CN112348275A - Regional ecological environment change prediction method based on online incremental learning - Google Patents

Regional ecological environment change prediction method based on online incremental learning Download PDF

Info

Publication number
CN112348275A
CN112348275A CN202011290431.8A CN202011290431A CN112348275A CN 112348275 A CN112348275 A CN 112348275A CN 202011290431 A CN202011290431 A CN 202011290431A CN 112348275 A CN112348275 A CN 112348275A
Authority
CN
China
Prior art keywords
data
incremental
time
ecological environment
learning machine
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.)
Withdrawn
Application number
CN202011290431.8A
Other languages
Chinese (zh)
Inventor
徐鹤
张澳生
苗冬冬
季一木
王汝传
李鹏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202011290431.8A priority Critical patent/CN112348275A/en
Publication of CN112348275A publication Critical patent/CN112348275A/en
Withdrawn legal-status Critical Current

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
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Primary Health Care (AREA)
  • Educational Administration (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Physiology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to the field of ecological environment prediction models, and discloses a regional ecological environment change prediction method based on online incremental learning. The invention combines a convolutional neural network, a long-short term memory neural network and a fully-connected neural network with an increment limit learning machine based on an adaptive genetic algorithm. A traditional incremental extreme learning machine is optimized, and the number of hidden layer optimal nodes of a model is obtained by using an adaptive genetic algorithm. And the change of the regional ecological environment is predicted by using an incremental learning method, the method can be used for carrying out on-line training on the model according to the change of new environment data, and the method can also be used for greatly reducing the cost of the training model and improving the prediction precision. The regional ecological environment change prediction method based on online incremental learning can realize accurate prediction of regional ecological environment change.

