CN107466816B - Irrigation method based on dynamic multilayer extreme learning machine - Google Patents
Irrigation method based on dynamic multilayer extreme learning machine Download PDFInfo
- Publication number
- CN107466816B CN107466816B CN201710605810.3A CN201710605810A CN107466816B CN 107466816 B CN107466816 B CN 107466816B CN 201710605810 A CN201710605810 A CN 201710605810A CN 107466816 B CN107466816 B CN 107466816B
- Authority
- CN
- China
- Prior art keywords
- learning machine
- layer
- extreme learning
- data
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01G—HORTICULTURE; CULTIVATION OF VEGETABLES, FLOWERS, RICE, FRUIT, VINES, HOPS OR SEAWEED; FORESTRY; WATERING
- A01G25/00—Watering gardens, fields, sports grounds or the like
- A01G25/16—Control of watering
- A01G25/167—Control by humidity of the soil itself or of devices simulating soil or of the atmosphere; Soil humidity sensors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
Abstract
The invention relates to an irrigation method based on a dynamic multilayer extreme learning machine, which comprises the steps of firstly, collecting a plurality of groups of training data, wherein the training data comprises soil environment data, meteorological data and crop coefficients of irrigated crops; then, each group of training data is normalized to form a training set; then training a multilayer extreme learning machine by adopting a training set to obtain a final model; and finally, acquiring test data, normalizing the test data, inputting the normalized test data into the final model to obtain the predicted irrigation water demand, and irrigating according to the predicted irrigation water demand. The invention adopts a 'simultaneous existence and difference' strategy according to the calculation precision, namely, if the calculation result of the model on the re-input data meets the precision requirement, the model is output, otherwise, incremental learning training is carried out on the basis of the existing model to obtain a dynamically adjusted model.
Description
Technical Field
The invention belongs to the field of intelligent irrigation of agricultural Internet of things, and relates to an irrigation method based on a dynamic multilayer extreme learning machine.
Background
The agricultural Internet of things is a high integration and comprehensive application of a new generation of information technology in the agricultural field, has an important leading effect on the development of agricultural informatization in China, changes the traditional agricultural production mode, and promotes the conversion of agriculture to the direction of intellectualization and refinement. A large number of sensor nodes are used for collecting real-time information of crop production environment, a monitoring system is formed by a network technology, farmers are helped to find problems in time, and the positions where the problems occur are accurately determined. The production mode originally depending on isolated machinery is changed to an intelligent production mode taking information and software as centers, so that the aim of high-efficiency agricultural production is fulfilled.
In the field of intelligent irrigation of the agricultural Internet of things, the irrigation water demand is accurately calculated, the water demand law of crops is known, and the method is a basis for formulating a scientific and reasonable irrigation system, determining the irrigation water consumption of an irrigation area and implementing fine irrigation; the method is an effective means and a basic guarantee for achieving the purposes of water saving, high yield and high efficiency and realizing the sustainable development of water resources in irrigation areas; the method is a basic basis for formulating the fields of drainage basin planning, regional water conservancy planning, water resource utilization planning, irrigation and drainage engineering planning, design, management and the like. The method for determining the irrigation method through accurate calculation has very important significance for reducing the water consumption of the crop in the growth period, improving the water utilization rate and developing water-saving agriculture.
At present, a great deal of research achievements exist in the field of agricultural irrigation methods, so that visual decision-making basis is provided for irrigation management layers and decision makers, irrigation in a proper and proper time is guided to irrigate irrigated areas, and the reasonable utilization rate of irrigation water resources of the irrigated areas is improved. The invention patent CN 201610951658.X discloses a method for calculating the water demand for irrigation of crops under future climatic conditions, which comprises the steps of constructing a response model of crop sowing date and crop growth period length to temperature according to growth period data and accumulated temperature formula of crop field test; calculating the daily water demand of the crops by utilizing a Peneman formula in combination with a single crop coefficient method and a soil water stress coefficient; and calculating the daily irrigation water requirement of the crops based on the crop irrigation system and the water balance principle. Although the processing result of the method is relatively accurate, the method needs more collected information, such as soil heat flux related to Peneman formula, and has small application range and difficult popularization. The invention patent CN 201710020805.6 provides a method and a device for calculating the amount of irrigation water for farmland, and proposes to calculate the evapotranspiration of crops irrigated in a target area according to a mechanism model and then calculate the water demand. Although the method has accurate processing results, the method also has the problems that the soil parameters such as soil moisture content are expensive to obtain, special equipment is required for monitoring, and the use cost is high. The invention patent CN 201611093504.8 provides a crop transpiration prediction method based on an improved extreme learning machine, and the method adopts a particle swarm optimization algorithm to optimize an input weight and a threshold between a network input layer and a hidden layer of the extreme learning machine, so that the calculation accuracy of the transpiration is improved, but the time consumed in an iterative optimization process is relatively longer than that of the method in the patent, and the method cannot realize rapid processing on large-scale data.
Therefore, the irrigation method has the advantages of rapid data processing, large data processing scale and capability of accurately calculating the water demand of crops in a short time under the condition of less data parameter types, and has a wide application prospect.
Disclosure of Invention
The invention aims to solve the problems of small data processing scale, low data processing speed, less types of given training data and the like in the prior art, and provides an irrigation method which has the advantages of rapid data processing and large data processing scale and can accurately calculate the water demand of crops in a short time under the condition of less types of given data parameters.
