Intelligent agricultural information management platform based on cloud computing

Abstract

The invention discloses an intelligent agricultural information management platform based on cloud computing, and relates to the technical field of agricultural information management.The intelligent agricultural information management platform is characterized in that space and time similarity parameters are preset, an exponential decay weighting algorithm is adopted, and missing and original data are efficiently integrated; capturing a spatial dependency relationship by using a graph neural network, and mining a data time sequence dynamic state in combination with an LSTM model to realize an accurate decision; the edge real-time response unit monitors field abnormity through an automatic encoder, and once deviation is detected, the deviation is quickly fed back and adjusted through a cloud edge coordination mechanism, so that the operation error of the automatic agricultural machine is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural information management, and in particular to a smart agricultural information management platform based on cloud computing. Background Art

[0002] With the promotion of digital agriculture, smart agricultural information management platforms are being used more and more. These platforms rely on IoT sensors, remote sensing images and field survey data to build a monitoring system covering the entire farmland to assist decision-making and intelligent scheduling.

[0003] In actual applications, some areas may experience gaps or deviations in data collection due to sensor aging, network fluctuations or environmental interference, and the platform may no longer obtain complete information. Some traditional solutions introduce backup sensors or timed calibration, but in actual scheduling decisions, the discontinuity of information can easily lead to judgment bias, affecting the monitoring effect of crop growth status; at the same time, automated agricultural machinery operations rely more on instructions issued by the cloud, and data lags or information transmission deviations will cause agricultural machinery to respond with delays or deviations during the operation process, affecting the accuracy and efficiency of agricultural production; for this reason, some traditional platforms adopt fault-tolerant mechanisms and data interpolation algorithms, but the incomplete local information will still cause the platform to be unable to quickly adjust the operation strategy in an emergency, which restricts the effectiveness of smart agriculture in actual production; therefore, how to dynamically capture and correct local abnormal data in real time while ensuring the overall stability of the system has become a problem that needs to be broken through urgently. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides a cloud computing-based smart agricultural information management platform to solve the problems of inaccurate real-time decision-making caused by local data loss in current agricultural information management solutions, as well as a certain degree of error in the real-time response of automated agricultural machinery.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The embodiment of the present invention provides a cloud computing-based smart agriculture information management platform, which includes:

[0008] A data acquisition unit that integrates multiple methods such as the Internet of Things, remote sensing, and field surveys to obtain farmland environmental data, including soil, weather, and crop growth status;

[0009] The cloud-based data pre-processing unit performs distributed cleaning, alignment, and standardization on environmental data to form an initial data set;

[0010] The spatiotemporal data completion module dynamically fills in the missing parts of the dataset based on preset spatial distance and time series similarity parameters, and outputs the completed data;

[0011] The data fusion unit performs multi-scale weighted fusion on the completed data and the initial data set to generate a fused data set;

[0012] A cloud-based decision support unit, which combines a graph neural network and a time series data model to comprehensively analyze the fused data set to form decision information;

[0013] The edge real-time response unit uses a lightweight model to monitor farmland field data in real time and provide feedback when environmental anomalies are detected;

[0014] The command dispatch unit dispatches decision information and edge real-time response information through the cloud-edge collaborative communication mechanism and sends it to the automated agricultural machinery execution unit;

[0015] The automated agricultural machinery execution unit performs agricultural operations according to the instructions issued by the instruction scheduling unit.

[0016] As a preferred solution of the cloud computing-based smart agricultural information management platform described in the present invention, the spatiotemporal data completion module uses a preset interpolation algorithm to fill in local data missing areas and adjusts the weights during the data fusion process.

[0017] As a preferred solution of the cloud computing-based smart agricultural information management platform described in the present invention, in which: in the spatiotemporal data completion module, the step of dynamically filling the local missing parts in the data set is: defining the completion value formula of the target missing point at time t and position (x, y) as

[0018] Among them, t is the target time, x is the horizontal coordinate of the target point, and y is the vertical coordinate of the target point. represents the time neighborhood set, t i is each moment in the time neighborhood, represents a set of spatial neighborhoods, (x j ,y j ) is the coordinate point in the spatial neighborhood, w ij (t,x,y) is the weight coefficient of the corresponding neighborhood data, D(t i ,x j ,y j ) is the known data value in the neighborhood; the weighted coefficient calculation formula is:

[0019] w ij (t,x,y)=exp(-α·d s((x,y),(x j ,y j )))·exp(-β·d t (t,t i )),

[0020] Among them, w ij (t,x,y) represents the weighted coefficient between the target missing point and the neighborhood data, α is the spatial similarity parameter, d s ((x,y),(x j ,y j )) is the spatial distance between the target point and the neighboring points, β is the temporal similarity parameter, d t (t,t i ) is the time difference between the target time and the neighboring time.

