Big data driven double special water and fertilizer alternate irrigation method and system for mountain camellia oleifera

Abstract

The application discloses a big data driven mountainous Camellia oleifera dual-purpose water and fertilizer alternate irrigation method and system, relates to the field of big data, and comprises the following steps: acquiring local meteorological data, current phenological period label and soil humidity data of a Camellia oleifera planting area; determining an ideal soil humidity interval of the Camellia oleifera planting area in the current phenological period, comparing the soil humidity data with the ideal soil humidity interval, and calculating a phenological period water demand matching degree parameter; determining a current fruiting state of a big year and a small year according to historical flower bud differentiation rate data and historical fruit setting rate data of the Camellia oleifera planting area; inputting the phenological period water demand matching degree parameter, the big year and small year fruiting state and the local meteorological data into an alternate irrigation prediction model array, taking the flower bud differentiation rate and the fruit setting rate as dual optimization objectives, and dynamically outputting an alternate irrigation strategy; and applying dual water and fertilizer alternately to the Camellia oleifera planting area according to the alternate irrigation strategy. The application solves the technical problems of low adaptability and poor irrigation effect of the alternate irrigation strategy in the prior art.

Description

Technical Field

[0001] This invention relates to the field of big data, specifically to a big data-driven method and system for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas. Background Technology

[0002] In mountainous camellia oleifera cultivation scenarios, there is a complex coupling relationship between the current phenological stage label of camellia oleifera trees and local meteorological data. This causes the ideal soil moisture range under the current phenological stage to change dynamically with meteorological conditions, making it impossible to accurately determine using a fixed mapping table. At the same time, the differentiated water and fertilizer requirements due to alternate bearing create a natural trade-off between flower bud differentiation rate and fruit set rate. That is, in a bumper year, priority should be given to ensuring flower bud differentiation rate, which may sacrifice fruit set rate; in a lean year, priority should be given to ensuring fruit set rate, which may sacrifice flower bud differentiation rate.

[0003] Existing irrigation methods cannot dynamically output alternating irrigation strategies that simultaneously consider flower bud differentiation rate and fruit set rate based on the matching degree of water requirements during phenological periods and the fruit-bearing status of alternate bearing. As a result, the timing and amount of switching between special water and fertilizer types cannot accurately match the real-time physiological needs of camellia trees. Summary of the Invention

[0004] This invention provides a big data-driven method and system for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas, which solves the technical problem that existing irrigation methods cannot be adapted to the growth conditions of camellia oleifera, resulting in low adaptability of alternating irrigation strategies and poor irrigation effects.

[0005] In view of the above problems, the present invention provides a big data-driven method and system for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas.

[0006] In a first aspect, the present invention provides a big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas, the method comprising:

[0007] Acquire local meteorological data, current phenological stage labels, and soil moisture data for camellia oleifera planting areas; Based on the current phenological period label, determine the ideal soil moisture range of the camellia oleifera planting area during the current phenological period, and compare the soil moisture data with the ideal soil moisture range to calculate the phenological period water requirement matching parameter. Based on the historical flower bud differentiation rate data and historical fruit setting rate data of the camellia oleifera planting area, the current alternate bearing status of the camellia oleifera planting area is determined; The phenological water requirement matching parameters, the alternate bearing status of alternate bearing and the local meteorological data are input into the pre-trained alternating irrigation prediction model array. With flower bud differentiation rate and fruit setting rate as dual optimization objectives, the alternating irrigation strategy is dynamically output. The alternating irrigation strategy includes special water and fertilizer type, application amount and application time. The special water and fertilizer type includes flower-promoting water and fertilizer in high-yield years and fruit-strengthening water and fertilizer in low-yield years. According to the alternating irrigation strategy, the camellia oleifera planting area is subjected to alternating application of water and fertilizer.

[0008] Secondly, this invention provides a big data-driven dual-purpose water and fertilizer alternating irrigation system for camellia oleifera in mountainous areas, the system comprising: The raw data acquisition module is used to acquire local meteorological data, current phenological period labels, and soil moisture data of the camellia oleifera planting area; The matching degree parameter calculation module is used to determine the ideal soil moisture range of the camellia oleifera planting area in the current phenological period based on the current phenological period label, and compare the soil moisture data with the ideal soil moisture range to calculate the phenological period water requirement matching degree parameter; The alternate bearing fruit status judgment module is used to determine the current alternate bearing fruit status of the camellia oleifera planting area based on the historical flower bud differentiation rate data and historical fruit setting rate data of the camellia oleifera planting area. The alternating irrigation strategy output module is used to input the phenological water requirement matching degree parameter, the alternate bearing status and the local meteorological data into the pre-trained alternating irrigation prediction model array, and dynamically output the alternating irrigation strategy with flower bud differentiation rate and fruit setting rate as dual optimization objectives. The alternating irrigation strategy includes special water and fertilizer type, application amount and application time. The special water and fertilizer type includes flower-promoting water and fertilizer in high-yield years and fruit-strengthening water and fertilizer in low-yield years. The alternating water and fertilizer application module is used to apply alternating water and fertilizer to the camellia oleifera planting area according to the alternating irrigation strategy.

[0009] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a big data-driven method for alternating water and fertilizer irrigation of camellia oleifera in mountainous areas. First, it acquires local meteorological data, current phenological stage labels, and soil moisture data, providing a real-time data foundation for the subsequent dynamic coupling of phenological stages and meteorological conditions. Second, it determines the ideal soil moisture range based on the current phenological stage label and compares it with measured soil moisture to calculate the phenological stage water requirement matching parameter, solving the problem of not being able to dynamically reflect changes in the ideal moisture range with meteorological conditions. Third, it determines the current alternate bearing state based on historical flower bud differentiation rate data and historical fruit setting rate data, providing a decision-making basis for subsequent state-specific optimization of irrigation strategies. Finally, it combines the phenological stage water requirement matching parameter with the alternate bearing state data... The annual fruit-bearing status and local meteorological data are input into an alternating irrigation prediction model array. With flower bud differentiation rate and fruit set rate as dual optimization targets, the model dynamically outputs an alternating irrigation strategy that includes the type, amount, and timing of application of specialized water and fertilizer. The model array simultaneously weighs the mutual constraints between flower bud differentiation rate and fruit set rate, and automatically adjusts the irrigation direction according to the current alternate bearing situation, prioritizing flower promotion in high-yield years and fruit strengthening in low-yield years, to achieve precise matching of the timing and amount of specialized water and fertilizer switching. Finally, the alternating irrigation strategy is applied to the camellia oleifera planting area, and the optimal strategy output by the model is transformed into actual irrigation operation, improving the adaptability of the alternating irrigation strategy to the growth status of camellia oleifera and the irrigation effect. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the big data-driven alternating water and fertilizer irrigation method for camellia oleifera in mountainous areas provided by this invention. Figure 2 This is a logical schematic diagram of the big data-driven alternating water and fertilizer irrigation method for camellia oleifera in mountainous areas provided by the present invention. Figure 3 This is a preferred example of the correspondence table between the special water and fertilizer types and the water and fertilizer content values ​​in the big data-driven alternating water and fertilizer irrigation method for camellia oleifera in mountainous areas provided by the present invention. Figure 4 This is a schematic diagram of the structure of the big data-driven dual-specific water and fertilizer alternating irrigation system for mountain camellia oleifera provided by the present invention.

[0011] In the attached diagram, the components represented by each number are as follows: The module includes: raw data acquisition module 11, matching degree parameter calculation module 12, alternate bearing status judgment module 13, alternating irrigation strategy output module 14, and alternating water and fertilizer application module 15. Detailed Implementation

[0012] This invention provides a big data-driven method and system for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas, which solves the technical problem that existing irrigation methods cannot be adapted to the growth conditions of camellia oleifera, resulting in low adaptability of alternating irrigation strategies and poor irrigation effects.

[0013] The present invention will now be described in detail with reference to the accompanying drawings.

[0014] Example 1, as Figure 1 , Figure 2 As shown, this invention provides a big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas, the method comprising: S10: Obtain local meteorological data, current phenological stage labels, and soil moisture data for the camellia oleifera planting area; In this embodiment of the invention, local meteorological data refers to local small-scale meteorological parameters collected by meteorological stations deployed in the camellia oleifera planting area, which can reflect the real meteorological conditions of the microenvironment in which the camellia oleifera trees are located; the current phenological stage label is the specific developmental stage of the camellia oleifera tree in its growth cycle, and each phenological stage corresponds to different physiological needs and water and fertilizer sensitivities; soil moisture data is the volumetric water content or relative water content measured by soil moisture sensors buried in the active layer of the camellia oleifera root system, which is used to represent the water content in the soil available for absorption by the camellia oleifera trees.

[0015] Specifically, the camellia oleifera planting area is determined, and then the local meteorological data corresponding to the area, the local phenological period label of the current camellia oleifera planting area, and the soil moisture data are obtained.

[0016] Step S10 in the method of this embodiment of the invention includes: Acquire local meteorological data collected by local meteorological stations deployed in camellia oleifera planting areas, wherein the local meteorological data includes at least rainfall data, air temperature data, and air humidity data; Obtain the current phenological stage label of the camellia oleifera planting area, wherein the current phenological stage label is one of the following: flower bud differentiation stage label, fruit enlargement stage label, oil conversion stage label, or dormancy stage label; Obtain soil moisture data collected by soil moisture sensors deployed in the camellia oleifera planting area.

[0017] In this embodiment of the invention, firstly, the local meteorological station area where the camellia oleifera planting area is located is obtained. Local meteorological data is collected at a fixed frequency within the local meteorological station, including at least rainfall data, air temperature data, and air humidity data. Specifically, rainfall affects soil moisture replenishment; air temperature affects transpiration and phenological processes; and air humidity affects stomatal opening and the risk of disease occurrence.

[0018] Secondly, based on the different developmental stages of the camellia oleifera tree, the current phenological stage labels for the camellia oleifera planting area are determined. These labels can be derived from different developmental stages within the annual growth cycle of the camellia oleifera tree, including four types: flower bud differentiation stage, fruit enlargement stage, oil conversion stage, and dormancy stage. Specifically, the flower bud differentiation stage is the period from the formation of flower primordia to the differentiation of flower organs, and it is sensitive to water and phosphorus and potassium fertilizers; the fruit enlargement stage is the rapid growth stage of young fruits, requiring a large amount of water and increasing the demand for nitrogen and calcium fertilizers; the oil conversion stage is the stage of oil accumulation within the fruit, requiring moderate water and high potassium fertilizers; and the dormancy stage is the stage of leaf abscission and metabolic stagnation, requiring extremely low water.

[0019] Finally, soil moisture sensors were buried in the main distribution layer of the camellia root system in the camellia planting area, and the soil moisture sensors collected soil moisture data at a frequency of hourly or higher.

[0020] In this embodiment of the invention, local meteorological data is obtained through local meteorological stations in the camellia oleifera planting area; the current phenological stage is obtained according to the growth and development stage of the camellia oleifera trees; and soil moisture sensors are deployed to detect soil moisture data, providing sufficient raw data for subsequent processing.