In order to achieve the purpose, the invention adopts the technical scheme that:
an irrigation method based on a dynamic multilayer extreme learning machine comprises the steps of collecting data related to known irrigation water demand, conducting normalization processing, training the multilayer extreme learning machine to obtain a final prediction model, conducting normalization processing on the data related to the irrigation water demand to be required, inputting the data into the final prediction model to obtain predicted irrigation water demand, and conducting irrigation, wherein input and output of an upper layer in the multilayer extreme learning machine are simultaneously used as input of a lower layer, a human brain re-consolidation learning mechanism is simulated, the dynamic multilayer extreme learning machine means that the number of hidden layer nodes of the multilayer extreme learning machine is dynamically updated in the training process, the dynamic updating of the number of hidden layer nodes is mainly based on the fact that a certain relation exists between prediction precision and the hidden layer nodes, and therefore dynamic adjustment can be conducted according to the prediction result of current data, and the steps are as follows:
(1) collecting a plurality of groups of training data, wherein each group of training data comprises soil environment data, meteorological data and crop coefficients of irrigated crops;
(2) carrying out normalization processing on each group of training data, wherein all groups of training data form a training set;
(3) training a multi-layer extreme learning machine by using a training set to obtain a final prediction model;
(4) collecting a plurality of groups of test data, wherein each group of test data comprises soil environment data, meteorological data and crop coefficients of irrigated crops;
(5) normalizing each group of test data, inputting the normalized test data into a final prediction model to obtain predicted irrigation water demand, and irrigating according to the predicted irrigation water demand;
the specific steps of training the multi-layer extreme learning machine by adopting the training set are as follows:
1) dividing all groups of training data into n data blocks according to the size of a sliding window in equal parts according to the acquisition time sequence and numbering the data blocks in sequence, wherein the unit of the sliding window is a group;
2) training a multi-layer extreme learning machine (DELM) by using a data block 1 to obtain a model M1, inputting the data block 1 into a first-layer extreme learning machine to obtain a first-layer prediction result in the training process, inputting the data block 1 and the first-layer prediction result into a second-layer extreme learning machine as input parameters at the same time to obtain a second-layer prediction result, inputting the data block 1, the first-layer prediction result and the second-layer prediction result into a third-layer extreme learning machine as input parameters at the same time to obtain a third-layer prediction result, and so on to obtain a model M1;
3) let j equal 2;
4) inputting the data block j into an integrated prediction model C (j-2) to obtain the predicted irrigation water demand and calculating the prediction precision piAnd a prediction result decision coefficient R2The integrated prediction model C (j-2) is a model obtained by training the data block (j-1), and the integrated prediction model C0 is a model M1;
5) judging the prediction accuracy piWhether the prediction precision is more than or equal to the set prediction precision E1 or not is judged, if yes, an identity-seeking strategy is adopted to output a model M (j-1) which is an integrated prediction model C (j-1), otherwise, an exclusive-or-storage strategy is adopted to enter the next step;
6) determining the coefficient R according to the prediction result2UpdatingNumber of implied tier nodes of a multi-tier extreme learning machine (DELM);
7) updating an output weight matrix of a multi-layer extreme learning machine (DELM) by adopting an incremental learning mechanism according to a data block j to obtain an incremental multi-layer extreme learning machine (IDELM), and when data is newly added, not reconstructing all knowledge bases, but only updating caused by newly added data on the basis of an original knowledge base, thereby conforming to the thinking principle of human brain;
8) training an incremental multi-layer extreme learning machine (IDELM) by using a data block j to obtain a model Mj, wherein the model Mj is an integrated prediction model C (j-1);
9) let j equal j + 1;
10) and (4) circulating the steps from 4) to 8) to j ═ n, and obtaining the integrated prediction model C (n-1), namely the final prediction model.
As a preferred technical scheme:
the irrigation method based on the dynamic multilayer extreme learning machine comprises the following specific steps: the unit of the temperature is DEG C, the unit of the humidity is%, and the unit of the daily average solar total radiation is MJm-2day-1The unit of the wind speed is m.s-1The unit of the pressure is KPa; the crop coefficient refers to an empirical value of each growth period given by experts in different growth periods according to the type of the irrigated crop.
The dynamic multi-layer extreme learning machine-based irrigation method is characterized in that the normalization processing in the step (2) and the step (5) is to normalize all collected data to [ -1, 1], and the normalization formula is as follows:
in the formula, X is normalized data, X is data needing to be normalized at present, and XminFor the minimum in the data that needs to be normalized, xmaxFor data to be normalizedIs measured.
The irrigation method based on the dynamic multi-layer extreme learning machine comprises 4000-5000 groups of training data in the training set, and the size of the sliding window is 500 groups.
The irrigation method based on the dynamic multi-layer extreme learning machine is characterized in that the multi-layer extreme learning machine is a double-layer extreme learning machine.