[0021] As a preferred solution of the cloud computing-based smart agricultural information management platform described in the present invention, in the step of dynamically filling the locally missing parts of the data set:

[0022] The spatial distance is calculated using Euclidean distance, which is expressed as:

[0023] Among them, d s ((x,y),(x j ,y j )) represents the Euclidean distance between the target point and the neighboring points, x is the horizontal coordinate of the target point, y is the vertical coordinate of the target point, x j is the horizontal coordinate of the neighborhood point, y j is the ordinate of the neighborhood point;

[0024] The absolute difference is used to calculate the time difference, which is expressed as:

[0025] d t (t,t i )=|tt i |,

[0026] Among them, d t (t,t i ) represents the absolute time difference between the target time and the neighboring time, t is the target time, t i is the time in the neighborhood; the value of the spatial and temporal similarity parameter α=1 / σ s , where α is the spatial similarity parameter, σ s is the standard deviation of spatial distance, β=1 / σ t Among them, β is the time similarity parameter, σ t is the standard deviation of the time difference.

[0027] As an optimal solution of the cloud computing-based smart agricultural information management platform described in the present invention, in which: in the cloud-based decision support unit, a graph neural network is constructed to obtain spatial structure information, and a time series model is introduced to capture the temporal changes of data. Finally, the two parts of features are fused to generate decision information.

[0028] As a preferred solution of the cloud computing-based smart agricultural information management platform described in the present invention, in which: in the cloud decision support unit, a graph neural network GNN is constructed to construct the fusion data into a graph structure. In the graph G = (V, E), the node v i Corresponding to the sampling points at each spatial position in the fusion data, each node has the initial features The update formula for defining the l-th layer node features in the graph network is:

[0029]

[0030] in, Represents node v i The implicit representation at layer l is Represents node v i The neighbor set of A ij Represents node v in the graph i With v j The normalized adjacency weight between (l) represents the weight matrix of the lth layer, σ represents the activation function;

[0031] After all layers are iterated, the aggregation operation is used to construct the graph-level features, which can be expressed as:

[0032]

[0033] Among them, L is the number of network layers, Readout represents the node feature aggregation function, and H G Represents the spatial characteristics of the output;

[0034] In the cloud-based decision support unit, a long short-term memory (LSTM) network is used to model the temporal dynamics of the fused data. For a node or global data, the time series model uses the following recursive formula:

[0035] h t =LSTM(x t ,h t-1 ; Θ),

[0036] Among them, x t represents the input features at time t, h t represents the hidden state at time t, Θ represents the LSTM model parameters, including weights and biases;

[0037] Aggregate all hidden states at all times to obtain the global temporal features:

[0038]

[0039] Where T represents the number of time steps; H T is the time series feature representation after aggregation;

[0040] In the cloud-based decision support unit, the spatial features H output by the graph neural network are G The time series features H output by the time series model T Perform fusion and construct the joint feature vector z:

[0041] z=Concat(H G ,H T );

[0042] The fused features are mapped to the decision information space through the fully connected layer:

[0043] I=φ(W dz +b d ),

[0044] Among them, Concat represents the feature vector concatenation operation, W d with b d are the weight matrix and bias vector of the decision layer respectively, φ represents the activation function, softmax and sigmoid are selected, and I represents the final decision information output.

[0045] As a preferred solution of the cloud computing-based smart agricultural information management platform described in the present invention, in the cloud decision support unit, the data set is divided into a training set and a validation set, and the training process includes:

[0046] Initialize the weights of each layer of the graph neural network {W (l)}, time series model parameters Θ and decision layer parameters W d ,b d ;

[0047] For each training sample, the graph structure is first constructed and the initial node features are input. After passing through the multi-layer graph neural network, the spatial features H are obtained using Readout. G ;

[0048] Input the data at each moment into LSTM, use the recursive formula to obtain the time series hidden state sequence, and aggregate to obtain the time series feature H T ;

[0049] H G With H T After splicing, the output I is generated through the fully connected decision layer;

[0050] Calculate the error between the model output and the true label according to the predefined loss function, and use the backpropagation algorithm to update all parameters;

[0051] Repeat the iterations until the loss converges, and monitor the model performance on the validation set.

