Straw disposal optimization system and method based on yield prediction model

Abstract

The invention relates to the technical field of agricultural information processing, in particular to a straw disposal optimization system and method based on a yield prediction model.The straw disposal optimization system comprises a yield prediction subsystem, a supply and demand matching subsystem and a logistics planning subsystem, the farmland basic data, the farmland environment data and the crop growth data are used as input, and a straw yield prediction value is output; the supply and demand matching subsystem collects straw resource data and straw demand data, and establishes a supply and demand matching scheme of straw resources; and the logistics planning subsystem optimizes and deploys logistics paths and transportation resources based on the supply and demand matching scheme. According to the invention, full-process automatic management from yield prediction to supply and demand matching to logistics planning can be realized, many problems in the straw disposal process are effectively solved, and the informatization management level in the agricultural production process is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural information processing, and in particular to a straw disposal optimization system and method based on a yield prediction model. Background Art

[0002] Straw refers to the remaining stems and leaves of crops after harvesting the seeds. It is a by-product of agricultural production, mainly from crops such as wheat, rice, corn, soybeans, rapeseed, etc. According to empirical data, the amount of straw produced per mu varies depending on the crop type. Generally, about 1000 - 1500 catties of dry straw can be produced per mu of rice, 800 - 1600 catties of dry straw per mu of wheat, about 3000 catties of dry straw per mu of corn, about 600 - 1300 catties of dry straw per mu of soybeans, and about 750 catties of straw can be produced per mu of cotton. In the past, straw was considered an agricultural waste. To avoid affecting subsequent planting plans, farmers mostly used the method of drying and burning for treatment. The large amount of particulate matter and soot generated during the burning process will cause serious pollution to the environment. However, with the progress of modern sustainable agriculture and resource management technologies, extensive reuse value has been discovered for straw, and it can be processed into feed, organic fertilizer, biomass energy, building materials, pulp, boards, fiber products, etc., thus generating certain economic value.

[0003] However, under the existing technical conditions, there are still various problems in the reprocessing and utilization process of straw. For example, the yield of crops is affected by factors such as meteorological conditions, soil data, crop type, fertilizer application, and pests and diseases, resulting in certain fluctuations in the yield of straw. If the corresponding processing equipment and transportation resources are allocated according to the actual yield of straw, the statistical workload is large, the statistical results may not be accurate enough, and at the same time, the allocation plan may have serious lag, affecting the progress of straw disposal and subsequent crop planting plans. On the other hand, the reprocessing and utilization of straw involves various downstream products and corresponding types of processing units. The collection and transportation of straw resources in different regions have different difficulties. For example, the transportation difficulty of straw on slopes is significantly increased compared to that in plain areas. How to accurately collect the demand information of each processing unit, accurately match the straw resources with the actual processing needs, and establish a traceable and transparent straw trading mechanism is also an issue that needs to be considered. Summary of the Invention

[0004] The present invention provides a straw disposal optimization system and method based on a yield prediction model, which can realize the full-process automated management from yield prediction to supply-demand matching and then to logistics planning, effectively solve many problems in the straw disposal process, improve the information management level in the agricultural production process, and also provide strong technical support for realizing the sustainable development of agriculture.

[0005] The present application provides the following technical solutions:

[0006] The straw disposal optimization system based on the yield prediction model includes a yield prediction subsystem, a supply-demand matching subsystem, and a logistics planning subsystem. Among them, the yield prediction subsystem constructs a yield prediction model based on machine learning algorithms, uses basic farmland data, farmland environment data, and crop growth data as inputs, and outputs the predicted straw yield value; the supply-demand matching subsystem collects straw resource data and straw demand data, and establishes a supply-demand matching plan for straw resources; the logistics planning subsystem optimizes and allocates the logistics path and transportation resources based on the supply-demand matching plan.

[0007] Technical principle: The yield prediction subsystem uses basic farmland data (such as soil type, cultivated land area), farmland environment data (such as temperature, precipitation), and crop growth data (such as growth cycle, pest and disease situation), and constructs a yield prediction model through machine learning algorithms to output the predicted straw yield value. Secondly, the supply-demand matching subsystem collects straw resource data (and demand data), establishes a supply-demand matching plan to ensure the reasonable allocation of resources; the logistics planning subsystem optimizes and allocates the logistics path and transportation resources based on the supply-demand matching plan, using the improved Dijkstra algorithm or VRP model to reduce the total transportation cost.

[0008] Beneficial effects: Through multi-dimensional data analysis and intelligent optimization, the system realizes the full-process automated management from yield prediction to supply-demand matching and then to logistics planning. The comprehensive system design not only improves the information management level in the agricultural production process, but also provides strong technical support for the realization of agricultural sustainable development. Through scientific prediction, precise matching, and efficient logistics planning, the system can effectively solve many problems faced in the straw disposal process.

[0009] Furthermore, the yield prediction subsystem includes a data collection module, a data storage module, a data preprocessing module, a feature extraction module, a prediction model construction module, and a prediction output module, where:

[0010] The data collection module is used to obtain remote sensing data and ground sensor data;

[0011] The data storage module, the data preprocessing module, the feature extraction module, the prediction model construction module, and the prediction output module are all set on the cloud server. The data storage module obtains remote sensing data through a data interface, receives the data of the ground sensor network through a wireless data network, integrates the received remote sensing data and ground sensor data, and establishes an original data set in units of plots through a unified time axis;

[0012] The data preprocessing module performs data cleaning and data preprocessing on the original data set, and converts the original data set into standardized time series data;

[0013] The feature extraction module includes a feature selection unit and a feature construction unit. The feature selection unit screens out features highly correlated with straw yield based on the correlation analysis method, and the feature construction unit optimizes and constructs the feature data; a feature data set is established according to the original data set, and the feature data set is divided into a training set, a validation set, and a test set according to a ratio;

[0014] The prediction model construction module constructs a yield prediction model based on a hybrid model combining long short-term memory network (LSTM) and random forest (RF), and uses the training set to train the constructed yield prediction model. During the training process, the stochastic gradient descent method is used to optimize the model parameters;

[0015] The prediction output module uses the model with the optimal model parameters saved as the actual straw yield prediction model, uses the feature data set as the input quantity, and the output quantity is the straw prediction yield prediction of a single plot.

[0016] Beneficial effects: Collect data related to yield from multiple channels to provide a rich information basis for accurate yield prediction; the collected data is stored for a long time to establish a historical data warehouse, providing rich historical materials for subsequent data analysis and model training, and helping to discover the laws and trends of yield changes; data from different data sources are further integrated to eliminate conflicts and inconsistencies between data; by deeply analyzing and processing the data, representative and stable features are selected, which helps to improve the generalization ability of the model, enabling it to maintain a good prediction effect in different data sets and scenarios and reducing the risk of overfitting of the model.

[0017] Further, the supply-demand matching subsystem establishes a supply-demand matching plan for straw based on the multi-attribute decision-making method.