The irrigation method based on the dynamic multi-layer extreme learning machine is characterized in that the multi-layer extreme learning machine is constructed by the following steps:
(1) initializing input parameters of a multi-layer extreme learning machine network, and selecting an activation function g (x) of a hidden layer; the input layer of the multilayer extreme learning machine network comprises soil environment data, meteorological data and crop coefficients, and the output layer comprises irrigation water demand; the input parameters comprise an input layer number, a hidden layer node number and an output layer number, wherein the input layer number is the number of input data types, the output layer number is 1, and an initialization formula of the hidden layer node number is as follows:
in the formula (I), the compound is shown in the specification,
for the number of hidden layer nodes of the k-th layer limit learning machine,
for the number of input level nodes of the k-th level limit learning machine,
the number of nodes of an output layer of the k-th layer limit learning machine is counted;
the activation function g (x) is used to calculate the output weights between the hidden layer and the output layer, and the activation function g (x) here takes the equation of the sigmod function as follows:
wherein x is an independent variable, and in the extreme learning machine network, x is specifically Wv·Xs+bv，WvAs a connection weight between the input layer and the hidden layer, bvIs a threshold value, and is,
Xsis the input vector, e is a natural constant,
number of nodes of hidden layer;
(2) respectively and randomly initializing the connection weight W between the input layer and the hidden layer of the network of the multilayer extreme learning machine according to the number of nodes of the input layer and the number of nodes of the hidden layervAnd a threshold value bvThe initialization range is [ -0.5, 0.5 [)]；
(3) According to the connection weight W between the input layer and the hidden layervThreshold bvAnd an activation function g (x) for calculating a hidden layer output matrix H, specifically: assuming a given N sets of samples { (X) in an extreme learning machine networks，ts) N, assuming that the number of hidden layer nodes of the basic Extreme Learning Machine (ELM) is 1
The extreme learning machine network structure is as follows:
in the formula, Xs＝[Xs1，Xs2，...，Xsn]T∈Rn，ts＝[ts1，ts2，...，tsm]T∈Rm，XsAnd tsRespectively representing input variables and corresponding output variables, Wv＝[Wv1,Wv2,...,Wvn]TIs an input weight vector connecting the nodes of the v-th hidden layer with the n input layers, bvIs the threshold of the v-th hidden layerThe vector of the vector is then calculated,
βv＝[βv1,βv2,...,βvm]Tis a weight vector, W, connecting the nodes of the v-th hidden layer with the m output layersv·XsExpressing and solving an inner product;
the above equation transforms to obtain:
Hβ＝T；
in the formula, H is a hidden layer output matrix, β is an output weight matrix, T is an expected output matrix, and elements in T are the crop water demand calculated by multiplying the FAO-56PM equation by the crop coefficient, where:
w is the determined value because the input weight and the threshold between the input layer and the hidden layer are determined valuesvAnd bvDetermining the value, and calculating a hidden layer output matrix H;
(4) and calculating an output weight matrix β of the multi-layer extreme learning machine according to the hidden layer output matrix H, wherein the formula is as follows:
in the formula (I), the compound is shown in the specification,
the Moore-Penrose generalized inverse of the matrix H is expressed, lambda is a regularization coefficient, I is an identity matrix, T is an expected output matrix, and elements in T are the crop water demand calculated by multiplying the FAO-56PM equation by the crop coefficient.
The irrigation method based on the dynamic multi-layer extreme learning machine has the prediction precision piThe calculation formula of (a) is as follows:
in the formula, ETPM56(i) The unit is mm-day obtained by multiplying the calculated value of the water demand of the group i of the crops by the corresponding crop coefficient through an FAO-56PM equation-1，ETpredicted(i) Predicted irrigation water demand in mm day for group i data crops-1N is the number of groups of input samples;
the value range of the set prediction precision E1 is more than 0.90, and the prediction precision is determined according to the prediction precision requirement of the actual situation;
the determination coefficient R2The calculation formula of (a) is as follows:
in the formula, meanETpredictedCalculating the average of the predicted values of the N groups of data;
the ETPM56(i) The calculation formula of (a) is as follows:
in the formula, delta is the slope of the function relation curve of saturated water vapor pressure-temperature, and the unit is KPa DEG C-1，RnFor net radiation input into the canopy, the unit is MJm-2day-1G is the soil heat flux, here neglected, in units of MJm-2day-1，esIs saturated water vapor pressure with KPa, eaThe actual water vapor pressure is expressed in KPa, and gamma represents thermometer constant expressed in KPa-deg.C-1Q is the daily average temperature in degrees Celsius and u is the wind speed at 2m height in m.s-1Crop coefficient, Crop coefficient; the calculation formula for Δ is as follows:
the irrigation method based on the dynamic multi-layer extreme learning machine is characterized in that the coefficient R is determined according to the prediction result2Updating multi-level limit learningThe number of hidden layer nodes of a machine (DELM) is determined by determining a coefficient R from the prediction result2Defining an updating mechanism, wherein a specific formula is as follows:
in the formula, c is a positive integer and is 5; ε is a constant close to 0, 0.01 is taken to prevent the divisor from being 0; r2To determine the coefficients, the coefficients R are determined2The closer to 1, the higher the prediction accuracy.