Description

Regional ecological environment change prediction method based on online incremental learning
Technical Field
The invention relates to the field of ecological environment prediction models, in particular to a regional ecological environment change prediction method based on online incremental learning.
Background
With the development of economy and science, environmental issues are becoming increasingly the topic of interest and enthusiasm. Because environmental issues directly concern our physical and mental health and sustainable development of green ecology, especially the complicated problem of regional ecological environment change, government departments and experts in the environment are engaged in research and analysis. However, the regional ecological environment change can be influenced by various environmental factors and the mutual influence between adjacent regions, so the invention provides a regional ecological environment change prediction method based on online incremental learning.
The traditional prediction method for studying regional ecological environment change can be roughly divided into two methods, namely statistics and classical machine learning. However, because the data volume of the existing ecological environment, the data volume of the existing ecological environment and the large data volume and the related influence factors are very various, and the traditional method for predicting the regional ecological environment change consumes a great deal of time and financial resources, the invention provides a method for adopting incremental learning to solve the problem.
Disclosure of Invention
In order to solve the problems, the invention provides a regional ecological environment change prediction method based on online incremental learning, and compared with the traditional statistical method and machine learning method, the method can better improve the prediction accuracy and reduce the cost spent in model training.
The invention is realized by the following technical scheme: a regional ecological environment change prediction method based on online incremental learning comprises the following steps:
step 1, preprocessing data;
step 2: establishing a convolutional neural network layer;
and step 3: adding a time window and establishing an incremental limit learning machine model based on an adaptive genetic algorithm;
and 4, step 4: and establishing a long-term and short-term memory neural network layer and a full-connection neural network layer.
Further, the step 1 comprises the following specific steps:
step 1-1, data source: acquiring a time sequence composed of characteristics which may influence the change of the ecological environment of a certain area through a plurality of monitoring points arranged in the area;
step 1-2, discrete characteristic numeralization coding: processing the collected character features in the time sequence in a one-hot coding mode to obtain digital information;
step 1-3, missing value processing: filling missing data of not more than two time intervals by using the previous data, and filling missing data of more than two time intervals by adopting a linear interpolation method;
step 1-4, abnormal value processing: for the time sequence with the characteristic value of a few time periods obviously higher than the characteristic values of the preceding and following time periods, performing first-order difference operation on the characteristic value of the data of the preceding time interval and the characteristic value of the data of the preceding time interval; setting a threshold value to be 0.1, if the threshold value does not exceed 0.1, determining the data to be reasonable mutation data, otherwise, determining the data to be abnormal data; the abnormal data is processed by regarding the data as missing values, namely, the step 1-3 is operated;
step 1-5, standardization treatment: because the normalization processing is easily influenced by extreme values and has poor robustness, the data is normalized;
1-6, dividing a data set: the processed data is divided into 80% of training set according to the length of the time sequence, and the rest 20% of training set is testing set.
Further, the step 2 comprises the following specific steps:
2-1, establishing a convolutional neural network, and adding a convolutional layer and a pooling layer;
2-2, converting the data obtained through preprocessing into an n-dimensional matrix with a time sequence, wherein n-dimension refers to the characteristic quantity of the data;
and 2-3, inputting the n-dimensional matrix into a CNN structure for training and extracting important features.
Further, the step 3 comprises the following specific steps:
step 3-1, setting a proper time window size k;
step 3-2, the time sequence with the extracted important features enters an incremental extreme learning machine in batches through a time window;
3-3, setting a time input sequence as X;
3-4, establishing an expected threshold value as epsilon in the incremental extreme learning machine;
3-5, setting the number of initial nodes of a hidden layer in the incremental extreme learning machine to be 1;
3-6, setting the maximum value of the hidden layer node as M in the incremental extreme learning machine;
3-7, establishing a self-adaptive genetic algorithm model, which is specifically divided into a fitness calculation part, a selection operator part, a self-adaptive crossover operator part and a self-adaptive mutation operator part;
step 3-8, in the fitness calculation, the fitness of the ith sample F (i) is 1/E (i), and E (i) is an error function of the ith sample;
in the step 3-9 and the adaptive cross factor calculation, the calculation of the adaptive cross probability P (c) of the ith sample is divided into two cases:
(1) when f (i) < f (mean), f (mean) is the average fitness value of all samples;
at this time, p (c) ═ p (c) max, and p (c) max is the maximum cross probability of all samples;
(2) when F (i) > F (mean), P (c) min is the minimum cross probability in all samples;
at this time, P (c) ═ P (c) max- [ (P (c) max-P (c) min)
Enpochs (max) ]. Enpochs (i), Enpochs (max) and Enpochs (i)
Respectively the maximum iteration times of the samples and the iteration times of the ith sample;
3-10, in the calculation of the adaptive mutation factor, the calculation of the adaptive mutation probability P (m) of the ith sample is divided into two cases:
(1) when F (i) < F (mean), P (m) min is the minimum variation probability of all samples;
