CN109919374A - Prediction of Stock Price method based on APSO-BP neural network - Google Patents
Prediction of Stock Price method based on APSO-BP neural network Download PDFInfo
- Publication number
- CN109919374A CN109919374A CN201910154140.7A CN201910154140A CN109919374A CN 109919374 A CN109919374 A CN 109919374A CN 201910154140 A CN201910154140 A CN 201910154140A CN 109919374 A CN109919374 A CN 109919374A
- Authority
- CN
- China
- Prior art keywords
- particle
- neural network
- prediction
- apso
- price
- 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
Abstract
The invention discloses a kind of Prediction of Stock Index methods based on APSO-BP neural network, including data prediction, building APSO-BP neural network prediction model.It goes forward side by side line number Data preprocess firstly, extracting validity feature in the stock certificate data of collection, forms characteristic sample set.Then, suitable BP neural network model is constructed.Optimize the initial weight and threshold value of the BP neural network using TSP question particle swarm algorithm (APSO) later, to obtain final APSO-BP neural network prediction model.The precision of prediction of the method for the present invention is higher, and will not fall into local extremum, and fast convergence rate, predicted time is short, thereby executing more efficient.
Description
Technical field
The invention belongs to the fields machine learning (Machine Learning, ML), relate generally to a kind of Prediction of Stock Price
Method is based especially on the Prediction of Stock Price method of APSO-BP neural network.
Background technique
With the persistently overheating of machine learning research temperature, machine learning has also obtained extensively in financial time series field
Application have very strong capability of fitting because the ability of artificial neural network processing nonlinear problem is prominent, there is good prediction
Property and practicability, obtain comparing accurately prediction result in certain fields, so more and more researcher is by it and finance
Time series combines, and applies in stock trend prediction.Machine learning can be arranged from a large amount of historical data, be excavated
The rule and valuable information that time series itself implies obtain difference profit for investor and provide effective technical support.
Although present Financial Time Series Forecasting technology has very much, what existing technology still cannot be simple and efficient reaches expected
Effect.In addition, the economic development of entire society and the well development in financial market are closely related, China's Financial market is also in hair
The exhibition stage.Therefore, machine learning is applied to finance data, is capable of the inherent law in Correct Analysis financial market, constantly holds
Its dynamic change has important practical significance.
Machine learning techniques can imitate the stock market of fluctuation, can produce than conventional method in stock trend prediction
Raw better prediction result.Traditional BP neural network is that the work side of neuron is imitated from physiological angle to a certain extent
Formula can arbitrarily approach non-linear complicated function, but it is easily trapped into local extremum, and convergence rate is slow, so as to cause prediction
Model performance does not reach expection.
Summary of the invention
To solve the problems of the above-mentioned prior art, the present invention proposes a kind of based on the BP for improving particle swarm algorithm optimization
Neural network prediction model, proposes a kind of TSP question grain at the characteristics of realizing simple, fast convergence rate using particle swarm algorithm
What subgroup optimization algorithm (Adaptive Particle Swarm Optimization, APSO) and BP neural network combined
APSO-BP algorithm, the algorithm introduce variation particle to adjust the weight and threshold value of BP neural network, and will nerve net after optimization
Network weight and threshold value, which substitute into BP network, is tested, and is outputed test data, is adjusted the weight of BP network, so that prediction stock
Admission fee lattice direction is more accurate, and the problem of overcame adaptation, avoids falling into local optimum.
In order to achieve the above objectives, the present invention takes following technical scheme:
S1: obtaining stock certificate data, carries out Feature Selection and data prediction；
S2: the network structure of BP neural network, including input layer number, output layer number of nodes, hidden layer node are determined
Number, activation primitive and total number of particles；
S3 constructs the BP neural network model optimized based on APSO, specifically:
S3.1, initiation parameter: particle group velocity, position, inertial factor w, Studying factors c1,c2, maximum number of iterations,
The learning rate η and aimed at precision g of neural network；
S3.2 calculates particle and adapts to, using the mean square error of stock advance-decline prediction value and actual value as fitness function,
Formula is as follows:
Wherein, djFor desired output；yjFor reality output；M is number of samples；
S3.3, it is optimal to find out each particle individual, sets local optimum for the optimal location under particle current iteration；
S3.4 finds out the global optimum of entire group, compares the mean square error of each particle in population, obtains the overall situation most
It is excellent；
S3.5 updates particle rapidity, formula are as follows:
Wherein Vk、Xk、Respectively correspond speed of certain particle in kth time iteration, position, optimal location；Generation
The global optimum position of table population in kth time iterative process, r1,r2∈ [0,1], is uniformly distributed at random；
If V > Vmax, then V=Vmax；If V <-Vmax, then V=-Vmax, otherwise constant；
S3.6 gives random variation for non-optimal particle, updates particle position:
Calculation formula are as follows: Xk+1=Xk+Vk+1；
S3.7 judges whether that reaching maximum cycle or fitness value reaches target value, if then entering S3.8, otherwise
Return to S3.2；
S3.8, export optimal particle, be up to requirement particle correspond BP neural network initial connection weight with
Threshold value；
S3.9 inputs training data, training BP neural network；
S3.10, input test sample；
S3.11 exports prediction result, and is compared with actual value.