The irrigation method based on the dynamic multi-layer extreme learning machine is characterized in that the incremental multi-layer extreme learning machine learns new knowledge from new data according to the model M (j-1) and the data block j, the data which is processed before does not need to be processed repeatedly, and compared with the most basic extreme learning machine, the output weight matrix of the multi-layer extreme learning machine needs to be updated according to the new data block every time the new data block is added:
according to the basic extreme learning machine, the 1 st data block output weight matrix calculation formula is as follows:
the output weight matrix calculation formula of the 2 nd data block of the incremental learning mechanism is as follows:
in the formula (I), the compound is shown in the specification,
by analogy, the output weight matrix β of the incremental multi-layer extreme learning machine when the jth data block is input can be obtainedjThe following were used:
wherein j is not less than 2, HjHidden layer output matrix of incremental multi-layer extreme learning machine for j-th data block input, TjDesired output matrix of incremental multi-level extreme learning machine for j-th data block input, β1Output weight matrix of the multi-level extreme learning machine for the input of the 1 st data block, H1Hidden layer output matrix of multi-layer extreme learning machine for input of 1 st data block, T1The expected output matrix of the multi-layer extreme learning machine is input for the 1 st data block.
Has the advantages that:
(1) according to the irrigation method based on the dynamic multilayer extreme learning machine, the intelligent dynamic multilayer extreme learning machine algorithm is adopted to calculate the irrigation water demand, the multilayer extreme learning machine algorithm is provided, and the algorithm has better generalization performance and data processing stability in terms of large-scale data processing effect than the standard extreme learning machine algorithm;
(2) according to the irrigation method based on the dynamic multilayer extreme learning machine, the number of hidden layer nodes is dynamically adjusted according to the calculation processing result, the learning speed is high, the time loss is low, and the calculation accuracy of irrigation water demand is high by improving the algorithm of the extreme learning machine;
(3) the irrigation method based on the dynamic multilayer extreme learning machine divides the acquired data into data blocks, and adopts a 'simultaneous existence and difference' strategy according to the calculation precision, namely, if the calculation result of the model for inputting data again meets the precision requirement, the model is directly output, otherwise, incremental learning training is carried out on the basis of the existing model to obtain a dynamically adjusted model, so that the purpose of learning new knowledge from the new data can be realized, and the data processed before does not need repeated training.
Drawings
FIG. 1 is a block diagram of the steps of a dynamic multi-tier extreme learning machine (DELM) based irrigation method of the present invention;
FIG. 2 is a block diagram of the construction steps of the multi-tiered extreme learning machine of the present invention;
FIG. 3 is a schematic diagram of a network architecture for predicting crop irrigation water demand in accordance with the present invention;
FIG. 4 is a schematic diagram of a sigmoid activation function of the present invention;
FIG. 5 is a graph comparing actual crop water demand values, basic Extreme Learning Machine (ELM) predicted crop water demand, and DELM predicted crop water demand;
FIG. 6 is a graph of the absolute error of the water demand for the crops predicted by ELM and DELM of the present invention.
Detailed Description
The invention will be further illustrated with reference to specific embodiments. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Further, it should be understood that various changes or modifications of the present invention may be made by those skilled in the art after reading the teaching of the present invention, and such equivalents may fall within the scope of the present invention as defined in the appended claims.
Example 1
An irrigation method based on a dynamic double-layer extreme learning machine is shown in fig. 1, and comprises the steps of collecting data related to known irrigation water demand, carrying out normalization processing on the data, training the double-layer extreme learning machine to obtain a final prediction model, carrying out normalization processing on the data related to the irrigation water demand to be required, inputting the data into the final prediction model to obtain predicted irrigation water demand, and carrying out irrigation, wherein the specific steps are as follows:
(1) the method comprises the steps of collecting soil environment data and meteorological data of a farmland, wherein the collected soil environment data and meteorological data come from agricultural Internet of things equipment researched and developed by the information center of the agricultural Committee of Shanghai city, and the agricultural Internet of things equipment comprises daily average temperature and humidity of soil, daily average temperature and humidity of air, daily average total solar radiation, wind speed at a height of 2m, atmospheric pressure and corresponding crop coefficients, wherein the unit of temperature is ℃, the unit of humidity is, and the unit of daily average total solar radiation is MJm-2day-1Wind speed ofThe bit is m.s-1The unit of pressure is KPa; inquiring the type of the irrigated crops, and giving an empirical value of each growth period according to experts in different growth periods;
(2) normalizing all the data collected in the step (1) to obtain a training set; normalization means to normalize all collected data to [ -1, 1], and the normalization formula is as follows:
in the formula, x is normalized data, x is data needing to be normalized at present, and xminFor the minimum in the data that needs to be normalized, xmaxThe maximum value in the data needing normalization is obtained;
(3) and (3) training a double-layer extreme learning machine by adopting a training set to obtain a final prediction model:
1) 4000 groups of data are selected, the data are arranged according to a time sequence, the 500 groups of data are used as a data block partition, the data are divided into 8 data blocks and are numbered in sequence;
2) the method comprises the following steps of training a double-layer extreme learning machine by using a data block 1 to obtain a model M1, inputting the data block 1 into a first-layer extreme learning machine to obtain a first-layer prediction result in the training process, and inputting the data block 1 and the first-layer prediction result into a second-layer extreme learning machine as input parameters to obtain a model M1, wherein the basic steps of the construction of the double-layer extreme learning machine are shown in FIG. 2:
i) initializing input parameters of a double-layer extreme learning machine network, and selecting an activation function g (x) of a hidden layer; the input parameters comprise an input layer number, a hidden layer node number and an output layer number, wherein the input layer number is the number of input data types, the output layer number is 1, and an initialization formula of the hidden layer node number is as follows:
in the formula (I), the compound is shown in the specification,