[0021] S20: Based on the current phenological period label, determine the ideal soil moisture range of the camellia oleifera planting area during the current phenological period, and compare the soil moisture data with the ideal soil moisture range to calculate the phenological period water requirement matching parameter; In this embodiment of the invention, the ideal soil moisture range is the range of soil moisture that can optimally promote the normal physiological activities of camellia trees under the current phenological period and current meteorological conditions; the phenological period water requirement matching degree parameter is used to indicate the degree and direction of the current actual soil moisture deviating from the ideal moisture range.

[0022] Specifically, based on the current phenological stage label of the camellia oleifera planting area, the corresponding ideal soil moisture range is obtained, and the soil moisture data of the camellia oleifera planting area is compared with the ideal soil moisture range to calculate the phenological stage water requirement matching parameter, which reflects the degree to which the soil moisture deviates from the ideal soil moisture.

[0023] Step S20 in the method of this embodiment of the invention includes: The current phenological period label is subjected to one-hot encoding to obtain the phenological period encoding vector; The rainfall, air temperature, and air humidity data in the local meteorological data are normalized to obtain normalized rainfall, normalized air temperature, and normalized air humidity values. The normalized rainfall, normalized air temperature, and normalized air humidity values ​​are then combined to form a meteorological feature vector. The phenological period encoding vector and the meteorological feature vector are concatenated along the channel dimension to obtain the phenological period-meteorological joint feature vector; The phenological-meteorological joint feature vector is input into the pre-trained ideal soil moisture prediction model, and the predicted ideal soil moisture value and the predicted soil moisture standard deviation are output. Subtracting the product of the predicted soil moisture standard deviation and 1.5 from the predicted ideal soil moisture value yields the lower limit of ideal soil moisture. Adding the product of the predicted soil moisture standard deviation and 1.5 to the predicted ideal soil moisture value yields the upper limit of ideal soil moisture. The closed interval formed by the lower limit of ideal soil moisture and the upper limit of ideal soil moisture is defined as the ideal soil moisture interval; When the soil moisture data is less than the ideal soil moisture lower limit, the difference between the ideal soil moisture lower limit and the soil moisture data is calculated, and the difference is divided by the ideal soil moisture lower limit to obtain the phenological period water demand matching degree parameter. When the soil moisture data is greater than the upper limit of the ideal soil moisture, the difference between the soil moisture data and the upper limit of the ideal soil moisture is calculated, the difference is divided by the upper limit of the ideal soil moisture to obtain the phenological water demand matching degree parameter, and the phenological water demand matching degree parameter is negative. When the soil moisture data is greater than or equal to the lower limit of the ideal soil moisture and less than or equal to the upper limit of the ideal soil moisture, the phenological period water requirement matching degree parameter is set to zero.

[0024] In this embodiment of the invention, firstly, the current phenological stage label is subjected to one-hot encoding. One-hot encoding is an encoding method that converts discrete categorical variables into binary vectors. Assuming there are K possible phenological stage categories, the encoded result is a vector of length K, where the position corresponding to the category is 1 and the remaining positions are 0. After one-hot encoding, a phenological stage encoding vector is obtained. For the four phenological stages, the order is flower bud differentiation stage, fruit enlargement stage, oil conversion stage, and dormancy stage. For example, if the current stage is flower bud differentiation stage, the encoding vector is [1,0,0,0].

[0025] Secondly, the rainfall, air temperature, and air humidity data from the local meteorological data are normalized to obtain normalized rainfall, normalized air temperature, and normalized air humidity values. Specifically, the maximum-minimum value normalization method can be used, where the normalized value = (initial value - minimum value) / (maximum value - minimum value). The normalized rainfall, normalized air temperature, and normalized air humidity values ​​are then combined into a meteorological feature vector, for example, (0.83, 0.48).

[0026] Furthermore, the phenological period encoding vector and the meteorological feature vector are concatenated in the channel dimension in the order of their positions to obtain the phenological period-meteorological joint feature vector. After concatenation, the phenological period encoding vector of length 4 is concatenated with the meteorological feature vector of length 3 to obtain the phenological period-meteorological joint feature vector of length 7, for example [1,0,0,0,0.83,0.48].

[0027] Furthermore, the phenological-meteorological joint feature vector is input into the pre-trained ideal soil moisture prediction model, and the outputs are the predicted ideal soil moisture value and the predicted soil moisture standard deviation. The ideal soil moisture prediction model is a pre-trained regression model whose input is the phenological-meteorological joint feature vector, and whose outputs are the predicted ideal soil moisture value and the predicted ideal soil moisture standard deviation.

[0028] Furthermore, subtracting the product of the predicted soil moisture standard deviation and 1.5 from the predicted ideal soil moisture value yields the lower limit of ideal soil moisture: Lower limit of ideal soil moisture = (Predicted ideal soil moisture value - Predicted soil moisture standard deviation × 1.5). Adding the product of the predicted soil moisture standard deviation and 1.5 to the predicted ideal soil moisture value yields the upper limit of ideal soil moisture: Upper limit of ideal soil moisture = (Predicted ideal soil moisture value + Predicted soil moisture standard deviation × 1.5), where 1.5 is an empirical coefficient; based on the assumption of a normal distribution, approximately 1.5 times the standard deviation corresponds to an approximately 86.6% confidence interval.

[0029] Furthermore, the closed interval formed by the lower limit of ideal soil moisture and the upper limit of ideal soil moisture is defined as the ideal soil moisture interval, i.e., [lower limit of ideal soil moisture, upper limit of ideal soil moisture].

[0030] Furthermore, when the soil moisture data is less than the ideal lower limit of soil moisture, the difference between the ideal lower limit of soil moisture and the soil moisture data is calculated. This difference is then divided by the ideal lower limit of soil moisture to obtain the phenological water demand matching parameter. The phenological water demand matching parameter = (ideal lower limit of soil moisture - soil moisture data) / ideal lower limit of soil moisture, and the result is a positive value.

[0031] Furthermore, when the soil moisture data exceeds the upper limit of the ideal soil moisture value, the difference between the soil moisture data and the upper limit of the ideal soil moisture value is calculated. This difference is then divided by the upper limit of the ideal soil moisture value to obtain the phenological water demand matching parameter. This parameter is then negative. The phenological water demand matching parameter is calculated as: -(ideal soil moisture data - upper limit of soil moisture value) / upper limit of ideal soil moisture value, with a negative result.

[0032] Furthermore, when soil moisture data is greater than or equal to the lower limit of ideal soil moisture and less than or equal to the upper limit of ideal soil moisture, falling between the lower and upper limits of ideal soil moisture, the phenological period water demand matching parameter is set to zero. The phenological period water demand matching parameter reflects the degree and direction of deviation of the current actual soil moisture from the ideal moisture range. A positive value indicates soil water shortage, a negative value indicates excessive soil moisture, and zero indicates that it is within the ideal range. A larger positive value indicates a more severe water shortage, and a larger absolute negative value indicates a more severe excessive moisture.

[0033] In the method of this invention embodiment, the pre-training process of the ideal soil moisture prediction model includes: A historical phenological period-meteorological-soil moisture sample set is obtained. Each historical phenological period-meteorological-soil moisture sample set contains a historical phenological period label, a set of historical meteorological data, and a historical optimal soil moisture label set. The historical optimal soil moisture label set contains a first optimal soil moisture value and a second optimal soil moisture value. The first optimal soil moisture value is the soil moisture data recorded when the highest historical flower bud differentiation rate is obtained under the same historical phenological period label, and the second optimal soil moisture value is the soil moisture data recorded when the highest historical fruit setting rate is obtained under the same historical phenological period label. All historical phenological period labels are subjected to unique thermal encoding to obtain historical phenological period encoding vectors. All historical meteorological data are normalized and combined to form historical meteorological feature vectors. The historical phenological period encoding vectors and historical meteorological feature vectors are concatenated to form a training input sample set. The historical optimal soil moisture label set is used to form a training label set. An initial ideal soil moisture prediction model is constructed, which includes a phenological period coding layer, a meteorological feature coding layer, a feature splicing layer, and a regression output layer. The initial ideal soil moisture prediction model is trained in a supervised manner using the training input sample set and the training label set. The training objective is to minimize the weighted sum of the first mean square error between the predicted ideal soil moisture value and the first optimal soil moisture value output by the initial ideal soil moisture prediction model and the second mean square error between the predicted ideal soil moisture value and the second optimal soil moisture value, until the weighted sum meets the preset convergence condition, thus obtaining the trained ideal soil moisture prediction model.

[0034] In this embodiment of the invention, firstly, a historical phenological period-meteorological-soil moisture sample set is obtained. This sample set is extracted from historical data across multiple past growth cycles and used to train the model. Each set of historical phenological period-meteorological-soil moisture samples includes a historical phenological period label, a set of historical meteorological data, and a historically optimal soil moisture label set. The historically optimal soil moisture label set includes a first optimal soil moisture value and a second optimal soil moisture value. The first optimal soil moisture value is the soil moisture data recorded when the highest historical flower bud differentiation rate was achieved under the same historical phenological period label, and the second optimal soil moisture value is the soil moisture data recorded when the highest historical fruit set rate was achieved under the same historical phenological period label.

[0035] We collected historical data on camellia oleifera cultivation accumulated over many years, including daily phenological stage labels, historical meteorological parameters, and historical soil moisture values. Then, we grouped the historical soil moisture values ​​according to phenological stage and meteorological conditions. Within each group, we identified the historical soil moisture value corresponding to the highest flower bud differentiation rate as the first optimal soil moisture value, and the historical soil moisture value corresponding to the highest fruit set rate as the second optimal soil moisture value. Each group formed a training sample.

[0036] Secondly, all historical phenological period labels are subjected to unique thermal encoding to obtain historical phenological period encoding vectors; all historical meteorological data are normalized and combined into historical meteorological feature vectors; the historical phenological period encoding vectors and historical meteorological feature vectors are concatenated to form historical phenological period-meteorological joint feature vectors, which constitute the training input sample set; and the historical best soil moisture label set constitutes the training label set.

[0037] Next, an initial ideal soil moisture prediction model was constructed, which includes a phenological period coding layer, a meteorological feature coding layer, a feature splicing layer, and a regression output layer. The training set, validation set, and test set were divided in an 8:1:1 ratio. A multilayer perceptron (MLP), a feedforward artificial neural network, was employed, capable of nonlinear transformations and approximating arbitrarily complex functions.

[0038] For example, the phenological period encoding layer receives and encodes the input phenological period labels, converting them into digital form, and includes 4 nodes; the meteorological feature encoding layer receives and encodes meteorological features, and includes 3 nodes; the feature concatenation layer concatenates the phenological period encoding vector and the meteorological feature vector to form a phenological-meteorological joint feature vector, and includes 7 nodes; subsequently, three fully connected hidden layers are set: the first hidden layer contains 64 neurons and uses the ReLU activation function to perform nonlinear mapping on the joint feature vector to extract the interaction features between the phenological period labels and meteorological features; the second hidden layer contains 32 neurons and uses the ReLU activation function to reduce the dimensionality and abstract the output of the first hidden layer; the third hidden layer contains 16 nodes and uses the ReLU activation function to provide feature representation for the regression output layer. The regression output layer includes 2 nodes, uses a linear activation function, and outputs the predicted ideal soil moisture value and the predicted ideal soil moisture standard deviation.