[0018] Beneficial effects: The multi-attribute decision-making method can comprehensively consider various factors of straw supply and demand. By quantitatively analyzing each attribute, it can accurately match the supply and demand sides of straw, ensure the comprehensiveness of the plan, and improve the resource allocation efficiency; at the same time, the attribute weights can be flexibly adjusted according to market dynamics to quickly adapt to supply and demand changes and enhance the adaptability of the system.

[0019] Further, the logistics planning subsystem establishes a logistics distribution plan according to the supply-demand matching plan, optimizes several logistics paths from the straw collection point to the reprocessing factory, comprehensively considers factors such as road distance cost, transport vehicle load, volume limit, and road traffic restriction regulations, and realizes the minimization of the total transportation cost. It includes a path planning module and a multi-vehicle path collaborative optimization module. The path planning module performs path planning according to the improved Dijkstra algorithm, and the multi-vehicle path collaborative optimization module solves the problem of multiple straw supply sources delivering to the same factory or multiple factories having different complex requirements through the vehicle routing problem VRP model.

[0020] Beneficial effects: According to the supply-demand matching plan, it closely adheres to the actual business needs. During the planning process, various factors are comprehensively considered, and an improved path planning algorithm and VRP model are used to solve complex distribution requirements, improving the transportation efficiency and effectively reducing the transportation cost at the same time.

[0021] Furthermore, the straw disposal optimization system further includes a resource conversion and trading subsystem. The resource conversion and trading subsystem establishes a straw trading and logistics management platform based on the consortium blockchain architecture to realize the transparent management of straw resource trading, including the following modules:

[0022] Node creation module: Creates nodes for all users in the consortium blockchain, builds the underlying platform based on Hyperledger Fabric, and defines independent communication channels for different business processes;

[0023] Smart contract module: Used to develop and deploy smart contracts to define trading rules and operation logics;

[0024] Permission management module: Used to manage the permissions of different roles in the consortium blockchain;

[0025] Data encryption module: Used to encrypt the key data in the consortium blockchain before it is uploaded to the chain;

[0026] Consensus verification module: Used to perform consensus verification on the update of each transaction information;

[0027] Visual query module: Used to provide a visual query interface for users with permissions.

[0028] Beneficial effects: Utilize the immutable feature of the blockchain to realize the transparent management of straw trading and transportation, achieve full tracking of logistics, and is conducive to controlling the transportation status and delivery nodes.

[0029] Furthermore, the straw disposal optimization system further includes a monitoring and maintenance subsystem. The monitoring and maintenance subsystem includes a monitoring data collection module, an early warning module, and a remote maintenance module, where:

[0030] The monitoring data collection module is used to collect monitoring data for each step of the straw disposal link. The collected data is transmitted to the system server through a wireless data network or the Internet of Things. The early warning module monitors the disposal process data according to the set threshold rules. When the monitoring data deviates from the threshold range, the system triggers an early warning message and notifies relevant personnel to handle it in a timely manner through text messages, system pop-ups, etc. The remote maintenance module maintains the software and hardware of the straw disposal optimization system to ensure the stable operation of the system and also provides a firmware upgrade function.

[0031] Beneficial effects: It can provide real-time feedback on the system operation status, quickly respond to abnormal data for problem handling, and ensure the stable operation of the straw disposal optimization system through continuous monitoring and maintenance of the system, avoiding service interruption caused by failures.

[0032] On the other hand, the present application provides a straw disposal optimization method based on a yield prediction model, including the following steps:

[0033] S1. Establish a yield prediction model to output straw yield prediction data in units of plots. The specific steps are as follows:

[0034] S11. Obtain satellite remote sensing data, unmanned aerial vehicle (UAV) remote sensing data, and sensor data;

[0035] S12. The cloud server stores the data obtained in S11 and establishes an original data set;

[0036] S13. Perform data cleaning and preprocessing on the original data set, including removing outliers, filling in missing values, and normalizing;

[0037] S14. Perform feature engineering on the preprocessed data, extract feature data highly correlated with straw yield through correlation analysis methods, and construct a feature data set; divide the feature data set into a training set, a validation set, and a test set;

[0038] S15. Build a yield prediction model based on a deep learning model. The yield prediction model includes a front-end LSTM layer and a back-end random forest regressor;

[0039] S16. Use the training set to train the constructed yield prediction model, and optimize the model parameters using the stochastic gradient descent method during the training process;

[0040] S2. Based on the yield prediction result of S1, establish a supply-demand matching plan for straw;

[0041] S3. Establish a straw logistics distribution plan according to the final matching plan, including the following sub-steps:

[0042] S31. Perform path planning according to the improved Dijkstra algorithm;

[0043] S32. Solve complex situations where multiple straw supply sources deliver to the same manufacturer or multiple manufacturers have different demands based on the vehicle routing problem (VRP) model.

[0044] Furthermore, S16 includes the following sub-steps:

[0045] S161. Randomly initialize the weight parameters in LSTM and the random forest;

[0046] S162. Divide the training data set into several small batches;

[0047] S163. For the selected small batch of data, perform forward propagation to calculate the predicted values through the model, then calculate the loss value according to the mean square error (MSE) loss function between the predicted values and the true values, and calculate the gradient of the loss function with respect to the model parameters through the backpropagation algorithm;

[0048] S164. Update the model parameters according to the calculated gradient at the selected learning rate;

[0049] S165. After each epoch of training, calculate the loss value on the validation set and compare it with the previously recorded minimum validation loss value. If the current validation loss value is less than the minimum validation loss value, update the minimum validation loss value and save the current model parameters; if the validation loss values for 5 consecutive epochs are not less than the minimum validation loss value, it is considered that the model has started overfitting, and at this time, stop training and save the current optimal model parameters.

[0050] S166. Use the model with the optimal model parameters saved as the actual straw yield prediction model. The input feature parameters of the model include crop type, plot area, land type, soil humidity, product of ambient temperature and precipitation, ambient light amount, pests and diseases, crop planting time, normalized difference vegetation index NDVI, and the output is the straw yield of the plot in one planting cycle.

[0051] Further, the S2 includes the following sub-steps:

[0052] S21. Collect straw resource information and reprocessing factory demand information, and establish complete and accurate straw resource data and straw demand data. The straw resource data includes straw type, straw yield, straw collection time period, straw resource location, and the straw demand data includes straw type, quality standards (water content, fiber length, etc.), demand quantity, expected receipt time, factory location, etc.;

[0053] S22. Determine the attribute weights of the straw resource data and the straw demand data through the analytic hierarchy process;

[0054] S23. Calculate the comprehensive matching degree of the straw resource data and the straw demand data according to the attribute weights, and generate a preliminary matching relationship list according to the comprehensive matching degree score. The content of the preliminary matching list includes information such as the name of the reprocessing factory, factory location, required straw type, demand quantity, expected receipt time, recommended straw supply source serial number, straw resource location, straw type, estimated straw yield, estimated collectible time, and comprehensive matching degree score, etc.;

[0055] S24. Feed the preliminary matching list back to the reprocessing factory and the straw supply source, collect the feedback from both parties, correct the data, adjust the attribute weights or optimize the matching algorithm according to the feedback, and perform the matching again until both parties can accept the matching result to form the final matching plan.