for the number of hidden layer nodes of the k-th layer limit learning machine,
for the number of input level nodes of the k-th level limit learning machine,
the number of nodes of an output layer of the k-th layer limit learning machine is counted;
the activation function g (x) is shown in FIG. 4, and its equation is as follows:
wherein X is an independent variable, assuming a given set of N samples { (X) in the extreme learning machine networks，ts) N, where x is specifically W in the extreme learning machinev·Xs+bv，WvAs a connection weight between the input layer and the hidden layer, bvIs a threshold value, and is,
Xsis the input vector, e is the natural constant;
ii) respectively and randomly initializing the connection weight W between the input layer and the hidden layer of the double-layer extreme learning machine network according to the number of the nodes of the input layer and the number of the nodes of the hidden layervAnd a threshold value bvThe initialization range is [ -0.5, 0.5 [)]；
iii) calculating a hidden layer output matrix H according to the input weight and the threshold and the activation function g (x), specifically: suppose that N sets of samples { (X) are given in an extreme learning machine networks，ts) N, with hidden layer nodes of 1
The extreme learning machine network structure is as follows:
in the formula, Xs＝[Xs1，Xs2，...，Xsn]T∈Rn，ts＝[ts1，ts2，...，tsm]T∈Rm，XsAnd tsRespectively representing input variables and corresponding output variables, Wv＝[Wv1，Wv2，...，Wvn]TIs an input weight vector connecting the nodes of the v-th hidden layer with the n input layers, bvIs the threshold vector of the v-th hidden layer, βv＝[βv1，βv2，...，βvm]TIs a weight vector, W, connecting the nodes of the v-th hidden layer with the m output layersv·XsExpressing and solving an inner product;
the above equation transforms to obtain:
Hβ＝T；
in the formula, H is a hidden layer output matrix, β is an output weight matrix, T is an expected output matrix, and elements in T are the crop water demand calculated by multiplying the FAO-56PM equation by the crop coefficient, where:
w is the determined value because the input weight and the threshold between the input layer and the hidden layer are determined valuesvAnd bvDetermining the value, and calculating a hidden layer output matrix H;
iv) calculating a hidden layer output matrix H according to an activation function g (x) of the hidden layer of the extreme learning machine network, and calculating an extreme learning machine output weight matrix β, wherein the formula for calculating the extreme learning machine output weight matrix β is as follows:
in the formula (I), the compound is shown in the specification,
Moore-Penrose generalized inverse of matrix H, λ being the regularization coefficient, IIs an identity matrix;
3) inputting the data block 2 into a model M1 to obtain the predicted irrigation water demand and calculating the prediction precision piAnd a prediction result decision coefficient R2The calculation process is shown in FIG. 3, wherein
In the formula, ETPM56(i) The unit is mm-day, the calculated value of the water demand of the ith particle crop by the FAO-56PM equation is multiplied by the corresponding crop coefficient to obtain the crop irrigation water demand-1，ETpredicted(i) The unit of the training predicted value of the limit learning machine for the water demand of the ith particle crop is mm-day-1N is the number of groups of input samples;
water requirement ET for crop irrigationPM56The calculation formula of (a) is as follows:
in the formula, delta is the slope of the function relation curve of saturated water vapor pressure-temperature, and the unit is KPa DEG C-1，RnFor net radiation input into the canopy, the unit is MJm-2day-1G is the soil heat flux, here neglected, in units of MJm-2day-1，esIs saturated water vapor pressure with KPa, eaThe actual water vapor pressure is expressed in KPa, and gamma represents thermometer constant expressed in KPa-deg.C-1Q is the daily average temperature in degrees Celsius and u is the wind speed at 2m height in m.s-1(ii) a crop coefficient is crop coefficient;
the calculation formula for Δ is as follows:
determining the coefficient R2The closer to 1, the higher the prediction accuracy, and the coefficient R is determined2The calculation formula of (2) is as follows:
4) judging the prediction accuracy piWhether the prediction accuracy is greater than or equal to a set prediction accuracy E1(0.93) or not, if so, the output model M1 is the integrated prediction model C1, and if not, the next step is carried out;
5) determining the coefficient R according to the prediction result2Updating the number of hidden layer nodes of the double-layer extreme learning machine, wherein
Wherein c is 5 and epsilon is 0.01;
6) updating the output weight matrix of the double-layer extreme learning machine by adopting an incremental learning mechanism according to the data block 2 to obtain the incremental double-layer extreme learning machine, and inputting the jth data block into the output weight matrix β of the incremental double-layer extreme learning machinejThe following were used:
wherein j is not less than 2, HjHidden layer output matrix of incremental double-layer extreme learning machine for j-th data block input, TjInputting the expected output matrix of the incremental double-layer extreme learning machine for the jth data block, when j is 2, βj-1Output weight matrix of double-layer extreme learning machine for input of 1 st data block, H1Hidden layer output matrix of double-layer extreme learning machine for 1 st data block input, T1Inputting the expected output matrix of the double-layer extreme learning machine for the 1 st data block;
7) training an incremental double-layer extreme learning machine by using a data block 2 to obtain a model M2, wherein the model M2 is an integrated prediction model C1, and the training step is the same as the step 2);
8) inputting the data block 3 into an integrated prediction model C1 to obtain the predicted irrigation water demand, and calculating the prediction precision and the prediction result to determine a coefficient R2The calculation step is the same as the step 3);
9) circulating the steps 3) to 7) until the data block 8 is input, wherein the obtained integrated prediction model C7 is the final prediction model;
(4) collecting five groups of test data, wherein each group of test data comprises soil environment data, meteorological data and crop coefficients of irrigated crops; normalizing each group of test data as shown in step (2), inputting the normalized test data into a final prediction model C7 to obtain a final predicted irrigation water demand, carrying out irrigation according to the predicted irrigation water demand, wherein the predicted irrigation water demand and the error between the predicted irrigation water demand and a true value are shown in FIGS. 5 and 6, and a prediction index decision coefficient R is determined2＝0.98244。
Comparative example 1
The data of the irrigation method based on the standard extreme learning machine are the same as those of the irrigation method 1, the finally obtained predicted irrigation water demand and the error of the predicted irrigation water demand and the actual value are shown in figures 5 and 6, and the prediction index decision coefficient R of the irrigation method is shown in figure 520.9473, it can be seen from fig. 5 and 6 that the dynamic multi-layer extreme learning machine of the present invention can calculate the irrigation water demand more accurately and realize reasonable irrigation compared with the standard extreme learning machine.
