A crop monitoring method, system and device based on digital twin technology

Abstract

The present invention discloses a crop monitoring method, system and device based on digital twin technology, which obtains real-time monitoring data of crops in a target farm area; inputs the real-time monitoring data into a preset farm digital twin model, and obtains fused feature data based on the farm digital twin model; inputs the fused feature data and historical fused feature data into a monitoring prediction model to obtain crop status prediction results; determines control instructions based on the crop status prediction results, and adjusts the growth environment of crops in the target farm area according to the control instructions. This application obtains real-time monitoring data of crops, processes the real-time monitoring data using a constructed farm digital twin model and a monitoring prediction model, determines the control instructions that are optimal for crop growth, optimizes the growth environment of crops, and accurately monitors and controls crops, thereby increasing crop yields.

Description

Technical Field

[0001] The present invention relates to the field of information automation control technology, and in particular to a crop monitoring method, system and device based on digital twin technology. Background Art

[0002] Existing crop monitoring primarily relies on single technologies such as remote sensing, the Internet of Things, image recognition, and machine vision to monitor crop physiology and pest and disease characteristics. While this monitoring approach can achieve some success, crop growth is affected by multiple factors, including weather, temperature, humidity, and pests and diseases. A single data source or dimension of information is insufficient to fully reflect crop growth in complex environments, making it impossible to obtain accurate monitoring information and, consequently, to achieve precise and scientific crop management.

[0003] Therefore, the prior art needs further improvement. Summary of the Invention

[0004] In view of the shortcomings of the above-mentioned related technologies, the purpose of the present invention is to provide a crop monitoring method, system and device based on digital twin technology to overcome the defect that the existing technology cannot achieve comprehensive and accurate monitoring of crops.

[0005] The technical solutions adopted by the present invention to solve the technical problems are as follows:

[0006] In a first aspect, this embodiment discloses a crop monitoring method based on digital twin technology, which includes:

[0007] Obtain real-time monitoring data of crops in the target farm area;

[0008] Inputting real-time monitoring data into a preset farm digital twin model, and obtaining fused feature data based on the farm digital twin model;

[0009] Inputting the fused feature data and the historical fused feature data into a trained monitoring and prediction model to obtain a crop state prediction result output by the monitoring and prediction model;

[0010] Control instructions are determined based on the crop status prediction results, and the growth environment of the crops in the target farm area is adjusted according to the control instructions.

[0011] Optionally, the real-time monitoring data includes: environmental monitoring data, crop growth data, farm equipment data, geographic location information and weather data of the target farm, and market dynamics data of crops.

[0012] Using information collection equipment deployed in the target farm area to collect environmental monitoring data of crops in the target farm;

[0013] Using smart devices to collect crop growth data of crops in target farms;

[0014] using crop growth control equipment to obtain farm equipment data of crops on a target farm;

[0015] Use GIS system to obtain the geographical location information of target farms;

[0016] Obtain weather data for the current target farm area;

[0017] Use web crawler technology to obtain market dynamics data of crops in target farms.

[0018] Optionally, after the step of obtaining real-time monitoring data of crops in the target farm area, the method further includes:

[0019] Preprocessing the real-time monitoring data to obtain preprocessed real-time monitoring data;

[0020] The pre-processed real-time monitoring data is sequentially normalized, logarithmically and one-hot encoded to obtain the converted monitoring data;

[0021] The converted monitoring data is subjected to feature engineering processing to obtain real-time monitoring data after feature selection and feature fusion.

[0022] Optionally, before the step of inputting the real-time monitoring data into a preset farm digital twin model and obtaining fused feature data based on the farm digital twin model, the step further includes:

[0023] Construct a spatial characteristic model of the target farm area based on the geographical location information of the target farm;

[0024] Build a crop entity feature model based on the target farm's crop environmental monitoring data and the operating status and performance data obtained by growth control equipment;

[0025] Constructing a crop activity characteristic model based on the growth status data of crops in the target farm;

[0026] Build a regular feature model of crops based on the market dynamics data of crops in the target farm;

[0027] According to the hierarchical relationship and spatial constraints among the spatial feature model, entity feature model, activity feature model and rule feature model, the spatial feature model, entity feature model, activity feature model and rule feature model are fused and assembled into a farm digital twin model.

[0028] Optionally, the step of inputting the real-time monitoring data into a preset farm digital twin model and obtaining fused feature data based on the farm digital twin model includes:

[0029] Input the real-time monitoring data into the preset farm digital twin model to obtain an updated digital twin model;

[0030] Comparing the simulated data output by the updated farm digital twin model with the real-time monitoring data to obtain a comparison result;

[0031] The updated farm digital twin model is corrected based on the comparison results, and the corrected fusion feature data is output.

[0032] Optionally, before the step of inputting the fused feature data and the historical fused feature data into a trained monitoring and prediction model to obtain a crop state prediction result output by the monitoring and prediction model, the step includes:

[0033] Obtain historical monitoring data of crops and construct a training dataset based on the historical monitoring data;

[0034] The training data set is used to train the preset long short-term memory network multi-task learning model to obtain a trained monitoring prediction model.