[0052] As a preferred solution of the cloud computing-based smart agricultural information management platform described in the present invention, the edge real-time response unit adopts a lightweight deep learning model to perform predictive analysis on real-time data of farmland sites. When data anomalies are detected, the local decision parameters are immediately corrected and the correction information is fed back to the instruction scheduling unit.

[0053] As a preferred solution of the cloud computing-based smart agricultural information management platform described in the present invention, a cloud-edge collaborative communication mechanism is formed between the cloud decision support unit and the edge real-time response unit to perform two-way real-time transmission of data and instructions, so that while performing global analysis in the cloud, on-site operations can maintain a rapid response to local mutations.

[0054] As a preferred solution of the cloud computing-based smart agricultural information management platform described in the present invention, in which: in the edge real-time response unit, the step of providing feedback when an environmental anomaly is detected is:

[0055] On the edge device, the collected farmland data is used to form the input vector u, which is processed by the autoencoder model;

[0056] The autoencoder consists of two parts: the encoder and the decoder. The encoder maps the input data to the latent space, and the decoder reconstructs the latent vector. The calculation steps include:

[0057] The mapping formula of the encoder is: z = σ(au + b), where z represents the latent vector, a is the encoder weight matrix, b is the encoder bias vector, and σ is the activation function;

[0058] The decoder reconstructs the input data based on the latent vector. The process is: in, represents the reconstructed data, c is the decoder weight matrix, and d is the decoder bias vector;

[0059] To determine whether the data is abnormal, the reconstruction error is defined as the Euclidean distance between the input and the reconstructed data. The formula is: Wherein, e(u) represents the reconstruction error. When e(u) exceeds the preset threshold τ, the field data is considered abnormal;

[0060] The feedback signal is used to verify the feedback mechanism. The verification formula is:

[0061] f(u)=γ·(e(u)-τ),

[0062] When e(u)>τ, the feedback signal f(u) is greater than zero, indicating the abnormal intensity. If e(u)≤τ, then f(u)=0, and there is no abnormal feedback, where τ is the abnormality detection threshold and γ is the feedback signal scaling factor;

[0063] The feedback signal f(u), anomaly detection results, and field data identification information are transmitted to the instruction scheduling unit via the edge device through a low-latency communication protocol for real-time feedback;

[0064] The channel uses a lightweight encryption mechanism to ensure transmission security, while supporting two-way data transmission to ensure that feedback information can drive local adjustments in a timely manner.

[0065] The beneficial effects of the present invention are as follows: the present invention presets spatial and temporal similarity parameters, adopts an exponential decay weighted algorithm, and efficiently integrates missing and original data; uses a graph neural network to capture spatial dependencies, and combines the LSTM model to mine data time series dynamics to achieve accurate decision-making; the edge real-time response unit monitors on-site anomalies through an automatic encoder. Once a deviation is detected, it quickly feeds back and adjusts through the cloud-edge collaborative mechanism to reduce errors in automated agricultural machinery operations.

[0066] This invention significantly improves data integrity and real-time response capabilities, provides a more robust and intelligent technical guarantee for precise agricultural management, and effectively solves the shortcomings of traditional platforms in local data loss and response delays. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 any creative work.

[0068] Figure 1 This is a schematic diagram of the framework of the cloud computing-based smart agriculture information management platform in Example 1. DETAILED DESCRIPTION

[0069] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0070] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0072] Example 1, with reference to Figure 1 This embodiment provides a cloud computing-based smart agriculture information management platform, including:

[0073] The data acquisition unit integrates multiple methods such as the Internet of Things, remote sensing, and field surveys to obtain farmland environmental data, including soil, meteorological, and crop growth status;

[0074] The cloud-based data pre-processing unit performs distributed cleaning, alignment, and standardization on environmental data to form an initial data set;

[0075] The spatiotemporal data completion module dynamically fills in the missing parts of the data set based on the preset spatial distance and time series similarity parameters, and outputs the completed data;

