Crop growth prediction and regulation system based on multiple sensors

Abstract

The invention discloses a crop growth prediction and regulation system based on multiple sensors, and relates to the technical field of artificial intelligence, and the system comprises the following steps: 1, obtaining monitoring data uploaded by a sensor assembly, the sensor assembly comprises a plurality of sensors, and the plurality of sensors are respectively arranged in different positions of a greenhouse; step 2, inputting the preprocessed monitoring data into an SSA-CNN-LSTM model so as to output a crop growth prediction result; and step 3, based on the crop growth prediction result, realizing regulation and control of the crop growth environment. According to the crop growth prediction and regulation and control system, the growth environment of crops can be comprehensively and accurately monitored based on detection data uploaded by a plurality of sensors.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a technology for predicting and regulating crop growth based on multi-sensors. Background Art

[0002] In modern agricultural production, the growth environment and state of crops directly affect the yield and quality. Traditional manual observation methods have problems of low efficiency and low accuracy. At the same time, the single-sensor monitoring method also has certain limitations and cannot comprehensively and accurately reflect the growth of crops. These problems seriously affect the efficiency and quality of agricultural production. First, traditional manual observation methods usually rely on manual data collection, which has problems such as large workload, long cycle, and being easily affected by subjective factors. Due to the limitations of manual data collection, it is often impossible to timely and accurately understand the growth of crops, resulting in the inability to take effective control measures in a timely manner and affecting the growth and development of crops. Second, although the single-sensor monitoring method can monitor some parameters, due to the limitations of a single sensor, it is often impossible to comprehensively and accurately monitor the growth environment of crops. For example, a single sensor can only provide some data parameters and cannot comprehensively understand key parameters such as soil humidity and light intensity, resulting in incomplete and inaccurate data collection.

[0003] The patent application with the publication number CN117730702A discloses a method and system for crop planting management based on big data, which uses multiple sensors to collect crop growth environment information and crop growth state information in saline-alkali land in real time; based on a preset threshold or the standard growth environment information of crops, the currently collected growth environment information is evaluated in real time, and a control instruction is sent to the corresponding environment adjustment device according to the evaluation result; the environment adjustment device adjusts the growth environment of crops according to the received control instruction; it is judged whether the crops are damaged by pests and diseases based on the crop growth state information, and pest protection is carried out for different disease types. This document is about the environmental management of crops, and the final result is the information of the growth environment data, rather than the information of the crop growth stage.

[0004] The patent application with the publication number CN117972337A discloses an agricultural meteorological disaster monitoring and prediction method based on multi-modal deep learning, including: collecting data such as meteorological factors, crop growth conditions, and soil conditions; preprocessing text and image data respectively, extracting data features and reducing dimensions; inputting the extracted features into a Transformer model for pre-training; using the pre-trained Transformer model as a basic model and then fine-tuning the model in new data; optimizing and adjusting the multi-modal deep learning model by methods such as ten-fold cross-validation and adjusting hyperparameters. The prediction object of this literature is meteorological data, and the prediction result is also meteorological data, without predicting crop growth data through meteorological data. Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to comprehensively and accurately monitor the growth environment of crops.

[0006] The present invention solves the above technical problem through the following technical means: A crop growth prediction and regulation system based on multiple sensors, including multiple sensors, a data processing module, an SSA-CNN-LSTM model, and an intelligent regulation module. The multiple sensors are respectively arranged at different positions in the greenhouse. The prediction and regulation steps using this system include:

[0007] Step 1: The data processing module obtains the monitoring data uploaded by the sensor component;

[0008] Step 2: The data processing module inputs the preprocessed monitoring data into the SSA-CNN-LSTM model to output the crop growth prediction result;

[0009] Step 3: The intelligent regulation module realizes the regulation of the crop growth environment based on the crop growth prediction result.

[0010] As a further optimized technical solution, the multiple sensors in the sensor component are deployed based on a network topology structure.