[0039] Finally, the initial ideal soil moisture prediction model is trained in a supervised manner using the training input sample set and the training label set. The training objective is to minimize the weighted sum of the first mean square error between the predicted ideal soil moisture value and the first optimal soil moisture value, and the second mean square error between the predicted ideal soil moisture value and the second optimal soil moisture value, and the weighted sum is used as the total loss.

[0040] Total loss = First mean square error × First mean square error weighting coefficient + Second mean square error × Second mean square error weighting coefficient, where the weighting coefficients can be determined based on the influence of the first and second mean square errors on the total loss. This process continues until the weighted sum meets the preset convergence condition, resulting in a trained ideal soil moisture prediction model.

[0041] For example, an initial learning rate of 0.001 is set, and the Adam optimizer is used for parameter updates. The loss function is set to the weighted mean squared error function. The training batch size is set to 32, and the maximum number of iterations is set to 500. The phenological-meteorological joint feature vector is transmitted to the hidden layer via forward propagation for nonlinear transformation. Subsequently, the gradient of the total loss with respect to the weights and biases of each layer is calculated via backpropagation, and the Adam optimizer uses the gradient to update the parameters. After each epoch, the total loss is calculated on the validation set. If the validation loss does not decrease for 30 consecutive epochs, training is stopped.

[0042] After training, the test set is input to calculate the root mean square error and the coefficient of determination between the predicted mean and the first and second optimal labels. If the coefficient of determination is greater than 0.85 and the root mean square error is less than 2%, the model is considered to have passed the validation. After meeting the validation criteria, the ideal soil moisture prediction model that has been trained is obtained.

[0043] In this embodiment of the invention, the phenological period encoding vector and the meteorological feature vector are concatenated to obtain the phenological period-meteorological joint feature vector; then, it is input into the ideal soil moisture prediction model for prediction, which solves the defect that the ideal soil moisture cannot be mapped and determined under different meteorological conditions, and improves the accuracy and response rate of the prediction; an ideal soil moisture range is constructed, and the soil moisture data is compared with the ideal soil moisture range to obtain the phenological period water demand matching degree parameter, which provides the urgency and direction of subsequent irrigation.

[0044] S30: Based on the historical flower bud differentiation rate data and historical fruit setting rate data of the camellia oleifera planting area, determine the current alternate bearing status of the camellia oleifera planting area; In this embodiment of the invention, the historical flower bud differentiation rate data is the statistical value of the proportion of the number of flower buds to the total number of buds in one or more past growth cycles; the historical fruit setting rate data is the statistical value of the proportion of the number of flower buds that eventually develop into fruit to the total number of flower buds in one or more past growth cycles; the alternate bearing state is the physiological state of the periodic fluctuation of yield of the camellia tree in different years due to the dynamic imbalance between vegetative growth and reproductive growth, including the state of a bumper year and the state of a lean year.

[0045] Specifically, historical flower bud differentiation rate data and historical fruit set rate data from the previous growth cycle and multiple consecutive growth cycles are obtained. Then, the deviations in flower bud differentiation and fruit set are calculated to determine the current alternate bearing status in the camellia oleifera planting area. In a bumper year, the tree's nutrients are largely consumed in fruit development, resulting in insufficient flower bud differentiation; therefore, priority should be given to applying flower-promoting fertilizers to encourage flower bud formation. In a lean year, the fruit load is low, and the tree is recovering its nutrient accumulation; therefore, priority should be given to applying fruit-enhancing fertilizers to promote fruit enlargement and tree vigor recovery.

[0046] Step S30 in the method of this embodiment includes: Obtain historical flower bud differentiation rate data and historical fruit setting rate data of the tea oleifera planting area in the previous growth cycle; Obtain the historical flower bud differentiation rate data set and historical fruit setting rate data set of the tea oleifera planting area over the past multiple consecutive growth cycles; Calculate the mean of the historical flower bud differentiation rate data set and the mean of the historical fruit setting rate data set. Use the mean of the historical flower bud differentiation rate data set as the benchmark value for high yield due to flower bud differentiation and the mean of the historical fruit setting rate data set as the benchmark value for high yield due to fruit setting. The ratio of the historical flower bud differentiation rate data in the previous growth cycle to the benchmark value of high yield due to flower bud differentiation is calculated to obtain the flower bud differentiation deviation. The ratio of the historical fruit setting rate data in the previous growth cycle to the benchmark value of high yield due to fruit setting is calculated to obtain the fruit setting deviation. When both the flower bud differentiation deviation and the fruit setting deviation are greater than the preset high-yield deviation threshold, the current alternate bearing state of the camellia oleifera planting area is determined to be a high-yield state. When at least one of the flower bud differentiation deviation and the fruit setting deviation is less than or equal to the preset high-yield deviation threshold, the current alternate bearing state of the camellia oleifera planting area is determined to be an alternate bearing state.

[0047] In this embodiment of the invention, firstly, after the flower bud differentiation period, the number of flower buds and the total number of buds are counted; and before the fruit matures and is harvested, the number of fruits and the initial number of flower buds are counted. Historical flower bud differentiation rate data and historical fruit set rate data for the Camellia oleifera planting area in the previous growth cycle are calculated. The growth cycle refers to the complete process of Camellia oleifera from flower bud differentiation to fruit harvest. Flower bud differentiation rate = number of flower buds / total number of buds. Fruit set rate = number of fruits that develop into mature fruits / total number of initial flower buds.

[0048] Secondly, extract the flower bud differentiation rate and fruit setting rate records of multiple consecutive growth cycles in the past to obtain the historical flower bud differentiation rate data set and historical fruit setting rate data set of the camellia oleifera planting area in the past multiple consecutive growth cycles.

[0049] Next, the mean values ​​of the historical flower bud differentiation rate dataset and the historical fruit set rate dataset were calculated. The mean value of the historical flower bud differentiation rate dataset was used as the benchmark value for high yield due to flower bud differentiation, representing the normal level of flower bud differentiation in camellia trees in this region under normal years. The mean value of the historical fruit set rate dataset was used as the benchmark value for high yield due to fruit set, representing the normal level of fruit set.

[0050] Furthermore, the ratio of historical flower bud differentiation rate data from the previous growth cycle to the baseline value for high-yield flower bud differentiation is calculated to obtain the flower bud differentiation deviation: Flower bud differentiation deviation = Historical flower bud differentiation rate data from the previous growth cycle / Baseline value for high-yield flower bud differentiation. Similarly, the ratio of historical fruit set rate data from the previous growth cycle to the baseline value for high-yield fruit set is calculated to obtain the fruit set deviation: Fruit set deviation = Historical fruit set rate data from the previous growth cycle / Baseline value for high-yield fruit set. These two deviations reflect the degree of high yield in the previous cycle relative to the long-term average in both flower bud differentiation and fruit set dimensions. A value greater than 1 indicates yield higher than the long-term average, and a value less than 1 indicates yield lower than the long-term average.

[0051] Furthermore, a preset threshold for deviation from a high yield is set to determine whether the actual performance deviates from the normal level. This threshold can be set based on the target for yield fluctuations in camellia oleifera from alternate bearing to biennial bearing, with a value close to 1 but slightly less than 1, for example, 0.8 or 0.85.

[0052] When both the deviation of flower bud differentiation and the deviation of fruit setting are greater than the preset high-yield deviation threshold, it is considered a high-yield or normal yield, and the current alternate bearing status of the camellia oleifera planting area is determined to be a high-yield state.

[0053] Furthermore, when at least one of the deviations in flower bud differentiation and fruit setting is less than or equal to the preset high-yield deviation threshold, it is considered a poor yield, and the current alternate bearing status of the camellia oleifera planting area is determined to be a low-yield state.

[0054] By determining the alternate bearing fruit status, key decision-driving parameters are provided for the alternating irrigation prediction model array. The alternating irrigation prediction model dynamically adjusts the number of activation branches of the flower bud differentiation rate prediction model set and the fruit setting rate prediction model set according to the alternate bearing fruit status. In a year with abundant fruit, more emphasis is placed on the prediction accuracy of flower bud differentiation rate to ensure the flowering effect, while in a year with deficient fruit, more emphasis is placed on the prediction accuracy of fruit setting rate to ensure the fruit enlargement effect, thereby achieving an adaptive trade-off between the two optimization objectives.

[0055] In this embodiment of the invention, the deviation of flower bud differentiation and the deviation of fruit setting are calculated respectively. The fruit setting status of alternate bearing is determined by comparing the preset high yield deviation with the deviation. The overall high yield of the camellia tree in the previous cycle is reflected from two dimensions, avoiding misjudgment from a single dimension and providing a physiological basis for the differentiated optimization of irrigation strategies.

[0056] S40: Input the phenological water requirement matching degree parameter, the alternate bearing status and the local meteorological data into the pre-trained alternating irrigation prediction model array, with flower bud differentiation rate and fruit setting rate as dual optimization objectives, and dynamically output the alternating irrigation strategy. The alternating irrigation strategy includes special water and fertilizer type, application amount and application time. The special water and fertilizer type includes flower-promoting water and fertilizer in high-yield years and fruit-strengthening water and fertilizer in low-yield years. In this embodiment of the invention, the alternating irrigation prediction model array is a composite model structure composed of multiple sub-models. After pre-training with historical data, it can output the optimal irrigation strategy based on input conditions. The specific water and fertilizer types include a flowering-promoting water and fertilizer for bumper years and a fruit-enhancing water and fertilizer for leaner years. The application rate is the amount of water and fertilizer used per unit area or per plant. The application timing refers to whether irrigation is initiated at the current moment and the duration or interval of irrigation. Preferred examples of specific values ​​for flowering-promoting water and fertilizer in bumper years and fruit-enhancing water and fertilizer in leaner years are as follows: Figure 3 As shown.

[0057] Specifically, the phenological water requirement matching parameters, the fruit-bearing status in alternate years, and local meteorological data are input into the alternating irrigation prediction model array. The alternating irrigation prediction model array predicts the fusion flower bud differentiation rate, the fusion fruit set rate, and the trading irrigation strategy, respectively. Alternating irrigation can be applied according to the alternating irrigation strategy, providing a strategy for subsequent implementation.