[0076] The spatiotemporal data completion module uses a preset interpolation algorithm to fill in the local data missing areas and adjust the weights during the data fusion process;

[0077] In the spatiotemporal data completion module, the steps for dynamically filling the local missing parts in the data set are as follows: define the completion value formula of the target missing point at time t and position (x, y) as

[0078]

[0079] Among them, t is the target time, x is the horizontal coordinate of the target point, and y is the vertical coordinate of the target point. represents the time neighborhood set, t i is each moment in the time neighborhood, represents a set of spatial neighborhoods, (x j ,y j ) is the coordinate point in the spatial neighborhood, w ij (t,x,y) is the weight coefficient of the corresponding neighborhood data, D(t i ,x j ,y j ) is the known data value in the neighborhood; the weighted coefficient calculation formula is:

[0080]

[0081] Among them, w ij(t,x,y) represents the weighted coefficient between the target missing point and the neighborhood data, α is the spatial similarity parameter, d s ((x,y),(x j ,y j )) is the spatial distance between the target point and the neighboring points, β is the temporal similarity parameter, d t (t,t i ) is the time difference between the target time and the neighboring time;

[0082] In the steps of dynamically filling in the local missing parts of the data set:

[0083] The spatial distance is calculated using Euclidean distance, which is expressed as:

[0084] Among them, d s ((x,y),(x j ,y j )) represents the Euclidean distance between the target point and the neighboring points, x is the horizontal coordinate of the target point, y is the vertical coordinate of the target point, x j is the horizontal coordinate of the neighborhood point, y j is the ordinate of the neighborhood point;

[0085] The absolute difference is used to calculate the time difference, which is expressed as:

[0086] d t (t,t i )=|tt i |,

[0087] Among them, d t (t,t i ) represents the absolute time difference between the target time and the neighboring time, t is the target time, t i is the time in the neighborhood; the value of the spatial and temporal similarity parameter α=1 / σ s , where α is the spatial similarity parameter, σ s is the standard deviation of spatial distance, β=1 / σ t Among them, β is the time similarity parameter, σ t is the standard deviation of the time difference;

[0088] Specifically, a dynamic interpolation method is used to adaptively fill in missing data areas. By combining the spatial distribution and temporal variation characteristics of the data in the neighborhood, the Euclidean distance and absolute time difference between the target point and the neighboring points are calculated, and an exponential decay function is used to assign weights so that closer data contributes more to the filling result. The dynamic completion of local data is achieved by fully utilizing spatiotemporal information.

[0089] The data fusion unit performs multi-scale weighted fusion on the completed data and the initial data set to generate a fused data set;

[0090] A cloud-based decision support unit, which combines graph neural networks and time series data models to conduct comprehensive analysis of fused data sets and generate decision information;

[0091] In the cloud-based decision support unit, a graph neural network is constructed to obtain spatial structure information, and a time series model is introduced to capture the temporal changes of the data. Finally, the two features are fused to generate decision information.

[0092] In the cloud decision support unit, a graph neural network GNN is constructed to construct the fusion data into a graph structure. In the graph G = (V, E), node v i Corresponding to the sampling points at each spatial position in the fusion data, each node has the initial features The update formula for defining the l-th layer node features in the graph network is:

[0093]

[0094] in, Represents node v i The implicit representation at layer l is Represents node v i The neighbor set of A ij Represents node v in the graph i With v j The normalized adjacency weight between (l) represents the weight matrix of the lth layer, σ represents the activation function;

[0095] After all layers are iterated, the aggregation operation is used to construct the graph-level features, which can be expressed as:

[0096]

[0097] Among them, L is the number of network layers, Readout represents the node feature aggregation function, and H G Represents the spatial characteristics of the output;

[0098] In the cloud-based decision support unit, a long short-term memory (LSTM) network is used to model the temporal dynamics of the fused data. For a node or global data, the time series model uses the following recursive formula:

[0099] h t =LSTM(x t ,h t-1 ; Θ),

[0100] Among them, x t represents the input features at time t, h t represents the hidden state at time t, Θ represents the LSTM model parameters, including weights and biases;

[0101] Aggregate all hidden states at all times to obtain the global temporal features:

[0102]

[0103] Where T represents the number of time steps; H T is the time series feature representation after aggregation;

[0104] In the cloud-based decision support unit, the spatial features H output by the graph neural network are G The time series features H output by the time series model T Perform fusion and construct the joint feature vector z:

[0105] z=Concat(H G ,H T );

[0106] The fused features are mapped to the decision information space through the fully connected layer:

[0107] I=φ(W dz +b d ),

[0108] Among them, Concat represents the feature vector concatenation operation, W d with b d are the weight matrix and bias vector of the decision layer, φ represents the activation function, softmax and sigmoid are selected, and I represents the final decision information output;

[0109] In the cloud-based decision support unit, the data set is divided into a training set and a validation set. The training process includes:

[0110] Initialize the weights of each layer of the graph neural network {W (l)}, time series model parameters Θ and decision layer parameters W d ,b d ;

[0111] For each training sample, the graph structure is first constructed and the initial node features are input. After passing through the multi-layer graph neural network, the spatial features H are obtained using Readout. G ;

[0112] Input the data at each moment into LSTM, use the recursive formula to obtain the time series hidden state sequence, and aggregate to obtain the time series feature H T ;

[0113] H G With H T After splicing, the output I is generated through the fully connected decision layer;

[0114] Calculate the error between the model output and the true label according to the predefined loss function, and use the backpropagation algorithm to update all parameters;

[0115] Repeat the iteration until the loss converges and monitor the model performance on the validation set;

[0116] Specifically, the cloud-based decision support unit designed in this step uses a graph neural network and a time series model to work together, extracting the spatial dependencies between sampling points in the fused data while capturing the dynamic characteristics of the data over time. The graph neural network aggregates multi-layer neighbor information to transfer local information into a global representation, while the time series model fully exploits the temporal correlation of the data using the LSTM structure. The two are then combined and fed into the decision layer, achieving information complementarity. The resulting decision information can reflect complex spatiotemporal dynamics.

[0117] The edge real-time response unit uses a lightweight model to monitor farmland field data in real time and provide feedback when environmental anomalies are detected;

[0118] The edge real-time response unit uses a lightweight deep learning model to perform predictive analysis on real-time farmland data. When data anomalies are detected, it immediately corrects local decision parameters and feeds the correction information back to the command scheduling unit.

[0119] A cloud-edge collaborative communication mechanism is established between the cloud-based decision support unit and the edge real-time response unit, enabling two-way real-time transmission of data and instructions. This allows for global analysis in the cloud while enabling on-site operations to quickly respond to local sudden changes.

[0120] In the edge real-time response unit, the steps for providing feedback when an environmental anomaly is detected are:

[0121] On the edge device, the collected farmland data is used to form the input vector u, which is processed by the autoencoder model;

[0122] The autoencoder consists of two parts: the encoder and the decoder. The encoder maps the input data to the latent space, and the decoder reconstructs the latent vector. The calculation steps include:

[0123] The mapping formula of the encoder is: z = σ(au+b), where z represents the latent vector, a is the encoder weight matrix, b is the encoder bias vector, and σ is the activation function;

[0124] The decoder reconstructs the input data based on the latent vector. The process is: in, represents the reconstructed data, c is the decoder weight matrix, and d is the decoder bias vector;

[0125] To determine whether the data is abnormal, the reconstruction error is defined as the Euclidean distance between the input and the reconstructed data. The formula is: Wherein, e(u) represents the reconstruction error. When e(u) exceeds the preset threshold τ, the field data is considered abnormal;

[0126] The feedback signal is used to verify the feedback mechanism. The verification formula is:

[0127] f(u)=γ·(e(u)-τ),

[0128] When e(u)>τ, the feedback signal f(u) is greater than zero, indicating the abnormal intensity. If e(u)≤τ, then f(u)=0, and there is no abnormal feedback, where τ is the abnormality detection threshold and γ is the feedback signal scaling factor;

[0129] The feedback signal f(u), anomaly detection results, and field data identification information are transmitted to the instruction scheduling unit via the edge device through a low-latency communication protocol for real-time feedback;

[0130] This channel uses a lightweight encryption mechanism to ensure transmission security and supports two-way data transmission, ensuring that feedback information can drive local adjustments in a timely manner.

[0131] Specifically, an anomaly detection mechanism based on an autoencoder is designed here, which can monitor changes in farmland field data in real time. The encoder is used to map the input data into a latent space, and the decoder is used to reconstruct the original data. Data anomalies are determined by comparing the error between the input and reconstructed data. The reconstruction error is used as an anomaly indicator. When it exceeds a preset threshold, a feedback signal is triggered.