[0035] Optionally, the step of determining a control instruction according to the crop state prediction result, and adjusting the growth environment of the crops in the target farm area according to the control instruction includes:

[0036] Based on the crop status prediction results, an optimized decision plan is generated by minimizing the weighted loss function and regularization term;

[0037] Generate corresponding control instructions based on the optimized decision-making plan and the updated farm digital twin model;

[0038] The growing environment of crops in the target farm area is adjusted according to the control instructions.

[0039] Optionally, the optimization decision function corresponding to the optimization decision scheme is:

[0040] D(y pred )=argmind(L(y pred ,d)+λR(d));

[0041] Among them, D(y pred ) is the optimization decision function, d is the optimal decision solution output by the optimization decision function; L is the loss function, which is used to measure the difference between the current decision solution and the predicted result; R is the regularization term; λ is the trade-off factor.

[0042] In a second aspect, the present application discloses a crop monitoring system based on digital twin technology, which includes:

[0043] A data acquisition module is used to obtain real-time monitoring data of crops in the target farm area;

[0044] A digital twin module, configured to input real-time monitoring data into a preset farm digital twin model and obtain fused feature data based on the farm digital twin model;

[0045] A prediction module, configured to input the fused feature data and the historical fused feature data into a trained monitoring prediction model to obtain a crop state prediction result output by the monitoring prediction model;

[0046] The control execution module is used to determine the control instructions according to the crop state prediction result, and adjust the growth environment of the crops in the target farm area according to the control instructions.

[0047] In a third aspect, the present application also discloses a crop monitoring device based on digital twin technology, which includes: a data acquisition device and a data processing device;

[0048] The data collection equipment is arranged in the target farm area and is used to collect real-time monitoring data of crops in the target farm area;

[0049] The data processing device is used to receive real-time monitoring data collected by the data acquisition device, input the real-time monitoring data into a preset farm digital twin model, and obtain fused feature data based on the farm digital twin model, input the fused feature data and historical fused feature data into a trained monitoring and prediction model, obtain the crop status prediction results output by the monitoring and prediction model, and determine control instructions based on the crop status prediction results, and adjust the growth environment of crops in the target farm area according to the control instructions.

[0050] Beneficial effects:

[0051] This embodiment discloses a crop monitoring method based on digital twin technology, which obtains real-time monitoring data of crops in a target farm area; inputs the real-time monitoring data into a preset farm digital twin model, and obtains fused feature data based on the farm digital twin model; inputs the fused feature data and historical fused feature data into a trained monitoring prediction model to obtain a crop state prediction result output by the monitoring prediction model; determines control instructions based on the crop state prediction result, and adjusts the growth environment of crops in the target farm area based on the control instructions. The method of this embodiment obtains real-time monitoring data of crops, processes the real-time monitoring data using the constructed farm digital twin model and monitoring prediction model, determines the control instructions that are optimal for crop growth, optimizes the growth environment of crops, and accurately monitors and controls crops, thereby increasing crop yields. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flowchart of a crop monitoring method based on digital twin technology according to an embodiment of the present invention;

[0053] Figure 2 This is a principle block diagram of real-time monitoring data collection in an embodiment of the present invention;

[0054] Figure 3 This is a flow chart of the steps for constructing a farm digital twin model in an embodiment of the present invention;

[0055] Figure 4 is a flowchart of the steps for predicting crop status in an embodiment of the present invention;

[0056] Figure 5 This is a principle block diagram of a crop monitoring system based on digital twin technology in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0058] Crop monitoring is used to monitor crop growth, identify pests and diseases, optimize planting management, and improve agricultural production efficiency. Existing machine learning-based crop monitoring methods generally employ single technologies, such as remote sensing, the Internet of Things, image recognition, or machine vision. However, because crop growth is influenced by multiple factors, the use of a single technology cannot fully capture crop information, and thus cannot enable comprehensive, accurate, and systematic monitoring of crops based on this information.

[0059] In order to overcome the defects in the above-mentioned existing technologies, the present application provides a crop monitoring method, system and device based on digital twin technology. First, through the application of multiple and multi-dimensional technologies, all-round acquisition of crop information is achieved. Secondly, the real-time monitoring data of crops is integrated and processed using the constructed farm digital twin model, and then input into the monitoring and prediction model. The corresponding control instructions are obtained using the monitoring and prediction model. Finally, the growth conditions and growth environment of crops are adjusted based on the control instructions to improve the production efficiency of crops.

[0060] The following is a further description of a crop monitoring method, system, and device based on digital twin technology disclosed in this embodiment with reference to the accompanying drawings.

[0061] In the first aspect, this embodiment discloses a crop monitoring method based on digital twin technology, such as Figure 1 As shown, the crop monitoring method includes:

[0062] Step S1: Acquire real-time monitoring data of crops in a target farm area.

[0063] Since the growth of crops is not only related to themselves, but also affected by many factors, such as environmental temperature, humidity, pests and diseases, and geographical location, in order to accurately obtain the growth environment information, growth status information and related information of growth control equipment of crops in the target farm area, this step uses a multi-dimensional information acquisition method to obtain information from different aspects, and then obtain real-time monitoring data of crops.