[0058] Step S40 in the method of this embodiment includes: The phenological water requirement matching degree parameter, the alternate bearing fruit status and the local meteorological data are input into the flower bud differentiation rate prediction model set in the alternating irrigation prediction model array, and the flower bud differentiation rate prediction model set is output by the fused flower bud differentiation rate; The phenological water requirement matching degree parameter, the alternate bearing fruit status and the local meteorological data are input into the fruit set rate prediction model set in the alternating irrigation prediction model array, and the fruit set rate prediction model set outputs the fused fruit set rate. The phenological water requirement matching degree parameter, the alternate bearing state and the local meteorological data are input into the strategy selection model in the alternating irrigation prediction model array. The strategy selection model outputs multiple sets of candidate alternating irrigation strategies. Each set of candidate alternating irrigation strategies includes candidate special water and fertilizer type, candidate application amount and candidate application time. Using the maximum weighted harmonic function value of the fusion flower bud differentiation rate and the fusion fruit setting rate as the selection objective, a candidate alternating irrigation strategy is selected from the multiple candidate alternating irrigation strategies as the alternating irrigation strategy.

[0059] In this embodiment of the invention, firstly, the phenological water requirement matching degree parameter, alternate bearing fruit status, and local meteorological data are used as input data and input into the flower bud differentiation rate prediction model set in the alternating irrigation prediction model array. The flower bud differentiation rate is predicted by the flower bud differentiation rate prediction model set, and the fused flower bud differentiation rate is output.

[0060] Secondly, the phenological water demand matching degree parameter, alternate bearing fruit status and local meteorological data are used as input data and input into the set of fruit setting rate prediction models in the alternating irrigation prediction model array to predict the fruit setting rate. The fused fruit setting rate is then output by the set of fruit setting rate prediction models.

[0061] Next, the phenological water requirement matching parameters, alternate bearing fruit status, and local meteorological data are input into the strategy selection model in the alternating irrigation prediction model array. After forward calculation, the strategy selection model outputs multiple sets of candidate alternating irrigation strategies. Each set of candidate alternating irrigation strategies includes candidate specific water and fertilizer types, candidate application rates, and candidate application timing. Among them, the specific water and fertilizer types include flowering-promoting types for bumper years and fruit-strengthening types for alternate bearing years; the candidate application timings include immediate application, a 2-hour delay, or periodic intervals.

[0062] Ultimately, with the goal of maximizing the weighted harmonic function value that combines flower bud differentiation rate and fruit set rate, a candidate alternating irrigation strategy was selected from multiple candidate strategies. The selection objective refers to choosing the strategy that maximizes the weighted harmonic function value; the weighted harmonic function is a comprehensive evaluation index used to balance two potentially conflicting objectives.

[0063] The characteristic of a weighted harmonic function is that if either indicator is low, the overall score will be lowered. The optimization process must consider both indicators simultaneously, and cannot maximize one at the expense of the other. Weighted harmonic functions, for example... Where S1 is the fusion flower bud differentiation rate, S2 is the fusion fruit set rate, and β is the weighting coefficient. When β > 1, the fusion flower bud differentiation rate is given more importance; when β < 1, the fusion fruit set rate is given more importance. β is the weighting coefficient used to adjust the relative importance of flower bud differentiation rate to fruit set rate. Its value can be dynamically determined based on the alternate bearing conditions, allowing the selection of alternating irrigation strategies to automatically adapt to the current physiological needs of the camellia trees. When β = 1, flower bud differentiation rate and fruit set rate are equally important; when β > 1, flower bud differentiation rate is more important, applicable to bumper years, requiring priority to promote flowering; when β < 1, fruit set rate is more important, applicable to lean years, requiring priority to ensure fruit development.

[0064] In the method of this embodiment of the invention, the pre-training process of the alternating irrigation prediction model array includes: Obtain a historical irrigation sample set. Each historical irrigation sample set includes a historical phenological period water requirement matching degree parameter, a historical alternate bearing status, a set of historical meteorological data, a historical flower bud differentiation rate label value, a historical fruit setting rate label value, and a historical optimal irrigation strategy label. The historical optimal irrigation strategy label includes the historical optimal special water and fertilizer type, the historical optimal application amount, and the historical optimal application time. An initial flower bud differentiation rate prediction model set is constructed, which includes multiple initial flower bud differentiation rate prediction branch models. Each initial flower bud differentiation rate prediction branch model is trained in a supervised manner using the historical irrigation sample set. The training objective is to minimize the mean square error between the output predicted flower bud differentiation rate and the historical flower bud differentiation rate label value until the verification convergence is obtained, thus obtaining the trained flower bud differentiation rate prediction model set. Construct an initial fruit set prediction model set, which includes multiple initial fruit set prediction branch models. Use the historical irrigation sample set to perform supervised training on each initial fruit set prediction branch model. The training objective is to minimize the mean square error between the output predicted fruit set rate and the historical fruit set rate label value until the verification convergence is obtained, thus obtaining the trained fruit set prediction model set. An initial strategy selection model is constructed using a multilayer perceptron structure. The initial strategy selection model is then trained in a supervised manner using the historical irrigation sample set. The training objective is to minimize the weighted sum of the classification cross-entropy loss between the predicted special water and fertilizer type and the historical optimal special water and fertilizer type, the first mean square error between the predicted application amount and the historical optimal application amount, and the second mean square error between the predicted application timing and the historical optimal application timing, until the model is verified to converge, thus obtaining the trained strategy selection model. The trained flower bud differentiation rate prediction model set, the trained fruit set prediction model set, and the trained strategy selection model are combined to form an alternating irrigation prediction model array.

[0065] In this embodiment of the invention, firstly, historical data of the camellia oleifera planting area over multiple past growth cycles is collected. Water requirement matching parameters are calculated, alternate bearing conditions are determined, and the irrigation strategy that actually optimizes flower bud differentiation rate or fruit set rate in each time period is selected as the historical optimal irrigation strategy label. A historical irrigation sample set is obtained, which is a set of labeled data used for supervised learning. Each historical irrigation sample set includes a historical phenological water requirement matching parameter, a historical alternate bearing fruit set condition, a set of historical meteorological data, a historical flower bud differentiation rate label value, a historical fruit set rate label value, and a historical optimal irrigation strategy label. The historical optimal irrigation strategy label includes the historical optimal type of dedicated water and fertilizer, the historical optimal application rate, and the historical optimal application timing.

[0066] Secondly, a set of prediction models for the initial flower bud differentiation rate was constructed, which included multiple branch models with the same structure but different initial parameters.

[0067] M initial flower bud differentiation rate prediction models with identical initial structures are constructed, where M ≥ 3 (e.g., 10). Specifically, a fully connected neural network is used to simultaneously construct the M initial flower bud differentiation rate prediction models. Each initial flower bud differentiation rate prediction model consists of an input layer, a hidden layer, and an output layer. The input layer of each model receives matching degree parameters, alternate year / biennial bearing conditions, and meteorological data, and includes 3 nodes. The hidden layer performs complex nonlinear transformations on the data to extract deep features. Nonlinear mapping is performed using an activation function, and the bias is calculated; this layer includes 16 nodes. The output layer includes 1 node, used to output the predicted flower bud differentiation rate.

[0068] M initial flower bud differentiation rate prediction models with identical initial structures were trained simultaneously. Supervised training was performed on each initial flower bud differentiation rate prediction branch model using a historical irrigation sample set. The historical irrigation sample set was used as input to the set of initial flower bud differentiation rate prediction models, and the training, validation, and test sets were divided in a 7:2:1 ratio. The training objective was to minimize the mean squared error between the predicted flower bud differentiation rate and the label of the historical optimal irrigation strategy. The mean squared error was used as the loss function during training. An initial learning rate of 0.001 was set, and the Adam optimizer was used for training. Through forward propagation, the historical irrigation sample set was weighted and summed with weights and biases, and then subjected to multiple nonlinear transformations by the activation function. Subsequently, the loss gradient was calculated through backpropagation, and the weights and biases in the network were updated using gradient descent to continuously reduce the prediction error of the model. After training, the performance of the model on the reserved validation set was evaluated. If the mean squared error between the predicted flower bud differentiation rate and the actual result was within 0.05, the requirement was met, and the M initially identical flower bud differentiation rate prediction models were obtained.

[0069] Because the initial weights of each branch model are randomized differently, they will converge to different local optima after training, resulting in a set of trained flower bud differentiation rate prediction models. The arithmetic mean of the output values ​​of all flower bud differentiation rate prediction models is the fused flower bud differentiation rate.

[0070] Furthermore, an initial fruit set prediction model set is constructed, which contains multiple initial fruit set prediction branch models. Each initial fruit set prediction branch model is trained in a supervised manner using a historical irrigation sample set.

[0071] Similarly, N initial fruit set rate prediction models with identical structures are constructed, where N ≥ 3, for example, 10. Specifically, N initial fruit set rate prediction models are constructed simultaneously using a fully connected neural network. The structure of each initial fruit set rate prediction model includes an input layer, a hidden layer, and an output layer. The input layer of each model receives matching degree parameters, alternate year status, and meteorological data, and also includes 3 nodes. The hidden layer performs complex nonlinear transformations on the data to extract deep features. Nonlinear mapping is performed through activation functions, and bias is calculated, including 16 nodes. The output layer includes 1 node, used to output the predicted fruit set rate.

[0072] N identical initial fruit set rate prediction models were trained simultaneously, with each branch model trained under supervision using a historical irrigation sample set. The historical irrigation sample set was used as input to the set of initial fruit set rate prediction models, and the training, validation, and test sets were divided in a 7:2:1 ratio. The training objective was to minimize the mean squared error (MSE) between the predicted fruit set rate and the label of the historical optimal irrigation strategy, using MSE as the loss function. An initial learning rate of 0.001 was set, and the Adam optimizer was used for training. Forward propagation involved weighted summation of the historical irrigation sample set through weights and biases, followed by multiple nonlinear transformations using the activation function. Backpropagation was then used to calculate the loss gradient, and gradient descent was used to update the weights and biases in the network, continuously reducing the model's prediction error. After training, the model's performance on the reserved validation set was evaluated. If the MSE between the predicted fruit set rate and the actual result was within 0.04, the requirement was met, and the N identical initial fruit set rate prediction models were obtained.

[0073] Because the initial weights of each branch model are randomized differently, they will converge to different local optima after training, resulting in a set of trained fruit set prediction models. The arithmetic mean of the output values ​​of all fruit set prediction models is the fused fruit set rate.

[0074] Furthermore, a multilayer perceptron structure was used to construct an initial strategy selection model, which was then trained in a supervised manner using a historical irrigation sample set. The initial strategy selection model consists of an input layer, a hidden layer, and an output layer. The input layer receives phenological water demand matching parameters, alternate bearing conditions, and local meteorological data, and includes 3 nodes. The hidden layer performs nonlinear feature extraction on the input, containing 18 nodes, and uses the ReLU function as the activation function to alleviate the gradient vanishing problem, mapping the input to the feature space and extracting features. The output layer uses a linear activation function, contains 3 nodes, and outputs the specific water and fertilizer type, candidate application amount, and candidate application timing.