at this time, p (m) ═ p (m) min;
(2) when F (i) > F (mean), P (m) max is the maximum variation probability of all samples;
at this time, P (m) ═ P (m) min + [ (P (m) max-P (m) min)
/Enpochs(max)]*Enpochs(i)；
3-11, establishing a self-adaptive genetic algorithm model through the steps, and then obtaining the optimal hidden layer node number of the model through the model;
3-12, randomly assigning a weight W and a bias b to a hidden layer of the incremental extreme learning machine at the moment;
3-13, obtaining an input vector X1 ═ W × X + b of the activation function from the hidden layer of the incremental limit learning machine;
3-14, using an activation function sigmoid in the incremental limit learning machine, namely g (X) is 1/(1+ e ^ -X), namely the output obtained by the activation function is g (X1) and is recorded as H;
3-15, at the moment, the output weight W2 of the hidden layer node of the incremental limit learning machine is (E ^ H ^ T)/(H ^ T), and the initial value of E is the time sequence of the first batch of input incremental limit learning machines;
3-16, setting a residual error function E-W2H in the incremental extreme learning machine;
and 3-17, obtaining a final incremental limit learning machine model based on the adaptive genetic algorithm.
Further, the step 4 comprises the following specific steps:
step 4-1, establishing an LSTM deep learning network, wherein the network model mainly comprises an input layer, a hidden layer and an output layer;
4-2, the network optimizer uses an Adam algorithm, and the activation function uses a Relu function;
4-3, after the LSTM network is established, establishing a pruning layer (Dropout) to inhibit the result of the prediction of the LSTM layer from being over-fitted;
and 4-4, finally, establishing a full-connection neural network, decoding the output result of the LSTM layer, and finally obtaining the prediction result.
Compared with the prior art, the invention has the following beneficial effects: the invention provides a regional ecological environment change prediction method based on online incremental learning, which utilizes the idea of combining incremental learning and deep learning. The method can retain part of old knowledge and learn new knowledge in the Learning process through Incremental Learning (Incremental Learning); the learning efficiency does not decrease with the increase of data; the knowledge of the old model is not used in the training process; the incremental learning method can not only reduce the cost of training a large amount of data, but also improve the accuracy of a prediction system. The prediction method in the application not only can analyze a large amount of environment data and automatically extract important features, but also utilizes the advantage of the idea of incremental learning. The method can realize the retention of partial important old data of the line environment and the learning of the newly added data, and can greatly reduce the calculation amount and the complexity of the data, thereby reducing the calculation cost. The method also uses CNN neural networks, which are models that excel in extracting features, and therefore is used here to extract important features of data. In addition, the LSTM deep learning network is added in the method, and the special gate structure of the network model is particularly suitable for the time series prediction problem and can solve the defects that the gradient disappearance and the gradient explosion of the circulating neural network can occur. Therefore, compared with the traditional deep learning model, the accuracy of the operation can be improved.
Drawings
FIG. 1 is an overall flow framework diagram of the present invention;
FIG. 2 is a flow diagram of data preprocessing of the present invention;
FIG. 3 is a schematic diagram of the convolutional neural network structure of the present invention;
FIG. 4 is a schematic structural diagram of an incremental limit learning machine model based on an adaptive genetic algorithm according to the present invention;
FIG. 5 is a schematic diagram of a long term memory network according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Moreover, the technical solutions in the embodiments of the present invention may be combined with each other, but it is necessary to be able to be realized by a person skilled in the art, and when the technical solutions are contradictory or cannot be realized, the combination of the technical solutions should be considered to be absent, and is not within the protection scope of the present invention.
The invention provides a regional ecological environment change prediction method based on online Incremental Learning, which comprises the specific steps of preprocessing data, constructing a feature extraction layer by using a Convolutional Neural Network (CNN), putting the processed data into the CNN for training, putting a time sequence containing important features obtained by training into an Incremental Extreme Learning Machine (IELM) optimized based on an adaptive genetic algorithm in batches according to the size of a set time window for training and Learning, putting an output result into a Long Short-Term Memory Network (LSTM) for training, and decoding the obtained result by a Full Connected Neural Network (FCNN) to obtain a final prediction result.
A regional ecological environment change prediction method based on online incremental learning comprises the following steps:
step 1, preprocessing data;
step 2: establishing a convolutional neural network layer;
and step 3: adding a time window and establishing an incremental limit learning machine model based on an adaptive genetic algorithm;
and 4, step 4: and establishing a long-term and short-term memory neural network layer and a full-connection neural network layer.
Further, the step 1 comprises the following specific steps:
step 1-1, data source: acquiring a time sequence composed of characteristics which may influence the change of the ecological environment of a certain area through a plurality of monitoring points arranged in the area;
step 1-2, discrete characteristic numeralization coding: processing the collected character features in the time sequence in a one-hot coding mode to obtain digital information;
step 1-3, missing value processing: filling missing data of not more than two time intervals by using the previous data, and filling missing data of more than two time intervals by adopting a linear interpolation method;