[0011] As a further optimized technical solution, the multiple sensors are deployed based on a star network, specifically:

[0012] All sensor nodes are connected to a central node. The central node is responsible for receiving and processing the data of the sensor nodes. An Internet of Things wireless repeater is used to expand the diffusion and area coverage of the wireless network signal. The Internet of Things wireless repeater is installed at the edge position of the transmission distance from the central node;

[0013] Multiple paths are introduced in the star network to connect the same destination. When a certain path fails, the monitoring data obtained by the sensor nodes at that destination can continue to be transmitted to the central node through other paths bypassing the faulty node;

[0014] Communication between sensor nodes adopts an encryption protocol. At the same time, a unique identity and key are assigned to each sensor node in the sensor component for security authentication.

[0015] As a further optimized technical solution, the sensor component includes two or more of the following sensors: air temperature and humidity sensor, wind speed sensor, light sensor, carbon dioxide sensor, and soil pH sensor.

[0016] As a further optimized technical solution, in step two, the preprocessing process of the monitoring data is specifically as follows:

[0017] Data collection;

[0018] Perform data normalization on the collected monitoring data, divide the normalized monitoring data into multiple segment data according to the time series, perform outlier detection on the segment data at each moment based on the LSTM model, and mark the segments detected as outliers;

[0019] Data fusion: Perform fusion calculation on the monitoring data based on the extended Kalman filter algorithm and the weighting method. Compensate for the outliers generated in the measurement process of the monitoring data through the extended Kalman filter algorithm, replace the segment data of the original outliers with the compensated values, and input them into the SSA-CNN-LSTM model together with other normal segment data.

[0020] As a further optimized technical solution, in the outlier detection of the segment data at each moment based on the LSTM model and marking the segments detected as outliers, specifically:

[0021] Assume that it is necessary to perform outlier detection on the sensor data at time k. First, predict the segment data at time k based on the detection data input before time k to obtain the predicted value x k0 ;

[0022] Subtract the predicted value x k0 from the true collected value x of the sensor at time k k to obtain the difference m;

[0023] If the difference m is greater than the pre-set threshold, mark the true collected value x of the sensor at time k k as an outlier;

[0024] Output the marked true collected value x k to vacate the position at time k, and the vacated position is dynamically compensated through data fusion;

[0025] Loop the above operations to mark the outliers corresponding to all segment data.

[0026] As a further optimized technical solution, in data fusion, it specifically includes:

[0027] Assume that it is necessary to perform dynamic compensation on the detection data at the k-th moment. First, initialize the state equation, covariance, and noise variance. The state equation is shown in Equation 1, where x k is the state at the k-th moment (current moment), u k is the controller vector, and w k is the process noise; the covariance represents the relationship between the measurement values obtained by sensor sampling and is used to represent the change relationship between two variables. If the covariance is positive, it means that the change trends of the sampling values of these two independent sensors are the same. If the covariance is negative, it means that the change trends of the sensor sampling values are opposite. For a certain sensor, the noise variance represents the degree to which the measurement value at a certain moment deviates from other measurement values. Combining the optimal estimated value of the sensor at the (k - 1)-th moment retained by the system, the predicted value of the sensor at the k-th moment is obtained by solving.

[0028]

[0029] Calculate the covariance of the predicted value of the sensor at the k-th moment based on the predicted value of the sensor at the k-th moment.

[0030] Based on the covariance of the predicted value of the sensor at the k-th moment, combined with the state equation and covariance of the corrected value of the sensor at the (k - 1)-th moment retained, the Kalman gain can be obtained through calculation, and the optimal estimated value of the segment data at the k-th moment is determined.

[0031] Loop the above operations to obtain the optimal estimated values corresponding to all the marked segment data.

[0032] As a further optimized technical solution, the SSA-CNN-LSTM model includes a CNN model and an LSTM model connected to the output end of the CNN model. Among them, a sparrow search algorithm is set at the input end of the CNN-LSTM model. The preprocessed monitoring data is respectively corresponded to the weights in the CNN-LSTM model. Based on the sparrow search algorithm, the optimal search for the preprocessed monitoring data is performed, and the obtained optimal result is used as the optimal weight of the CNN-LSTM model. Based on the optimal weight, the preprocessed monitoring data is processed to obtain the crop growth prediction result.