Preferably, in step S1, the feature of screening includes opening price, closing price, highest price, lowest price and exchange hand；Institute
State data prediction specifically: be normalized to stock certificate data, normalize formula are as follows:
Wherein, certain time series x={ x1,x2,...,xn, xnormIt is the value after normalization, xminIt is minimum value in sequence,
xmaxIt is the maximum value in sequence.
Preferably, in step S2, input layer includes opening price, closing price, highest price, lowest price and the exchange hand of stock,
The ups and downs situation that output layer is next day；Node in hidden layer is by formulaIt determines, wherein q is hidden layer section
Points；N is input layer number；M is output layer number of nodes；A is the constant between 0~10；The activation primitive usesTotal number of particles S=n × q+q × the m+q+m.
Compared with the prior art, the present invention have it is following the utility model has the advantages that
1, it is slow to solve BP neural network convergence rate for APSO-BP neural net prediction method proposed by the present invention, nerve
The problem of futile-iteration of network, consuming time is long；
2, APSO algorithm proposed by the present invention introduces variation thought, and particle is enable to jump out previous optimal value position,
Carry out search in bigger space, while having stronger local search ability, ability of searching optimum is also promoted obviously, is overcome
The problem of being easily trapped into local extremum；
3, APSO algorithm high degree proposed by the present invention finds optimal BP neural network initial weight and threshold value goes to replace
Random value improves precision of prediction.
4, the present invention improves the precision of prediction of share price, this is significant to investor, and moves to financial market is held
State variation has realistic meaning.
Detailed description of the invention
Fig. 1 is system model figure of the invention；
Fig. 2 is flow chart of the invention；
Fig. 3 (a) is the convergence process figure of Apple Inc.'s mean square error in present example；
Fig. 3 (b) is the convergence process figure of Amazon Company's mean square error in present example；
Fig. 3 (c) is the convergence process figure of Microsoft's mean square error in present example.
Specific embodiment
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 show forecast system model of the invention, and the data used in the prediction model are a certain stock respectively
Opening price, highest price, lowest price, closing price, exchange hand, next day trend.Herein, next day trend is contingency table
Label, tag along sort is allocated according to closing price, Ct+1T+1 days closing prices are represented, then in the t days training label yt
It can be calculated by the following formula:
If next day closing price is greater than the closing price of today, it is judged to rising, use 1 indicates, it is on the contrary then be judged to falling,
It is indicated with -1.
After inputting stock certificate data, then with APSO algorithm go each of Optimized BP Neural Network weight and threshold value.Then
Determined, if ineligible, continues to be optimized with APSO, the BP neural network after otherwise being optimized.
