Predicting the blooming season of ornamental plants is significant for guiding adjustments in production decisions and providing viewing periods and routes. The current strategies for observation of ornamental plant booming periods are mainly based on manpower and experience, which have problems such as inaccurate recognition time, time-consuming and energy sapping. Therefore, this paper proposes a neural network-based method for predicting the flowering phase of pear tree. Firstly, based on the meteorological observation data of Shijiazhuang Meteorological Station from 2000 to 2019, three principal components (the temperature factor, weather factor, and humidity factor) with high correlation coefficient with the flowering phase of pear tree are obtained by using the principal component analysis method. Then, the three components are used as input factors for the BP neural network. A BP neural network prediction model is constructed based on genetic algorithm optimization. The crossover operator and mutation operator in the adaptive genetic algorithm are improved. Finally, the meteorological sample data from 2013 to 2019 are used to test and verify the algorithm in this paper. The results demonstrate that, the model can solve the local optimization problem of the BP neural network model. The prediction results of the flowering phase of pear tree are evaluated in terms of relevance and prediction accuracy. Both are superior to the traditional effective accumulated temperature and the prediction results of the stepwise regression method. This method can provide more reliable forecast information for the blooming period, which can provide decision-making reference for improving the development of tourism industry.
Flowering period data of ornamental plants are important for the organization and development of flower-viewing tourism activities. These data can provide guidance for tourists to formulate flower viewing tourism plans, and provide reference for the planning of flower-viewing activities in scenic locations. Ornamental plants have a strong seasonal flowering period, short duration, and are easily affected by the external environment. The earlier the flowering period is monitored, the earlier best views and routs can be got. Traditional manual observation is time-consuming and laborious. It cannot effectively predict and release information in a timely manner.
There are many studies on the relationships among the plant flowering period and climate, climate change and prediction technology of flowering period. Gonsamo et al. [
With the development of computational science, neural networks have been widely investigated and applied in various fields of science and engineering, and relatively satisfactory prediction results have been achieved. Long et al. [
Aimed at forecasting the flowering phase of ornamental plants, existing research focuses on cherry blossoms, while there are few simulations of the flowering phase of other ornamental plants. This paper propose a prediction method based on neural network for flowering phase of pear trees. Through principal component analysis, three principal components with higher correlation coefficient with the flowering phase of pear trees were obtained as input factors of the neural network. The problem of low accuracy of the forecasted flowering period is improved. This paper is organized as follows: In the second section, the sources of pear phenological data were introduced, and three principal components were obtained by using principal component analysis. In the third section, a BP neural network model is established based on genetic algorithm optimization, and the model parameters are instantiated. In the fourth section, the model is trained and tested, and compared with the traditional effective accumulated temperature method. The fifth section uses the trained model to forecast the flowering stage of pear trees in 2020. In the last, a conclusion is provided.
The phenological data of pear flowers were obtained from the observation data of pear trees planting area provided by Shijiazhuang Meteorological Bureau from 2009 to 2020. The observation basis and standards are the “China Phenological Observation Network” observation standards and the China Meteorological Administration “Agricultural Meteorological Observation Specifications.” The initial flowering period of a plant is defined as the data when the first fully open flower begins to appear on the observation plant. The full bloom period is defined as the data when more than half of the flower buds on the observation plant unfold and show the petals [
The flowering day ordinal number is the conversion of the observation date into a diurnal sequence. January 1 is taken as the beginning data of the flowering diurnal sequence, and the diurnal number is 1, and so on, such as February 5, the diurnal number is 36 [
By analyzing the location observation phenological data and parallel observation meteorological data of the pear trees planting area from 2009 to 2020, it is calculated that the average initial flowering time of pear trees is April 3; the average blooming period is April 10; and the average end of flowering is April 20. The earliest flowering date is March 20, and the latest data is April 16. The flowering phase begins as early as March 28 and as late as April 24. The earliest date of the end of flowering is April 8, and the latest date is May 4. The flowering phase of pear trees lasts for approximately 20 days on average from the beginning to the end of flowering. According to years of observation and analysis, pear blossoms will enter a suitable viewing period after 2–3 days of initial flowering [
Flowering index | Flowering state | Viewing period | Viewing suggestions | |
---|---|---|---|---|
W1 | Beginning of flowering | Flowering rate 0%–10% | Viewing not recommended | |
W2 | Blooming | Flowering rate 50%–80% | Best viewing period, suitable | |
W3 | End of flowering | Flowering rate 80%-declining flower day | Worst viewing period, not suitable |
Meteorological factors, including temperature, precipitation, sunshine, etc., impose obvious constraints on the growth of pear trees. However, the restrictive effects of different meteorological factors are divided into strong and weak factors that are relatively independent.
This paper selects the daily maximum temperature, daily minimum temperature, daily average temperature, daily precipitation, daily sunshine duration, daily average ground temperature and daily average relative humidity from 2000 to 2019 at Shijiazhuang City Meteorological Station. These data have been strictly controlled by artificial quality. The average values of each meteorological element in winter from 2001 to 2019 are obtained by calculation.