Example 2
The method is basically the same as that in embodiment 1, but is different in that the multi-layer extreme learning machine is a three-layer extreme learning machine, 5000 groups of selected data are divided into 10 data blocks, and finally, a prediction index decision coefficient R of the predicted irrigation water demand and the actual demand value is calculated2The prediction results are all averages of 20 runs of the program, 0.97405.
Claims (9)
1. An irrigation method based on a dynamic multilayer extreme learning machine is characterized in that data related to known irrigation water demand is collected and normalized, then the multilayer extreme learning machine is trained to obtain a final prediction model, the data related to the irrigation water demand to be required is input to the final prediction model after being normalized to obtain predicted irrigation water demand, and then irrigation is carried out, wherein the input and the output of the upper layer in the multilayer extreme learning machine are simultaneously used as the input of the lower layer, the dynamic multilayer extreme learning machine means that the number of nodes of the hidden layer in the training process of the multilayer extreme learning machine is dynamically updated, and the method comprises the following steps:
(1) collecting a plurality of groups of training data, wherein each group of training data comprises soil environment data, meteorological data and crop coefficients of irrigated crops;
(2) carrying out normalization processing on each group of training data, wherein all groups of training data form a training set;
(3) training a multi-layer extreme learning machine by using a training set to obtain a final prediction model;
(4) collecting a plurality of groups of test data, wherein each group of test data comprises soil environment data, meteorological data and crop coefficients of irrigated crops;
(5) normalizing each group of test data, inputting the normalized test data into a final prediction model to obtain predicted irrigation water demand, and irrigating according to the predicted irrigation water demand;
the specific steps of training the multi-layer extreme learning machine by adopting the training set are as follows:
1) dividing all groups of training data into n data blocks according to the size of a sliding window in equal parts according to the acquisition time sequence and numbering the data blocks in sequence, wherein the unit of the sliding window is a group;
2) training a multi-layer extreme learning machine by using a data block 1 to obtain a model M1;
3) let j equal 2;
4) inputting the data block j into an integrated prediction model C (j-2) to obtain the predicted irrigation water demand and calculating the prediction precision piAnd a prediction result decision coefficient R2The integrated prediction model C (j-2) is a model obtained by training the data block (j-1), and the integrated prediction model C0 is a model M1;
5) judging the prediction accuracy piWhether the prediction accuracy E1 is larger than or equal to the set prediction accuracy, if so, the output model M (j-1) is the integrated prediction model C (j-1), otherwise, the next step is carried out;
6) determining the coefficient R according to the prediction result2Updating the number of hidden layer nodes of the multilayer extreme learning machine;
7) updating an output weight matrix of the multilayer extreme learning machine by adopting an incremental learning mechanism according to the data block j to obtain the incremental multilayer extreme learning machine;
8) training an incremental multi-layer extreme learning machine by using a data block j to obtain a model Mj, wherein the model Mj is an integrated prediction model C (j-1);
9) let j equal j + 1;
10) and (4) circulating the steps from 4) to 8) to j ═ n, and obtaining the integrated prediction model C (n-1), namely the final prediction model.
2. The irrigation method based on the dynamic multi-layer extreme learning machine as claimed in claim 1, wherein the soil environment data and the meteorological data are specifically: the unit of the temperature is DEG C, the unit of the humidity is%, and the unit of the daily average solar total radiation is MJm-2day-1The unit of the wind speed is m.s-1The unit of the pressure is KPa; the crop coefficient refers to an empirical value of each growth period given by experts in different growth periods according to the type of the irrigated crop.
3. The irrigation method based on the dynamic multi-layer extreme learning machine as claimed in claim 1, wherein the normalization process in step (2) and step (5) is to normalize all collected data to [ -1, 1], and the normalization formula is as follows:
in the formula, X is normalized data, X is data needing to be normalized at present, and XminFor the minimum in the data that needs to be normalized, xmaxIs the maximum value in the data that needs to be normalized.
4. The irrigation method based on the dynamic multi-layer extreme learning machine as claimed in claim 1, wherein the training set comprises 4000-5000 sets of training data, and the size of the sliding window is 500 sets.
5. The irrigation method based on a dynamic multi-layer extreme learning machine as claimed in claim 1, wherein the multi-layer extreme learning machine is a double-layer extreme learning machine.