[0056] Further, it also includes:

[0057] S4. Establish a straw trading and logistics management platform for all users based on the consortium blockchain architecture, including the following sub-steps:

[0058] S41. Create nodes for all users, where the users include straw suppliers, straw demanders, and regulatory departments;

[0059] S42. Define independent communication channels for different business processes;

[0060] S43. Develop and deploy smart contracts to define trading rules and operation logics;

[0061] S44. Establish permission management for different roles in the consortium blockchain;

[0062] S45. Encrypt the key data to be uploaded to the blockchain;

[0063] S46. Perform consensus verification on the update of trading information;

[0064] S47. Establish a visual query interface to display the whole process information from the straw source collection to the final processed product for the query needs of users with permissions.

[0065] S5. Collect the process data in the straw disposal link, judge the process data according to the set threshold rules, and generate warning information when the monitoring data is within the threshold range; maintain the hardware and software of the straw disposal optimization system. Description of the Drawings

[0066] Figure 1 It is a schematic diagram of straw for the straw disposal optimization system based on the yield prediction model;

[0067] Figure 2 It is a method flow chart of the straw disposal optimization system based on the yield prediction model. Detailed Description of the Invention

[0068] The following is a further detailed description through specific embodiments:

[0069] Embodiment 1

[0070] This embodiment provides a straw disposal optimization system based on a yield prediction model, including a yield prediction subsystem, a supply-demand matching subsystem, a logistics planning subsystem, a resource conversion and trading subsystem, and a monitoring and maintenance subsystem, as Figure 1 shown. Among them, the yield prediction subsystem is used to construct a yield prediction model, using basic farmland data, farmland environment data, and crop growth data as inputs, and outputting a predicted straw yield value; the supply-demand matching subsystem collects the predicted straw yield values of multiple farmland blocks and the straw demand values of multiple reprocessing units, and establishes a matching relationship between the reprocessing units and the farmland blocks that produce straw; the logistics planning subsystem optimally allocates logistics paths and transportation resources based on the matching relationship between straw supply data and straw demand data; the resource conversion and trading subsystem establishes a straw conversion trading platform based on blockchain technology to achieve transparent management of straw resource transactions; the monitoring and maintenance subsystem real-time monitors the status of various sensors and devices, creates charts and reports for the collected data, and implements an early warning mechanism to maintain the crop production process.

[0071] The yield prediction subsystem includes a data collection module, a data storage module, a data preprocessing module, a feature extraction module, a prediction model construction module, and a prediction output module, where:

[0072] The data collection module is used to obtain various input parameters of the yield prediction subsystem, and the sources of the input parameters include remote sensing data and ground sensor data set on the plots.

[0073] Remote sensing technology obtains information on the Earth's surface through different methods such as multispectral, hyperspectral, or synthetic aperture radar. In the application scenarios of agricultural remote sensing monitoring, remote sensing data can provide multi-dimensional and multi-scale information, monitor the crop planting range and crop growth status, and accurately identify the changes and distribution of crop planting straw according to the spectral characteristics of the crops.

[0074] Remote sensing data can be divided into satellite remote sensing data and unmanned aerial vehicle (UAV) monitoring data. Satellite remote sensing can provide remote sensing images of a large area, which is suitable for monitoring vast farmlands. Through the analysis of satellite remote sensing images and combined with geographic information system (GIS) technology, the actual planting distribution and crop production of farmlands can be dynamically monitored, ensuring the accuracy and integrity of planting data, and laying a foundation for accurate yield prediction. The satellite remote sensing analysis data mainly includes:

[0075] 1) Land use and soil cover monitoring: Through the acquisition and analysis of satellite remote sensing data, the monitoring and classification of land types are realized. For example, for paddy fields, remote sensing monitoring can be carried out using the spectral characteristics of water bodies, and for forests and grasslands, vegetation indices can be used for monitoring, which is beneficial to accurately grasp the distribution and planting area of crops;

[0076] 2) Monitoring of crop growth status: By analyzing the vegetation index and spectral characteristics in satellite remote sensing data, the growth status and health of crops can be monitored in real time, problem areas can be identified, and timely management suggestions can be provided to farmers. The main monitoring indicators for crop production include leaf area index, chlorophyll content, water use efficiency, biomass, etc.;

[0077] 3) Monitoring of pests and diseases: By analyzing the vegetation index and spectral characteristics in satellite remote sensing data, the pest and disease situation in farmland can be monitored in real time, the early symptoms of crop pests and diseases can be identified, and the areas where pests and diseases occur can be discovered in a timely manner. A pest and disease warning system is established to guide farmers in precise prevention and control. The pest and disease indicators include leaf area index, disease spot area, and pest density. Among them, the disease spot area can be calculated through image classification and segmentation algorithms, and the pest density can be analyzed through hyperspectral characteristics.

[0078] On the other hand, unmanned aerial vehicle (UAV) remote sensing can provide higher-resolution and more detailed images, which is suitable for monitoring small areas and specific regions. Agricultural UAVs are usually equipped with a variety of data collection devices, including multispectral cameras, thermal imaging cameras, RGB cameras, etc. Multispectral cameras can provide spectral information in multiple bands to evaluate the health status of crops; thermal imaging cameras can detect the temperature of the crop canopy to evaluate water stress, irrigation requirements, and the impact of pests and diseases; RGB cameras can capture standard color images for visually observing the growth status of crops, identifying weed and disease areas. The remote sensing data of UAVs can be transmitted to the cloud server through a wireless data network or by reading the on-board memory card.

[0079] At the same time, a variety of soil sensor networks and meteorological data collection networks integrated with the cloud server through the Internet of Things platform are arranged in the farmland to comprehensively collect soil, meteorological, and water quality data that affect crop growth. Generally, soil data includes soil humidity, soil temperature, soil conductivity, soil pH value, meteorological data includes temperature, humidity, wind speed, wind direction, precipitation, solar radiation, and water quality data includes water quality pH value, water quality conductivity, dissolved oxygen, etc. In addition, ground sensors can also collect various types of data such as light, pests and diseases, and carbon dioxide concentration. The specific details of various sensors can be implemented based on existing technologies and will not be elaborated here. The data collected by ground sensors is transmitted to the cloud server through the MQTT protocol and wireless communication technologies. Wireless communication technologies usually include LoRa, NB-IoT, Wi-Fi, etc. At the same time, the data collection module also accesses the data interface of the real-time meteorological monitoring station to track the current meteorological status in real time and focus on early warning information of sudden extreme weather (such as heavy rain, drought, hail, cold wave).