step 1-4, abnormal value processing: for the time sequence with the characteristic value of a few time periods obviously higher than the characteristic values of the preceding and following time periods, performing first-order difference operation on the characteristic value of the data of the preceding time interval and the characteristic value of the data of the preceding time interval; setting a threshold value to be 0.1, if the threshold value does not exceed 0.1, determining the data to be reasonable mutation data, otherwise, determining the data to be abnormal data; the abnormal data is processed by regarding the data as missing values, namely, the operation of the step 1-3 is carried out;
step 1-5, standardization treatment: because the normalization processing is easily influenced by extreme values and has poor robustness, the data is normalized;
1-6, dividing a data set: the processed data is divided into 80% of training set according to the length of the time sequence, and the rest 20% of training set is testing set.
The step 2 comprises the following specific steps:
2-1, establishing a convolutional neural network, and adding a convolutional layer and a pooling layer;
2-2, converting the data obtained through preprocessing into an n-dimensional matrix with a time sequence, wherein n-dimension refers to the characteristic quantity of the data;
and 2-3, inputting the n-dimensional matrix into a CNN structure for training and extracting important features.
The step 3 comprises the following specific steps:
step 3-1, setting a proper time window size k;
step 3-2, the time sequence with the extracted important features enters an incremental extreme learning machine in batches through a time window;
3-3, setting a time input sequence as X;
3-4, establishing an expected threshold value as epsilon in the incremental extreme learning machine;
3-5, setting the number of initial nodes of a hidden layer in the incremental extreme learning machine to be 1;
3-6, setting the maximum value of the hidden layer node as M in the incremental extreme learning machine;
3-7, establishing a self-adaptive genetic algorithm model, which is specifically divided into a fitness calculation part, a selection operator part, a self-adaptive crossover operator part and a self-adaptive mutation operator part;
step 3-8, in the fitness calculation, the fitness of the ith sample F (i) is 1/E (i), and E (i) is an error function of the ith sample;
in the step 3-9 and the adaptive cross factor calculation, the calculation of the adaptive cross probability P (c) of the ith sample is divided into two cases:
(1) when f (i) < f (mean), f (mean) is the average fitness value of all samples;
at this time, p (c) ═ p (c) max, and p (c) max is the maximum cross probability of all samples;
(2) when F (i) > F (mean), P (c) min is the minimum cross probability in all samples;
at this time, P (c) ═ P (c) max- [ (P (c) max-P (c) min)
Enpochs (max) } enpochs (i), enpochs (max) and enpochs (i) are the maximum number of iterations of the sample and the number of iterations up to the ith sample, respectively;
3-10, in the calculation of the adaptive mutation factor, the calculation of the adaptive mutation probability P (m) of the ith sample is divided into two cases:
(1) when F (i) < F (mean), P (m) min is the minimum variation probability of all samples;
at this time, p (m) ═ p (m) min;
(2) when F (i) > F (mean), P (m) max is the maximum variation probability of all samples;
at this time, P (m) ═ P (m) min + [ (P (m) max-P (m) min)
/Enpochs(max)]*Enpochs(i)；
3-11, establishing an adaptive genetic algorithm model through the steps, and obtaining the adaptive genetic algorithm model through the model
The number of nodes of the optimal hidden layer of the model;
3-12, randomly assigning a weight W and a bias b to a hidden layer of the incremental extreme learning machine at the moment;
3-13, obtaining an input vector X1 ═ W × X + b of the activation function from the hidden layer of the incremental limit learning machine;
3-14, using an activation function sigmoid in the incremental limit learning machine, namely g (X) is 1/(1+ e ^ -X), namely the output obtained by the activation function is g (X1) and is recorded as H;
3-15, wherein the output weight W2 of the hidden layer node of the incremental limit learning machine is (E x H ^ T)/(H x H ^ T), and the initial value of E is the time sequence of entering the IELM of the first batch;
3-16, setting a residual error function E-W2H in the incremental extreme learning machine;
and 3-17, obtaining a final incremental limit learning machine model based on the adaptive genetic algorithm.
The step 4 comprises the following specific steps:
step 4-1, establishing an LSTM deep learning network, wherein the network model mainly comprises an input layer, a hidden layer and an output layer;
4-2, the network optimizer uses an Adam algorithm, and the activation function uses a Relu function;
4-3, after the LSTM network is established, establishing a pruning layer (Dropout) to inhibit the result of the prediction of the LSTM layer from being over-fitted;
and 4-4, finally, establishing a full-connection neural network, decoding the output result of the LSTM layer, and finally obtaining the prediction result.
In the step 3 of the invention, an incremental limit learning machine model based on an adaptive genetic algorithm is used, and the improved method can quickly find the number of nodes of the optimal hidden layer of the model and also has the advantage of incremental learning. The method has the advantages that the characteristics of reserving partial important old data and learning newly added data are realized, and compared with the traditional increment limit learning machine, the calculation cost is reduced, and the over-fitting phenomenon is also inhibited.
The invention uses the idea of combining various deep learning algorithms and incremental learning algorithms to build a prediction system. The method comprises the steps of extracting features by using a convolutional neural network, enabling a time sequence with the features to enter an incremental model in batches by using a time window, and training and learning the time sequence by using a long-short term memory network and a fully-connected neural network. By the method, the change of the regional ecological environment can be accurately predicted.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims (5)