[0075] During training, the optimal irrigation strategy labels from the historical irrigation sample set are used as supervision signals. The training objective is to minimize the weighted sum of three factors: the classification cross-entropy loss between the predicted specific water and fertilizer type and the historical optimal specific water and fertilizer type; the first mean square error between the predicted application amount and the historical optimal application amount; and the second mean square error between the predicted application timing and the historical optimal application timing. The weight coefficients are dynamically set according to the priority given to flower bud differentiation rate and fruit set rate under the current alternate bearing condition. In a high-yield year, the weight of the predicted specific water and fertilizer type and application timing loss is increased to ensure flower bud differentiation; in a low-yield year, the weight of the predicted application amount loss is increased to ensure fruit set rate. The weighted sum is used as the total loss, and the gradient of the total loss is reduced through backpropagation training until convergence is verified, resulting in the trained strategy selection model.

[0076] Finally, the trained flower bud differentiation rate prediction model set, the trained fruit set prediction model set, and the trained strategy selection model are combined to form an alternating irrigation prediction model array. During runtime, each sub-model can be called independently to complete the alternating irrigation prediction.

[0077] In the method of this embodiment of the invention, the phenological water requirement matching degree parameter, the alternate bearing fruit status, and the local meteorological data are input into the flower bud differentiation rate prediction model set in the alternating irrigation prediction model array, and the flower bud differentiation rate prediction model set outputs a fused flower bud differentiation rate, including: Obtain the total number of flower bud differentiation rate prediction branch models in the flower bud differentiation rate prediction model set, and multiply the absolute value of the phenological water requirement matching degree parameter by the total number of flower bud differentiation rate prediction branch models to obtain the initial number of flower bud differentiation branches; When the alternate bearing state is a bumper year, the number of preliminary branches of flower bud differentiation is multiplied by the preset flower bud branch adjustment coefficient for bumper years, and the result is rounded to obtain the number of predicted branches activated for flower bud differentiation. When the alternate bearing state is a low-yield state, the number of preliminary branches of flower bud differentiation is multiplied by the preset low-yield flower bud branch adjustment coefficient, and the result is rounded to obtain the predicted number of activated branches of flower bud differentiation. The high-yield flower bud branch adjustment coefficient is greater than the low-yield flower bud branch adjustment coefficient. Randomly select flower bud differentiation rate prediction branch models from the flower bud differentiation rate prediction model set, with the number of flower bud differentiation rate prediction branch models equal to the number of activated flower bud differentiation rate prediction branches. The selected flower bud differentiation rate prediction branch models output the predicted flower bud differentiation rate respectively. Calculate the arithmetic mean of the predicted flower bud differentiation rates output by all selected flower bud differentiation rate prediction branch models to obtain the fused flower bud differentiation rate.

[0078] In this embodiment of the invention, firstly, the total number of flower bud differentiation rate prediction branch models in the flower bud differentiation rate prediction model set is obtained. Then, the absolute value of the phenological water requirement matching parameter is multiplied by the total number of flower bud differentiation rate prediction branch models, and rounded up to obtain the initial number of flower bud differentiation branches. The initial number of flower bud differentiation branches (rounded up) = total number of flower bud differentiation rate prediction branch models × |phenological water requirement matching parameter|. A larger phenological water requirement matching parameter results in more activated branch models, allowing for the consideration of more model opinions during prediction fusion and improving prediction reliability under extreme conditions.

[0079] Furthermore, when the alternate bearing state is a "bumper year," the number of initial flower bud branches is multiplied by a preset "bumper year" flower bud branch adjustment coefficient, and then rounded to obtain the predicted number of activated flower bud branches. The predicted number of activated flower bud branches (rounded up) = the number of initial flower bud branches × the preset "bumper year" flower bud branch adjustment coefficient.

[0080] The preset adjustment coefficient for flower bud branching in bumper years is a pre-set value. It can be set based on the mean square error of the predicted flower bud differentiation rate under different irrigation strategies in historical data. The larger the mean square error, the larger the value, in order to increase the number of branches participating in the fusion. For example: obtain a historical bumper year irrigation sample set, which consists of all samples from the historical alternate bearing sample set that are in a bumper year state; obtain a preset set of candidate adjustment coefficients, which contains multiple different candidate adjustment coefficient values.

[0081] For each candidate adjustment coefficient value in the candidate adjustment coefficient set, the following calculations are performed: The candidate adjustment coefficient value is used as the temporary flower bud branch adjustment coefficient. The number of activated temporary flower bud branches is calculated based on the temporary flower bud branch adjustment coefficient. A flower bud differentiation rate prediction branch model equal to the number of activated temporary flower bud branches is randomly selected from the flower bud differentiation rate prediction model set. The selected flower bud differentiation rate prediction branch model processes the samples in the historical high-yield irrigation sample set and outputs the predicted flower bud differentiation rate. The arithmetic mean of the predicted flower bud differentiation rates output by all selected flower bud differentiation rate prediction branch models is calculated to obtain the temporary fused flower bud differentiation rate. The mean square error between this rate and the historical flower bud differentiation rate label value of the corresponding sample in the historical high-yield irrigation sample set is calculated and used as the prediction error corresponding to the candidate adjustment coefficient value. The prediction errors corresponding to all candidate adjustment coefficient values ​​are compared, and the candidate adjustment coefficient value with the smallest prediction error is used as the high-yield flower bud branch adjustment coefficient.

[0082] Furthermore, when the fruit-bearing state is a low-yield year, the number of initial flower bud branches is multiplied by a preset low-yield flower bud branch adjustment coefficient, and rounded to obtain the number of flower bud differentiation predicted branch activations. The number of flower bud differentiation predicted branch activations (rounded up) = number of initial flower bud branches × preset low-yield flower bud branch adjustment coefficient. Since a high-yield year requires activating more flower bud branches for more accurate flower bud differentiation prediction, the high-yield flower bud branch adjustment coefficient is greater than the low-yield flower bud branch adjustment coefficient. The preset low-yield flower bud branch adjustment coefficient is set similarly to the preset high-yield flower bud branch adjustment coefficient, based on the predicted mean square error of the flower bud differentiation rate in historical data under low-yield conditions. The smaller the mean square error, the smaller the value. For example, on a historical sample set under low-yield conditions, the sample mean square error of the flower bud differentiation rate output by all branch models can be calculated for each sample input, and the candidate adjustment coefficient value with the smallest prediction error can be used as the low-yield flower bud branch adjustment coefficient.

[0083] Furthermore, a number of flower bud differentiation rate prediction branch models, equal to the number of activated flower bud differentiation prediction branches, are randomly selected from the set of flower bud differentiation rate prediction models. Each of the selected flower bud differentiation rate prediction branch models outputs its predicted flower bud differentiation rate. The arithmetic mean of the predicted flower bud differentiation rates output by all selected flower bud differentiation rate prediction branch models is then calculated to obtain the fused flower bud differentiation rate. Here, random selection involves randomly drawing a specified number of branch models from the set without replacement to ensure that different combinations of sub-models may be used for each prediction.

[0084] In the method of this invention embodiment, the phenological water requirement matching degree parameter, the alternate bearing fruit status, and the local meteorological data are input into the fruit set rate prediction model set in the alternating irrigation prediction model array, and the fruit set rate prediction model set outputs the fused fruit set rate, including: Obtain the total number of fruit set rate prediction branch models in the set of fruit set rate prediction models, and multiply the absolute value of the phenological period water requirement matching degree parameter by the total number of fruit set rate prediction branch models to obtain the initial number of fruit set rate branches. When the alternate bearing state is a bumper year, the number of preliminary branches of the fruit set rate is multiplied by the preset adjustment coefficient for the fruit set branches of a bumper year, and the result is rounded to obtain the number of predicted branches activated for the fruit set rate. When the alternate bearing state is a low-yield state, the number of preliminary branches of the fruit setting rate is multiplied by the preset low-yield fruit setting branch adjustment coefficient, and the result is rounded to obtain the number of predicted branches activated for the fruit setting rate. The low-yield fruit setting branch adjustment coefficient is greater than the high-yield fruit setting branch adjustment coefficient. Randomly select a number of fruit set rate prediction branch models from the set of fruit set rate prediction models that are equal to the number of activated fruit set rate prediction branches. The selected fruit set rate prediction branch models output the predicted fruit set rate respectively. Calculate the arithmetic mean of the predicted fruit set rates output by all selected fruit set rate prediction branch models to obtain the fused fruit set rate.

[0085] In this embodiment of the invention, firstly, the total number of fruit set rate prediction branch models in the fruit set rate prediction model set is obtained. Then, the absolute value of the phenological period water requirement matching degree parameter is multiplied by the total number of fruit set rate prediction branch models to obtain the preliminary number of fruit set rate branches. The preliminary number of fruit set rate branches = |phenological period water requirement matching degree parameter| × the total number of fruit set rate prediction branch models.

[0086] Furthermore, when the fruit-bearing state is a "bumper year," the initial number of branches for fruit set rate is multiplied by a preset adjustment coefficient for fruit set branches in a "bumper year," and the result is rounded up to obtain the number of activated branches for predicting fruit set rate. The number of activated branches for predicting fruit set rate (rounded up) = the initial number of branches for fruit set rate × the preset adjustment coefficient for fruit set branches in a "bumper year." The preset adjustment coefficient for fruit set branches in a "bumper year" can be set based on the prediction mean square error of the historical data output by each branch model in the fruit set prediction model set under the "bumper year" condition. The sample mean square error of the fruit set rate output by each branch model is calculated for each sample in the historical sample set under the "bumper year" condition; finally, the minimum value of each sample variance is calculated according to its respective state and used as the preset adjustment coefficient for fruit set branches in a "bumper year."

[0087] Furthermore, when the fruit-bearing state is a "short year," the initial number of branches for fruit set rate is multiplied by a preset adjustment coefficient for short year fruit set branches, and the result is rounded up to obtain the number of activated branches for fruit set rate prediction. The number of activated branches for fruit set rate prediction (rounded up) = the initial number of branches for fruit set rate × the preset adjustment coefficient for short year fruit set branches. The preset adjustment coefficient for short year fruit set branches can be set based on the historical mean square error of the prediction results of each branch model in the fruit set rate prediction model set under the short year state. The larger the mean square error, the larger the value, to enhance the stability of the integrated prediction. Similarly, the mean square error of the fruit set rate sample output of each branch model is calculated for each sample in the historical sample set under the short year state. Finally, the minimum value of the calculated mean square error of each sample prediction is taken according to its state, and this minimum value is used as the preset short year fruit set branch coefficient.

[0088] In particular, since more fruit-setting branches need to be activated in a low-yield year to make a more accurate prediction of the fruit-setting rate, the adjustment coefficient of the fruit-setting branches in a low-yield year is greater than that in a high-yield year.

[0089] Furthermore, a number of fruit set prediction branch models are randomly selected from the set of fruit set prediction models, equal to the number of activated fruit set prediction branches. Each selected fruit set prediction branch model outputs a predicted fruit set rate, with each selected model outputting one predicted fruit set rate. The arithmetic mean of the predicted fruit set rates output by all selected fruit set prediction branch models is then calculated to obtain the fused fruit set rate.