[0080] The data storage module, data preprocessing module, feature extraction module, prediction model construction module, and prediction output module of the yield prediction subsystem are all set in the cloud server. The data storage module obtains remote sensing data through a data interface, receives data from the ground sensor network through a wireless data network, integrates the received remote sensing data and ground sensor data, and establishes an original dataset in units of plots through a unified time axis. The original dataset contains time series data in multiple dimensions, which can comprehensively reflect the basic data of the plot, the environmental data of the plot, and the historical change trends of crop growth data. Among them, the basic data of the plot includes land type, plot area, plot geographical coordinates, soil humidity, soil temperature, soil pH value, etc., the environmental data of the plot includes environmental temperature, environmental humidity, environmental wind speed, environmental wind direction, environmental light, precipitation, solar radiation, water quality pH value, pest and disease data, etc., and the crop growth data includes crop type, historical crop planting cycle, historical crop yield, normalized difference vegetation index NDVI, enhanced vegetation index EVI. Due to the large amount of data, a distributed database (such as MongoDB, Cassandra) or cloud storage service (such as AWS S3, Google Cloud Storage) can be used to implement it.

[0081] In this embodiment, an original dataset for yield prediction is established with towns A, B, and C in a certain county in the south as observation cases. The crops in this county are triple-cropping per year, and the crop types include rice, corn, soybeans, rapeseed, and wheat. The cultivated land types include paddy fields, irrigated land, dry land, sloping land, and terraced fields. A number of plots in different areas are selected in towns A, B, and C to establish the original dataset. Among them, 13 plots are selected in town A, 15 plots are selected in town B, and 16 plots are selected in town C. The historical straw yield data of all observed plots for 3 years are collected, and the original dataset contains a total of 396 pieces of data.

[0082] The data preprocessing module performs data cleaning and data preprocessing on the original dataset, and converts the original dataset into standardized time series data, including the following steps:

[0083] Remove the outliers in the original dataset; since the sensors may malfunction during operation or the data network may malfunction, resulting in abnormal transmission of sensor data, data outside the normal range may be generated in the original dataset, and abnormal data needs to be deleted.

[0084] For partially missing values, linear interpolation or mean filling method based on historical data of the same period is used for filling. For example, due to deleting outliers or sensor network failures that cause some data not to be uploaded in time, linear interpolation method or mean filling method based on historical data of the same period is needed to ensure the continuity of the data sequence. In meteorological data, individual extremely abnormal temperature or precipitation values may be caused by instrument failures. The box plot method or 3σ principle based on statistics is used to identify and reasonably correct these outliers to prevent them from interfering with subsequent model training.

[0085] Standardize or normalize the numerical features in the original dataset to improve the convergence speed and stability of the prediction model;

[0086] For time series data related to meteorology and planting, rich features are extracted. In addition to conventional time tags such as year, month, day, and quarter, features such as moving average, moving standard deviation, and seasonal indicators are also calculated. For example, calculate the moving average of the average temperature in the past month to reflect the impact of the short-term change trend of temperature on the growth rhythm of crops; extract the seasonal cycle features of precipitation data through Fourier transform to identify the peak and trough periods of precipitation during the crop growth season, and assist the model to capture the periodic relationship between meteorological factors and straw yield.

[0087] The feature extraction module includes a feature selection unit and a feature construction unit.

[0088] As mentioned above, the original dataset includes data of multiple dimensions. It is necessary to screen out the feature data highly correlated with straw yield from these feature data to improve the processing efficiency of the yield prediction model, eliminate redundant features with weak correlation, reduce the model complexity, and avoid the risk of overfitting. The feature selection unit screens out the feature data highly correlated with straw yield based on methods such as Pearson correlation coefficient and Spearman rank correlation coefficient. The specific steps are as follows:

[0089] Calculate the Pearson correlation coefficient R for each feature data;

[0090] Judge the correlation between variables according to the calculated Pearson correlation coefficient R. The value of R ranges from -1 to 1, reflecting the strength of the linear relationship. In this embodiment, the threshold of the significance level is set to 0.05.

[0091] After calculation, the correlation coefficients of crop type, plot area, land type, soil moisture, environmental temperature, precipitation, environmental light amount, pests and diseases, crop planting time, and normalized difference vegetation index NDVI are all less than 0.05, meeting the requirements of the significance level threshold. Therefore, the above features are defined as the feature data highly correlated with straw yield, together with the straw yield and straw maturity time of the corresponding crop production cycle as the feature dataset of the yield prediction model.

[0092] The feature dataset is divided into a training set, a validation set, and a test set according to a ratio of 8:1:1. The training set includes 318 data records, the validation set includes 39 data records, and the test set includes 39 data records.

[0093] The feature construction unit optimizes and constructs the feature data to help the yield prediction model better discover the potential trends in the data and simplify the model input. For example, crop growth is significantly affected by seasons, and climate conditions such as temperature, precipitation, and sunlight in different months have important impacts on crop yields. Therefore, the actual date is converted into month data. Additionally, temperature and precipitation are two key meteorological factors affecting crop growth, and different crops have different requirements for temperature and moisture during their growth cycles. By combining these two variables, the actual physiological requirements of crops can be better simulated. Therefore, in this embodiment, the product of temperature and precipitation is used to quantify the feature data.

[0094] The prediction model construction module constructs a yield prediction model based on a deep learning model. Considering that straw yield is affected by the interaction of multiple complex factors and has certain time series characteristics, a hybrid model architecture combining Long Short-Term Memory (LSTM) and Random Forest (RF) is selected.

[0095] During the crop growth cycle, the changes of meteorological factors (such as temperature, precipitation, sunlight, etc.) and soil factors (such as changes in soil nutrient content, etc.) over time have a cumulative impact on straw yield. LSTM is good at dealing with long-term dependencies in time series data and can effectively capture the cumulative impact of meteorological and soil factors on straw yield over time during the crop growth cycle;

[0096] Although LSTM performs well in dealing with time series dependencies, it has certain limitations in complex feature interaction modeling. Based on the idea of ensemble learning, Random Forest has a strong fitting ability for non-linear combinations of features and can make up for the deficiencies of LSTM in complex feature interaction modeling. Therefore, the feature representation output by LSTM is further processed and used as the input of Random Forest. Multiple decision trees are used to divide and combine the features, thus making up for the deficiencies of LSTM and improving the overall prediction accuracy of the model.

[0097] The prediction model includes a front-end LSTM layer and a back-end Random Forest regressor. The feature data output by the feature construction unit is used as the input data of the front-end LSTM layer. The overall structure of the LSTM layer is a series connection of multiple LSTM units. Each unit is responsible for processing the data of one time step and passing the result to the next unit. At each time step, the LSTM unit receives the current input x t and the hidden state h of the previous moment t-1 . First, the input gate i t, forget gate f t and output gate o t perform a linear transformation with the input x t and h t-1 and calculate the gating value through an activation function (such as the sigmoid function). The cell state C t determines how much of the previous cell state C t should be retained according to the forget gate f t-1 , and is updated in combination with the input gate i t and the current input information. Finally, the output gate o t controls the output of the cell state C t , and the current hidden state h t is obtained through an activation function (such as the tanh function). This hidden state h t contains important information from the time series up to the current moment, and serves as the input for the next time step and the input features for the backend random forest. The number of hidden units in the front-end LSTM layer is crucial for the performance of the prediction model. More hidden units can learn more complex patterns, but may lead to overfitting; fewer hidden units may not be able to fully capture the information in the data. In this embodiment, the number of hidden units is set to 64.