1.一种基于在线增量学习的区域生态环境变化预测方法，其特征在于：所述方法包括如下步骤：1. a regional ecological environment change prediction method based on online incremental learning, is characterized in that: described method comprises the steps: 步骤1:对数据进行预处理；Step 1: preprocess the data; 步骤2：建立卷积神经网络层；Step 2: Build a convolutional neural network layer; 步骤3：添加时间窗口和建立基于自适应遗传算法的增量极限学习机模型；Step 3: Add time window and build an incremental extreme learning machine model based on adaptive genetic algorithm; 步骤4：建立长短期记忆神经网络层和全连接神经网络层。Step 4: Build a long short-term memory neural network layer and a fully connected neural network layer. 2.根据权利要求1所述的一种基于在线增量学习的区域生态环境变化预测方法，其特征在于：所述步骤1中包括如下具体步骤：2. a kind of regional ecological environment change prediction method based on online incremental learning according to claim 1, is characterized in that: described step 1 comprises following concrete steps: 步骤1-1、数据来源：通过在某个区域设立的多个监测点获取到可能会影响到该区域生态环境变化的特征所组成的时间序列；Step 1-1. Data source: obtain a time series composed of features that may affect the changes of the ecological environment in a certain area through multiple monitoring points established in a certain area; 步骤1-2、离散特征数值化编码：把收集到的时间序列中的字符特征通过独热编码方式处理变成数字信息；Step 1-2, numerical encoding of discrete features: the character features in the collected time series are processed into digital information by one-hot encoding; 步骤1-3、缺失值处理：对于不超过两个时间间隔的缺失数据使用前面的数据进行填充，对于超过两个时间间隔的缺失数据采用线性插值法进行填充；Step 1-3. Missing value processing: Use the previous data to fill in the missing data of no more than two time intervals, and use linear interpolation to fill the missing data of more than two time intervals; 步骤1-4、异常值处理：对于少数时间段的特征值明显高于前后时间段的特征值的时间序列，让其与前一时间间隔的数据的特征值进行一阶差分运算；设立阈值为0.1，如果未超过0.1则认为为合理的突变数据，否则则认为为异常数据；异常数据的处理方式为把这些数据当作缺失值处理，即进行步骤1-3的操作；Step 1-4, outlier processing: For a time series whose eigenvalues in a few time periods are significantly higher than those in the previous and previous time periods, perform first-order difference operation with the eigenvalues of the data in the previous time interval; 0.1, if it does not exceed 0.1, it is considered to be reasonable mutation data, otherwise it is considered to be abnormal data; the processing method of abnormal data is to treat these data as missing values, that is, perform steps 1-3; 步骤1-5、标准化处理：由于使用归一化处理会容易受到极值的影响，鲁棒性比较差，因此这里把数据进行标准化处理；Step 1-5, normalization processing: Since normalization processing is easily affected by extreme values, the robustness is relatively poor, so the data is standardized here; 步骤1-6、数据集的划分：把处理后的数据按照时间序列的长度划分为80％为训练集，剩下的20％为测试集。Steps 1-6, data set division: Divide the processed data into 80% as the training set according to the length of the time series, and the remaining 20% as the test set. 3.根据权利要求1所述的一种基于在线增量学习的区域生态环境变化预测方法，其特征在于：所述步骤2中包括如下具体步骤：3. a kind of regional ecological environment change prediction method based on online incremental learning according to claim 1, is characterized in that: in described step 2, comprises following concrete steps: 步骤2-1、建立卷积神经网络，并且添加一层卷积层和一层池化层；Step 2-1, build a convolutional neural network, and add a convolutional layer and a pooling layer; 步骤2-2、把经过预处理得到的数据转化为具有时间序列的n维矩阵，这里的n维指的是数据具有的特征数量；Step 2-2: Convert the preprocessed data into an n-dimensional matrix with time series, where n-dimensional refers to the number of features the data has; 步骤2-3、再把这n维矩阵输入到CNN结构中进行训练提取重要特征。