Principal Component Analysis (PCA) aims to analyze the characteristics of the covariance matrix to reduce the dimensionality of the data while maintaining the largest contribution to the variance of the data set [
Step 1: Select the meteorological factors that are related to the flowering period of pear blossoms, and perform z-score normalization (zero-mean normalization) on the original data. The formula is expressed as follows:
In the formula,
Step 2: Calculate the correlation coefficient matrix R of the standardized meteorological factor data:
In the formula,
Step 3: According to the meteorological factor correlation coefficient matrix R, find the eigenvalue, the contribution rate of the principal component and the contribution rate of the cumulative variance. Determine the number of principal components, and solve the characteristic equation:
Find the eigenvalue
Cumulative contribution rate:
According to the principle of selecting the number of principal components, eigenvalues greater than 1 and cumulative contribution rates greater than 90% are selected.
The meteorological observation data of meteorological stations from 2000 to 2019 were used, and the eigenvalues of the correlation matrix and the contribution rate of each principal component were obtained. The results are shown in
Component | Characteristic root | Contribution rate/% | Cumulative contribution rate/% |
---|---|---|---|
1 | 3.8978 | 48.75 | 48.75 |
2 | 2.8867 | 30.65 | 79.40 |
3 | 1.5892 | 12.54 | 91.94 |
4 | 0.5093 | 4.09 | 96.03 |
The eigenvectors of the three principal component factors are listed in
Weather factor | 1st principal component | 2nd principal component | 3rd principal component |
---|---|---|---|
0.4123 | 0.3998 | 0.0011 | |
0.3923 | −0.0301 | 0.1270 | |
0.3911 | 0.1500 | 0.0776 | |
0.2011 | −0.2298 | 0.0423 | |
−0.0889 | 0.4788 | 0.1199 | |
0.4678 | −0.1007 | −0.1102 | |
0.2153 | −0.5064 | 0.4019 |
According to the characteristic vector of the principal components (
In addition, studies have pointed out that environmental parameters such as soil temperature, cold demand and hourly accumulated temperature, as well as management measures such as tree age, fertilization and irrigation, are also related to the flowering phase of trees. This article forecasts the overall flowering period in the region from the perspective of weather forecasting services, instead of a single fruit tree or orchard. Taking into account the availability of data, these parameters are not involved.
The research and application of neural networks in various fields of science and engineering are extensive, and relatively satisfactory prediction results have been achieved. The BP neural network, which is the most commonly employed from of artificial neural network [
The BP neural network is generally composed of an input layer, a hidden layer and an output layer. And a large number of neurons are connected to each other as network nodes [
The error function between the output and the expected value is:
where
where
Because
where
Traditional genetic algorithms use fixed crossover probability and mutation probability [
where
where
To improve the stability of the BP algorithm and solve the problem of local optimal solutions, this paper constructs a BP neural network model that is based on genetic algorithm optimization [
(1) Initialize the population
(2) Calculate the fitness function. The fitness function F, which is also known as the evaluation function, can realize the measurement of the pros and cons of individuals in the group. The fitness function can be expressed by the sum of the absolute value of the error between the predicted output of the BP neural network and the actual output. The calculation formula is
(3) The selection operation is performed after completing the evaluation of the fitness function. The commonly employed selection methods of genetic algorithms include the fitness ratio method, partial selection method and roulette selection method. This article chooses the roulette selection method. The probability of an individual that is selected in this method is proportional to the fitness value that is calculated in Step (2). Assume that the fitness of individual
where
(4) Crossover operation. Crossover operation refers to the operation of partially reorganizing the structure of the two selected first-generation individuals according to the principle of cross-exchange of biological chromosomes to form a new individual, and the crossover operation acts on the selected group. The crossover probability is
(5) In the mutation operation, the mutation operator judges the mutation probability of the individual in the group and then randomly selects the mutation position of the mutated individual to obtain the result of the mutation. The mutation operation is applied to the selected population. The mutation probability is
(6) After the population
(7) If the termination condition is met, proceed to the decoding operation; otherwise, return to Step (3).
The overall design process is shown in
This study selects a 3-layer BP neural network and uses principal component analysis to obtain three principal component factors that affect the flowering phase of pear trees as the input layer of the BP neural network. Coupled with the input layer threshold, the model has 4 input vectors. The daily ordinal number of the full-blossom period of pear trees (best viewing period) is utilized as the output vector of the output layer, and thus the number of nodes in the output layer is 1, and the output vector is 1. The determination of the number of neurons in the hidden layer is related to the actual problem that needs to be solved, and the optimal number of neurons is generally determined by a formula.
where
Number of nodes | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
---|---|---|---|---|---|---|---|
Training error | 0.2323 | 0.2047 | 0.1868 | 0.1998 | 0.2211 | 0.3645 | 0.4367 |
The relevant parameters of the genetic algorithm are listed as follows: the number of iterations is 10; the randomly generated population size is 100; the crossover rate
This article uses the MATLAB R2014a neural network development toolbox and Sheffield genetic algorithm toolbox for training. In the process of training the determined network structure, to prevent “transition training,” the network adopts the following convergence rules: if the samples with an absolute error of less than 2 reach 85% of the total samples, the training is stopped. Otherwise, the maximum number of training iterations (10,000 iterations) is specified, and the samples are used to verify the network when the training meets the requirements.