[0098] The hidden state output by the LSTM layer contains long-term dependency information in the time series, and these features are further analyzed and combined by the random forest regressor. The random forest regressor consists of multiple decision trees. When constructing each decision tree, a part of the samples (sampling with replacement) is randomly drawn from the training dataset, and a part of the features is randomly selected. For each node, by calculating metrics such as information gain or Gini coefficient, the optimal feature and splitting point are selected to divide the data into left and right child nodes until a certain stopping condition is met (such as the number of samples in the node is less than a certain threshold or the depth of the tree reaches the upper limit). Each decision tree is trained independently, and the final prediction result of the random forest is the average of the prediction results of all decision trees. In this way, the random forest can reduce the variance of the model and improve the stability and generalization ability of the prediction.

[0099] The prediction model construction module uses the training set to train the constructed yield prediction model, and adopts the stochastic gradient descent method to optimize the model parameters during the training process, including the following steps:

[0100] Randomly initialize the weight parameters in the LSTM and the random forest;

[0101] Divide the training dataset into several small batches, which are divided into 6 batches in this embodiment;

[0102] For the selected small batch of data, forward propagation calculates the predicted values through the model, then calculates the loss value according to the mean square error (MSE) loss function between the predicted values and the true values, and calculates the gradient of the loss function with respect to the model parameters through the backpropagation algorithm;

[0103] According to the calculated gradient, update the model parameters according to the selected learning rate. The learning rate selected in this embodiment is 0.01;

[0104] After each epoch of training, calculate the loss value on the validation set and compare it with the previously recorded minimum validation loss value. If the current validation loss value is less than the minimum validation loss value, update the minimum validation loss value and save the current model parameters; if the validation loss values for 5 consecutive epochs are not less than the minimum validation loss value, it is considered that the model has started overfitting, and at this time, stop training and save the current optimal model parameters. Here, one epoch means that each sample in the dataset is used for training the model once.

[0105] Use the model with the optimal model parameters saved as the actual straw yield prediction model. The input feature parameters of the model include crop type, plot area, land type, soil humidity, product of ambient temperature and precipitation, ambient light amount, pest damage, crop planting time, and normalized difference vegetation index NDVI, and the output is the straw yield of the plot in one planting cycle.

[0106] In this embodiment, the yield prediction model adopts a more flexible deployment method, including using cloud services for model deployment, and at the same time deploying the lightweight model to a mobile device that establishes a data connection with the cloud server, which can ensure the requirements of supporting large-scale data processing and real-time monitoring at the same time.

[0107] After obtaining the straw yield prediction results, since the straw yields are distributed in multiple plots in different regions, and the crop maturity times of different plots are different, and correspondingly the straw generation times are also different, therefore, in order to improve the straw disposal efficiency, realize the efficient and transparent transfer of straw from the collection point to the reprocessing factory, and improve the comprehensive utilization efficiency of straw resources, this embodiment provides a supply-demand matching subsystem and a logistics planning subsystem to complete the integrated management function of straw supply information, demand information, and logistics allocation plan.

[0108] The supply-demand matching subsystem establishes a supply-demand matching plan for straw based on the multi-attribute decision-making method, including the following steps:

[0109] First, collect the straw resource information and the requirements information of reprocessing manufacturers, and establish complete and accurate straw resource data and straw demand data. The straw resource data includes straw types, straw yields, straw collection time periods, and straw resource locations. The straw demand data includes straw types, quality standards (such as water content, fiber length, etc.), demand quantities, expected receipt times, factory locations, etc. Among them, the straw yield data is mainly based on the yield prediction model, and data can also be supplemented by establishing connections with agricultural cooperatives. The straw resource data and straw demand data need to be normalized and standardized: for numerical attributes such as straw yields and manufacturer demand quantities, the linear normalization method is used to map them to the interval from 0 to 1. For categorical attributes such as straw types and required straw types, the one-hot encoding method is used for standardization. For example, assume there are three types of straw: wheat straw, corn straw, and rice straw. If a straw supply source is wheat straw, its one-hot encoding is [1, 0, 0]; if the straw type required by the reprocessing manufacturer is corn straw, its one-hot encoding is [0, 1, 0]; for time attributes such as the expected collection time and the expected receipt time, they can be converted into the number of days from a fixed reference time (such as January 1 of the current year), and then normalized to be on the same scale as other attributes.

[0110] Secondly, determine the attribute weights through the analytic hierarchy process: Invite experts from fields such as agriculture, logistics, and reprocessing industries to form an expert group. The members of the expert group have rich industry experience and professional knowledge and can accurately judge the relative importance of each attribute in the matching of straw resources and reprocessing manufacturer requirements. The number of experts selected in this embodiment is 10, which can be adjusted according to the actual situation. For the main attributes such as straw type matching degree, quality matching degree, transportation distance, and time matching degree, design a questionnaire, and the members of the expert group compare these attributes with each other and score independently. The scoring standard uses the 1-9 scale method, and the larger the value, the more important the attribute. For example, if an expert believes that the straw type matching degree is slightly more important than the quality matching degree, then fill in 3 in the corresponding judgment matrix cell. Each expert fills in the judgment matrix according to their own judgment. After collecting the judgment matrices of all experts, calculate the average judgment matrix. Conduct a consistency test on the average judgment matrix to ensure the rationality of expert judgments. The consistency test index is the consistency ratio (CR). By calculating the maximum eigenvalue and the corresponding eigenvector of the judgment matrix, and normalizing the eigenvector, the weight vector of each attribute is obtained.

[0111] Calculate the comprehensive matching degree of straw resource data and straw demand data: Compose the attribute vectors of each attribute value of the straw supply sources after standardization processing, and calculate the comprehensive matching degree scores of each straw supply source and the reprocessing manufacturers. According to the calculated comprehensive matching degree scores, sort the matching combinations of all straw supply sources and reprocessing manufacturers in descending order. Recommend no less than 3 most matching straw resources for each reprocessing manufacturer, and organize the recommendation results into a detailed preliminary matching list for users to view and manage. The content of the preliminary matching list includes information such as the name of the reprocessing manufacturer, factory location, required straw types, demand quantity, expected receipt time, recommended straw supply source serial number, straw resource location, straw types, estimated straw output, estimated collectible time, and comprehensive matching degree score, etc.

[0112] Finally, feedback the preliminary matching list to the reprocessing manufacturers and straw supply sources, and collect the feedback opinions of both parties. If there are dissatisfied situations, it may be due to inaccurate data, unreasonable weight settings, or defects in the matching algorithm, etc. According to the feedback opinions, correct the data, adjust the attribute weights, or optimize the matching algorithm, and perform the matching again until both parties can accept the matching result to form the final matching plan.