[0033] As a further optimized technical solution, the CNN-LSTM model is optimized by the sparrow algorithm as follows:

[0034] (2) Initial parameter setting: First, define and initialize the network structure of the CNN-LSTM model, including the settings of the convolutional layer and pooling layer of CNN, the hidden state dimension and number of layers of LSTM, and the number of heads of the attention mechanism; (2) Use the sparrow algorithm to optimize the parameters of the CNN-LSTM model. During the optimization process, minimize the loss function by adjusting parameters such as the weights and biases of the model; (3) After being optimized by the sparrow algorithm, the optimized CNN-LSTM model is obtained.

[0035] As a further optimized technical solution, in step three, the intelligent control module realizes the regulation of the crop growth environment based on the crop growth prediction result through the cloud platform, realizes the state monitoring of the remote monitoring system. The supplementary lighting system for controlling the light uses high-pressure sodium lamps, and the OK-LUX1 light sensor is used to measure the light value. The greenhouse ventilation adopts natural ventilation, and the OK-FS1 wind speed sensor is used to measure the wind speed, observe the measurement values of the CO 2 sensor and the temperature and humidity sensor, and then adjusts it to the optimal wind speed for crop growth through the ventilation system. The irrigation system uses the PH-CT-SFJ automatic water and fertilizer integrated machine.

[0036] The advantages of a crop growth prediction and regulation system based on multi-sensors provided by the present invention are as follows:

[0037] In the structure of the present invention, a crop growth prediction and regulation system based on multi-sensors can comprehensively and accurately monitor the crop growth environment based on the detection data uploaded by multiple sensors;

[0038] When processing multiple data, it is necessary to consider the combination and interaction relationships between different data. For example, temperature may be affected by light; usually, more complex models are needed to process multiple data to better capture the correlations between data; when processing multiple data, it may be necessary to integrate and fuse different data to construct a more comprehensive dataset to provide richer information for the model.

[0039] The intelligent control module realizes the regulation of the crop growth environment according to the prediction result, including but not limited to irrigation control and greenhouse ventilation control.

[0040] A document with the publication number CN117730702A is about the environmental management of crops, and does not involve predicting crop growth through historical environmental data. The final result is the information of the growth environment data, rather than the information of the crop growth stage. For example, predicting the growth height of the next stage of crops through historical environments such as temperature and humidity.

[0041] The literature with the publication number CN117972337A uses a model structure that is very different from that of the present invention, and its prediction object is meteorological data, and the prediction result is also meteorological data. The present invention predicts crop growth data through meteorological data, belonging to predicting different objects. The present invention can instantaneously adjust environmental data through the predicted crop growth data, which is not involved in this literature. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic flowchart of the method for predicting and regulating crop growth based on multi-sensors according to the present invention;

[0043] Figure 2 It is a schematic structural diagram of the system for predicting and regulating crop growth based on multi-sensors according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0044] Next, the technical solution of the present invention will be described in detail through specific embodiments. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0045] As Figure 1 and 2 shown, a method for predicting and regulating crop growth based on multi-sensors proposed by the present invention includes the following steps:

[0046] Step 1: Obtain the monitoring data uploaded by the sensor component. The sensor component includes multiple sensors of various types, and the multiple sensors are respectively arranged at different positions in the greenhouse;

[0047] Using multiple sensors, including temperature sensors, humidity sensors, soil humidity sensors, etc., to achieve comprehensive monitoring of the crop growth environment and improve the comprehensiveness and accuracy of data collection.

[0048] The multiple sensors in the sensor component are deployed based on a network topology structure; the network topology structure determines the connection method and communication path between nodes. Common network topology structures include star networks, mesh networks, and tree networks, etc. Selecting a network topology structure suitable for a specific application scenario can improve the stability and performance of the sensor network.