[0090] In this embodiment of the invention, the deviation of flower bud differentiation and the deviation of fruit set are calculated separately. The alternate bearing state is determined by comparing the deviation with a preset high-yield deviation, reflecting the overall high-yield level of the camellia oleifera tree in the previous cycle from two dimensions. This avoids misjudgment based on a single dimension and provides a physiological basis for the differentiated optimization of irrigation strategies. Furthermore, a set of flower bud differentiation rate prediction models, a set of trained fruit set prediction models, and a combination of trained strategy selection models are constructed to form an alternating irrigation prediction model array. Then, the alternating irrigation strategy is output through the alternating irrigation prediction model array, and the optimization target is adaptively adjusted based on the alternate bearing state. Simultaneously, the flower bud differentiation rate prediction model set outputs a fused flower bud differentiation rate; the fruit set prediction model set outputs a fused fruit set rate. Through dynamic activation and random selection, adaptive adjustment is achieved, improving the accuracy of the fused flower bud differentiation rate and fused fruit set rate predictions.

[0091] S50: According to the alternating irrigation strategy, the camellia oleifera planting area is subjected to alternating application of water and fertilizer.

[0092] In this embodiment of the invention, the alternating application of water and fertilizer is based on the strategy output by the model, which alternates between water and fertilizer that promotes flowering in good years and water and fertilizer that promotes fruit growth in bad years at different time points or different phenological stages.

[0093] Specifically, the output alternating irrigation strategy will be used to precisely irrigate and fertilize the camellia planting area through integrated water and fertilizer irrigation equipment, such as drip irrigation systems and sprinkler irrigation systems with fertilizer pumps, according to the specified special water and fertilizer types, application amounts and application times.

[0094] In this embodiment of the invention, alternating water and fertilizer application is used to meet the differentiated physiological needs under different alternate bearing conditions, and to avoid the decline in physiological indicators caused by a single type of water and fertilizer.

[0095] Through the above specific implementation methods, the embodiments of the present invention achieve the following technical effects: In this embodiment of the invention, local meteorological data is first obtained through local meteorological stations in the camellia oleifera planting area; the current phenological stage is obtained according to the growth and development stage of the camellia oleifera trees; at the same time, soil moisture sensors are deployed to detect soil moisture data, providing sufficient raw data for subsequent processing.

[0096] Secondly, the phenological period encoding vector and the meteorological feature vector are concatenated to obtain the phenological period-meteorological joint feature vector. This vector is then input into the ideal soil moisture prediction model for prediction, which solves the problem that the ideal soil moisture cannot be mapped and determined under different meteorological conditions, and improves the accuracy and response rate of the prediction. An ideal soil moisture range is constructed, and the soil moisture data is compared with the ideal soil moisture range to obtain the phenological period water demand matching degree parameter, which provides the urgency and direction of subsequent irrigation.

[0097] Furthermore, the deviation of flower bud differentiation and the deviation of fruit setting are calculated separately. The fruit setting status of alternate bearing is determined by comparing the preset high yield deviation with the deviation. The overall high yield of the camellia tree in the previous cycle is reflected from two dimensions, avoiding misjudgment from a single dimension and providing physiological state basis for the differentiated optimization of irrigation strategies.

[0098] Furthermore, a set of flower bud differentiation rate prediction models, a set of trained fruit set prediction models, and a combination of trained strategy optimization models are constructed to form an alternating irrigation prediction model array. Subsequently, the alternating irrigation prediction model array outputs an alternating irrigation strategy, adaptively adjusting the optimization objective based on alternate bearing conditions. Simultaneously, the flower bud differentiation rate prediction model set outputs a fused flower bud differentiation rate, and the fruit set rate prediction model set outputs a fused fruit set rate. Through dynamic activation and random selection, adaptive adjustment is achieved, improving the accuracy of the fused flower bud differentiation rate and fused fruit set rate predictions.

[0099] Ultimately, by alternating water and fertilizer application, the differentiated physiological needs under different alternate bearing conditions can be met, avoiding the decline in physiological indicators caused by a single type of water and fertilizer.

[0100] Example 2, as Figure 4 As shown, based on the same inventive concept as the big data-driven dual-specific water and fertilizer alternating irrigation method for mountain camellia oleifera provided in Embodiment 1, this embodiment of the invention also provides a big data-driven dual-specific water and fertilizer alternating irrigation system for mountain camellia oleifera, the system comprising: The raw data acquisition module 11 is used to acquire local meteorological data, current phenological period labels, and soil moisture data of the camellia oleifera planting area; The matching degree parameter calculation module 12 is used to determine the ideal soil moisture range of the camellia planting area in the current phenological period based on the current phenological period label, and compare the soil moisture data with the ideal soil moisture range to calculate the phenological period water requirement matching degree parameter; The alternate bearing fruit status judgment module 13 is used to determine the current alternate bearing fruit status of the camellia oleifera planting area based on the historical flower bud differentiation rate data and historical fruit setting rate data of the camellia oleifera planting area. The alternating irrigation strategy output module 14 is used to input the phenological water requirement matching degree parameter, the alternate bearing status and the local meteorological data into the pre-trained alternating irrigation prediction model array, and dynamically output the alternating irrigation strategy with flower bud differentiation rate and fruit setting rate as dual optimization objectives. The alternating irrigation strategy includes special water and fertilizer type, application amount and application time. The special water and fertilizer type includes flower-promoting water and fertilizer in high-yield years and fruit-strengthening water and fertilizer in low-yield years. The alternating water and fertilizer application module 15 is used to apply alternating water and fertilizer to the camellia oleifera planting area according to the alternating irrigation strategy. In one embodiment, the raw data acquisition module 11 is used for: Acquire local meteorological data collected by local meteorological stations deployed in camellia oleifera planting areas, wherein the local meteorological data includes at least rainfall data, air temperature data, and air humidity data; Obtain the current phenological stage label of the camellia oleifera planting area, wherein the current phenological stage label is one of the following: flower bud differentiation stage label, fruit enlargement stage label, oil conversion stage label, or dormancy stage label; Obtain soil moisture data collected by soil moisture sensors deployed in the camellia oleifera planting area.

[0101] In one embodiment, the matching degree parameter calculation module 12 is used for: The current phenological period label is subjected to one-hot encoding to obtain the phenological period encoding vector; The rainfall, air temperature, and air humidity data in the local meteorological data are normalized to obtain normalized rainfall, normalized air temperature, and normalized air humidity values. The normalized rainfall, normalized air temperature, and normalized air humidity values ​​are then combined to form a meteorological feature vector. The phenological period encoding vector and the meteorological feature vector are concatenated along the channel dimension to obtain the phenological period-meteorological joint feature vector; The phenological-meteorological joint feature vector is input into the pre-trained ideal soil moisture prediction model, and the predicted ideal soil moisture value and the predicted soil moisture standard deviation are output. Subtracting the product of the predicted soil moisture standard deviation and 1.5 from the predicted ideal soil moisture value yields the lower limit of ideal soil moisture. Adding the product of the predicted soil moisture standard deviation and 1.5 to the predicted ideal soil moisture value yields the upper limit of ideal soil moisture. The closed interval formed by the lower limit of ideal soil moisture and the upper limit of ideal soil moisture is defined as the ideal soil moisture interval; When the soil moisture data is less than the ideal soil moisture lower limit, the difference between the ideal soil moisture lower limit and the soil moisture data is calculated, and the difference is divided by the ideal soil moisture lower limit to obtain the phenological period water demand matching degree parameter. When the soil moisture data is greater than the upper limit of the ideal soil moisture, the difference between the soil moisture data and the upper limit of the ideal soil moisture is calculated, the difference is divided by the upper limit of the ideal soil moisture to obtain the phenological water demand matching degree parameter, and the phenological water demand matching degree parameter is negative. When the soil moisture data is greater than or equal to the lower limit of the ideal soil moisture and less than or equal to the upper limit of the ideal soil moisture, the phenological period water requirement matching degree parameter is set to zero.

[0102] The pre-training process of the ideal soil moisture prediction model includes: A historical phenological period-meteorological-soil moisture sample set is obtained. Each historical phenological period-meteorological-soil moisture sample set contains a historical phenological period label, a set of historical meteorological data, and a historical optimal soil moisture label set. The historical optimal soil moisture label set contains a first optimal soil moisture value and a second optimal soil moisture value. The first optimal soil moisture value is the soil moisture data recorded when the highest historical flower bud differentiation rate is obtained under the same historical phenological period label, and the second optimal soil moisture value is the soil moisture data recorded when the highest historical fruit setting rate is obtained under the same historical phenological period label. All historical phenological period labels are subjected to unique thermal encoding to obtain historical phenological period encoding vectors. All historical meteorological data are normalized and combined to form historical meteorological feature vectors. The historical phenological period encoding vectors and historical meteorological feature vectors are concatenated to form a training input sample set. The historical optimal soil moisture label set is used to form a training label set. An initial ideal soil moisture prediction model is constructed, which includes a phenological period coding layer, a meteorological feature coding layer, a feature splicing layer, and a regression output layer. The initial ideal soil moisture prediction model is trained in a supervised manner using the training input sample set and the training label set. The training objective is to minimize the weighted sum of the first mean square error between the predicted ideal soil moisture value and the first optimal soil moisture value output by the initial ideal soil moisture prediction model and the second mean square error between the predicted ideal soil moisture value and the second optimal soil moisture value, until the weighted sum meets the preset convergence condition, thus obtaining the trained ideal soil moisture prediction model.

[0103] In one embodiment, the alternate bearing fruit status determination module 13 is used for: Obtain historical flower bud differentiation rate data and historical fruit setting rate data of the tea oleifera planting area in the previous growth cycle; Obtain the historical flower bud differentiation rate data set and historical fruit setting rate data set of the tea oleifera planting area over the past multiple consecutive growth cycles; Calculate the mean of the historical flower bud differentiation rate data set and the mean of the historical fruit setting rate data set. Use the mean of the historical flower bud differentiation rate data set as the benchmark value for high yield due to flower bud differentiation and the mean of the historical fruit setting rate data set as the benchmark value for high yield due to fruit setting. The ratio of the historical flower bud differentiation rate data in the previous growth cycle to the benchmark value of high yield due to flower bud differentiation is calculated to obtain the flower bud differentiation deviation. The ratio of the historical fruit setting rate data in the previous growth cycle to the benchmark value of high yield due to fruit setting is calculated to obtain the fruit setting deviation. When both the flower bud differentiation deviation and the fruit setting deviation are greater than the preset high-yield deviation threshold, the current alternate bearing state of the camellia oleifera planting area is determined to be a high-yield state. When at least one of the flower bud differentiation deviation and the fruit setting deviation is less than or equal to the preset high-yield deviation threshold, the current alternate bearing state of the camellia oleifera planting area is determined to be an alternate bearing state.