Step 2-3, then input the n-dimensional matrix into the CNN structure for training to extract important features. 4.根据权利要求1所述的一种基于在线增量学习的区域生态环境变化预测方法，其特征在于：所述步骤3中包括如下具体步骤：4. a kind of regional ecological environment change prediction method based on online incremental learning according to claim 1, is characterized in that: described step 3 comprises following concrete steps: 步骤3-1、设立合适的时间窗口大小k；Step 3-1. Establish a suitable time window size k; 步骤3-2、把提取到重要特征的时间序列通过时间窗口分批进入增量极限学习机中；Step 3-2. Enter the time series extracted into the important features into the incremental extreme learning machine in batches through the time window; 步骤3-3、设时间输入序列为X；Step 3-3, set the time input sequence as X; 步骤3-4、在增量极限学习机中设立期望阈值为ε；Step 3-4, in the incremental extreme learning machine, set the expected threshold as ε; 步骤3-5、在增量极限学习机中设立隐藏层的初始节点个数为1；Steps 3-5, the initial number of nodes in the hidden layer in the incremental extreme learning machine is 1; 步骤3-6、在增量极限学习机中设立隐藏层节点最大值为M；Steps 3-6, in the incremental extreme learning machine, set the maximum value of hidden layer nodes as M; 步骤3-7、建立自适应遗传算法模型，具体分为适应度计算、选择算子、自适应交叉算子、自适应变异算子这几个部分；Steps 3-7, establish an adaptive genetic algorithm model, which is specifically divided into several parts: fitness calculation, selection operator, adaptive crossover operator, and adaptive mutation operator; 步骤3-8、在适应度计算中，第i个样本适应度F(i)＝1/E(i)，E(i)为第i个样本的误差函数；Step 3-8, in the fitness calculation, the fitness of the ith sample is F(i)=1/E(i), and E(i) is the error function of the ith sample; 步骤3-9、自适应交叉因子计算中，第i个样本的自适应交叉概率P(c)的计算分为两种情况：Step 3-9. In the calculation of the adaptive cross factor, the calculation of the adaptive cross probability P(c) of the ith sample is divided into two cases: (1)当F(i)<F(mean),F(mean)为所有样本的平均适应度值；(1) When F(i)<F(mean), F(mean) is the average fitness value of all samples; 此时P(c)＝P(c)max,P(c)max为所有样本最大交叉概率；At this time, P(c)=P(c)max, and P(c)max is the maximum crossover probability of all samples; (2)当F(i)>F(mean)，P(c)min为所有样本中最小交叉概率；(2) When F(i)>F(mean), P(c)min is the minimum crossover probability in all samples; 此时P(c)＝P(c)max-[(P(c)max-P(c)min)/Enpochs(max)]*Enpochs(i),Enpochs(max)和Enpochs(i)分别为样本最大迭代次数和到第i个样本时的迭代次数；At this time P(c)=P(c)max-[(P(c)max-P(c)min)/Enpochs(max)]*Enpochs(i), Enpochs(max) and Enpochs(i) are respectively The maximum number of iterations of the sample and the number of iterations when the ith sample is reached; 步骤3-10、在自适应变异因子计算中，第i个样本的自适应变异概率P(m)的计算分为两种情况：Step 3-10. In the calculation of the adaptive variation factor, the calculation of the adaptive variation probability P(m) of the ith sample is divided into two cases: (1)当F(i)<F(mean)，P(m)min为所有样本最小变异概率；(1) When F(i)<F(mean), P(m)min is the minimum variation probability of all samples; 此时P(m)＝P(m)min；At this time P(m)=P(m)min; (2)当F(i)>F(mean)，P(m)max为所有样本最大变异概率；(2) When F(i)>F(mean), P(m)max is the maximum variation probability