[0064] Furthermore, to collect real-time crop monitoring data, sensors, networks, and other data collection equipment must first be deployed within the target farm area. Specifically, these sensors can be temperature sensors, illumination sensors, humidity sensors, or wind direction sensors, measuring temperature, light intensity, ambient humidity, or wind direction, respectively. These sensors will establish a connection with data processing equipment, transmitting the collected data to the data processing equipment for further processing.

[0065] In one embodiment, the real-time monitoring data includes: environmental monitoring data, crop growth data, farm equipment data, geographic location information and weather data of the target farm, and market dynamics data of the crops.

[0066] Specifically, combined Figure 2 As shown, this step includes:

[0067] Environmental monitoring data of crops in the target farm is collected using information collection equipment deployed in the target farm area; the environmental monitoring data includes: data collected in real time using sensors, monitoring equipment or network technology on the environment such as light, temperature and humidity, soil conditions, water resources, wind speed and direction.

[0068] Use smart devices to collect crop growth data of crops in the target farm; among them, crop growth data is data related to the growth status of crops, including: crop types, growth conditions, pests and diseases, etc. obtained through monitoring equipment such as satellites, drone remote sensing, and ground monitoring networks.

[0069] The farm equipment data of crops in the target farm is obtained by using the crop growth control equipment; the farm equipment data includes: the operating status and performance data of equipment management systems such as agricultural machinery, irrigation systems, and storage facilities.

[0070] The geographic location information of the target farm is obtained using the GIS system; the geographic location information includes: topography, land use, soil type and other geographic information collected through the GIS system.

[0071] Obtain the meteorological data of the current target farm area; the meteorological data is obtained by querying historical weather and real-time weather through the API interface provided by the meteorological website.

[0072] Web crawler technology is used to obtain market dynamic data of crops in the target farm, where the market dynamic data includes: collecting data such as agricultural production costs, agricultural product market demand, prices and sales channels through web crawler technology, and obtaining market dynamic data related to crops.

[0073] Furthermore, after the corresponding data are obtained through the above methods, in order to facilitate more efficient data processing, the collected data needs to be processed to improve the quality of the collected data and obtain more accurate processing results.

[0074] Specifically, after this step, the following steps are also included:

[0075] Step S110: pre-process the real-time monitoring data to obtain pre-processed real-time monitoring data.

[0076] Data preprocessing is to remove missing data, abnormal data, or duplicate data to avoid deviations in the data processing results caused by missing data, abnormal data, or duplicate data. Specifically, data preprocessing mainly includes:

[0077] Process missing values ​​of original data: Dmissing = fill(D,S);

[0078] Outlier processing: Doutliers = clip(Dmissing, lower, upper);

[0079] And, deduplication processing: Dunique=unique(Doutliers).

[0080] Among them, fill is a missing value processing function used to fill missing values ​​according to S (such as mean, median, front and back value filling, etc.); clip is an exception handling function used to limit data to the range of [lower, upper]. Values ​​out of the range are set to boundary values; unique is a deduplication function used to remove duplicate data records.

[0081] Step S120: The pre-processed real-time monitoring data is standardized, logarithmically and one-hot encoded in sequence to obtain converted monitoring data.

[0082] In order to achieve better data processing effect, the pre-processed real-time monitoring data is further converted in this step to facilitate subsequent data analysis. Specifically, in this step, data conversion includes: data standardization conversion, logarithmic conversion and unique hot encoding conversion. The data conversion is to standardize the data. scaled = scale(Dunique), logarithmic transformation D log =log(D scaled ) and one-hot encoding D onehot =onehot(D log )deal with.

[0083] Step S130: performing feature engineering processing on the converted monitoring data to obtain real-time monitoring data after feature selection and feature fusion.

[0084] After the real-time monitoring data is converted in the above step S120, feature engineering processing is also performed on it to extract more feature information from the data, or in order to better apply the real-time monitoring data to the model to improve model performance, increase model robustness and generalization ability, in this step, feature engineering processing is also performed on the preprocessed data to obtain real-time monitoring data after feature selection and feature fusion.

[0085] Feature engineering is the process of selecting features from data. features D=select(D log , features) and feature fusion D′=fuse(D features) operation. Select is a feature selection function used to select the most important features for analysis; fuse is a feature fusion function used to weight and fuse multiple features into a comprehensive feature through principal component analysis (PCA) to obtain the fused real-time monitoring data.

[0086] Step S2: input the real-time monitoring data into a preset farm digital twin model, and obtain fused feature data based on the farm digital twin model.

[0087] The real-time monitoring data obtained in the above step S1 is input into the preset farm digital twin model, and the fused feature data is obtained through the farm digital twin model.

[0088] The farm digital twin model is a three-dimensional model constructed based on farm-related data. It creates a digital farm model within the digital world that fully corresponds to the real (physical) world. This model records and displays comprehensive real-world farm information, enabling real-time monitoring of crops within the farm area, enabling digital planting and improving production efficiency.

[0089] In this step, the real-time monitoring data is input into the farm digital twin model to enable the farm digital twin model to correct and display the farm digital twin model according to the real-time monitoring data. The received real-time monitoring data is also analyzed and processed into fused feature data.