[0104] In one embodiment, the alternating irrigation strategy output module 14 is used for: The phenological water requirement matching degree parameter, the alternate bearing fruit status and the local meteorological data are input into the flower bud differentiation rate prediction model set in the alternating irrigation prediction model array, and the flower bud differentiation rate prediction model set is output by the fused flower bud differentiation rate; The phenological water requirement matching degree parameter, the alternate bearing fruit status and the local meteorological data are input into the fruit set rate prediction model set in the alternating irrigation prediction model array, and the fruit set rate prediction model set outputs the fused fruit set rate. The phenological water requirement matching degree parameter, the alternate bearing state and the local meteorological data are input into the strategy selection model in the alternating irrigation prediction model array. The strategy selection model outputs multiple sets of candidate alternating irrigation strategies. Each set of candidate alternating irrigation strategies includes candidate special water and fertilizer type, candidate application amount and candidate application time. Using the maximum weighted harmonic function value of the fusion flower bud differentiation rate and the fusion fruit setting rate as the selection objective, a candidate alternating irrigation strategy is selected from the multiple candidate alternating irrigation strategies as the alternating irrigation strategy.

[0105] The pre-training process of the alternating irrigation prediction model array includes: Obtain a historical irrigation sample set. Each historical irrigation sample set includes a historical phenological period water requirement matching degree parameter, a historical alternate bearing status, a set of historical meteorological data, a historical flower bud differentiation rate label value, a historical fruit setting rate label value, and a historical optimal irrigation strategy label. The historical optimal irrigation strategy label includes the historical optimal special water and fertilizer type, the historical optimal application amount, and the historical optimal application time. An initial flower bud differentiation rate prediction model set is constructed, which includes multiple initial flower bud differentiation rate prediction branch models. Each initial flower bud differentiation rate prediction branch model is trained in a supervised manner using the historical irrigation sample set. The training objective is to minimize the mean square error between the output predicted flower bud differentiation rate and the historical flower bud differentiation rate label value until the verification convergence is obtained, thus obtaining the trained flower bud differentiation rate prediction model set. Construct an initial fruit set prediction model set, which includes multiple initial fruit set prediction branch models. Use the historical irrigation sample set to perform supervised training on each initial fruit set prediction branch model. The training objective is to minimize the mean square error between the output predicted fruit set rate and the historical fruit set rate label value until the verification convergence is obtained, thus obtaining the trained fruit set prediction model set. An initial strategy selection model is constructed using a multilayer perceptron structure. The initial strategy selection model is then trained in a supervised manner using the historical irrigation sample set. The training objective is to minimize the weighted sum of the classification cross-entropy loss between the predicted special water and fertilizer type and the historical optimal special water and fertilizer type, the first mean square error between the predicted application amount and the historical optimal application amount, and the second mean square error between the predicted application timing and the historical optimal application timing, until the model is verified to converge, thus obtaining the trained strategy selection model. The trained flower bud differentiation rate prediction model set, the trained fruit set prediction model set, and the trained strategy selection model are combined to form an alternating irrigation prediction model array.

[0106] Specifically, the phenological water requirement matching degree parameter, the alternate bearing fruit status, and the local meteorological data are input into the flower bud differentiation rate prediction model set in the alternating irrigation prediction model array. The flower bud differentiation rate prediction model set outputs a fused flower bud differentiation rate, including: Obtain the total number of flower bud differentiation rate prediction branch models in the flower bud differentiation rate prediction model set, and multiply the absolute value of the phenological water requirement matching degree parameter by the total number of flower bud differentiation rate prediction branch models to obtain the initial number of flower bud differentiation branches; When the alternate bearing state is a bumper year, the number of preliminary branches of flower bud differentiation is multiplied by the preset flower bud branch adjustment coefficient for bumper years, and the result is rounded to obtain the number of predicted branches activated for flower bud differentiation. When the alternate bearing state is a low-yield state, the number of preliminary branches of flower bud differentiation is multiplied by the preset low-yield flower bud branch adjustment coefficient, and the result is rounded to obtain the predicted number of activated branches of flower bud differentiation. The high-yield flower bud branch adjustment coefficient is greater than the low-yield flower bud branch adjustment coefficient. Randomly select flower bud differentiation rate prediction branch models from the flower bud differentiation rate prediction model set, with the number of flower bud differentiation rate prediction branch models equal to the number of activated flower bud differentiation rate prediction branches. The selected flower bud differentiation rate prediction branch models output the predicted flower bud differentiation rate respectively. Calculate the arithmetic mean of the predicted flower bud differentiation rates output by all selected flower bud differentiation rate prediction branch models to obtain the fused flower bud differentiation rate.

[0107] Specifically, the phenological water requirement matching degree parameter, the alternate bearing fruit status, and the local meteorological data are input into the fruit set rate prediction model set in the alternating irrigation prediction model array. The fruit set rate prediction model set outputs a fused fruit set rate, including: Obtain the total number of fruit set rate prediction branch models in the set of fruit set rate prediction models, and multiply the absolute value of the phenological period water requirement matching degree parameter by the total number of fruit set rate prediction branch models to obtain the initial number of fruit set rate branches. When the alternate bearing state is a bumper year, the number of preliminary branches of the fruit set rate is multiplied by the preset adjustment coefficient for the fruit set branches of a bumper year, and the result is rounded to obtain the number of predicted branches activated for the fruit set rate. When the alternate bearing state is a low-yield state, the number of preliminary branches of the fruit setting rate is multiplied by the preset low-yield fruit setting branch adjustment coefficient, and the result is rounded to obtain the number of predicted branches activated for the fruit setting rate. The low-yield fruit setting branch adjustment coefficient is greater than the high-yield fruit setting branch adjustment coefficient. Randomly select a number of fruit set rate prediction branch models from the set of fruit set rate prediction models that are equal to the number of activated fruit set rate prediction branches. The selected fruit set rate prediction branch models output the predicted fruit set rate respectively. Calculate the arithmetic mean of the predicted fruit set rates output by all selected fruit set rate prediction branch models to obtain the fused fruit set rate.

[0108] Compared to existing technologies, this invention first acquires local meteorological data through local meteorological stations in the camellia oleifera planting area; obtains the current phenological stage label based on the growth and development stage of the camellia oleifera trees; and deploys soil moisture sensors to detect soil moisture data, providing sufficient raw data for subsequent processing. Secondly, the phenological stage encoding vector and meteorological feature vector are concatenated to obtain a phenological-meteorological joint feature vector; this vector is then input into an ideal soil moisture prediction model for prediction, solving the defect that ideal soil moisture cannot be accurately mapped under different meteorological conditions, thus improving the accuracy and response rate of prediction. An ideal soil moisture range is constructed, and the soil moisture data is compared with the ideal soil moisture range to obtain the phenological stage water requirement matching degree parameter, providing information on the urgency and direction of subsequent irrigation. Furthermore, the deviation degree of flower bud differentiation and fruit setting is calculated separately, and the alternate bearing status is determined by comparing the preset high-yield deviation with this deviation, reflecting the overall high-yield level of the camellia oleifera trees in the previous cycle from two dimensions, avoiding misjudgment based on a single dimension, and providing a physiological basis for differentiated optimization of irrigation strategies. Furthermore, a set of flower bud differentiation rate prediction models, a set of trained fruit set prediction models, and a combination of trained strategy optimization models were constructed to form an alternating irrigation prediction model array. Subsequently, the alternating irrigation prediction model array was used to output an alternating irrigation strategy, adaptively adjusting the optimization objective based on alternate bearing conditions. Simultaneously, the flower bud differentiation rate prediction model set outputs a fused flower bud differentiation rate, and the fruit set rate prediction model set outputs a fused fruit set rate. Through dynamic activation and random selection, adaptive adjustment was achieved, improving the accuracy of the fused flower bud differentiation rate and fused fruit set rate predictions. Finally, through alternating water and fertilizer application, the differentiated physiological needs under different alternate bearing conditions were met, avoiding the decline in physiological indicators caused by a single water and fertilizer type.

Claims

1. A big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas, characterized in that: The method includes: Acquire local meteorological data, current phenological stage labels, and soil moisture data for camellia oleifera planting areas; Based on the current phenological period label, determine the ideal soil moisture range of the camellia oleifera planting area during the current phenological period, and compare the soil moisture data with the ideal soil moisture range to calculate the phenological period water requirement matching parameter. Based on the historical flower bud differentiation rate data and historical fruit setting rate data of the camellia oleifera planting area, the current alternate bearing status of the camellia oleifera planting area is determined; The phenological water requirement matching parameters, the alternate bearing status of alternate bearing and the local meteorological data are input into the pre-trained alternating irrigation prediction model array. With flower bud differentiation rate and fruit setting rate as dual optimization objectives, the alternating irrigation strategy is dynamically output. The alternating irrigation strategy includes special water and fertilizer type, application amount and application time. The special water and fertilizer type includes flower-promoting water and fertilizer in high-yield years and fruit-strengthening water and fertilizer in low-yield years. According to the alternating irrigation strategy, the camellia oleifera planting area is subjected to alternating application of water and fertilizer.

2. The big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas according to claim 1, characterized in that, Obtain local meteorological data, current phenological stage labels, and soil moisture data for the camellia oleifera planting area, including: Acquire local meteorological data collected by local meteorological stations deployed in camellia oleifera planting areas, wherein the local meteorological data includes at least rainfall data, air temperature data, and air humidity data; Obtain the current phenological stage label of the camellia oleifera planting area, wherein the current phenological stage label is one of the following: flower bud differentiation stage label, fruit enlargement stage label, oil conversion stage label, or dormancy stage label; Obtain soil moisture data collected by soil moisture sensors deployed in the camellia oleifera planting area.

3. The big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas according to claim 1, characterized in that, Based on the current phenological stage label, determine the ideal soil moisture range for the camellia oleifera planting area during the current phenological stage, including: The current phenological period label is subjected to one-hot encoding to obtain the phenological period encoding vector; The rainfall, air temperature, and air humidity data in the local meteorological data are normalized to obtain normalized rainfall, normalized air temperature, and normalized air humidity values. The normalized rainfall, normalized air temperature, and normalized air humidity values ​​are then combined to form a meteorological feature vector. The phenological period encoding vector and the meteorological feature vector are concatenated along the channel dimension to obtain the phenological period-meteorological joint feature vector; The phenological-meteorological joint feature vector is input into the pre-trained ideal soil moisture prediction model, and the predicted ideal soil moisture value and the predicted soil moisture standard deviation are output. Subtracting the product of the predicted soil moisture standard deviation and 1.5 from the predicted ideal soil moisture value yields the lower limit of ideal soil moisture. Adding the product of the predicted soil moisture standard deviation and 1.5 to the predicted ideal soil moisture value yields the upper limit of ideal soil moisture. The closed interval formed by the lower limit of ideal soil moisture and the upper limit of ideal soil moisture is defined as the ideal soil moisture interval; When the soil moisture data is less than the ideal soil moisture lower limit, the difference between the ideal soil moisture lower limit and the soil moisture data is calculated, and the difference is divided by the ideal soil moisture lower limit to obtain the phenological period water demand matching degree parameter. When the soil moisture data is greater than the upper limit of the ideal soil moisture, the difference between the soil moisture data and the upper limit of the ideal soil moisture is calculated, the difference is divided by the upper limit of the ideal soil moisture to obtain the phenological water demand matching degree parameter, and the phenological water demand matching degree parameter is negative. When the soil moisture data is greater than or equal to the lower limit of the ideal soil moisture and less than or equal to the upper limit of the ideal soil moisture, the phenological period water requirement matching degree parameter is set to zero.