6. The irrigation method based on the dynamic multi-layer extreme learning machine as claimed in claim 1, wherein the multi-layer extreme learning machine is constructed by the following steps:
(1) initializing input parameters of a multi-layer extreme learning machine network, and selecting an activation function g (x) of a hidden layer; the input parameters comprise an input layer number, a hidden layer node number and an output layer number, wherein the input layer number is the number of input data types, the output layer number is 1, and an initialization formula of the hidden layer node number is as follows:
in the formula (I), the compound is shown in the specification,
for the number of hidden layer nodes of the k-th layer limit learning machine,
for the number of input level nodes of the k-th level limit learning machine,
the number of nodes of an output layer of the k-th layer limit learning machine is counted;
the equation for the activation function g (x) is as follows:
wherein X is an independent variable, assuming a given set of N samples { (X) in the extreme learning machine networks，ts) N, assuming that the number of hidden layer nodes of the basic extreme learning machine is 1
X here is specifically referred to as W in the extreme learning machinev·Xs+bv，WvAs a connection weight between the input layer and the hidden layer, bvIs a threshold value, and is,
Xsis the input vector, e is the natural constant;
(2) respectively and randomly initializing the connection weight W between the input layer and the hidden layer of the network of the multilayer extreme learning machine according to the number of nodes of the input layer and the number of nodes of the hidden layervAnd a threshold value bvThe initialization range is [ -0.5, 0.5 [)]；
(3) According to the connection weight W between the input layer and the hidden layervThreshold bvAnd an activation function g (x) for computing a hidden layer output matrix H, the expression of the hidden layer output matrix H being as follows:
(4) and calculating an output weight matrix β of the multi-layer extreme learning machine according to the hidden layer output matrix H, wherein the formula is as follows:
in the formula (I), the compound is shown in the specification,
Moore-Penrose generalized inverse of matrix H, where λ is regularization coefficient, I is identity matrix, T is expected output matrix, and T isThe element is the crop water demand calculated by multiplying the FAO-56PM equation by the crop coefficient.
7. The irrigation method based on the dynamic multi-layer extreme learning machine as claimed in claim 1, wherein the prediction precision p isiThe calculation formula of (a) is as follows:
in the formula, ETPM56(i) The unit is mm-day obtained by multiplying the calculated value of the water demand of the group i of the crops by the corresponding crop coefficient through an FAO-56PM equation-1，ETpredicted(i) Predicted irrigation water demand in mm day for group i data crops-1N is the number of groups of input samples;
the value range of the set prediction precision E1 is more than 0.90;
the determination coefficient R2The calculation formula of (a) is as follows:
in the formula, meanETpredictedCalculating the average of the predicted values of the N groups of data;
ETPM56the calculation formula of (a) is as follows:
in the formula, delta is the slope of the function relation curve of saturated water vapor pressure-temperature, and the unit is KPa DEG C-1，RnFor net radiation input into the canopy, the unit is MJm-2day-1G is the soil heat flux, here neglected, in units of MJm-2day-1，esIs saturated water vapor pressure with KPa, eaThe actual water vapor pressure is expressed in KPa, and gamma represents thermometer constant expressed in KPa-deg.C-1Q is RipingThe average air temperature is expressed in the unit of DEG C, u is the air speed at a height of 2 meters and is expressed in the unit of m & s-1Crop coefficient, Crop coefficient; the calculation formula for Δ is as follows:
8. the irrigation method based on dynamic multi-tier extreme learning machine as claimed in claim 1, wherein the coefficient R is determined according to the prediction result2The step of updating the number of hidden layer nodes of the multilayer extreme learning machine refers to determining a coefficient R according to a prediction result2Defining an updating mechanism, wherein a specific formula is as follows:
wherein c is 5 and e is 0.01.