[0049] For scenarios that require a large number of sensor nodes, the star network topology is a good choice. When multiple sensors are deployed based on the star network, all sensor nodes are connected to a central node, which is responsible for receiving and processing the data of the sensor nodes. This structure is simple and easy to manage, but the central node becomes a single point of failure risk. In a specific embodiment of the present invention, the star topology is adopted, and the specific sensor node layout is as follows:

[0050] First of all, properly and reasonably arranging sensor nodes can achieve effective coverage and communication between nodes:

[0051] (a1) To increase the transmission distance of the star topology and extend the coverage of the network, a WL-R-A IoT wireless repeater is used to expand the diffusion and area coverage of the wireless network signal. The IoT wireless repeater is installed at the edge position of the transmission distance from the central node;

[0052] (a2) Use multi-path transmission. Multi-path transmission can enhance the network stability and reliability, and reduce the single point of failure risk. It is achieved by adding loop redundancy. Multiple paths are introduced in the network topology to connect the same destination, thus forming a loop structure. When a certain path fails, the data can bypass the faulty node through other paths and continue to be transmitted, ensuring the connectivity and availability of the network;

[0053] (a3) Consideration of obstacles. When arranging sensor nodes, it is necessary to consider the influence of obstacles in the surrounding environment on signal transmission. Obstacles such as walls and large equipment may block the signal, resulting in some areas not being covered or unable to communicate normally. By reasonably selecting the node positions, try to avoid or reduce obstacles to obtain better signal transmission effects.

[0054] (a4) Node energy management. Sensor nodes are usually powered by batteries, and energy is the key to the continuous operation of the sensor network. Node energy management is a key factor to ensure the long-term stable operation of the network;

[0055] (a5) Low-power design. During the hardware and software design process of sensor nodes, it should be considered to minimize energy consumption. Reasonably select low-power chips and modules, optimize software algorithms, and reduce the energy consumption of nodes. For example, reduce energy consumption by adopting energy-saving modes, reducing transmission power, and reasonably setting the sensor sampling frequency, etc.;

[0056] (a6) Energy replenishment and collection. For nodes remotely deployed in the sensor network, the batteries cannot be replaced frequently, and it is necessary to consider ways of energy replenishment and collection. Supplement energy through means such as solar panels and kinetic energy generators to extend the service life of the nodes. In addition, design an energy collection module to convert the environmental energy (such as temperature difference, vibration, etc.) around the sensor nodes into electrical energy to supplement the energy of the nodes.

[0057] Secondly, when arranging sensor nodes, security protection should be considered:

[0058] The communication between sensor nodes should adopt an encryption protocol to prevent data from being stolen or tampered with. Encryption algorithms such as AES and RSA are used. When deploying the network, it should be ensured that each node can correctly configure and use the encryption protocol;

[0059] Assign a unique identity and key to each node in the sensor network for security authentication to ensure that only legitimate nodes can join the network. Digital certificates, digital signatures, etc. are used for node identity verification.

[0060] Finally, suitable sensors need to be selected:

[0061] In this embodiment, OK-KWLD1 air temperature and humidity sensor, OK-FS1 wind speed sensor, OK-LUX1 light sensor, OK-CO 2 carbon dioxide sensor and OK-TPH1 soil pH sensor are used.

[0062] Step 2: Input the preprocessed monitoring data into the SSA-CNN-LSTM model to output the prediction results of crop growth, specifically including:

[0063] Analyze and process the data collected by the sensors through the data processing module and the SSA-CNN-LSTM model to achieve accurate prediction of the crop growth status and improve the accuracy and reliability of the prediction.

[0064] Among them, (b) an LSTM-based outlier or failure value marker is added to the data processing module, effectively solving the situation that when the extended Kalman filter algorithm performs data fusion calculation, if a sensor fails or an outlier is received, the entire prediction system will make a wrong judgment or even collapse. The preprocessing process of the detection data by the data processing module is as follows:

[0065] (b1) Data collection: This is the first step of data processing. First, transmit the monitoring data collected by the sensors to the data processing module in chronological order for data analysis and processing;

[0066] (b2) Data normalization: Monitoring stations (sensors) often generate detection data with different patterns and characteristics, resulting in a large amount and variability of the detection data. Therefore, these detection data are first normalized to make the features have the same measurement scale.

[0067] The maximum-minimum normalization function is selected to map the detected data points to the interval [0, 1] to improve the prediction accuracy of SSA-CNN-LSTM. To detect whether the detected data uploaded by the sensor is an outlier, the detected data is divided into multiple segments according to the time series to obtain segment data, so as to minimize the influence of data distribution. The divided time step is set to 30 times the sensor sampling time interval.

[0068] Then, outlier detection is performed on the segment data at each moment based on the LSTM model. Suppose it is necessary to perform outlier detection on the sensor data at time k. First, based on the segment data input before the k-th moment, the segment data at the k-th moment is predicted to obtain the predicted value x k0 , and the predicted value x k0 is subtracted from the true collected value x k of the sensor at time k to obtain the difference value m; if the difference value m is greater than the pre-set threshold, the true collected value x k of the sensor at time k is marked as an outlier, and the marked true collected value x k is output to vacate the position at the k-th moment, which is convenient for the next extended Kalman filter algorithm to dynamically fit the predicted value to fill the outlier. That is to say, the vacated position is dynamically compensated by fitting the predicted value through the extended Kalman filter algorithm in data fusion; the above operations are looped to mark the outliers corresponding to all segment data.

[0069] (b3) Data fusion

[0070] The monitoring data of the air quality sensor is fused and calculated based on the extended Kalman filter algorithm and the weighting method. The extended Kalman filter algorithm can compensate for the outliers (dynamic errors) generated in the measurement process. The compensated value replaces the segment data of the original outliers and is input into the SSA-CNN-LSTM model together with other normal segment data, which can effectively process the dynamic noise data of the sensor, and then realize the high-precision measurement of the target state of the sensor, so as to obtain the estimated value closest to the true air quality data.

[0071] Suppose it is necessary to perform dynamic compensation on the segment data at the k-th moment. First, the state equation, covariance, and noise variance are initialized. The state equation is shown in Equation 1, where x k is the state at time k (the current moment), u k is the controller vector, and w kis the process noise; the covariance represents the relationship between the measurement values obtained by sensor sampling and is used to represent the change relationship between two variables. If the covariance is positive, it means that the change trends of the two independent sensor sampling values are the same. If the covariance is negative, it means that the change trends of the sensor sampling values are opposite. The noise variance, for a certain sensor, represents the degree to which the measurement value at a certain moment deviates from other measurement values. Combining the optimal estimated value of the sensor at the (k - 1)-th moment retained by the system, the predicted value of the sensor at the k-th moment is solved; based on the predicted value of the sensor at the k-th moment, the covariance of the predicted value of the sensor at the k-th moment is calculated; based on the covariance of the predicted value of the sensor at the k-th moment, combining the state equation and covariance of the corrected value of the sensor at the (k - 1)-th moment retained, the optimal estimated value of the segment data at the k-th moment is calculated; loop the above operations to obtain the optimal estimated values corresponding to all the marked segment data.

[0072] (c) The SSA-CNN-LSTM model includes a CNN model and an LSTM model connected to the output end of the CNN model. The input end of the CNN-LSTM model is set with the sparrow search algorithm. The preprocessed monitoring data is respectively corresponded to the weights in the CNN-LSTM model. Based on the sparrow search algorithm, the optimal search is carried out on the preprocessed monitoring data, and the obtained optimal result is used as the optimal weight of the CNN-LSTM model. Based on the optimal weight, the preprocessed monitoring data is processed to obtain the crop growth prediction result;

[0073] Compared with the basic sparrow search algorithm, the sparrow search algorithm based on elite opposition-based learning is mainly improved in that after the sparrow update, 10% of the sparrow fitness rankings are taken as the elite solutions. At the same time, the dynamic boundaries of the elite sparrows are obtained, and the opposition-based learning strategy is used to solve the opposition-based solutions. Compare the sparrows before and after the update. If it is better, replace the previous sparrow. The improved sparrow search algorithm process is as follows: (1) Initialize the population, the number of iterations, the proportions of predators and joiners; (2) Apply the elite opposition-based learning strategy to the initialization stage. The basic idea of opposition-based learning is based on the current solution, and the corresponding opposition-based solution is found through the opposition-based learning mechanism, and then evaluated and compared to save the better solution. The introduction of the opposition-based solution can expand the search area of the algorithm, but opposition-based learning has a certain degree of blindness, and the search space where the opposition-based solution is located may not be more conducive to the search space of the current solution. In view of this situation, an elite strategy is added, elite individuals are introduced, and opposition-based learning is carried out through the elite individuals, making full use of the effective information of the elite individuals to generate elite opposition-based solutions, guiding the search process to approach the optimal solution, and selecting excellent individuals from the current solution and the elite opposition-based solutions as the next generation population elite opposition-based learning. Let be a solution of the i-th sparrow in the J-dimensional search space at the t-th iteration, and its opposition-based solution is (x i ') t . F(x) is the fitness function. When At this time, it is called the elite individual in the t-th iteration; when At this time, it is called the ordinary individual in the t-th iteration. The definition of the elite reverse solution is shown in Equation 1, where k is a random number between 0 and 1, and are the upper and lower bounds constructed for the elite individual; (3) Calculate the fitness value and sort; (4) Use Formula 2 to update the predator position. Q is the normal distribution mean, L is a 1*d matrix, ST is the safety value, and R 2 is the warning value. When R 2 < ST, it means it is safe, and at this time the search range of the predator is relatively large; when R 2 ≤ ST, it means that there are a certain number of predators, and at this time it is necessary to move to a safe area; (5) Use Formula 3 to update the joiner position. X worst is the current global optimal value, X t+1 P is the best position of the discoverer. Here, n is the number of joiners, not the population size; (6) Use Equation 4 to update the vigilant position. X t best is the current global optimal position, K is a random number in [-1, 1], f i is the current fitness value, f g is the best fitness value, f w is the worst fitness value, and β is a random number of normal distribution; (7) Calculate the fitness value and update the sparrow position. Here, i is the number of sparrows that are aware of danger during traversal, not the entire population; (8) Obtain the elite sparrows, calculate the dynamic change of their boundaries, and use the elite reverse learning strategy to update the elite sparrows; (9) Calculate the fitness value and update the sparrow position; (10) Whether the stop condition is met. If it is met, exit and output the result. Otherwise, repeat steps 3-9.

[0074] The method for optimizing the CNN-LSTM model by the sparrow algorithm is as follows: (1) Initial parameter setting: First, define and initialize the network structure of the CNN-LSTM model. This includes the settings of the convolutional layer and pooling layer of the CNN, the hidden state dimension and number of layers of the LSTM, the number of heads of the attention mechanism, etc. (2) Use the sparrow algorithm to optimize the parameters of the CNN-LSTM model. During the optimization process, the loss function can be minimized by adjusting parameters such as the weights and biases of the model to improve the prediction performance of the model on multivariate time series. (3) After being optimized by the sparrow algorithm, an optimized CNN-LSTM model is obtained. By inputting the historical data of the sensor, the growth of crops is predicted.

[0075] Step 3: Based on the crop growth prediction results, realize the regulation of the crop growth environment.

[0076] The intelligent control module realizes the intelligent regulation of the crop growth environment according to the prediction results, can take effective regulation measures in time, and improve the growth stability and yield of crops.

[0077] The intelligent control module can realize the automatic regulation of the growth environment without manual intervention. When the sensor monitors that the environmental parameters deviate from the preset range or the growth model predicts that the crop needs specific conditions, the intelligent control module will automatically adjust the relevant parameters to ensure that the crop growth is in the best state.

[0078] The main way for the intelligent control module to achieve is through the cloud platform, to realize the status monitoring of the remote monitoring system and conduct intelligent early warning. When growing crops in a greenhouse, water, carbon dioxide, temperature, humidity and plant nutrient supply can be precisely controlled. Similarly, light can also be precisely controlled, and the artificial light system is used to control light. There are several plant pigments in plants that can absorb light energy and provide plant photosynthesis, but the main plant pigment is chlorophyll.

[0079] For the supplementary lighting system that controls light, the spectral diagrams of the 600W and 1000W agricultural sodium lamps, which are the most popular in commercial glass greenhouses, are shown for high-pressure sodium lamps. High-pressure sodium lamps have very good application effects in greenhouse sheds for growing crops because high-pressure sodium lamps have very high light output. For example, the Philips 1000W high-pressure sodium lamp can reach a light output of more than 150,000 lumens and a PPFD of more than 2100 μmol / (m2s), and the output maintenance rate is also very good. After 10,000h of use, it can still ensure an output characteristic of more than 95%. However, high-pressure sodium lamps also have some disadvantages, such as a large increase in heat with the increase of power and less blue spectral components. In this embodiment, the OK-LUX1 light sensor is used to measure the light value.

[0080] To meet the needs of greenhouse crops for temperature, humidity and CO 2 concentration control, the design of the ventilation system is of course important. Natural ventilation should be adopted first for greenhouse ventilation. When the natural ventilation system cannot meet the greenhouse design requirements, a fan ventilation system should be set up. When designing the ventilation system, the type, size and position of the ventilation air vents and the type, quantity and layout of the fans should be determined according to the geographical location, climate type, greenhouse structure type and use of the greenhouse. The wind speed is measured according to the OK-FS1 wind speed sensor, and the measured values of sensors such as CO 2 sensors and temperature and humidity sensors are observed, and then the ventilation system is used to regulate to the best wind speed for crop growth.

[0081] The PH-CT-SFJ automatic water and fertilizer integrated machine adopted by the irrigation system is a modern advanced agricultural technology developed by combining drip irrigation with fertilization. This technology mainly relies on the drip irrigation system, uses the water in the drip irrigation system as a carrier, and applies fertilizer during irrigation to achieve the integrated utilization and management of water and fertilizer, enabling water and fertilizer to be supplied to crops for absorption and utilization in an optimized combined state in the soil.

[0082] Through steps one to three, based on the detection data uploaded by multiple sensors, the growth environment of crops can be comprehensively and accurately monitored. When processing multiple data, it is necessary to consider the combination and interaction relationships between different data. For example, temperature may be affected by light; processing multiple data usually requires the use of more complex models to better capture the correlations between data; when processing multiple data, it may be necessary to integrate and fuse different data to construct a more comprehensive dataset to provide richer information for the model. The intelligent regulation module realizes the regulation of the crop growth environment according to the prediction results, including but not limited to irrigation control and greenhouse ventilation control.

[0083] As described above, only the preferred specific embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, should be covered by the protection scope of the present invention.

Claims

1. A crop growth prediction and control system based on multiple sensors, characterized in that: It includes multiple sensors, data processing modules, SSA-CNN-LSTM models and intelligent control modules. Multiple sensors are set in different positions of the greenhouse. The prediction and control steps of the system include: Step 1: The data processing module obtains the monitoring data uploaded by the sensor component; Step 2: The data processing module inputs the preprocessed monitoring data into the SSA-CNN-LSTM model to output the crop growth prediction results; Step 3: The intelligent control module controls the crop growth environment based on the crop growth prediction results.

2. The multi-sensor based crop growth prediction and control system according to claim 1, characterized in that: The multiple sensors in the sensor component are deployed based on a network topology.

3. The multi-sensor based crop growth prediction and control system according to claim 2, characterized in that: Multiple sensors are deployed based on a star network, specifically: All sensor nodes are connected to a central node, which is responsible for receiving and processing the data of the sensor nodes. The IoT wireless repeater is used to expand the diffusion and regional coverage of the wireless network signal. The IoT wireless repeater is installed at the edge of the transmission distance from the central node. Introducing multiple paths to connect the same destination in a star network, when a path fails, the monitoring data acquired by the sensor node at the destination can be transmitted to the central node through other paths bypassing the failed node; The communication between sensor nodes uses encryption protocol, and each sensor node in the sensor component is assigned a unique identity and key for security authentication.

4. The crop growth prediction and control system based on multiple sensors according to claim 1, characterized in that: The sensor assembly includes two or more of the following sensors: an air temperature and humidity sensor, a wind speed sensor, a light sensor, a carbon dioxide sensor, and a soil pH sensor.

5. The crop growth prediction and control system based on multiple sensors according to claim 1, characterized in that: In step 2, the preprocessing process of monitoring data is as follows: Data collection; Perform data normalization on the collected monitoring data, divide the normalized monitoring data into multiple fragments according to the time series, perform outlier detection on the fragment data at each moment based on the LSTM model, and mark the fragments detected as outliers; Data fusion: The monitoring data is fused and calculated based on the extended Kalman filter algorithm and the weighted method. The abnormal values ​​generated in the monitoring data measurement process are compensated by the extended Kalman filter algorithm. The compensated values ​​replace the original abnormal value fragment data and are input into the SSA-CNN-LSTM model together with other normal fragment data.