In this paper, the flowering phenology data of pear trees from 2000 to 2012 in the pear planting area of Shijiazhuang are employed as the training input samples of the BP neural network. A trial report is conducted using the 2013–2019 sample. When using the three main component factors as the network input, the comprehensive situation of the meteorological factors is considered, the main factor is enlarged and the secondary factor is weakened. Network training and testing were carried out according to the forecast model, and the training effect map (
Comparing
Using the data samples of the flowering phase of pear trees from 2016 to 2019, the improved BP neural network method proposed in this article, the traditional effective accumulated temperature law method and stepwise regression method are verified. The verification results of the three methods are shown in
Years | Actual early flowering period | Effective accumulated temperature prediction | Stepwise regression prediction | BP neural network model optimized based on genetic algorithm |
---|---|---|---|---|
2016 | March 24 | March 26 | March 25 | March 24 |
2017 | March 23 | March 27 | March 20 | March 25 |
2018 | March 28 | March 24 | March 24 | March 30 |
2019 | March 25 | March 22 | March 28 | March 24 |
To intuitively compare the applicability and accuracy of the method in this paper for the prediction of the flowering phase of pear trees in Shijiazhuang, a combination of internal inspection and cross-checking is employed. First, internal inspection is carried out. The parameters fitted from 2000 to 2019 are used to simulate the phenological sequence of the flowering phase of pear trees. The observed sequence of the flowering phase of pear trees is compared with the simulated sequence, and the variance (
where
We compare and analyze the actual observation sequence of the flowering phase of pear trees with the simulated sequence of a certain year. If the difference between the two sequences is less than 3 days, the prediction is considered accurate and recorded as “1”; otherwise, the prediction is considered wrong and recorded as “0”. The accuracy of the forecast is the percentage of the number of accurate forecasts in the total number of years. It can be seen from
Model | Internal inspection accuracy/% | Cross-check accuracy/% |
---|---|---|
Effective accumulated temperature method | 81.22 | 81.22 |
Stepwise regression method | 76.89 | 76.89 |
Improved BP neural network proposed in this paper | 91.67 | 91.67 |
After the training is completed, the topology of the neural network is readjusted. The number of nodes in the input layer includes the number of weather factors, the number of days in the initial flowering period and the accumulated temperature since the initial flowering period. The number of nodes in the output layer is the number of forecasted objects (1, use 1/0 to indicate whether flowering occurs the next day).
Accumulated temperature refers to the daily average temperature that is accumulated in the corresponding period of time when a plant completes a certain period or all-growth period. Accumulated temperature is an important indicator to measure the requirements of crop growth for thermal conditions and to evaluate thermal resources. According to the statistics of the accumulated temperature from January 1 to the full flowering period of pear blossom in Shijiazhuang City, the daily average temperature is greater than or equal to the threshold temperature of pear trees (any value in the range of 0.1–20.0
A sample of a certain year is selected to form a calculation sample of the model in this paper within 10 days before and after flowering. The forecast value is 1 from the day before the flowering period, which indicates that the next day has entered the flowering period, and the number of days before that is set to 0. For example, the initial of flowering period of a certain year is March 20, the starting data of the sample is March 10–29, the forecast result before March 19 is 0, and the forecast result from March 19 is 1. The flowchart is shown in
Using the flowering period forecast model in this paper, the flowering phase of pear trees in 2020 in the Shijiazhuang area is forecasted. The forecast will be carried out every day beginning on March 1st. The maximum forecast time limit is 10 days, that is, the cycle in
This paper selects the pear trees planting area in Shijiazhuang as the target area for prediction. Three principal component factors that affect the flowering period of pear are obtained through principal component analysis, namely, the temperature factor, weather factor and humidity factor. The BP neural network optimized based on a genetic algorithm is applied to the forecasting of the pear flowering period, and the crossover operator and mutation operator in the adaptive genetic algorithm are improved. The experimental results show that the error of the flowering period forecast using the forecast model in this paper is 1 day, the value of the effective accumulated temperature forecast method is 3.25 days, and that of the stepwise regression forecast method is 2.75 days. It can be seen that the algorithm proposed in this paper achieves better results than the traditional methods in predicting the flowering period, and also verifies the correlation between the flowering data and meteorological factors. The algorithm presented in this paper can provide more reliable forecast information of the flowering phase and provide reasonable reference for the public’s travel arrangement.