[0113] After establishing the matching plan for straw supply data and demand data, the logistics planning subsystem establishes a logistics deployment plan according to the supply-demand matching plan, and optimizes several logistics paths from the straw collection points to the reprocessing manufacturers. The optimization process comprehensively considers factors such as road distance cost, transport vehicle load, volume limit, and road traffic restrictions to minimize the total transportation cost. It includes a path planning module and a multi-vehicle path collaborative optimization module, where:

[0114] The path planning module performs path planning according to the improved Dijkstra algorithm, including:

[0115] Initialize the starting point (straw resource location), starting point distance, shortest path point set (S), distances of all other points, and priority queue respectively;

[0116] Execute iterations according to the rules, and the rules include: Select the point (u) closest to the starting point from the point set not in (s) and add it to (s); For each adjacent point (v) of (u), update its distance (d[v]) to the starting point, and the calculation of the distance (d[v]) includes road distance cost, road congestion coefficient, vehicle traffic restriction data, vehicle load, and volume limit at the same time.

[0117] When the target point (reprocessing manufacturer) is added to (S), stop the iteration. At this time, the optimal logistics path from the starting point to the target point has been found.

[0118] The multi-vehicle route collaborative optimization module solves the situation where multiple straw supply sources deliver to the same manufacturer or multiple manufacturers have different demands through the Vehicle Routing Problem (VRP) model, including the following steps:

[0119] Determine all the straw collection points and manufacturer locations that need to be served;

[0120] Define the capacity limit (including load and volume) and service time window (if any) for each vehicle;

[0121] Jointly solve vehicle allocation and route planning with the goal of minimizing the total transportation cost. Specifically, allocate vehicles reasonably according to the demand at each collection point and the capacity limit of the vehicles. For example, if the demand at a certain collection point is large, multiple vehicles may be needed to serve it together; conversely, if the demand is small, one vehicle can serve multiple collection points. For each vehicle, use improved path planning algorithms such as the Dijkstra algorithm, A* algorithm, genetic algorithm, etc., and comprehensively consider various factors (such as road congestion coefficient, whether the vehicle meets the road traffic conditions, etc.) to generate the best path from the starting point to the end point. The total transportation cost can be decomposed into direct costs and indirect costs, where direct costs include fuel costs, driver salaries, vehicle depreciation, etc., and indirect costs include default fines due to time delays, etc.

[0122] Establish constraint conditions, and determine the driving route and service customer sequence of each vehicle on the premise of meeting the constraint conditions. The said constraint conditions include vehicle capacity limit, delivery time window, road traffic restrictions, and real-time traffic information.

[0123] The resource conversion and trading subsystem provides an efficient, transparent, and secure straw trading and logistics management platform for all users (straw suppliers, straw demanders, regulatory departments), and ensures the immutability and traceability of data based on the alliance chain architecture, including the following modules:

[0124] Node creation module, create nodes for all users in the alliance chain, including straw suppliers, reprocessing manufacturers, logistics enterprises, and regulatory departments, build the underlying platform based on Hyperledger Fabric, and define independent communication channels for different business processes to ensure data isolation and privacy protection;

[0125] Smart contract module, define trading rules and operation logics by developing and deploying smart contracts, such as contract signing, goods handover, payment confirmation, etc.;

[0126] Permission management module, used to manage the permissions of different roles in the alliance chain to ensure that only authorized users can access specific data and operations;

[0127] A data encryption module, which is used to encrypt the key data in the alliance chain before it is uploaded to the chain. These key data include transaction contract information and logistics transportation information. The transaction contract information includes the main body information of both parties to the contract, the type of straw traded, the trading quantity and price, and the agreed delivery time. The logistics transportation information includes vehicle trajectories, the time of cargo loading and unloading nodes, and driver information;

[0128] A consensus verification module, which is used to perform consensus verification on the update of each transaction information to ensure the integrity and immutability of the data. Multiple consensus algorithms such as Raft and Kafka can be used for consensus verification;

[0129] A visualization query module, which is used to provide a visualization query interface for users with permissions. By inputting the transaction order number or the batch number of goods, the system will display the full-process information from the straw source collection to the final processed product on the blockchain browser, including the timestamp, operator, geographical location, etc. of each link, realizing full-process transparent control.

[0130] The monitoring and maintenance subsystem includes a monitoring data collection module, an early warning module, and a remote maintenance module, where:

[0131] The monitoring data collection module is used to collect monitoring data for each step of the straw disposal link. The straw disposal link involves multiple nodes and requires coordinated operation of each link. Therefore, sensors need to be installed at key nodes to collect the data of the disposal process. For example, temperature and humidity sensors are installed at the straw collection point to collect the straw storage environment data, positioning devices and cameras are installed on the logistics vehicles to collect the transportation process status data, and cameras are installed in the factory workshop to collect the production and processing progress data, etc. The collected data is transmitted to the system server through a wireless data network or the Internet of Things.

[0132] The early warning module monitors the disposal process data according to the set threshold rules. When the monitoring data deviates from the threshold range, the system triggers an early warning message and notifies relevant personnel to handle it in a timely manner through text messages, system pop-ups, etc. The deviation of the monitoring data from the normal range may include various situations. For example, excessive humidity in the straw warehouse may cause mildew risks, and the vehicle driving route deviating from the planned path by a certain distance may cause delays.

[0133] The remote maintenance module maintains the software and hardware of the straw disposal optimization system to ensure the stable operation of the system. For software problems, technicians can connect to the remote device through a remote desktop tool and perform tasks such as updating, configuration adjustment, and troubleshooting; for hardware facilities, monitor their status through a network interface, diagnose possible problems, and perform operations such as restarting the device or calibrating the sensor. In addition, the remote maintenance module also provides a firmware upgrade function, which can still keep the device in the latest state even in remote locations.

[0134] Embodiment 2

[0135] This embodiment provides a straw disposal optimization method based on a yield prediction model, as Figure 2 shown, which includes the following steps:

[0136] S1. Establish a yield prediction model to output straw yield prediction data in units of plots, specifically as follows;

[0137] S11. Obtain satellite remote sensing data, UAV remote sensing data, and sensor data;

[0138] S12. The cloud server stores the data obtained in S11 and establishes an original data set;

[0139] S13. Clean and preprocess the original data set, including removing outliers, filling in missing values, and normalizing;

[0140] S14. Perform feature engineering on the preprocessed data, extract feature data highly correlated with straw yield through correlation analysis methods, and construct a feature data set; divide the feature data set into a training set, a validation set, and a test set;

[0141] S15. Build a yield prediction model based on a deep learning model. The yield prediction model includes a front-end LSTM layer and a back-end random forest regressor;

[0142] S16. Use the training set to train the constructed yield prediction model. During the training process, use the stochastic gradient descent method to optimize the model parameters, including the following sub-steps:

[0143] S161. Randomly initialize the weight parameters in the LSTM and the random forest;

[0144] S162. Divide the training data set into several small batches;

[0145] S163. For the selected small batch of data, perform forward propagation through the model to calculate the predicted value, then calculate the loss value according to the mean square error (MSE) loss function between the predicted value and the true value, and calculate the gradient of the loss function with respect to the model parameters through the backpropagation algorithm;

[0146] S164. Update the model parameters according to the calculated gradient according to the selected learning rate;

[0147] S165. After each epoch of training, calculate the loss value on the validation set and compare it with the previously recorded minimum validation loss value. If the current validation loss value is less than the minimum validation loss value, update the minimum validation loss value and save the current model parameters; if the validation loss values for 5 consecutive epochs are not less than the minimum validation loss value, it is considered that the model has started overfitting. At this time, stop training and save the current optimal model parameters.