[0090] In one embodiment, the farm digital twin model analyzes and processes real-time monitoring data to obtain the fused feature data using the following formula:

[0091] D′=fuse(select(log(scale(unique(fill(D,strategy),lower,upper),features))));

[0092] Among them, the real-time monitoring data is D, and the fused feature data obtained after processing is D′.

[0093] It is conceivable that before inputting real-time monitoring data into the pre-set farm digital twin model, the process also includes the following steps: Building the farm digital twin model. A digital twin model is a virtual model with multiple dimensions, multiple spatiotemporal scales, and multiple domain descriptions. It is dynamically constructed in a virtual space based on real-time collected data.

[0094] Specifically, combined Figure 3 As shown in the figure, the steps to build a farm digital twin model include:

[0095] S201: Construct a spatial feature model of the target farm area based on the geographical location information of the target farm.

[0096] A spatial model of the target farm is constructed based on point cloud (obtained through 3D scanning technology), elevation and terrain data.

[0097]

[0098] Where M represents the spatial model of the target farm, T represents the terrain surface model, P represents the point cloud data, x and y are horizontal coordinates, and z is the elevation value. Represents a model fusion operation.

[0099] S202: Constructing a physical feature model of the crops based on the environmental monitoring data of the crops in the target farm and the operating status and performance data acquired by the growth control equipment.

[0100] Taking the collected data of crops, machinery and facilities as input, combining the advantages of multiple regression, and using regularization technology to improve the generalization ability of the entity model, the entity model of the target farm is constructed.

[0101] f(x)=k T x+b;

[0102] Among them, x contains multiple feature input vectors, f(x) is the output of the model (predicted target variable), k T is the feature weight, and b is the intercept or bias of the model. That is, by learning k T and b make f(x) as close as possible to the actual observed value of the target farm, and output the predicted result for the farm entity.

[0103] S203: Constructing an activity characteristic model of crops based on the growth status data of crops in the target farm.

[0104] In order to effectively capture the dynamic changes and dependencies in time series data such as tillage, irrigation and harvesting during crop growth, the recurrent neural network (RNN) method is used to construct a digital twin activity feature model of the target farm.

[0105] F(x t )=h t =tanh(W xh x t +W hh h t -1+b h );

[0106] Among them, F is the RNN model, x t is the input feature vector at time step t, h t is the hidden state at time step t, W xh is the weight matrix input to the hidden state, W hhis the weight matrix from the previous time step hidden state to the current hidden state, b h is the bias vector of the hidden state, and tanh is the hyperbolic tangent activation function.

[0107] S204: Construct a regular feature model of crops based on the market dynamic data of crops in the target farm.

[0108] The rule model is generated by combining management strategies (such as farming methods, crop rotation, irrigation, etc.), empirical knowledge (such as crop yields over the years, weather patterns, frequency of pests and diseases, etc.) and market development data (such as market prices of crops, consumer demand, competitor behavior, etc.).

[0109]

[0110] Where R represents the output of the rule model, which is the expected yield, crop maturity, or other key performance indicators of the target farm; M is the management strategy vector, which contains different management practices and decisions; E is the experience vector, which contains historical data and knowledge. S is the market data vector, which contains price, demand, and other market factors. i , γ j and δ k represents the degree of influence of management strategy, experience and market data on farm performance. m, n and p are the number of features of management strategy, experience and market data, respectively.

[0111] S205: According to the hierarchical relationship and spatial constraints among the spatial feature model, entity feature model, activity feature model and rule feature model, the spatial feature model, entity feature model, activity feature model and rule feature model are integrated and assembled into a farm digital twin model.

[0112] To build a comprehensive digital twin model and achieve a comprehensive description and prediction of the target farm, we first fuse the spatial feature model, entity feature model, activity feature model, and rule feature model constructed in the above steps. The function expression is:

[0113] F(x,y,z,w)=α·S(x)+β·E(y)+γ·A(z)+δ·R(w);

[0114] Among them, S(x) represents the spatial model, E(y) represents the entity feature model, A(z) represents the activity feature model, R(w) represents the rule feature model, and α, β, γ, and δ are weight coefficients used to adjust the proportion of different models in the fusion process.

[0115] Then, according to the hierarchical relationship of the spatial feature model (S), entity feature model (E), activity feature model (A) and rule feature model (R) and the spatial constraints, they are assembled into a farm digital twin model (DTIS).

[0116] The functional formula of the farm twin digital model is: DTIS = f(S, E, A, R).

[0117] The specific steps of obtaining fused feature data based on the farm twin digital model in this step include:

[0118] The real-time monitoring data is input into a preset farm digital twin model to obtain an updated digital twin model; the simulation data output by the updated farm digital twin model is compared with the real-time monitoring data to obtain a comparison result; the updated farm digital twin model is corrected according to the comparison result, and the corrected fusion feature data is output.

[0119] During the specific implementation, the farm data collected in real time will be input into the farm digital twin model to conduct a real-life farm simulation, compare the simulation with the actual growth conditions, and correct the model through a feedback mechanism. Continuous learning and adaptation will then be carried out to improve the accuracy of the farm digital twin model's simulation of the real-world farm, thereby better guiding farm management decisions.