6. The multi-sensor based crop growth prediction and control system according to claim 5, characterized in that: In the process of performing outlier detection on the fragment data at each moment based on the LSTM model and marking the fragments detected as outliers, the specific steps are as follows: Assume that the sensor data at time k needs to be detected for anomalies. First, the segment data at time k is predicted based on the detection data input before time k to obtain the predicted value x k0 ; The predicted value x k0 and the actual sensor value x at time k k Make a difference and get the difference value m; If the difference m is greater than the preset threshold, mark the actual sensor collection value x at the kth moment k is an outlier; The actual collected value x after marking k Output, so as to vacate the position at the kth moment, and the vacated position is dynamically compensated through data fusion; The above operations are repeated to mark the abnormal values ​​corresponding to all fragment data.

7. The multi-sensor based crop growth prediction and control system according to claim 5, characterized in that: Data fusion includes: Assuming that the detection data at the kth moment needs to be dynamically compensated, the state equation, covariance and noise variance are first initialized. The state equation is shown in Formula 1, x k is the state at time k (current time), u k is the controller vector, w k is process noise; covariance represents the relationship between the measured values ​​obtained by sensor sampling, and is used to represent the change relationship between two variables. If the covariance is a positive value, it means that the change trends of the two independent sensor sampling values ​​are the same. If the covariance is a negative value, it means that the change trends of the sensor sampling values ​​are opposite. For a certain sensor, the noise variance is the degree to which the measured value at a certain moment deviates from other measured values. Combined with the optimal estimated value of the sensor at the k-1th moment retained by the system, the sensor prediction value at the kth moment is solved. The covariance of the sensor prediction value at the kth moment is calculated based on the sensor prediction value at the kth moment; Based on the covariance of the sensor prediction value at the kth moment, combined with the state equation and covariance of the retained sensor correction value at the k-1th moment, the Kalman gain can be calculated to determine the optimal estimate of the segment data at the kth moment; The above operations are repeated to obtain the optimal estimated values ​​corresponding to all the marked fragment data.

8. The multi-sensor based crop growth prediction and control system according to claim 5, characterized in that: The SSA-CNN-LSTM model includes a CNN model and an LSTM model connected to the output end of the CNN model, wherein a sparrow search algorithm is set at the input end of the CNN-LSTM model, and the preprocessed monitoring data are respectively matched with the weights in the CNN-LSTM model. The preprocessed monitoring data is optimally searched based on the sparrow search algorithm, and the optimal result is used as the optimal weight of the CNN-LSTM model. The preprocessed monitoring data is processed based on the optimal weight to obtain crop growth prediction results.

9. The multi-sensor based crop growth prediction and control system according to claim 8, characterized in that: The CNN-LSTM model is optimized by the Sparrow algorithm as follows: (1) Initial parameter setting: First, define and initialize the network structure of the CNN-LSTM model, including the convolutional layer and pooling layer settings of the CNN, the hidden state dimension and number of layers of the LSTM, and the number of heads of the attention mechanism; (2) Use the Sparrow algorithm to optimize the parameters of the CNN-LSTM model. During the optimization process, the loss function is minimized by adjusting the model's weights, biases and other parameters; (3) After optimization by the Sparrow algorithm, the optimized CNN-LSTM model is obtained.

10. The multi-sensor based crop growth prediction and control system according to claim 8, characterized in that: In step three, the intelligent control module controls the crop growth environment based on the crop growth prediction results. It realizes the status of the remote monitoring system through the cloud platform. The light supplement system uses high-pressure sodium lamps to control the light value. The OK-LUX1 light sensor is used to measure the light value. The greenhouse ventilation adopts natural ventilation. The wind speed is measured according to the OK-FS1 wind speed sensor. The measured values ​​of the CO2 sensor and the temperature and humidity sensor are observed, and then the ventilation system is used to control the optimal wind speed for crop growth. The irrigation system uses the PH-CT-SFJ automatic water and fertilizer integrated machine.

Images