Referring to Fig. 2, this implementation includes the following steps:
S1 obtains stock certificate data
S1.1, Feature Selection
Apple (Apple), Amazon of the stock certificate data collection that the present invention uses from the offer of the Yahoo website (Yahoo)
(Amazon), Microsoft (Microsoft) stock information.Using the data of 2010-04-01 to 2014-05-10 as training data,
The data of 2014-05-10 to 2014-10-12 are as test data.Use herein the opening price of stock, closing price, highest price,
Lowest price, exchange hand as input feature vector, next day ups and downs situation as output, partial data as indicated with 1:
1 Microsoft's stock certificate data of table
| Date | Opening price | Highest price | Lowest price | Closing price | Exchange hand | Stock trend |
| 2010/4/01 | 29.64 | 29.72 | 29.17 | 29.29 | 63760000 | -1 |
| 2010/4/02 | 29.35 | 29.54 | 28.62 | 29.16 | 74768100 | 1 |
| 2010/4/03 | 29.13 | 29.43 | 29.03 | 29.27 | 34331200 | 1 |
| 2010/4/04 | 29.15 | 29.58 | 28.98 | 29.32 | 47366800 | 1 |
| 2010/4/05 | 29.16 | 29.56 | 29.14 | 29.35 | 58318800 | 1 |
| 2010/4/06 | 29.32 | 29.98 | 29.30 | 29.92 | 63713800 | 1 |
| 2010/4/07 | 29.95 | 30.41 | 29.9 | 30.34 | 54752500 | -1 |
| 2010/4/08 | 30.25 | 30.49 | 30.21 | 30.32 | 37068800 | 1 |
| 2010/4/09 | 30.15 | 30.5 | 30.13 | 30.45 | 41374600 | 1 |
| 2010/4/10 | 30.79 | 31.0 | 30.66 | 30.82 | 68941200 | 1 |
| 2010/4/11 | 30.82 | 30.95 | 30.71 | 30.87 | 52745400 | -1 |
S1.2, data prediction
As shown in Table 1 above, the parameter that the present invention inputs is more, closing price, exchange hand, stock trend they have difference
Unit, it is therefore desirable to it is normalized, after first stock certificate data is normalized, then inputs corresponding mould
Type.It is as follows to normalize formula:
Wherein assume certain time series x={ x1,x2,...,xn, xnormIt is the value after normalization, xminIt is minimum in sequence
Value, xmaxIt is the maximum value in sequence.
S2 determines the network structure of BP neural network.
S2.1, the determination of input layer and output layer number of nodes
The present invention chooses opening price, closing price, highest price, lowest price and the exchange hand of stock as input feature vector；Under
One day ups and downs situation is output.
S2.2, the determination of node in hidden layer
Node in hidden layer has larger impact to nicety of grading；If number of nodes is very little, network cannot learn well, need
Increase frequency of training, trained precision also can be impacted；Conversely, net training time increases, network is easy over-fitting.
The empirical equation of BP neural network hidden layer number are as follows:
Wherein: nhFor node in hidden layer；N is input layer number；M is output layer number of nodes；A is between 0~10
Constant.In this example, input node 5, output node 1, it is 4~14 that range, which is arranged, in hidden layer.
S2.3, the determination of activation primitive
S2.4, the determination of total number of particles
S=n × q+q × m+q+m
Wherein: n is input layer number；Q is node in hidden layer；M is output layer number of nodes.N=5,4 in this example≤
Q≤14, m=1.
S3, in conjunction with the BP neural network model construction optimized based on TSP question particle swarm optimization algorithm (APSO), mistake
Journey is as follows:
S3.1, initiation parameter: particle group velocity, position, inertial factor w, Studying factors c1,c2, maximum number of iterations,
The learning rate η and aimed at precision g of neural network.
Particle swarm algorithm parameter c under normal circumstances1=c2=2, vmax=1, maximum number of iterations 100, learning rate is
0.1, aimed at precision 0.001, inertia weight w is the random value between 0~1.
Due to the parameter of particle swarm algorithm and the value of the neural network parameter prediction result last to algorithm have it is larger
Influence, accurate result in order to obtain, parameter is analyzed in this example selection table 2.
Table 2
S3.2 calculates particle fitness.The present invention is using the mean square error of stock advance-decline prediction value and actual value as suitable
Response function, formula are as follows:
Wherein, djFor desired output；yjFor reality output；M is number of samples.
It is optimal to find out each particle individual by S3.3.Local optimum is set by the optimal location under particle current iteration.?
In iterative process, certain random variation is given to the position and speed of non-global optimum's particle, i.e., is chosen from population at random
A particle is selected, then to this particle variations, formula is as follows:
f(r3> u) then xi k=r4x
Wherein r3、r4、r5、r6Indicate the equally distributed random number in 0~1, u, w are the constants between 0~1.It represents
The position reinitialized, expression reinitialize speed, and k represents kth time iteration.
When the speed of particle is zero, mutation operation is carried out to its speed, so that the particle stagnated obtains new speed again
Degree, in favor of finding optimal value, formula is as follows:
Wherein r7Indicate the equally distributed random number in 0~1, μ represents the direction of Particles Moving, and v indicates again initial
Change speed.But work as random number r7When greater than 0.5, μ is -1, and otherwise μ is 1.