[0120]

[0121] Among them, M corrected is the calibrated farm digital twin model; M initial This is the initial farm digital twin model. real It is real-time monitoring data, including real-time collection of crop growth, soil moisture, temperature, etc. sim This is simulated growth data generated by the initial digital twin model. After the calibration of the farm digital twin model, the final output of simulated growth data is fused feature data.

[0122] Step S3: input the fused feature data and the historical fused feature data into the trained monitoring prediction model to obtain the crop status prediction result output by the monitoring prediction model.

[0123] After the fused feature data is obtained in the above steps, the fused feature data and the historical fused feature data are combined and input into the trained monitoring and prediction model. The monitoring and prediction model processes the received fused feature data and the historical fused feature data to obtain the prediction results of the crop status.

[0124] It is conceivable that before inputting the fused feature data and the historical fused feature data into the monitoring prediction model, it is necessary to obtain the trained monitoring prediction model. Therefore, before this step, the following steps are also included:

[0125] Historical monitoring data of crops is obtained, a training data set is constructed based on the historical monitoring data, and the training data set is used to train a preset long short-term memory network multi-task learning model to obtain a trained monitoring prediction model.

[0126] like Figure 4 As shown, in a specific application embodiment, the steps of training a monitoring prediction model are as follows:

[0127] Step S301: extracting features of the crop monitoring data in the training dataset.

[0128] Obtain relevant data within the target farm area, perform feature extraction on the relevant data, and construct a training dataset. The extracted features can be expressed as:

[0129]

[0130] Among them, x1, x2,…, xn can include climate data (such as temperature, humidity, rainfall, etc.), soil conditions (such as pH value, nutrient content, etc.), crop growth stage information (such as growth cycle, growth rate, etc.), historical yield data, etc.

[0131] Step S302: Build a farm digital twin model.

[0132] In one implementation, a long short-term memory (LSTM) multi-task learning model is used as the preset network model. Since LSTM is good at processing time series data, it is suitable for tasks such as predicting crop growth trends. Specifically, the expression for constructing the preset network model is:

[0133]

[0134] Among them, M is the long short-term memory network (LSTM) multi-task learning model. L is the learning function, which uses the feature set and model parameters θ to train the model.

[0135] Step S303: define a loss function.

[0136] In the process of training the preset network model, it is also necessary to define the loss function and find the optimal model parameters.

[0137] In one embodiment, the loss function expression is defined as:

[0138]

[0139] Among them, y true is the true label vector, y pred is the label vector predicted by the preset network model, m is the number of training samples, and the loss function is the mean square error (MSE), which is used to measure the difference between the predicted value and the true value.

[0140] Step S304: performing algorithm optimization on the constructed model.

[0141] Find the optimal model parameters θ by minimizing the loss function * , the expression of the preset network model parameters is:

[0142]

[0143] Among them, y true is the true label vector, and θ is the original model parameter.

[0144] Step S305: Use the trained monitoring prediction model to predict crop growth.

[0145] Use the trained model parameters θ * , using data collected in real time from target farms The farm environment, crop growth, and pest and disease status of crops are predicted using the data as input.

[0146]

[0147] Step S4: determining control instructions according to the crop state prediction result, and adjusting the growth environment of the crops in the target farm area according to the control instructions.

[0148] After the above steps obtain the crop state prediction results based on the monitoring prediction model output, corresponding control instructions can be determined according to the crop state prediction results, and the control instructions can be executed to adjust the growth environment of the crops.

[0149] Specifically, based on the prediction results of the target farm status, an optimized decision plan is generated by minimizing the weighted loss function and the regularization term.

[0150] D(y pred )=argmind(L(y pred ,d)+λR(d))

[0151] Among them, D is the optimization decision function, which outputs the optimal decision plan d; L is the loss function, which measures the difference between the current decision plan and the predicted result; R is the regularization term, which includes factors such as resource consumption and disaster resistance to ensure the feasibility and sustainability of the decision plan; λ is the trade-off factor used to balance the influence between the loss function and the regularization term.

[0152] Based on the optimized decision function (D) and the updated data twin model (M), corresponding control instructions are generated to achieve precise control of the target farm.

[0153] C=argmin C {L(c,D)+λR(c)}

[0154] Where C is the set of generated control instructions; c is a single control instruction; L(c,D) is the loss function, which represents the difference between the control instruction c and the optimization decision D; R(c) is the regularization term, which represents the impact of the control instruction c on resource consumption; λ is a trade-off factor used to balance the loss function and the regularization term.

[0155] Control instructions include, but are not limited to: turning irrigation systems on and off and adjusting water flow; adjusting fertilization schedules, including fertilizer type and dosage; implementing pest and disease control measures, such as the timing and frequency of spraying; optimizing crop planting plans, including planting density and crop rotation; and arranging harvesting plans, including harvesting time and methods. These instructions are dynamically adjusted based on real-time farm data and monitoring and predictive models to ensure optimal farm operations. This approach enables efficient resource utilization, increased crop yield and quality, and reduced environmental impact.

[0156] It can be imagined that the farm digital twin model and the monitoring and prediction model disclosed in this embodiment are interrelated. The farm digital twin model outputs fused feature data processed based on real-time monitoring data to the monitoring and prediction model. After processing the fused feature data, the monitoring and prediction model obtains the crop status prediction result, and adjusts the growth environment of the crops based on the crop status prediction result. During the growth of crops, if the actual crop growth state caused by environmental changes is different from the crop status prediction result, the farm digital twin model can correct the model in real time according to the real-time monitoring data obtained from the real scene, and output the corrected fused feature data to the monitoring and prediction model. The monitoring and prediction model outputs a new crop status prediction result, and determines a new control instruction based on the new crop status prediction result, and then adjusts the strategy for adjusting the growth environment of the crops, and then realizes real-time update of the control instruction according to the actual situation, so as to realize the growth of crops in the most suitable environment, thereby improving the yield and quality of crops.

[0157] While disclosing the above method, this embodiment also proposes a crop monitoring system based on digital twin technology. Figure 5 Shown, including:

[0158] The data acquisition module 510 is used to acquire real-time monitoring data of crops in the target farm area; its function is as described in step S1.

[0159] The digital twin module 520 is used to input real-time monitoring data into a preset farm digital twin model and obtain fused feature data based on the farm digital twin model; its function is as described in step S2.

[0160] The prediction module 530 is used to input the fused feature data and the historical fused feature data into the trained monitoring prediction model to obtain the crop state prediction result output by the monitoring prediction model; its function is as described in step S3.

[0161] The control execution module 540 is used to determine the control instructions according to the crop state prediction result, and adjust the growth environment of the crops in the target farm area according to the control instructions. Its function is as described in step S4.

[0162] Furthermore, the real-time monitoring data includes: environmental monitoring data, crop growth data, farm equipment data, geographic location information and weather data of the target farm, and market dynamics data of crops.

[0163] The data acquisition module is used to use information collection equipment deployed in the target farm area to collect environmental monitoring data of crops in the target farm; use smart devices to collect crop growth data of crops in the target farm; use crop growth control equipment to obtain farm equipment data of crops in the target farm; use the GIS system to obtain the geographical location information of the target farm; obtain the meteorological data of the current target farm area; and use web crawler technology to obtain market dynamics data of crops in the target farm.

[0164] Furthermore, the system also includes: a data preprocessing module.

[0165] The data preprocessing module includes: preprocessing the real-time monitoring data to obtain preprocessed real-time monitoring data; standardizing, logarithmically and converting the preprocessed real-time monitoring data into one-hot encoding data to obtain converted monitoring data; and performing feature engineering on the converted monitoring data to obtain real-time monitoring data after feature selection and feature fusion.

[0166] Furthermore, the system also includes: a digital twin model construction module.

[0167] The digital twin model construction module is used to construct a spatial feature model of the target farm area based on the geographic location information of the target farm; to construct an entity feature model of the crops based on the environmental monitoring data of the crops in the target farm and the operating status and performance data obtained by the growth control equipment; to construct an activity feature model of the crops based on the growth status data of the crops in the target farm; to construct a rule feature model of the crops based on the market dynamic data of the crops in the target farm; and to fuse and assemble the spatial feature model, entity feature model, activity feature model and rule feature model into a farm digital twin model based on the hierarchical relationship and spatial constraints among the spatial feature model, entity feature model, activity feature model and rule feature model.

[0168] Furthermore, the digital twin module includes: a model update module, an information comparison module and a feature fusion module.

[0169] The model update module is used to input real-time monitoring data into the preset farm digital twin model to obtain an updated digital twin model.

[0170] The information comparison module is used to compare the simulation data output by the updated farm digital twin model with the real-time monitoring data to obtain a comparison result.

[0171] The feature fusion module is used to correct the updated farm digital twin model based on the comparison results and output the corrected fusion feature data.

[0172] Furthermore, the system also includes: a prediction model building module.

[0173] The prediction model construction module is used to obtain historical monitoring data of crops and construct a training data set based on the historical monitoring data; the training data set is used to train the preset long short-term memory network multi-task learning model to obtain a trained monitoring prediction model.

[0174] Furthermore, the control execution module includes: a decision generation module, an instruction generation module and an instruction execution module.

[0175] The decision generation module is used to generate an optimized decision plan based on the crop status prediction results by minimizing the weighted loss function and regularization term;

[0176] An instruction generation module is used to generate control instructions corresponding to items based on the optimized decision-making plan and the updated farm digital twin model;

[0177] The instruction execution module is used to adjust the growth environment of crops in the target farm area according to the control instruction. Specifically, the optimization decision function corresponding to the optimization decision scheme is:

[0178] D(ypred )=argmind(L(y pred ,d)+λR(d));

[0179] Among them, D(y pred ) is the optimization decision function, d is the optimal decision solution output by the optimization decision function; L is the loss function, which is used to measure the difference between the current decision solution and the predicted result; R is the regularization term; λ is the trade-off factor.

[0180] On the premise of providing the above-mentioned crop monitoring system, this embodiment further provides a crop monitoring device based on digital twin technology, and the monitoring device includes: a data acquisition device and a data processing device.

[0181] The data collection equipment is arranged in the target farm area and is used to collect real-time monitoring data of crops in the target farm area.

[0182] The data processing device is used to receive real-time monitoring data collected by the data acquisition device, input the real-time monitoring data into a preset farm digital twin model, and obtain fused feature data based on the farm digital twin model, input the fused feature data and historical fused feature data into a trained monitoring and prediction model, obtain the crop status prediction results output by the monitoring and prediction model, and determine control instructions based on the crop status prediction results, and adjust the growth environment of crops in the target farm area according to the control instructions.

[0183] Furthermore, the data processing device further includes:

[0184] Communication interface, used for communication between memory and processor.

[0185] Memory is used to store computer programs that can be run on the processor.

[0186] The memory may include a high-speed RAM memory and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0187] If the memory, processor, and communication interface are implemented independently, the communication interface, memory, and processor can be connected and communicate with each other via a bus. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be categorized as address buses, data buses, and control buses.

[0188] Optionally, in a specific implementation, if the memory, processor, and communication interface are integrated on a chip, the memory, processor, and communication interface can communicate with each other through an internal interface.

[0189] The processor may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0190] The method, system and device disclosed in this embodiment, first of all, realize the real-time data collection of multiple and multi-dimensional data such as the environment, crop growth, farm equipment, geographic information, meteorology, economy and society of the target farm, which can provide data support for the comprehensive monitoring of crops. Secondly, the farm digital twin model constructed by the present invention based on digital twin technology can more intuitively reflect the complex interactions and changes of the target farm at different time and space scales, and input the multiple and multi-dimensional real-time data into the farm digital twin model. The farm digital twin model can realize dynamic updating and continuous optimization of the model, which provides important technical guarantees for real-time and comprehensive monitoring of the target farm. Finally, based on historical experience data and real-time collected data, the present invention designs deep learning algorithms and monitoring prediction models to predict the growth trend of crops, and provide a basis for making advance pest and disease prevention and control measures and scientific farming operations (such as optimizing irrigation plans, fertilization plans, pest and disease prevention measures, etc.) decisions.

[0191] In addition, the present invention also involves a farm control method based on a digital twin platform. This method dynamically generates control instructions based on the farm's real-time data and predictive models, and automatically or semi-automatically adjusts the farm's irrigation system, fertilization strategy, pest and disease control measures, etc., to optimize the crop growth environment and improve crop yield and quality.

[0192] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0193] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A crop monitoring method based on digital twin technology, characterized in that: include: Obtain real-time monitoring data of crops in the target farm area; Inputting real-time monitoring data into a preset farm digital twin model, and obtaining fused feature data based on the farm digital twin model; Inputting the fused feature data and the historical fused feature data into a trained monitoring and prediction model to obtain a crop state prediction result output by the monitoring and prediction model; determining control instructions based on the crop state prediction results, and adjusting the growing environment of the crops in the target farm area according to the control instructions; Before the step of inputting the real-time monitoring data into the preset farm digital twin model and obtaining fusion feature data based on the farm digital twin model, the method further includes: Construct a spatial characteristic model of the target farm area based on the geographical location information of the target farm; Build a crop entity feature model based on the target farm's crop environmental monitoring data and the operating status and performance data obtained by growth control equipment; Constructing a crop activity characteristic model based on the growth status data of crops in the target farm; Build a regular feature model of crops based on the market dynamics data of crops in the target farm; According to the hierarchical relationship and spatial constraints among the spatial feature model, the entity feature model, the activity feature model and the rule feature model, the spatial feature model, the entity feature model, the activity feature model and the rule feature model are integrated and assembled into a farm digital twin model; The step of inputting the real-time monitoring data into a preset farm digital twin model and obtaining fused feature data based on the farm digital twin model includes: Input the real-time monitoring data into the preset farm digital twin model to obtain an updated farm digital twin model; Comparing the simulated data output by the updated farm digital twin model with the real-time monitoring data to obtain a comparison result; Correct the updated farm digital twin model based on the comparison results and output the corrected fusion feature data; The correction formula of the updated farm digital twin model is: ； in, It is the calibrated digital twin model of the farm; It is an updated digital twin model of the farm; It is real-time monitoring data; It is simulated growth data generated by the updated farm digital twin model; the simulated growth data finally output by the corrected farm digital twin model is fused feature data.

2. The crop monitoring method based on digital twin technology according to claim 1, characterized in that: The real-time monitoring data includes: environmental monitoring data, crop growth data, farm equipment data, geographic location information and weather data of the target farm, and market dynamics data of crops; The step of obtaining real-time monitoring data of crops in the target farm area includes: Using information collection equipment deployed in the target farm area to collect environmental monitoring data of crops in the target farm; Using smart devices to collect crop growth data of crops in target farms; using crop growth control equipment to obtain farm equipment data of crops on a target farm; Use GIS system to obtain the geographical location information of target farms; Obtain weather data for the current target farm area; Use web crawler technology to obtain market dynamics data of crops in target farms.

3. The crop monitoring method based on digital twin technology according to claim 2, characterized in that: After the step of obtaining real-time monitoring data of crops in the target farm area, the method further includes: Preprocessing the real-time monitoring data to obtain preprocessed real-time monitoring data; The pre-processed real-time monitoring data is sequentially normalized, logarithmically and one-hot encoded to obtain the converted monitoring data; The converted monitoring data is subjected to feature engineering processing to obtain real-time monitoring data after feature selection and feature fusion.

4. The crop monitoring method based on digital twin technology according to claim 1, characterized in that: Before the step of inputting the fused feature data and the historical fused feature data into the trained monitoring prediction model to obtain the crop state prediction result output by the monitoring prediction model, the method includes: Obtain historical monitoring data of crops and construct a training dataset based on the historical monitoring data; The training data set is used to train the preset long short-term memory network multi-task learning model to obtain a trained monitoring prediction model.

5. The crop monitoring method based on digital twin technology according to claim 1, characterized in that: The steps of determining a control instruction according to the crop state prediction result and adjusting the growth environment of the crops in the target farm area according to the control instruction include: Based on the crop status prediction results, an optimized decision plan is generated by minimizing the weighted loss function and regularization term; Generate corresponding control instructions based on the optimized decision-making plan and the updated farm digital twin model; The growing environment of crops in the target farm area is adjusted according to the control instructions.

6. The crop monitoring method based on digital twin technology according to claim 5, characterized in that: The optimization decision function corresponding to the optimization decision scheme is: ； in, is the optimization decision function, To optimize the optimal decision solution output by the decision function; is the loss function, which is used to measure the difference between the current decision plan and the predicted result; is the regularization term; is a trade-off factor.

7. A crop monitoring system based on digital twin technology, characterized in that: include: A data acquisition module is used to obtain real-time monitoring data of crops in the target farm area; A digital twin module, configured to input real-time monitoring data into a preset farm digital twin model and obtain fused feature data based on the farm digital twin model; A prediction module, configured to input the fused feature data and the historical fused feature data into a trained monitoring prediction model to obtain a crop state prediction result output by the monitoring prediction model; a control execution module, configured to determine control instructions based on the crop state prediction result, and adjust the growing environment of the crops in the target farm area according to the control instructions; The system further includes: a digital twin model construction module; The digital twin model construction module is used to construct a spatial feature model of the target farm area based on the geographic location information of the target farm; construct an entity feature model of the crops based on the environmental monitoring data of the crops in the target farm and the operating status and performance data obtained by the growth control equipment; construct an activity feature model of the crops based on the growth status data of the crops in the target farm; construct a rule feature model of the crops based on the market dynamic data of the crops in the target farm; and according to the hierarchical relationship and spatial constraint conditions among the spatial feature model, the entity feature model, the activity feature model and the rule feature model, the spatial feature model, the entity feature model, the activity feature model and the rule feature model are integrated and assembled into a farm digital twin model; The digital twin module includes: a model update module, an information comparison module and a feature fusion module; A model update module is used to input real-time monitoring data into a preset farm digital twin model to obtain an updated farm digital twin model; An information comparison module, used to compare the simulation data output by the updated farm digital twin model with the real-time monitoring data to obtain a comparison result; The feature fusion module is used to calibrate the updated farm digital twin model based on the comparison results and output the corrected fused feature data; The correction formula of the updated farm digital twin model is: ； in, It is the calibrated digital twin model of the farm; It is an updated digital twin model of the farm; It is real-time monitoring data; It is simulated growth data generated by the updated digital twin model; the simulated growth data finally output by the corrected farm digital twin model is fused feature data.

8. A crop monitoring device based on digital twin technology, characterized in that: include: Data acquisition equipment and data processing equipment; The data collection equipment is arranged in the target farm area and is used to collect real-time monitoring data of crops in the target farm area; The data processing device is used to receive real-time monitoring data collected by the data acquisition device, input the real-time monitoring data into a preset farm digital twin model, and obtain fused feature data based on the farm digital twin model, input the fused feature data and historical fused feature data into a trained monitoring prediction model, obtain crop status prediction results output by the monitoring prediction model, and determine control instructions based on the crop status prediction results, and adjust the growth environment of crops in the target farm area according to the control instructions; The data processing device includes: a digital twin model construction module and a digital twin module; The digital twin model construction module is used to construct a spatial feature model of the target farm area based on the geographic location information of the target farm; construct an entity feature model of the crops based on the environmental monitoring data of the crops in the target farm and the operating status and performance data obtained by the growth control equipment; construct an activity feature model of the crops based on the growth status data of the crops in the target farm; construct a rule feature model of the crops based on the market dynamic data of the crops in the target farm; and according to the hierarchical relationship and spatial constraint conditions among the spatial feature model, the entity feature model, the activity feature model and the rule feature model, the spatial feature model, the entity feature model, the activity feature model and the rule feature model are integrated and assembled into a farm digital twin model; The digital twin module includes: a model update module, an information comparison module and a feature fusion module; The model update module is used to input real-time monitoring data into the preset farm digital twin model to obtain an updated digital twin model; An information comparison module, used to compare the simulation data output by the updated farm digital twin model with the real-time monitoring data to obtain a comparison result; The feature fusion module is used to calibrate the updated farm digital twin model based on the comparison results and output the corrected fused feature data; The correction formula of the updated farm digital twin model is: ； in, It is the calibrated digital twin model of the farm; It is an updated digital twin model of the farm; It is real-time monitoring data; It is simulated growth data generated by the updated digital twin model; the simulated growth data finally output by the corrected farm digital twin model is fused feature data.

Images