of all samples; 此时P(m)＝P(m)min+[(P(m)max-P(m)min)/Enpochs(max)]*Enpochs(i)；At this time, P(m)=P(m)min+[(P(m)max-P(m)min)/Enpochs(max)]*Enpochs(i); 步骤3-11、通过以上步骤建立自适应遗传算法模型，然后通过该模型得到模型最优隐藏层节点数；Step 3-11, establish an adaptive genetic algorithm model through the above steps, and then obtain the optimal number of hidden layer nodes of the model through the model; 步骤3-12、此时增量极限学习机隐藏层随机赋值权重为W，偏置为b；Step 3-12, at this time, the hidden layer of the incremental extreme learning machine is randomly assigned a weight of W and a bias of b; 步骤3-13、从增量极限学习机隐藏层得到的激活函数的输入向量X1＝W*X+b；Step 3-13, the input vector X1=W*X+b of the activation function obtained from the hidden layer of the incremental extreme learning machine; 步骤3-14、在增量极限学习机中使用激活函数sigmoid，即g(x)＝1/(1+e^-x)，Step 3-14, use the activation function sigmoid in the incremental extreme learning machine, that is, g(x)=1/(1+e^-x), 即经过激活函数得到的输出为g(X1)记为H；That is, the output obtained through the activation function is g(X1) and is denoted as H; 步骤3-15、此时增量极限学习机隐藏层节点的输出权重W2＝(E*H^T)/(H*H^T)，E的初始值为第一批进入增量极限学习机的时间序列；Step 3-15. At this time, the output weight of the hidden layer node of the incremental extreme learning machine is W2=(E*H^T)/(H*H^T), and the initial value of E is the first batch to enter the incremental extreme learning machine. time series; 步骤3-16、在增量极限学习机中设立残差函数为E＝E-W2*H；Step 3-16, in the incremental extreme learning machine, set up the residual function as E=E-W2*H; 步骤3-17、得到最终的基于自适应遗传算法的增量极限学习机模型。Step 3-17, obtain the final incremental extreme learning machine model based on the adaptive genetic algorithm. 5.根据权利要求1所述的一种基于在线增量学习的区域生态环境变化预测方法，其特征在于：所述步骤4中包括如下具体步骤：5. a kind of regional ecological environment change prediction method based on online incremental learning according to claim 1, is characterized in that: in described step 4, comprises following concrete steps: 步骤4-1、建立LSTM深度学习网络，该网络模型主要由输入层、隐藏层和输出层组成；Step 4-1, establish a LSTM deep learning network, the network model is mainly composed of an input layer, a hidden layer and an output layer; 步骤4-2、并且该网络优化器使用Adam算法，激活函数使用Relu函数；Step 4-2, and the network optimizer uses the Adam algorithm, and the activation function uses the Relu function; 步骤4-3、建立完LSTM网络后，再设立剪枝层(Dropout),抑制LSTM层预测的结果发生过拟合；Step 4-3. After establishing the LSTM network, set up a pruning layer (Dropout) to suppress the over-fitting of the results predicted by the LSTM layer; 步骤4-4、最后再建立全连接神经网络，对LSTM层的输出结果进行解码，最后得到我们的预测结果。Step 4-4, finally build a fully connected neural network, decode the output of the LSTM layer, and finally get our prediction result.
CN202011290431.8A 2020-11-18 2020-11-18 Regional ecological environment change prediction method based on online incremental learning Withdrawn CN112348275A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011290431.8A CN112348275A (en) 2020-11-18 2020-11-18 Regional ecological environment change prediction method based on online incremental learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011290431.8A CN112348275A (en) 2020-11-18 2020-11-18 Regional ecological environment change prediction method based on online incremental learning