S3.4 finds out the global optimum of entire group.The mean square error for comparing each particle in population obtains the overall situation most
It is excellent.
S3.5 updates particle rapidity, formula are as follows:
Wherein Vk、Xk、Respectively correspond speed of certain particle in kth time iteration, position, optimal location；Generation
The global optimum position of table population in kth time iterative process, r1,r2∈ [0,1], is uniformly distributed at random.Update particle rapidity
Consider whether updated particle rapidity is limiting in range, if V > Vmax, then V=Vmax；If V <-Vmax, then V=-Vmax,
Otherwise constant.
S3.6 updates particle position, formula are as follows:
Xk+1=Xk+Vk+1
Wherein Vk、XkRespectively correspond speed of certain particle in kth time iteration, position.Particle position is calculated according to above-mentioned formula
It sets, certain random variation is given to the position of non-optimal particle, i.e., pick out a particle x from population at randomi k, then it is right
This particle random value, k represent kth time iteration, and population forgets the history desired positions of oneself and only remembers population most
Best placement, so that population keeps population diversity.
S3.7 judges whether that reaching maximum cycle or fitness value reaches target value, if then continuing S3.8, otherwise
Return to S3.2.
S3.8 exports optimal particle.Be up to requirement particle correspond BP neural network initial connection weight with
Threshold value.
S3.9 inputs training data, training BP neural network.
S3.10, input test sample.
S3.11 finally exports prediction result, and is compared with actual value.
According to above-mentioned steps, this example obtains with APSO-BP prediction model when different parameters are arranged different pre-
Survey precision.First three data of every Prediction of Stock Index performance of the model are finally obtained, as shown in table 3.
First three parameter combination of 3 ranking of table exports result
Apple Prediction of Stock Index precision average is 0.66 in table 3, and highest reaches 0.69, and parameter setting at this time is nerve
Network training number is 1000 times, step-length 0.1, factor of momentum 0.9, inertia weight 0.8, and error precision 0.001 implies
Layer neuronal quantity is 10.Amazon Prediction of Stock Index precision highest reaches 0.62, and parameter setting at this time is neural metwork training
Number is 8000 times, step-length 0.1, factor of momentum 0.8, inertia weight 0.7, error precision 0.001, hidden layer neuron
Quantity is 9.Microsoft's Prediction of Stock Index precision highest reaches 0.67, and parameter setting at this time is that neural metwork training number is 1000
Secondary, step-length 0.1, factor of momentum 0.3, inertia weight 0.8, error precision 0.001, hidden layer neuron quantity is 9.
By the above parameter setting, by taking the parameter of full accuracy as an example, traditional PS O-BP is respectively adopted to each branch stock and is calculated
Method and BP algorithm are predicted, are obtained shown in the convergence process figure such as 3 (a) -3 (c) of mean square error.It can be seen from the figure that through
After crossing APSO algorithm optimization BP neural network weight, the either decrease speed of mean square error gradient, or decline degree is all bright
The aobvious traditional PS O algorithm optimization BP mind that is better than is by network.The indices of each branch stock are as shown in table 4.
4 prediction data of table compares
The precision of prediction average value of APSO-BP prediction model is 0.66 as can be seen from Table 4, is better than PSO- from precision of prediction
BP prediction model and BP prediction model.Table 5 indicates that prediction model expends the time, and wherein the parameter setting of prediction model is all unified,
Practicing number is 1000 times, step-length 0.1, factor of momentum 0.8, inertia weight 0.8, error precision 0.001, hidden layer nerve
First quantity is 9.
PSO-BP prediction model is respectively adopted to time series and APSO-BP prediction model carries out algorithm model time-consuming ratio
Compared with.
5 prediction model time-consuming of table compares (unit: second s)
| Stock classification | PSO-BP algorithm | APSO-BP algorithm |
| Apple | 8.31 | 7.32 |
| Amazon | 8.22 | 7.20 |
| Microsoft | 8.32 | 7.21 |
| Average | 8.28 | 7.24 |
As shown in table 5, APSO-BP prediction model it is average it is time-consuming be 7.24s, and traditional PSO-BP prediction model disappears
Time-consuming is 8.28s, is primarily due to, and ability of searching optimum early period of APSO-BP prediction model is strong, and PSO-BP searching algorithm
In contrast, it is easily trapped into local extremum, increases local searching times, convergence rate is slow, expends more times.
Claims (3)
1. the Prediction of Stock Price method based on APSO-BP neural network, comprising the following steps:
S1: obtaining stock certificate data, carries out Feature Selection and data prediction；
S2: determining the network structure of BP neural network, including input layer number, output layer number of nodes, node in hidden layer, swashs
Function living and total number of particles；
S3 constructs the BP neural network model optimized based on APSO, specifically:
S3.1, initiation parameter: particle group velocity, position, inertial factor w, Studying factors c1,c2, maximum number of iterations, nerve
The learning rate η and aimed at precision g of network；
S3.2 calculates particle and adapts to, using the mean square error of stock advance-decline prediction value and actual value as fitness function, formula
It is as follows:
Wherein, djFor desired output；yjFor reality output；M is number of samples；
S3.3, it is optimal to find out each particle individual, sets local optimum for the optimal location under particle current iteration；
S3.4 finds out the global optimum of entire group, compares the mean square error of each particle in population, obtains global optimum；
S3.5 updates particle rapidity, formula are as follows:
Wherein Vk、Xk、Respectively correspond speed of certain particle in kth time iteration, position, optimal location；It represents
The global optimum position of population, r when kth time iterative process1,r2∈ [0,1], is uniformly distributed at random；
If V > Vmax, then V=Vmax；If V <-Vmax, then V=-Vmax, otherwise constant；
S3.6 gives random variation for non-optimal particle, updates particle position:
Calculation formula are as follows: Xk+1=Xk+Vk+1；
S3.7 judges whether that reaching maximum cycle or fitness value reaches target value, if then entering S3.8, otherwise returns
S3.2；
S3.8 exports optimal particle, and the particle for being up to requirement corresponds the initial connection weight and threshold value of BP neural network；
S3.9 inputs training data, training BP neural network；
S3.10, input test sample；
S3.11 exports prediction result, and is compared with actual value.
2. according to the method described in claim 1, it is characterized by: the feature of screening includes opening price, closing quotation in step S1
Valence, highest price, lowest price and exchange hand；The data prediction specifically: stock certificate data is normalized, is normalized
Formula are as follows:
Wherein, certain time series x={ x1,x2,...,xn, xnormIt is the value after normalization, xminIt is minimum value in sequence, xmaxIt is
Maximum value in sequence.
3. according to the method described in claim 1, it is characterized by: in step S2,
Input layer includes opening price, closing price, highest price, lowest price and the exchange hand of stock, the ups and downs that output layer is next day
Situation；
Node in hidden layer is by formulaIt determines,
Wherein, q is node in hidden layer；N is input layer number；M is output layer number of nodes；A is the constant between 0~10；
The activation primitive uses
Total number of particles S=n × q+q × the m+q+m.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910154140.7A CN109919374A (en) | 2019-03-01 | 2019-03-01 | Prediction of Stock Price method based on APSO-BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910154140.7A CN109919374A (en) | 2019-03-01 | 2019-03-01 | Prediction of Stock Price method based on APSO-BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109919374A true CN109919374A (en) | 2019-06-21 |
Family
ID=66962923
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910154140.7A Pending CN109919374A (en) | 2019-03-01 | 2019-03-01 | Prediction of Stock Price method based on APSO-BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919374A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111680107A (en) * | 2020-08-11 | 2020-09-18 | 南昌木本医疗科技有限公司 | Financial prediction system based on artificial intelligence and block chain |
| CN112580855A (en) * | 2020-11-27 | 2021-03-30 | 国网上海市电力公司 | Cable group steady-state temperature rise prediction method based on self-adaptive variation PSO-BP neural network |
| CN117574213A (en) * | 2024-01-15 | 2024-02-20 | 南京邮电大学 | APSO-CNN-based network traffic classification method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101276454A (en) * | 2007-12-05 | 2008-10-01 | 中原工学院 | Method for model building, forecasting and decision-making of stock market based on BP neural net |
| CN105913151A (en) * | 2016-04-12 | 2016-08-31 | 河海大学常州校区 | Photovoltaic power station power generation amount predication method based on adaptive mutation particle swarm and BP network |
| CN106600070A (en) * | 2016-12-20 | 2017-04-26 | 郭建峰 | Short-period share price prediction algorithm based on IPSO-BP neural network |