9. The irrigation method based on dynamic multi-layer extreme learning machine as claimed in claim 1, wherein the j-th data block is input into the output weight matrix β of the incremental multi-layer extreme learning machinejThe following were used:
wherein j is not less than 2, HjHidden layer output matrix of incremental multi-layer extreme learning machine for j-th data block input, TjInputting the expected output matrix of the incremental multi-layer extreme learning machine for the jth data block, when j is 2, βj-1For the 1 st data blockOutput weight matrix of input multi-layer extreme learning machine, H1Hidden layer output matrix of multi-layer extreme learning machine for input of 1 st data block, T1The expected output matrix of the multi-layer extreme learning machine is input for the 1 st data block.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710605810.3A CN107466816B (en) | 2017-07-24 | 2017-07-24 | Irrigation method based on dynamic multilayer extreme learning machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710605810.3A CN107466816B (en) | 2017-07-24 | 2017-07-24 | Irrigation method based on dynamic multilayer extreme learning machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107466816A CN107466816A (en) | 2017-12-15 |
| CN107466816B true CN107466816B (en) | 2020-04-07 |
Family
ID=60595891
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710605810.3A Active CN107466816B (en) | 2017-07-24 | 2017-07-24 | Irrigation method based on dynamic multilayer extreme learning machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107466816B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111369093B (en) * | 2018-12-26 | 2023-09-29 | 天云融创数据科技（北京）有限公司 | Irrigation method and device based on machine learning |
| CN110428106A (en) * | 2019-08-05 | 2019-11-08 | 山东农业大学 | A kind of crop water requirement prediction technique based on machine learning |
| CN110999766A (en) * | 2019-12-09 | 2020-04-14 | 怀化学院 | Irrigation decision method, device, computer equipment and storage medium |
| CN113497785B (en) * | 2020-03-20 | 2023-05-12 | 深信服科技股份有限公司 | Malicious encryption traffic detection method, system, storage medium and cloud server |
| CN111880489B (en) * | 2020-07-07 | 2022-12-09 | 北京理工大学 | Regression scheduling method for complex manufacturing system |
| CN112136667B (en) * | 2020-11-26 | 2021-02-12 | 江苏久智环境科技服务有限公司 | Intelligent sprinkling irrigation method and system based on edge machine learning |
| CN113050567B (en) * | 2021-03-17 | 2022-02-01 | 北京理工大学 | Dynamic scheduling method for intelligent manufacturing system |
| CN113994868B (en) * | 2021-09-27 | 2023-07-28 | 上海易航海芯农业科技有限公司 | Automatic irrigation method and system based on plant growth cycle |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570627A (en) * | 2016-11-02 | 2017-04-19 | 河海大学 | Crop irrigation water requirement calculation method on future climatic conditions |
| CN106651012A (en) * | 2016-12-02 | 2017-05-10 | 东华大学 | Crop transpiration prediction method based on improved extreme learning machine |
| CN106718695A (en) * | 2017-01-04 | 2017-05-31 | 吉林省沃特管业有限公司 | A kind of intelligent water-saving irrigates Internet of Things network control system |
-
2017
- 2017-07-24 CN CN201710605810.3A patent/CN107466816B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570627A (en) * | 2016-11-02 | 2017-04-19 | 河海大学 | Crop irrigation water requirement calculation method on future climatic conditions |
| CN106651012A (en) * | 2016-12-02 | 2017-05-10 | 东华大学 | Crop transpiration prediction method based on improved extreme learning machine |
| CN106718695A (en) * | 2017-01-04 | 2017-05-31 | 吉林省沃特管业有限公司 | A kind of intelligent water-saving irrigates Internet of Things network control system |
Non-Patent Citations (3)
| Title |
|---|
| Extreme Learning Machines: A new approach for prediction of reference evapotranspiration;Shafika Sultan Abdullah etc.;《Journal of Hydrology》;20151231;第527卷;184-195 * |
| Universal Approximation of Extreme Learning Machine with Adaptive Growth of Hidden Nodes;Zhang Rui etc.;《IEEE Transaction on Neural Networks and Learning Systems》;20121231;第23卷(第2期);365-371 * |
| 基于极限学习的深度学习算法;赵志勇 等;《计算机工程与设计》;20150430;第36卷(第4期);1022-1026 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107466816A (en) | 2017-12-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107466816B (en) | Irrigation method based on dynamic multilayer extreme learning machine | |
| CN103268366B (en) | A kind of combination wind power forecasting method suitable for distributing wind power plant | |
| Cai et al. | A method for modelling greenhouse temperature using gradient boost decision tree | |
| CN110084367A (en) | A kind of Forecast of Soil Moisture Content method based on LSTM deep learning model | |
| CN106651012A (en) | Crop transpiration prediction method based on improved extreme learning machine | |
| CN110163254B (en) | Cucumber greenhouse output intelligent prediction device based on recurrent neural network | |
| CN110084417A (en) | A kind of strawberry greenhouse environment parameter intelligent monitor system based on GRNN neural network | |
| CN106650784A (en) | Feature clustering comparison-based power prediction method and device for photovoltaic power station | |
| CN110069032B (en) | Eggplant greenhouse environment intelligent detection system based on wavelet neural network | |
| CN110119086B (en) | Tomato greenhouse environmental parameter intelligent monitoring device based on ANFIS neural network | |
| CN109376951A (en) | A kind of photovoltaic probability forecasting method | |
| CN110119169A (en) | A kind of tomato greenhouse temperature intelligent early warning system based on minimum vector machine | |
| CN110119767A (en) | A kind of cucumber green house temperature intelligent detection device based on LVQ neural network | |
| CN107909221A (en) | Power-system short-term load forecasting method based on combination neural net | |
| CN108762084A (en) | Irrigation system of rice field based on fuzzy control decision and method | |
| Wu et al. | Utilization of radial basis function neural network model for water production forecasting in seawater greenhouse units | |
| Li et al. | Prediction of grain yield in Henan Province based on grey BP neural network model | |
| CN110147825A (en) | A kind of strawberry greenhouse temperature intelligent detection device based on empirical mode decomposition model | |
| Tunalı et al. | Estimation of actual crop evapotranspiration using artificial neural networks in tomato grown in closed soilless culture system | |
| Liu et al. | Estimating models for reference evapotranspiration with core meteorological parameters via path analysis | |
| CN115453868B (en) | Full-growth-period light intensity regulation and control method based on tomato light response difference characteristics | |
| Chen et al. | Data-driven decision support scheme for multi-area light environment control in greenhouse | |
| Mimboro et al. | Weather monitoring system AIoT based for oil palm plantation using recurrent neural network algorithm | |
| Wang et al. | Application of grey systems in predicting the degree of cotton spider mite infestations | |
| CN114444399A (en) | XGboost-based greenhouse drip irrigation tomato transpiration amount calculation method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |