Agricultural planting environment data intelligent monitoring system based on deep learning model

Abstract

The invention relates to the technical field of agricultural data prediction and analysis, in particular to an agricultural planting environment data intelligent monitoring system based on a deep learning model. The system comprises a model training module used for obtaining sample data and training a prediction model by using the sample data to obtain an initial weight of a forgetting door; the change trend acquisition module is used for performing window division on the agricultural planting environment data to acquire windows with different time durations; respectively calculating data change trends corresponding to the short-term window, the medium-term window and the long-term window based on the data in the various windows; the change trend parameter acquisition module is used for calculating a change trend parameter at a prediction moment; the prediction module is used for adjusting the initial weight of the forgetting door by using the change trend parameter at the prediction moment; and obtaining a prediction result according to the prediction model after initial weight adjustment, and monitoring crop growth according to the prediction result. According to the invention, agricultural planting environment data can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural data prediction and analysis, and in particular to an intelligent monitoring system for agricultural planting environment data based on a deep learning model. Background Art

[0002] The monitoring of agricultural planting environment data plays a vital role in agricultural planting. Traditional agriculture relies on manual experience and a single monitoring method to manage environmental parameters such as moisture and temperature, which often has problems such as poor real-time performance, low efficiency, and inaccurate decision-making. The application of intelligent monitoring systems combined with deep learning models in the agricultural field provides technical support for real-time monitoring and accurate prediction of the planting environment, and has become a key research direction for agricultural informatization.

[0003] In the intelligent monitoring system of agricultural planting environment, predicting the future growth environment status of crops is a major problem. The traditional method is to use soil moisture and crop growth environment temperature as key features and use LSTM network to predict the crop growth environment status. By learning from historical data, this model can better capture the overall trend of key features on crop growth. However, this type of model usually has fixed input feature weights and does not fully consider the dynamic changes in feature importance under different environmental conditions. Especially in the case of large climate fluctuations or frequent extreme weather, the traditional LSTM lacks adaptive ability to process key features and is difficult to effectively cope with complex agricultural environments. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide an intelligent monitoring system for agricultural planting environment data based on a deep learning model. The technical solutions adopted are as follows: An embodiment of the present invention provides an intelligent monitoring system for agricultural planting environment data based on a deep learning model, the system comprising: The model training module is used to obtain historical agricultural planting environment data and preprocess it to form sample data; the sample data is used to train the prediction model to obtain the initial weight of the forget gate; A change trend acquisition module is used to divide the agricultural planting environment data before the prediction moment into windows with the previous moment of the prediction moment as the starting point, and obtain short-term windows, medium-term windows and long-term windows respectively; and calculate the data change trends corresponding to the short-term windows, medium-term windows and long-term windows respectively based on the data in various windows; A change trend parameter acquisition module is used to calculate the change trend parameters at the prediction time using the data change trends corresponding to the short-term window, the medium-term window and the long-term window; The prediction module is used to adjust the initial weight of the forget gate using the change trend parameters at the prediction moment; obtain the prediction result according to the prediction model after the initial weight adjustment, and monitor the crop growth according to the prediction result.

[0005] Preferably, the agricultural planting environment data includes: Agricultural planting environment data includes soil moisture and crop growth environment temperature.

[0006] Preferably, the prediction model comprises: The prediction model is an LSTM model, and the input of the prediction model is a two-dimensional matrix; the rows of the two-dimensional matrix represent the type of data, and the columns of the matrix represent the timestamp.

[0007] Preferably, obtaining the short-term window, the medium-term window and the long-term window includes: Each window takes the moment before the prediction moment as the starting point to traverse the previous data, and divides the agricultural planting environment data before the prediction moment into windows to obtain short-term windows, medium-term windows and long-term windows. The time length of the short-term window is the first time length, the time length of the medium-term window is the second time length, and the time length of the long-term window is the third time length, and traverses to; one prediction moment corresponds to one long-term window, four medium-term windows and seven short-term windows, one long-term window nests four medium-term windows, and seven short-term windows are nested in the medium-term window closest to the prediction moment.

[0008] Preferably, the data change trends corresponding to the short-term window, the medium-term window and the long-term window are calculated based on the data in various windows, including: For a window, the variance of the modulus of the data in the window is calculated; the absolute value of the difference between the corresponding moment of each data in the window and the predicted moment is obtained, which is recorded as the data distance; the data distance is negatively correlated and mapped using an exponential function with a natural constant as the base to obtain a mapping result; the modulus of the data in the window is multiplied by the corresponding mapping result and the sum is obtained, and the ratio of the sum result to the sum of the mapping results at each moment in the window is the weighted mean; a first weight corresponding to the weighted mean and a second weight corresponding to the variance are set, and the weighted mean and the variance are summed using the first weight and the second weight to obtain the data change trend of the window; thereby, the data change trends corresponding to the short-term window, the medium-term window and the long-term window are obtained.

[0009] Preferably, the calculation formula of the change trend parameter at the prediction moment is: , in, represents the change trend parameter at the prediction time; γ represents the third weight, δ represents the fourth weight, and ε represents the fifth weight; Represents the data change trend in the short-term window closest to the prediction moment; It represents the data change trend of the first short-term window from left to right in the medium-term window closest to the prediction time; Indicates the data change trend of the first medium-term window from left to right within the long-term window at the prediction time; Indicates the data trend in the medium-term window closest to the forecast moment; The data change trend of the long-term window closest to the prediction time; e represents a natural constant.

[0010] Preferably, the initial weight of the forget gate is adjusted using the change trend parameter at the prediction moment, including: The sixth weight is set, and the sixth weight is multiplied by the change trend parameter at the prediction moment, and the multiplication result is added to the initial weight of the forget gate to obtain the adjusted initial weight; when the change trend parameter at the prediction moment is greater than the set threshold, the value of the sixth weight is the first preset value, and when the change trend parameter at the prediction moment is less than or equal to the set threshold, the value of the sixth weight is the second preset value.

[0011] Preferably, the prediction result is obtained according to the prediction model after the initial weight adjustment, including: The data in the short-term window closest to the prediction time is used to form a two-dimensional matrix as the input of the prediction model after the initial weight adjustment, and the prediction result is output. The prediction result is the soil moisture and crop growth environment temperature at the prediction time.

[0012] The embodiments of the present invention have at least the following beneficial effects: the present invention trains a prediction model (LSTM) by using historical agricultural planting environment data to obtain the initial weight of the forget gate; further, the agricultural planting environment data before the prediction moment is divided into windows with the previous moment of the prediction moment as the starting point, and the data change trends corresponding to the short-term window, the medium-term window and the long-term window are respectively obtained, and the fluctuation of the agricultural planting environment data is analyzed by a multi-layer window nesting mechanism, and then the change trend parameters at the prediction moment are calculated, and the initial weight of the forget gate is adjusted according to the change trend parameters at the prediction moment, which can adjust the relative importance of soil moisture and crop growth environment temperature characteristics under different environmental conditions, thereby effectively reducing the prediction error caused by fixed weights, improving the adaptability of the prediction model to complex agricultural environments, and improving the accuracy of soil moisture and crop growth environment temperature prediction, so that the monitoring quality of agricultural planting environment data can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0014] Figure 1 A system block diagram of an agricultural planting environment data intelligent monitoring system based on a deep learning model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of an intelligent monitoring system for agricultural planting environment data based on a deep learning model proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0016] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0017] The following is a detailed description of a specific scheme of an intelligent monitoring system for agricultural planting environment data based on a deep learning model provided by the present invention in conjunction with the accompanying drawings.

[0018] Example: The main application scenario of the present invention is to improve the performance of the LSTM model in agricultural planting environment monitoring by introducing an adaptive weight mechanism. According to the changing trends of key environmental data such as soil moisture and crop growth environment temperature, its importance in predicting crop growth status is dynamically adjusted, so as to more accurately capture the impact of different environmental conditions on crop growth, reduce prediction errors, and improve the model's adaptability to complex agricultural environments and monitoring accuracy.

[0019] It should be noted that there are many types of agricultural planting environment data, including soil moisture, crop growth environment temperature, rainfall, light intensity, etc. In the embodiment of the present invention, the growth of crops is monitored by soil moisture and crop growth environment temperature.

[0020] See also Figure 1 , which shows a system block diagram of an agricultural planting environment data intelligent monitoring system based on a deep learning model provided by an embodiment of the present invention, the system includes the following modules: The model training module is used to obtain historical agricultural planting environment data and preprocess it to form sample data; the sample data is used to train the prediction model to obtain the initial weight of the forget gate.

[0021] Agricultural planting environment data is collected in real time through a sensor network deployed in the farmland. The data collection interval is 1 hour. The agricultural planting environment data includes soil moisture and crop growth environment temperature. These data are uploaded to the local data processing system via wireless transmission.

[0022] The collected agricultural planting environment data are classified and stored according to timestamps and sensor types to form a complete time series data set, providing a structured basis for subsequent analysis.

[0023] Furthermore, the collected raw data is preprocessed, including cleaning, removing noise, outliers or missing values, and repairing them by filling in interpolation. Then, the data is normalized to ensure that the numerical ranges of different types of features are consistent, which is convenient for model processing.

[0024] It should be noted that the historical agricultural planting environment data needs to be processed and then used as sample data for subsequent training, wherein the input in the sample data is a two-dimensional matrix, the rows of the matrix represent the type of data, and the columns of the matrix represent the timestamp. In an embodiment of the present invention, data of one day's length is an input, and the size of the formed two-dimensional input matrix is ​​2 rows and 24 columns, and its output is also a two-dimensional matrix with a size of 2 rows and 1 column.

[0025] The historical planting environment data is converted into sample data for training. Further, a prediction model is constructed, where the prediction model is an LSTM model. The formula of the LSTM model is: , Among them, represents the final feature output at time t, Represents the weight of the forget gate represents the forget gate, To control the last moment The degree of retention at the current moment, represents the sigmoid function, Indicates the hidden state of the previous moment and stores historical memory. Represents element-wise pointwise multiplication. Represents the candidate hidden state at the current moment, which is the current input and the previous hidden state Based on this, a candidate state is generated, and after the tanh activation function, the nonlinear change of the information flow is controlled. , Represents the learning parameters of the model, which is used to transform the input and the historical hidden state Mapped to the space of candidate states, b is the bias function, which adjusts the nonlinear transformation.

[0026] Furthermore, the prediction model is trained using sample data consisting of historical agricultural planting environment data. The loss function is the mean square error loss function. After the training is completed, the parameters in the prediction model can be obtained, including , , bAmong them is the weight of the forget gate, and the weight of the forget gate obtained after training is recorded as the initial weight of the forget gate.

[0027] The change trend acquisition module is used to divide the agricultural planting environment data before the prediction moment into windows with the previous moment before the prediction moment as the starting point, and obtain short-term windows, medium-term windows and long-term windows respectively; and calculate the data change trends corresponding to the short-term windows, medium-term windows and long-term windows based on the data in various windows.

[0028] By analyzing the time series changes of characteristics such as soil moisture and the temperature of the crop growth environment, the model identifies long-term trends and short-term fluctuations, and determines which characteristics have a more significant impact on crop growth in the current environment. Subsequently, based on these changing trends, the weights of the prediction model are redistributed so that short-term characteristics occupy a larger proportion in the model calculation, while medium- and long-term characteristics are appropriately weakened. The entire process relies on the dynamic changes of characteristic values. By adjusting the weights in real time to adapt to the nonlinear characteristics of the complex agricultural environment, the prediction accuracy and adaptability of the prediction model are improved.

[0029] Furthermore, the time series is divided into windows of different scales to capture the changing trends at different time scales. Specifically, the agricultural planting environment data before the prediction moment is divided into windows starting from the moment before the prediction moment, and the short-term windows are obtained respectively. , mid-term window and long-term windows , where each window starts from the moment before the prediction moment and traverses the previous data. The time length of the short-term window is the first time length, the time length of the medium-term window is the second time length, and the time length of the long-term window is the third time length. Specifically, the first time length is one day, the second time length is one week, and the third time length is one month.

[0030] Short-term windows can be used to capture rapid fluctuations and recent changes, medium-term windows can be used to capture trends and reflect seasonal changes, and long-term windows can be used to capture long-term trends and reflect changes in indicators within the crop growth cycle. The data at one moment in the window contains two data, namely the soil moisture at that moment and the temperature of the crop growth environment. It should be noted that one prediction moment corresponds to one long-term window, four medium-term windows, and seven short-term windows, one long-term window nests four medium-term windows, and seven short-term windows are nested in the medium-term window closest to the prediction moment. It should be noted that the time length of each window can be adjusted by the implementer according to the actual situation.

[0031] Based on the data in various windows, the data change trends corresponding to the short-term window, medium-term window and long-term window are calculated respectively. Specifically, for a window, the variance of the modulus of the data in the window is obtained; the absolute value of the difference between the corresponding moment of each data in the window and the predicted moment is obtained, which is recorded as the data distance; the data distance is negatively correlated and mapped using an exponential function with a natural constant as the base to obtain the mapping result; the modulus of the data in the window is multiplied by the corresponding mapping result and summed to obtain the sum result, and the ratio of the sum result to the sum of the mapping results at each moment in the window is the weighted mean; the first weight corresponding to the weighted mean and the second weight corresponding to the variance are set, and the weighted mean and variance are summed using the first weight and the second weight to obtain the data change trend of the window, thereby obtaining the data change trend corresponding to the short-term window, medium-term window and long-term window respectively. The specific calculation formula is:

[0032] in, Indicates the data change trend of a window corresponding to the prediction time t; i represents the i-th data in the window, or the i-th time; Represents the modulus of the i-th data in the window; The weighted mean represents the trend of data changes. The closer the data is to the prediction time, the greater the weight of the model is. This helps to capture trends and dynamic changes and reduce the deviation caused by sensor fluctuations or acquisition errors. This method can enhance the sensitivity to recent changes while weakening the impact of data far away from the current time point on the mean. W represents the time window, which is short-term, medium-term and long-term windows respectively. represents the average value of the modulus of the data in the window, By introducing a nonlinear transformation that is more sensitive to outliers and performing a high-order mapping on the difference, the contribution of abnormal fluctuations to volatility is emphasized. p represents a nonlinear index. Preferably, in the embodiment of the present invention, the value of p is 2. The impact of abnormal fluctuations (such as extreme humidity or temperature) is strengthened, and the nonlinear volatility of this formula is more sensitive to outliers, which helps to capture significant features under violent fluctuations. Since p is 2, the latter term can also be regarded as the variance of the modulus of the data in the window. Represents the absolute value of the difference between the corresponding time of each data in the window and the predicted time, recorded as data distance, Indicates the mapping result of the negative correlation mapping of data distance. α and β respectively represent the first weight corresponding to the weighted mean and the second weight corresponding to the variance, and in the embodiment of the present invention, the values ​​are 0.5 respectively, and e represents a natural constant. Thus, the data change trend corresponding to the short-term window, medium-term window and long-term window corresponding to the prediction time can be obtained.

[0033] The change trend parameter acquisition module is used to calculate the change trend parameters at the prediction time using the data change trends corresponding to the short-term window, the medium-term window and the long-term window.

[0034] After obtaining the changing trends of agricultural planting environment data in windows of different sizes, the data changing trends of the short-term window, medium-term window and long-term window closest to the prediction time are recorded as , , , and then obtain the data change trend of the first short-term window from left to right in the medium-term window closest to the prediction time, recorded as , get the data change trend of the first medium-term window from left to right in the long-term window, recorded as , from which the change trend parameters at the prediction time can be calculated, and the specific calculation formula is: , in, represents the change trend parameter at the prediction time (time t); γ represents the third weight, δ represents the fourth weight, and ε represents the fifth weight; Represents the data change trend in the short-term window closest to the prediction moment; It represents the data change trend of the first short-term window from left to right in the medium-term window closest to the prediction time; Indicates the data change trend of the first medium-term window from left to right within the long-term window at the prediction time; Indicates the data trend in the medium-term window closest to the forecast moment; The data change trend of the long-term window closest to the prediction time; e represents a natural constant.

[0035] The above formula uses the window nesting method to use the medium- and long-term windows as weights to adjust short-term data changes. γ represents the weight of the overall formula, which is to reduce the impact of short-term data change trends. For crop growth, the impact of changes in the growth environment on crop growth has a certain lag, so this weight is added. δ represents the weight coefficient of the data change trend under the nested window. It represents the average value of the data change trend under multi-layer nesting, and the value range is , Indicates the difference in data change trends between the long-term window and the medium-term window closest to the prediction time, with a value range of Preferably, in the embodiment of the present invention, γ can be 0.9 to adjust the impact of the data change trend in the medium and long-term window, and δ can be , ε can be taken When the trend parameter changes When it is greater than a set threshold, it can be determined that a mutation has occurred in the data in a short period of time. Preferably, the threshold value set in the present invention is 0.5.

[0036] The prediction module is used to adjust the initial weight of the forget gate using the change trend parameters at the prediction moment; obtain the prediction result according to the prediction model after the initial weight adjustment, and monitor the crop growth according to the prediction result.

[0037] After obtaining the change trend parameters at the prediction moment, the initial weight of the forget gate is adjusted using the change trend parameters at the prediction moment. Specifically, the sixth weight is set, and the sixth weight is multiplied by the change trend parameters at the prediction moment. The multiplication result is added to the initial weight of the forget gate to obtain the adjusted initial weight; when the change trend parameter at the prediction moment is greater than the set threshold, the value of the sixth weight is the first preset value, and when the change trend parameter at the prediction moment is less than or equal to the set threshold, the value of the sixth weight is the second preset value.

[0038] The calculation formula for the adjusted initial weight is: , in, Represents the initial weight after the forget gate is adjusted at the prediction moment; The initial weight of the forget gate at the prediction moment; φ represents the sixth weight; Indicates the change trend parameter at the time of prediction. When the data mutates, the weight of the forget gate should be enlarged to make the data before time t account for a larger proportion of the predicted data, that is, when the change trend parameter at the time of prediction When it is greater than 0.5 (the set threshold), The value of 0.2 (the first preset value) can be used to amplify the initial weight to amplify the trend of short-term changes. When the data changes are relatively stable (no mutation occurs), the weight of the forget gate should be reduced to reduce the proportion of data before time t, that is, when the change trend parameter at the prediction time is When it is less than or equal to 0.5 (set threshold), -0.2 (the second preset value) can be taken to reduce the trend of long-term changes, which can make the prediction value of the prediction model LSTM more accurate. The first preset value and the second preset value are 0.2 and -0.2 respectively, and the implementer can adjust them according to the actual situation.

[0039] Use the adaptive weight prediction model (LSTM model) to monitor the growth status of crops. During the growth process of crops, environmental factors such as soil moisture and the temperature of the crop growth environment have an important impact on the health of crops. The adaptive weight LSTM model can predict the growth status of crops based on the changes in these environmental variables (learned through historical data). The predicted value of the prediction model can be used to determine whether the crops are in a healthy growth state.

[0040] After adjusting the initial weight of the forget gate at the prediction time, the data before the prediction time is input into the prediction model with the initial weight adjustment of the forget gate. The input data is a two-dimensional matrix composed of the data before the prediction time. The elements in the two-dimensional matrix are the data in the short-term window closest to the prediction time. The rows of the matrix represent the type of data, and the columns represent the timestamp. The size is 2×24. From this, the prediction result can be obtained, that is, the soil moisture and crop growth environment temperature at the prediction time. When the soil moisture and crop growth environment temperature at the prediction time do not meet the conditions required for crop growth, intervention is required in advance to ensure the normal growth of crops.

[0041] By collecting environmental parameters closely related to crop growth, such as soil moisture and the temperature of the crop growth environment, these data are used to dynamically track the growth status of crops. By using LSTM with adaptive weights and combining historical data to model the growth patterns of crops, future growth status or potential risks can be predicted. For example, when soil moisture is too low, the model can provide early warning and help take intervention measures, such as irrigation. These prediction results are provided to farmers or agricultural managers to make scientific decisions.

[0042] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0043] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0044] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent monitoring system for agricultural planting environment data based on a deep learning model, characterized in that: The system includes: The model training module is used to obtain historical agricultural planting environment data and preprocess it to form sample data; the sample data is used to train the prediction model to obtain the initial weight of the forget gate; A change trend acquisition module is used to divide the agricultural planting environment data before the prediction moment into windows with the previous moment of the prediction moment as the starting point, and obtain short-term windows, medium-term windows and long-term windows respectively; and calculate the data change trends corresponding to the short-term windows, medium-term windows and long-term windows respectively based on the data in various windows; A change trend parameter acquisition module is used to calculate the change trend parameters at the prediction time using the data change trends corresponding to the short-term window, the medium-term window and the long-term window; The prediction module is used to adjust the initial weight of the forget gate using the change trend parameters at the prediction moment; obtain the prediction result according to the prediction model after the initial weight adjustment, and monitor the crop growth according to the prediction result.

2. According to claim 1, an intelligent monitoring system for agricultural planting environment data based on a deep learning model is characterized in that: The agricultural planting environment data include: Agricultural planting environment data includes soil moisture and crop growth environment temperature.

3. According to claim 1, an intelligent monitoring system for agricultural planting environment data based on a deep learning model is characterized in that: The prediction model comprises: The prediction model is an LSTM model, and the input of the prediction model is a two-dimensional matrix; the rows of the two-dimensional matrix represent the type of data, and the columns of the matrix represent the timestamp.

4. According to claim 1, an intelligent monitoring system for agricultural planting environment data based on a deep learning model is characterized in that: The obtaining of the short-term window, the medium-term window and the long-term window includes: Each window takes the moment before the prediction moment as the starting point to traverse the previous data, and divides the agricultural planting environment data before the prediction moment into windows to obtain short-term windows, medium-term windows and long-term windows. The time length of the short-term window is the first time length, the time length of the medium-term window is the second time length, and the time length of the long-term window is the third time length, and traverses to; one prediction moment corresponds to one long-term window, four medium-term windows and seven short-term windows, one long-term window nests four medium-term windows, and seven short-term windows are nested in the medium-term window closest to the prediction moment.

5. According to claim 1, an intelligent monitoring system for agricultural planting environment data based on a deep learning model is characterized in that: The data change trends corresponding to the short-term window, the medium-term window and the long-term window are calculated based on the data in various windows, including: For a window, the variance of the modulus of the data in the window is calculated; the absolute value of the difference between the corresponding moment of each data in the window and the predicted moment is obtained, which is recorded as the data distance; the data distance is negatively correlated and mapped using an exponential function with a natural constant as the base to obtain a mapping result; the modulus of the data in the window is multiplied by the corresponding mapping result and the sum is obtained, and the ratio of the sum result to the sum of the mapping results at each moment in the window is the weighted mean; a first weight corresponding to the weighted mean and a second weight corresponding to the variance are set, and the weighted mean and the variance are summed using the first weight and the second weight to obtain the data change trend of the window; thereby, the data change trends corresponding to the short-term window, the medium-term window and the long-term window are obtained.

6. The agricultural planting environment data intelligent monitoring system based on deep learning model according to claim 1 is characterized in that: The calculation formula of the change trend parameter at the prediction time is: ， in, represents the change trend parameter at the prediction time; γ represents the third weight, δ represents the fourth weight, and ε represents the fifth weight; Represents the data change trend in the short-term window closest to the prediction moment; It represents the data change trend of the first short-term window from left to right in the medium-term window closest to the prediction time; Indicates the data change trend of the first medium-term window from left to right within the long-term window at the prediction moment; Indicates the data change trend in the medium-term window closest to the forecast moment; The data change trend of the long-term window closest to the prediction time; e represents a natural constant.

7. The agricultural planting environment data intelligent monitoring system based on deep learning model according to claim 1 is characterized in that: The method of adjusting the initial weight of the forget gate by using the change trend parameter at the prediction moment includes: The sixth weight is set, and the sixth weight is multiplied by the change trend parameter at the prediction moment, and the multiplication result is added to the initial weight of the forget gate to obtain the adjusted initial weight; when the change trend parameter at the prediction moment is greater than the set threshold, the value of the sixth weight is the first preset value, and when the change trend parameter at the prediction moment is less than or equal to the set threshold, the value of the sixth weight is the second preset value.

8. The agricultural planting environment data intelligent monitoring system based on deep learning model according to claim 1 is characterized in that: The prediction result is obtained according to the prediction model after the initial weight adjustment, including: The data in the short-term window closest to the prediction time is used to form a two-dimensional matrix as the input of the prediction model after the initial weight adjustment, and the prediction result is output. The prediction result is the soil moisture and crop growth environment temperature at the prediction time.

Images