| CN106779145A (en) * | 2016-11-18 | 2017-05-31 | 北京信息科技大学 | A kind of stock trend forecasting method based on Artificial neural network ensemble |
-
2019
- 2019-03-01 CN CN201910154140.7A patent/CN109919374A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101276454A (en) * | 2007-12-05 | 2008-10-01 | 中原工学院 | Method for model building, forecasting and decision-making of stock market based on BP neural net |
| CN105913151A (en) * | 2016-04-12 | 2016-08-31 | 河海大学常州校区 | Photovoltaic power station power generation amount predication method based on adaptive mutation particle swarm and BP network |
| CN106779145A (en) * | 2016-11-18 | 2017-05-31 | 北京信息科技大学 | A kind of stock trend forecasting method based on Artificial neural network ensemble |
| CN106600070A (en) * | 2016-12-20 | 2017-04-26 | 郭建峰 | Short-period share price prediction algorithm based on IPSO-BP neural network |
Non-Patent Citations (4)
| Title |
|---|
| 周利军等: "自适应变异粒子群算法", 《计算机工程与应用》 * |
| 李洪英: "BP神经网络在股市预测模型中的应用——以上证股票价格收盘指数为例", 《中国证券期货》 * |
| 汪煜纯: "神经网络在股票预测中的应用", 《通讯世界》 * |
| 陈周林等: "改进PSO-BP网络预测模型在造纸能耗预测中的应用", 《轻工科技》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111680107A (en) * | 2020-08-11 | 2020-09-18 | 南昌木本医疗科技有限公司 | Financial prediction system based on artificial intelligence and block chain |
| CN111680107B (en) * | 2020-08-11 | 2020-12-08 | 上海竞动科技有限公司 | Financial prediction system based on artificial intelligence and block chain |
| CN112580855A (en) * | 2020-11-27 | 2021-03-30 | 国网上海市电力公司 | Cable group steady-state temperature rise prediction method based on self-adaptive variation PSO-BP neural network |
| CN117574213A (en) * | 2024-01-15 | 2024-02-20 | 南京邮电大学 | APSO-CNN-based network traffic classification method |
| CN117574213B (en) * | 2024-01-15 | 2024-03-29 | 南京邮电大学 | APSO-CNN-based network traffic classification method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102521656B (en) | Integrated transfer learning method for classification of unbalance samples | |
| CN110135630A (en) | The short term needing forecasting method with multi-step optimization is returned based on random forest | |
| CN110084424A (en) | A kind of Methods of electric load forecasting based on LSTM and LGBM | |
| CN109214449A (en) | A kind of electric grid investment needing forecasting method | |
| CN109118013A (en) | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based | |
| CN109063911A (en) | A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks | |
| CN111340273A (en) | Short-term load prediction method for power system based on GEP parameter optimization XGboost | |
| CN110059852A (en) | A kind of stock yield prediction technique based on improvement random forests algorithm | |
| CN110348075A (en) | A kind of grinding surface roughness prediction technique based on improvement algorithm of support vector machine | |
| CN109426889A (en) | Short-term load forecasting method based on KPCA in conjunction with improvement neural network | |
| CN109919374A (en) | Prediction of Stock Price method based on APSO-BP neural network | |
| CN109461025A (en) | A kind of electric energy substitution potential customers' prediction technique based on machine learning | |
| CN108171379A (en) | A kind of electro-load forecast method | |
| CN110443417A (en) | Multiple-model integration load forecasting method based on wavelet transformation | |
| CN113762387B (en) | Multi-element load prediction method for data center station based on hybrid model prediction | |
| CN109086900B (en) | Electric power material guarantee and allocation platform based on multi-target particle swarm optimization algorithm | |
| CN110348608A (en) | A kind of prediction technique for improving LSTM based on fuzzy clustering algorithm | |
| CN109583635A (en) | A kind of short-term load forecasting modeling method towards operational reliability | |
| CN109543741A (en) | A kind of FCM algorithm optimization method based on improvement artificial bee colony | |
| CN112524751B (en) | Dynamic air conditioning system energy consumption prediction model construction and prediction method and device | |
| CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
| CN110895772A (en) | Electricity sales amount prediction method based on combination of grey correlation analysis and SA-PSO-Elman algorithm | |
| CN111832839B (en) | Energy consumption prediction method based on sufficient incremental learning | |
| CN111008790A (en) | Hydropower station group power generation electric scheduling rule extraction method | |
| CN108268979A (en) | A kind of cut deal qualitative forecasting method based on evolution fuzzy association rules |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190621 |