | 2018-07-27 | 2018-12-25 | 福州大学 | Photovoltaic plant realtime power prediction technique based on SAGA-FCM-LSSVM model |
| CN109309382A (en) * | 2018-09-13 | 2019-02-05 | 广东工业大学 | A kind of Short-Term Load Forecasting Method |
| CN109595742A (en) * | 2018-11-29 | 2019-04-09 | 天津大学 | A kind of inexpensive adapting method of existing large public building air-conditioning system |
Non-Patent Citations (7)
| Title |
|---|
| KUN YANG,等: "Multi-criteria integrated evaluation of distributed energy system for community energy planning based on improved grey incidence approach: A case study in Tianjin", 《APPLIED ENERGY》 * |
| YAN DING,等: "Benchmark analysis of electricity consumption for complex campus buildings in China", 《APPLIED THERMAL ENGINEERING》 * |
| YOUSHAN ZHANG，QI LI: "A Regressive Convolution Neural network and upport Vector Regression Model for Electricity Consumption Forecasting", 《FICC 2019》 * |
| 刘敏，等: "办公建筑室内热源冷负荷计算修正模型", 《暖通空调》 * |
| 吴争荣，等: "基于配用电信息分区分类的短期空间负荷预测", 《电力系统及其自动化学报》 * |
| 徐步云，倪禾: "自组织神经网络和K-means聚类算法的比较分析", 《新型工业化》 * |
| 熊贇，等: "《大数据挖掘》", 30 April 2016 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110851762A (en) * | 2019-08-30 | 2020-02-28 | 北京中环合创环保能源科技有限公司 | Building dynamic simulation model building method of solar centralized heating system |
| CN111476455A (en) * | 2020-03-03 | 2020-07-31 | 中国南方电网有限责任公司 | Power grid operation section feature selection and online generation method based on two-stage structure |
| US11628848B2 (en) | 2020-03-31 | 2023-04-18 | Toyota Research Institute, Inc. | Systems and methods for training a neural network for estimating a trajectory of a vehicle |
| CN113610152A (en) * | 2021-08-06 | 2021-11-05 | 天津大学 | Air conditioning system flexible operation strategy making method based on load mode |
| CN113610152B (en) * | 2021-08-06 | 2023-07-18 | 天津大学 | Load mode-based air conditioning system flexibility operation strategy formulation method |
| CN114004513A (en) * | 2021-11-04 | 2022-02-01 | 浙江大学 | Demand prediction method, system and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110135623A (en) | Load prediction input feature vector screening technique based on hierarchical clustering and neural network | |
| CN103730006B (en) | A kind of combination forecasting method of Short-Term Traffic Flow | |
| CN105117602B (en) | A kind of metering device running status method for early warning | |
| CN109214566A (en) | Short-term wind power prediction method based on shot and long term memory network | |
| CN109754113A (en) | Load forecasting method based on dynamic time warping Yu length time memory | |
| CN105160444B (en) | Electrical equipment failure rate determining method and system | |
| CN109659933A (en) | A kind of prediction technique of power quality containing distributed power distribution network based on deep learning model | |
| CN104657791B (en) | A kind of wind farm group wind speed profile prediction technique based on correlation analysis | |
| CN103488869A (en) | Wind power generation short-term load forecast method of least squares support vector machine | |
| CN106503867A (en) | A kind of genetic algorithm least square wind power forecasting method | |
| CN109143408B (en) | Dynamic region combined short-time rainfall forecasting method based on MLP | |
| CN108053052B (en) | A kind of oil truck oil and gas leakage speed intelligent monitor system | |
| CN111369070A (en) | Envelope clustering-based multimode fusion photovoltaic power prediction method | |
| CN104899135A (en) | Software defect prediction method and system | |
| CN105701572A (en) | Photovoltaic short-term output prediction method based on improved Gaussian process regression | |
| CN109858700A (en) | BP neural network heating system energy consumption prediction technique based on similar screening sample | |
| CN104331737A (en) | Office building load prediction method based on particle swarm neural network | |
| CN115859837B (en) | Digital twin modeling-based fan blade dynamic impact detection method and system | |
| CN104850867A (en) | Object identification method based on intuitive fuzzy c-means clustering | |
| CN112232561A (en) | Power load probability prediction method based on constrained parallel LSTM quantile regression | |
| CN105930900A (en) | Method and system for predicting hybrid wind power generation | |
| CN104539601A (en) | Reliability analysis method and system for dynamic network attack process | |
| CN112288157A (en) | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning | |
| CN106980874B (en) | A kind of multi-time Scales dimension data fusion method towards distribution big data | |
| CN115600729A (en) | Grid load prediction method considering multiple attributes |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190816 |
|
| WD01 | Invention patent application deemed withdrawn after publication |