[0148] S166. Use the model with the optimal model parameters saved as the actual straw yield prediction model. The input feature parameters of the model include crop type, plot area, land type, soil humidity, product of ambient temperature and precipitation, ambient light amount, pest and disease damage, crop planting time, and normalized difference vegetation index NDVI. The output is the straw yield of the plot during a planting cycle.

[0149] S2. Based on the yield prediction results of S1, establish a supply-demand matching plan for straw, including the following sub-steps:

[0150] S21. Collect straw resource information and reprocessing factory demand information, and establish complete and accurate straw resource data and straw demand data. The straw resource data includes straw type, straw yield, straw collection time period, and straw resource location. The straw demand data includes straw type, quality standards (such as water content, fiber length, etc.), demand quantity, expected receipt time, factory location, etc.;

[0151] S22. Determine the attribute weights of the straw resource data and straw demand data through the analytic hierarchy process;

[0152] S23. Calculate the comprehensive matching degree of the straw resource data and straw demand data according to the attribute weights, and generate a preliminary matching relationship list according to the comprehensive matching degree score. The content of the preliminary matching list includes information such as reprocessing factory name, factory location, required straw type, demand quantity, expected receipt time, recommended straw supply source number, straw resource location, straw type, estimated straw yield, estimated collectible time, and comprehensive matching degree score;

[0153] S24. Feed back the preliminary matching list to the reprocessing factory and the straw supply source, collect the feedback opinions of both parties, correct the data, adjust the attribute weights or optimize the matching algorithm according to the feedback opinions, and re-match until both parties can accept the matching result to form the final matching plan.

[0154] S3. Establish a straw logistics deployment plan according to the final matching plan, including the following sub-steps:

[0155] S31. Perform path planning according to the improved Dijkstra algorithm, including:

[0156] S311. Initialize the starting point (the location of straw resources), the distance from the starting point, the set of shortest path points (S), the sum of distances to all other points, and the priority queue respectively;

[0157] S312. Execute iterations according to the iteration rules;

[0158] S313. When the target point (the reprocessing factory) is added to (S), stop the iteration.

[0159] S32. Solve the situation where multiple straw supply sources deliver to the same factory or multiple factories have different demands based on the Vehicle Routing Problem (VRP) model, including the following sub-steps:

[0160] S321. Determine all customer points (straw collection points) and target points (reprocessing factories) that need to be served;

[0161] S322. Define the capacity limit and service time window for each vehicle. The capacity limit includes the vehicle load and volume;

[0162] S323. Jointly solve vehicle allocation and route planning, with the goal of minimizing the total transportation cost. Use the route planning algorithm, comprehensively consider the road congestion coefficient and vehicle restriction conditions, and generate the best route from the starting point to the end point;

[0163] S324. Establish constraint conditions, and determine the driving route and service customer sequence of each vehicle on the premise of meeting the constraint conditions. The said constraint conditions include vehicle capacity limit, delivery time window, road restriction regulations, and real-time traffic information.

[0164] S4. Establish a straw trading and logistics management platform for all users based on the consortium blockchain architecture, including the following sub-steps:

[0165] S41. Create nodes for all users. The users include straw suppliers, straw demanders, and regulatory departments;

[0166] S42. Define independent communication channels for different business processes;

[0167] S43. Develop and deploy smart contracts, and define trading rules and operation logics;

[0168] S44. Establish permission management for different roles in the consortium blockchain;

[0169] S45. Encrypt the key data that needs to be put on the chain;

[0170] S46. Conduct consensus verification on the update of transaction information;

[0171] S47. Establish a visual query interface to display the whole process information from the straw source collection to the final processed product for the query requirements of authorized users;

[0172] S5. Collect the process data of the straw disposal link, judge the process data according to the set threshold rules, and generate a warning message when the monitoring data is within the threshold range; maintain the hardware and software of the straw disposal optimization system.

[0173] The specific implementation methods of each step can be referred to those described in Embodiment 1.

[0174] The above are only the embodiments of the present invention. The invention is not limited to the fields involved in this embodiment case. Common knowledge such as specific straws and their characteristics known in the solution is not described in detail here. It should be noted that for those skilled in the art, without departing from the straw of the present invention, several deformations and improvements can still be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application shall be subject to the content of its claims, and the specific implementation manners and the like recorded in the specification can be used to explain the content of the claims.

Claims

1. A straw disposal optimization system based on a yield prediction model, characterized in that: It includes a yield prediction subsystem, a supply and demand matching subsystem, and a logistics planning subsystem. The yield prediction subsystem builds a yield prediction model based on a machine learning algorithm, uses basic farmland data, farmland environment data, and crop growth data as input, and outputs straw yield prediction values; the supply and demand matching subsystem collects straw resource data and straw demand data, and establishes a supply and demand matching plan for straw resources; the logistics planning subsystem optimizes and allocates logistics routes and transportation resources based on the supply and demand matching plan.

2. The straw disposal optimization system based on the yield prediction model according to claim 1 is characterized in that: The yield prediction subsystem includes a data acquisition module, a data storage module, a data preprocessing module, a feature extraction module, a prediction model construction module and a prediction output module, wherein: The data acquisition module is used to obtain remote sensing data and ground sensor data; The data storage module, data preprocessing module, feature extraction module, prediction model building module and prediction output module are all set up in the cloud server. The data storage module obtains remote sensing data through the data interface, receives data from the ground sensor network through the wireless data network, integrates the received remote sensing data and ground sensor data, and establishes an original data set based on plots through a unified time axis; The data preprocessing module performs data cleaning and data preprocessing on the original data set, converting the original data set into standardized time series data; The feature extraction module includes a feature selection unit and a feature construction unit. The feature selection unit selects features that are highly correlated with straw yield based on a correlation analysis method, and the feature construction unit optimizes and constructs feature data; a feature data set is established based on the original data set, and the feature data set is divided into a training set, a validation set, and a test set according to a proportion; The prediction model building module builds a yield prediction model based on a hybrid model combining a long short-term memory network (LSTM) and a random forest (RF). The constructed yield prediction model is trained using a training set, and the model parameters are optimized using a stochastic gradient descent method during the training process. The prediction output module uses the model with the optimal model parameters as the actual straw yield prediction model, uses the feature data set as input, and outputs the straw yield prediction of a single plot.