4. The big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas according to claim 3, characterized in that, The pre-training process of the ideal soil moisture prediction model includes: A historical phenological period-meteorological-soil moisture sample set is obtained. Each historical phenological period-meteorological-soil moisture sample set contains a historical phenological period label, a set of historical meteorological data, and a historical optimal soil moisture label set. The historical optimal soil moisture label set contains a first optimal soil moisture value and a second optimal soil moisture value. The first optimal soil moisture value is the soil moisture data recorded when the highest historical flower bud differentiation rate is obtained under the same historical phenological period label, and the second optimal soil moisture value is the soil moisture data recorded when the highest historical fruit setting rate is obtained under the same historical phenological period label. All historical phenological period labels are subjected to unique thermal encoding to obtain historical phenological period encoding vectors. All historical meteorological data are normalized and combined to form historical meteorological feature vectors. The historical phenological period encoding vectors and historical meteorological feature vectors are concatenated to form a training input sample set. The historical optimal soil moisture label set is used to form a training label set. An initial ideal soil moisture prediction model is constructed, which includes a phenological period coding layer, a meteorological feature coding layer, a feature splicing layer, and a regression output layer. The initial ideal soil moisture prediction model is trained in a supervised manner using the training input sample set and the training label set. The training objective is to minimize the weighted sum of the first mean square error between the predicted ideal soil moisture value and the first optimal soil moisture value output by the initial ideal soil moisture prediction model and the second mean square error between the predicted ideal soil moisture value and the second optimal soil moisture value, until the weighted sum meets the preset convergence condition, thus obtaining the trained ideal soil moisture prediction model.

5. The big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas according to claim 1, characterized in that, Based on historical flower bud differentiation rate data and historical fruit set rate data of the camellia oleifera planting area, the current alternate bearing status of the camellia oleifera planting area is determined, including: Obtain historical flower bud differentiation rate data and historical fruit setting rate data of the tea oleifera planting area in the previous growth cycle; Obtain the historical flower bud differentiation rate data set and historical fruit setting rate data set of the tea oleifera planting area over the past multiple consecutive growth cycles; Calculate the mean of the historical flower bud differentiation rate data set and the mean of the historical fruit setting rate data set. Use the mean of the historical flower bud differentiation rate data set as the benchmark value for high yield due to flower bud differentiation and the mean of the historical fruit setting rate data set as the benchmark value for high yield due to fruit setting. The ratio of the historical flower bud differentiation rate data in the previous growth cycle to the benchmark value of high yield due to flower bud differentiation is calculated to obtain the flower bud differentiation deviation. The ratio of the historical fruit setting rate data in the previous growth cycle to the benchmark value of high yield due to fruit setting is calculated to obtain the fruit setting deviation. When both the flower bud differentiation deviation and the fruit setting deviation are greater than the preset high-yield deviation threshold, the current alternate bearing state of the camellia oleifera planting area is determined to be a high-yield state. When at least one of the flower bud differentiation deviation and the fruit setting deviation is less than or equal to the preset high-yield deviation threshold, the current alternate bearing state of the camellia oleifera planting area is determined to be an alternate bearing state.

6. The big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas according to claim 1, characterized in that, The phenological water requirement matching parameters, the alternate bearing fruit status, and the local meteorological data are input into a pre-trained alternating irrigation prediction model array. With flower bud differentiation rate and fruit set rate as dual optimization objectives, the alternating irrigation strategy is dynamically output, including: The phenological water requirement matching degree parameter, the alternate bearing fruit status and the local meteorological data are input into the flower bud differentiation rate prediction model set in the alternating irrigation prediction model array, and the flower bud differentiation rate prediction model set is output by the fused flower bud differentiation rate; The phenological water requirement matching degree parameter, the alternate bearing fruit status and the local meteorological data are input into the fruit set rate prediction model set in the alternating irrigation prediction model array, and the fruit set rate prediction model set outputs the fused fruit set rate. The phenological water requirement matching degree parameter, the alternate bearing state and the local meteorological data are input into the strategy selection model in the alternating irrigation prediction model array. The strategy selection model outputs multiple sets of candidate alternating irrigation strategies. Each set of candidate alternating irrigation strategies includes candidate special water and fertilizer type, candidate application amount and candidate application time. Using the maximum weighted harmonic function value of the fusion flower bud differentiation rate and the fusion fruit setting rate as the selection objective, a candidate alternating irrigation strategy is selected from the multiple candidate alternating irrigation strategies as the alternating irrigation strategy.

7. The big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas according to claim 6, characterized in that, The pre-training process for the alternating irrigation prediction model array includes: Obtain a historical irrigation sample set. Each historical irrigation sample set includes a historical phenological period water requirement matching degree parameter, a historical alternate bearing status, a set of historical meteorological data, a historical flower bud differentiation rate label value, a historical fruit setting rate label value, and a historical optimal irrigation strategy label. The historical optimal irrigation strategy label includes the historical optimal special water and fertilizer type, the historical optimal application amount, and the historical optimal application time. An initial flower bud differentiation rate prediction model set is constructed, which includes multiple initial flower bud differentiation rate prediction branch models. Each initial flower bud differentiation rate prediction branch model is trained in a supervised manner using the historical irrigation sample set. The training objective is to minimize the mean square error between the output predicted flower bud differentiation rate and the historical flower bud differentiation rate label value until the verification convergence is obtained, thus obtaining the trained flower bud differentiation rate prediction model set. Construct an initial fruit set prediction model set, which includes multiple initial fruit set prediction branch models. Use the historical irrigation sample set to perform supervised training on each initial fruit set prediction branch model. The training objective is to minimize the mean square error between the output predicted fruit set rate and the historical fruit set rate label value until the verification convergence is obtained, thus obtaining the trained fruit set prediction model set. An initial strategy selection model is constructed using a multilayer perceptron structure. The initial strategy selection model is then trained in a supervised manner using the historical irrigation sample set. The training objective is to minimize the weighted sum of the classification cross-entropy loss between the predicted special water and fertilizer type and the historical optimal special water and fertilizer type, the first mean square error between the predicted application amount and the historical optimal application amount, and the second mean square error between the predicted application timing and the historical optimal application timing, until the model is verified to converge, thus obtaining the trained strategy selection model. The trained flower bud differentiation rate prediction model set, the trained fruit set prediction model set, and the trained strategy selection model are combined to form an alternating irrigation prediction model array.

8. The big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas according to claim 6, characterized in that, The phenological water requirement matching degree parameter, the alternate bearing fruit status, and the local meteorological data are input into the flower bud differentiation rate prediction model set in the alternating irrigation prediction model array. The flower bud differentiation rate prediction model set outputs a fused flower bud differentiation rate, including: Obtain the total number of flower bud differentiation rate prediction branch models in the flower bud differentiation rate prediction model set, and multiply the absolute value of the phenological water requirement matching degree parameter by the total number of flower bud differentiation rate prediction branch models to obtain the initial number of flower bud differentiation branches; When the alternate bearing state is a bumper year, the number of preliminary branches of flower bud differentiation is multiplied by the preset flower bud branch adjustment coefficient for bumper years, and the result is rounded to obtain the number of predicted branches activated for flower bud differentiation. When the alternate bearing state is a low-yield state, the number of preliminary branches of flower bud differentiation is multiplied by the preset low-yield flower bud branch adjustment coefficient, and the result is rounded to obtain the predicted number of activated branches of flower bud differentiation. The high-yield flower bud branch adjustment coefficient is greater than the low-yield flower bud branch adjustment coefficient. Randomly select flower bud differentiation rate prediction branch models from the flower bud differentiation rate prediction model set, with the number of flower bud differentiation rate prediction branch models equal to the number of activated flower bud differentiation rate prediction branches. The selected flower bud differentiation rate prediction branch models output the predicted flower bud differentiation rate respectively. Calculate the arithmetic mean of the predicted flower bud differentiation rates output by all selected flower bud differentiation rate prediction branch models to obtain the fused flower bud differentiation rate.

9. The big data-driven method for alternating water and fertilizer irrigation for camellia oleifera in mountainous areas according to claim 6, characterized in that, The phenological water requirement matching degree parameter, the alternate bearing fruit status, and the local meteorological data are input into the fruit set rate prediction model set in the alternating irrigation prediction model array. The fruit set rate prediction model set outputs a fused fruit set rate, including: Obtain the total number of fruit set rate prediction branch models in the set of fruit set rate prediction models, and multiply the absolute value of the phenological period water requirement matching degree parameter by the total number of fruit set rate prediction branch models to obtain the initial number of fruit set rate branches. When the alternate bearing state is a bumper year, the number of preliminary branches of the fruit set rate is multiplied by the preset adjustment coefficient for the fruit set branches of a bumper year, and the result is rounded to obtain the number of predicted branches activated for the fruit set rate. When the alternate bearing state is a low-yield state, the number of preliminary branches of the fruit setting rate is multiplied by the preset low-yield fruit setting branch adjustment coefficient, and the result is rounded to obtain the number of predicted branches activated for the fruit setting rate. The low-yield fruit setting branch adjustment coefficient is greater than the high-yield fruit setting branch adjustment coefficient. Randomly select a number of fruit set rate prediction branch models from the set of fruit set rate prediction models that are equal to the number of activated fruit set rate prediction branches. The selected fruit set rate prediction branch models output the predicted fruit set rate respectively. Calculate the arithmetic mean of the predicted fruit set rates output by all selected fruit set rate prediction branch models to obtain the fused fruit set rate.

10. A big data-driven dual-purpose water and fertilizer alternating irrigation system for camellia oleifera in mountainous areas, characterized in that: For implementing the big data-driven dual-specific water and fertilizer alternating irrigation method for mountain camellia oleifera as described in any one of claims 1-9, the system comprises: The raw data acquisition module is used to acquire local meteorological data, current phenological period labels, and soil moisture data of the camellia oleifera planting area; The matching degree parameter calculation module is used to determine the ideal soil moisture range of the camellia oleifera planting area in the current phenological period based on the current phenological period label, and compare the soil moisture data with the ideal soil moisture range to calculate the phenological period water requirement matching degree parameter; The alternate bearing fruit status judgment module is used to determine the current alternate bearing fruit status of the camellia oleifera planting area based on the historical flower bud differentiation rate data and historical fruit setting rate data of the camellia oleifera planting area. The alternating irrigation strategy output module is used to input the phenological water requirement matching degree parameter, the alternate bearing status and the local meteorological data into the pre-trained alternating irrigation prediction model array, and dynamically output the alternating irrigation strategy with flower bud differentiation rate and fruit setting rate as dual optimization objectives. The alternating irrigation strategy includes special water and fertilizer type, application amount and application time. The special water and fertilizer type includes flower-promoting water and fertilizer in high-yield years and fruit-strengthening water and fertilizer in low-yield years. The alternating water and fertilizer application module is used to apply alternating water and fertilizer to the camellia oleifera planting area according to the alternating irrigation strategy.

Images