[0132] The command dispatch unit dispatches decision information and edge real-time response information through the cloud-edge collaborative communication mechanism and sends it to the automated agricultural machinery execution unit;

[0133] The automated agricultural machinery execution unit performs agricultural operations according to the instructions issued by the instruction scheduling unit.

[0134] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A cloud computing-based smart agriculture information management platform, characterized by: include, A data acquisition unit, used to obtain farmland environmental data, including soil, weather and crop growth status; The cloud-based data pre-processing unit performs distributed cleaning, alignment, and standardization on environmental data to form an initial data set; The spatiotemporal data completion module dynamically fills in the missing parts of the data set based on the preset spatial distance and time series similarity parameters, and outputs the completed data; The data fusion unit performs multi-scale weighted fusion on the completed data and the initial data set to generate a fused data set; A cloud-based decision support unit, which combines a graph neural network and a time series data model to comprehensively analyze the fused data set to form decision information; The edge real-time response unit uses a lightweight model to monitor farmland field data in real time and provide feedback when environmental anomalies are detected; The command dispatch unit dispatches decision information and edge real-time response information through the cloud-edge collaborative communication mechanism and sends it to the automated agricultural machinery execution unit; The automated agricultural machinery execution unit performs agricultural operations according to the instructions issued by the instruction scheduling unit.

2. The cloud computing-based smart agriculture information management platform according to claim 1, characterized in that: The spatiotemporal data completion module uses a preset interpolation algorithm to fill in the local data missing areas and adjusts the weights during the data fusion process.

3. The cloud computing-based smart agriculture information management platform according to claim 2, characterized in that: In the spatiotemporal data completion module, the step of dynamically filling the local missing parts in the data set is as follows: define the completion value formula of the target missing point at time t and position (x, y) as Among them, t is the target time, x is the horizontal coordinate of the target point, and y is the vertical coordinate of the target point. represents the time neighborhood set, t i is each moment in the time neighborhood, represents a set of spatial neighborhoods, (x j ,y j ) is the coordinate point in the spatial neighborhood, w ij (t,x,y) is the weight coefficient of the corresponding neighborhood data, D(t i ,x j ,y j ) is the known data value in the neighborhood; the weighted coefficient calculation formula is: w ij (t,x,y)=exp(-α·d s ((x,y),(x j ,y j )))·exp(-β·d t (t,t i )), Among them, w ij (t,x,y) represents the weighted coefficient between the target missing point and the neighborhood data, α is the spatial similarity parameter, d s ((x,y),(x j ,y j )) is the spatial distance between the target point and the neighboring points, β is the temporal similarity parameter, d t (t,t i ) is the time difference between the target time and the neighboring time.

4. The cloud computing-based smart agriculture information management platform according to claim 3, characterized in that: In the step of dynamically filling the locally missing parts in the data set: The spatial distance is calculated using Euclidean distance, which is expressed as: Among them, d s ((x,y),(x j ,y j )) represents the Euclidean distance between the target point and the neighboring points, x is the horizontal coordinate of the target point, y is the vertical coordinate of the target point, x j is the horizontal coordinate of the neighborhood point, y j is the ordinate of the neighborhood point; The absolute difference is used to calculate the time difference, which is expressed as: d t (t,t i )=|t-t i |, Among them, d t (t,t i ) represents the absolute time difference between the target time and the neighboring time, t is the target time, t i is the time in the neighborhood; the value of the spatial and temporal similarity parameter α=1 / σ s , where α is the spatial similarity parameter, σ s is the standard deviation of spatial distance, β=1 / σ t Among them, β is the time similarity parameter, σ t is the standard deviation of the time difference.

5. The cloud computing-based smart agriculture information management platform according to claim 1, characterized in that: In the cloud-based decision support unit, a graph neural network is constructed to obtain spatial structure information, and a time series model is introduced to capture the temporal changes of the data. Finally, the two parts of features are fused to generate decision information.