Publications (1)

Publication Number Publication Date
CN112348275A true CN112348275A (en) 2021-02-09

Family

ID=74364150

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011290431.8A Withdrawn CN112348275A (en) 2020-11-18 2020-11-18 Regional ecological environment change prediction method based on online incremental learning

Country Status (1)

Country Link
CN (1) CN112348275A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113780377A (en) * 2021-08-26 2021-12-10 重庆理工大学 Rainfall level prediction method and system based on Internet of Things data online learning

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113780377A (en) * 2021-08-26 2021-12-10 重庆理工大学 Rainfall level prediction method and system based on Internet of Things data online learning
CN113780377B (en) * 2021-08-26 2024-01-19 重庆理工大学 Rainfall grade prediction method and system based on online learning of data of Internet of things

Similar Documents

Publication Publication Date Title
CN111460728B (en) A method, device, storage medium and device for predicting remaining life of industrial equipment
CN108694502B (en) Self-adaptive scheduling method for robot manufacturing unit based on XGboost algorithm
CN110083125B (en) Machine tool thermal error modeling method based on deep learning
CN112733417A (en) Abnormal load data detection and correction method and system based on model optimization
CN112910690A (en) Network traffic prediction method, device and equipment based on neural network model
CN111046961B (en) Fault classification method based on bidirectional long short-term memory unit and capsule network
CN111507046B (en) Method and system for predicting remaining service life of electric gate valve
CN111274817A (en) An intelligent software cost measurement method based on natural language processing technology
CN110222387B (en) Multi-element drilling time sequence prediction method based on mixed leaky integration CRJ network
CN110956309A (en) Flow activity prediction method based on CRF and LSTM
CN115169430A (en) Cloud network end resource multidimensional time sequence anomaly detection method based on multi-scale decoding
CN112085619A (en) A Feature Selection Method for Distribution Network Data Optimization
CN116503118A (en) A Value Evaluation System for Waste Home Appliances Based on Classification Selection Enhanced Prediction Model
CN109787821B (en) An intelligent prediction method for large-scale mobile customer traffic consumption
WO2023231374A1 (en) Semi-supervised fault detection and analysis method and apparatus for mechanical device, terminal, and medium
CN117454743A (en) Deep foundation pit deformation prediction method, deep foundation pit deformation prediction device, deep foundation pit deformation prediction equipment and storage medium
CN117709540A (en) Short-term bus load prediction method and system for identifying abnormal weather
CN116628444A (en) Water quality early warning method based on improved meta-learning
CN113139570A (en) Dam safety monitoring data completion method based on optimal hybrid valuation
CN114124554B (en) Virtual network service chain throughput prediction method
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
CN112348275A (en) Regional ecological environment change prediction method based on online incremental learning
CN110288159A (en) A kind of smart city construction Measurement Method
CN113609766A (en) Soft measurement method based on depth probability latent model
CN117009752A (en) Power consumption demand prediction method based on POA-VMD-SSA-KELM model

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
WW01 Invention patent application withdrawn after publication

Application publication date: 20210209

WW01 Invention patent application withdrawn after publication