3. The straw disposal optimization system based on yield prediction model according to claim 1, characterized in that: The supply-demand matching subsystem establishes a straw supply-demand matching scheme based on a multi-attribute decision-making method.

4. The straw disposal optimization system based on yield prediction model according to claim 1, characterized in that: The logistics planning subsystem establishes a logistics allocation plan according to the supply and demand matching plan, optimizes several logistics routes from the straw collection point to the reprocessing factory, and comprehensively considers factors such as road distance cost, transport vehicle load, volume limit and road restriction regulations to minimize the total transportation cost. It includes a path planning module and a multi-vehicle path collaborative optimization module. The path planning module performs path planning according to the improved Dijkstra algorithm, and the multi-vehicle path collaborative optimization module uses the vehicle routing problem VRP model to solve the distribution of multiple straw supply sources to the same factory, or multiple factories have different complex needs.

5. The straw disposal optimization system based on the yield prediction model according to any one of claims 1 to 4, characterized in that: The straw disposal optimization system also includes a resource conversion and transaction subsystem, which establishes a straw transaction and logistics management platform based on the alliance chain architecture to achieve transparent management of straw resource transactions, including: The node creation module creates nodes for all users in the alliance chain, builds the underlying platform based on Hyperledger Fabric, and defines independent communication channels for different business processes; Smart contract module, used to develop and deploy smart contracts to define transaction rules and operation logic; The permission management module is used to manage the permissions of different roles in the alliance chain; Data encryption module, used to encrypt key data in the alliance chain before uploading it to the chain; The consensus verification module is used to verify the consensus of each update of transaction information; The visual query module is used to provide a visual query interface for authorized users.

6. The straw disposal optimization system based on yield prediction model according to claim 5 is characterized in that: The straw disposal optimization system further includes a monitoring and maintenance subsystem, which includes a monitoring data acquisition module, an early warning module and a remote maintenance module, wherein: The monitoring data acquisition module is used to collect monitoring data for each step of the straw disposal process. The collected data is transmitted to the system server through a wireless data network or the Internet of Things; the early warning module monitors the disposal process data according to the set threshold rules. When the monitoring data deviates from the threshold range, the system triggers an early warning message and notifies relevant personnel to handle it in a timely manner through text messages, system pop-ups, etc.; the remote maintenance module maintains the software and hardware of the straw disposal optimization system to ensure the stable operation of the system, and also provides firmware upgrade functions.

7. A straw disposal optimization method based on a yield prediction model, characterized in that: The following steps are involved: S1. Establish a yield prediction model and output straw yield prediction data based on plots. The specific steps are as follows: S11. Obtain satellite remote sensing data, drone remote sensing data and sensor data; S12, the cloud server stores the data obtained in S11 and establishes an original data set; S13, perform data cleaning and preprocessing on the original data set, including removing outliers, filling missing values, and normalizing; S14, performing feature engineering on the preprocessed data, extracting feature data highly correlated with straw yield through a correlation analysis method, and constructing a feature data set; Divide the feature dataset into training set, validation set and test set; S15. Building a yield prediction model based on the deep learning model, wherein the yield prediction model includes a front-end LSTM layer and a back-end random forest regressor; S16. Using the training set to train the constructed yield prediction model, and using the stochastic gradient descent method to optimize the model parameters during the training process; S2, based on the yield prediction results of S1, establish a straw supply and demand matching plan; S3. Establish a straw logistics allocation plan based on the final matching plan, including the following sub-steps: S31, performing path planning according to the improved Dijkstra algorithm; S32. A vehicle routing problem (VRP) model is used to solve the complex situation where multiple straw supply sources deliver to the same manufacturer or multiple manufacturers have different demands.

8. The straw disposal optimization method based on the yield prediction model according to claim 7 is characterized in that: The S16 comprises the following sub-steps: S161, randomly initialize the weight parameters in LSTM and random forest; S162, dividing the training data set into several small batches; S163. For the selected small batch of data, forward propagation is performed through the model to calculate the predicted value, and then the loss value is calculated according to the mean square error (MSE) loss function between the predicted value and the true value, and the gradient of the loss function with respect to the model parameters is calculated through the back propagation algorithm; S164, updating the model parameters according to the calculated gradient and the selected learning rate; S165. After each epoch of training, the loss value on the validation set is calculated and compared with the previously recorded minimum validation loss value. If the current validation loss value is less than the minimum validation loss value, the minimum validation loss value is updated and the current model parameters are saved. If the validation loss value of 5 consecutive epochs is not less than the minimum validation loss value, it is considered that the model has begun to overfit. At this time, the training is stopped and the current optimal model parameters are saved. S166. The model with the optimal model parameters is used as the actual straw yield prediction model. The input characteristic parameters of the model include crop type, plot area, land type, soil moisture, product of ambient temperature and precipitation, ambient light, pests and diseases, crop planting time, and Normalized Difference Vegetation Index (NDVI). The output is the straw yield of the plot within a planting cycle.

9. The straw disposal optimization method based on the yield prediction model according to claim 7, characterized in that: The S2 comprises the following sub-steps: S21. Collect straw resource information and reprocessing factory demand information to establish complete and accurate straw resource data and straw demand data, wherein the straw resource data includes straw type, straw yield, straw collection time period, and straw resource location; the straw demand data includes straw type, quality standard (water content, fiber length, etc.), demand, expected delivery time, factory location, etc.; S22, determining the attribute weights of straw resource data and straw demand data through the analytic hierarchy process; S23, calculating the comprehensive matching degree of the straw resource data and the straw demand data according to the attribute weights, and generating a preliminary matching relationship list according to the comprehensive matching degree score, wherein the preliminary matching list includes information such as the name of the reprocessing manufacturer, the factory location, the required straw type, the demand, the expected delivery time, the recommended straw supply source number, the straw resource location, the straw type, the straw yield estimate, the expected collection time, and the comprehensive matching degree score; S24. Feedback the preliminary matching list to the reprocessing manufacturer and the straw supply source, collect feedback from both parties, correct the data, adjust the attribute weights or optimize the matching algorithm based on the feedback, and re-match until both parties can accept the matching results and form a final matching plan.

10. The straw disposal optimization method based on the yield prediction model according to any one of claims 7 to 9, characterized in that: Also includes: S4. Establish a straw trading and logistics management platform for all users based on the alliance chain architecture, including the following sub-steps: S41. Create nodes for all users, including straw suppliers, straw demanders, and regulatory authorities; S42. Define independent communication channels for different business processes; S43. Develop and deploy smart contracts to define transaction rules and operation logic; S44. Establish authority management for different roles in the alliance chain; S45. Encrypt key data that needs to be uploaded to the blockchain; S46. Perform consensus verification on the update of transaction information; S47. Establish a visual query interface to display the entire process information from straw source collection to final processing into finished products for the query needs of authorized users. S5. Collect process data of the straw disposal link, judge the process data according to the set threshold rules, and generate warning information when the monitoring data is within the threshold range; Maintain the hardware and software of the straw disposal optimization system.

Images