6. The cloud computing-based smart agriculture information management platform according to claim 5, characterized in that: In the cloud decision support unit, a graph neural network GNN is constructed to construct the fusion data into a graph structure. In the graph G = (V, E), node v i Corresponding to the sampling points at each spatial position in the fusion data, each node has the initial features The update formula for defining the l-th layer node features in the graph network is: in, Represents node v i The implicit representation at layer l is Represents node v i The neighbor set of A ij Represents node v in the graph i With v j The normalized adjacency weight between (l) represents the weight matrix of the lth layer, σ represents the activation function; After all layers are iterated, the aggregation operation is used to construct the graph-level features, which can be expressed as: Among them, L is the number of network layers, Readout represents the node feature aggregation function, and H G Represents the spatial characteristics of the output; In the cloud-based decision support unit, a long short-term memory (LSTM) network is used to model the temporal dynamics of the fused data. For global data, the time series model uses the following recursive formula: h t =LSTM(x t ,h t-1 ;Θ), Among them, x t represents the input features at time t, h t represents the hidden state at time t, Θ represents the LSTM model parameters, including weights and biases; Aggregate all hidden states at all times to obtain the global temporal features: Where T represents the number of time steps; H T is the time series feature representation after aggregation; In the cloud-based decision support unit, the spatial features H output by the graph neural network are G The time series features H output by the time series model T Perform fusion and construct the joint feature vector z: z=Concat(H G ,H T ); The fused features are mapped to the decision information space through the fully connected layer: I=φ(W dz +b d ), Among them, Concat represents the feature vector concatenation operation, W d with b d are the weight matrix and bias vector of the decision layer respectively, φ represents the activation function, softmax and sigmoid are selected, and I represents the final decision information output.

7. The cloud computing-based smart agriculture information management platform according to claim 6, characterized in that: In the cloud-based decision support unit, the data set is divided into a training set and a validation set. The training process includes: Initialize the weights of each layer of the graph neural network {W (l) }, time series model parameters Θ and decision layer parameters W d ,b d ; For each training sample, the graph structure is first constructed and the initial node features are input. After passing through the multi-layer graph neural network, the spatial features H are obtained using Readout. G ; Input the data at each moment into LSTM, use the recursive formula to obtain the time series hidden state sequence, and aggregate to obtain the time series feature H T ; H G With H T After splicing, the output I is generated through the fully connected decision layer; Calculate the error between the model output and the true label according to the predefined loss function, and use the backpropagation algorithm to update all parameters; Repeat the iterations until the loss converges, and monitor the model performance on the validation set.

8. The cloud computing-based smart agriculture information management platform according to claim 1, characterized in that: The edge real-time response unit uses a lightweight deep learning model to perform predictive analysis on real-time data from farmland sites. When data anomalies are detected, the local decision parameters are immediately corrected and the correction information is fed back to the instruction scheduling unit.

9. The cloud computing-based smart agriculture information management platform according to claim 8, characterized in that: The cloud-side decision support unit and the edge real-time response unit form a cloud-edge collaborative communication mechanism to carry out two-way real-time transmission of data and instructions.

10. The cloud computing-based smart agricultural information management platform according to claim 9, characterized in that: In the edge real-time response unit, the steps of providing feedback when an environmental anomaly is detected are: On the edge device, the collected farmland data is used to form the input vector u, which is processed by the autoencoder model; The autoencoder consists of two parts: the encoder and the decoder. The encoder maps the input data to the latent space, and the decoder reconstructs the latent vector. The calculation steps include: The mapping formula of the encoder is: z = σ(au + b), where z represents the latent vector, a is the encoder weight matrix, b is the encoder bias vector, and σ is the activation function; The decoder reconstructs the input data based on the latent vector. The process is: in, represents the reconstructed data, c is the decoder weight matrix, and d is the decoder bias vector; To determine whether the data is abnormal, the reconstruction error is defined as the Euclidean distance between the input and the reconstructed data. The formula is: Wherein, e(u) represents the reconstruction error. When e(u) exceeds the preset threshold τ, the field data is considered abnormal; The feedback signal is used to verify the feedback mechanism. The verification formula is: f(u)=γ·(e(u)-τ), When e(u)>τ, the feedback signal f(u) is greater than zero, indicating the abnormal intensity. If e(u)≤τ, then f(u)=0, and there is no abnormal feedback, where τ is the abnormality detection threshold and γ is the feedback signal scaling factor; The feedback signal f(u), the anomaly detection results, and the field data identification information are transmitted to the instruction scheduling unit via the edge device through a low-latency communication protocol for real-time feedback.

Images