A flue-cured tobacco and mushroom integrated controller and a control method thereof

Abstract

The present application relates to the technical field of intelligent agricultural environment control, and discloses a flue-cured tobacco and mushroom integrated controller and a control method thereof, which comprises a multi-modal environment perception module, a mushroom growth stage intelligent identification module, a parameter coordination module, a collaborative execution control module and a self-adaptive feedback learning module. The controller realizes intelligent perception-analysis-execution closed-loop control of the mushroom planting process by real-time identification and parameter coordination of the differentiated requirements of temperature, humidity, light and CO2 concentration in different growth stages of the mushroom. The controller is suitable for staggered planting of flue-cured tobacco and mushrooms. During the flue-cured tobacco curing season, the traditional curing process is manually controlled by tobacco farmers according to the yellowing period, color fixing period and dry tendon period. The controller controls the curing process manually. During the mushroom growing season after the flue-cured tobacco season, the controller takes full responsibility for intelligent control of the mushrooms. The same facility is staggered and reused between the flue-cured tobacco season and the mushroom season. They are independent and do not interfere with each other.

Description

Technical Field

[0001] This invention relates to the field of intelligent agricultural environmental control technology, and in particular to an integrated controller and control method for flue-cured tobacco and mushrooms driven by artificial intelligence. Background Technology

[0002] In recent years, with the rapid development of facility agriculture and smart agriculture, crop environmental control technology within greenhouses has become a key link in improving the quality and efficiency of modern agriculture. In traditional tobacco cultivation and mushroom cultivation, tobacco curing chambers and mushroom cultivation rooms are typically built as separate facilities, equipped with their own temperature and humidity control systems and light control devices. This leads to the duplication of land resources and redundant investment in hardware. Furthermore, tobacco curing is highly seasonal; after the curing season ends each year from August to October, dedicated curing facilities often remain idle for extended periods, resulting in low annual utilization rates and low returns on investment. If mushroom cultivation could be carried out in the same facility at different times, allowing the tobacco curing facilities to be used for mushroom cultivation during the non-curing season, it would significantly improve facility utilization and expand the income sources for tobacco farmers.

[0003] From the perspective of control strategies, mushrooms have significantly different environmental parameter requirements at different growth stages: the mycelial culture stage requires a constant temperature of about 25°C and complete darkness; the primordia formation stage requires a strong stimulus of a sudden drop in temperature to about 15°C and the introduction of weak diffused light; the fruiting body growth stage requires a relatively cool environment of about 18.5°C combined with high humidity conditions of 85%RH to 95%RH; and the harvesting period requires maintaining suitable temperature and humidity and controlling... Concentration is crucial to ensure mushroom quality. This phased shift in demand makes it difficult for traditional rule-based controllers or simple PID control algorithms, which rely on fixed thresholds, to accurately and dynamically adjust environmental parameters according to the mushroom's growth status. This results in significant room for improvement in both the biological efficiency of mushrooms and the yield and quality of each batch. For flue-cured tobacco, since the process has been relatively mature through long-term practice, tobacco farmers typically use manual assistance to adjust the temperature and duration of the curing chamber according to the requirements of the yellowing, color-fixing, and drying stages. No additional intelligent judgment or closed-loop control is needed for the curing process, and any intelligent equipment added to the curing process must ensure that it does not interfere with the farmers' existing mature operational procedures.

[0004] Furthermore, existing environmental controllers for mushroom cultivation generally lack the ability to intelligently identify the real-time growth stages of mushrooms. They struggle to automatically adjust environmental parameters based on dynamic changes at different stages, such as mycelial growth, primordia formation, fruiting body growth, and harvesting, resulting in insufficient precision and adaptability in environmental control. At the execution end, the mushroom growth environment is highly sensitive to temperature, humidity, light, and... The coordination of multi-dimensional parameters such as concentration is highly required. If there is a lack of coordination mechanism between the actuators, it is easy for heating and heat dissipation, humidification and ventilation to cancel each other out, resulting in energy waste and oscillation of environmental parameters.

[0005] To address the aforementioned issues, there is an urgent need for an integrated intelligent controller and its control method that is centered on intelligent mushroom cultivation, possesses intelligent identification of mushroom growth stages, coordinates and controls environmental parameters at each stage of mushroom growth, and achieves refined regulation of mushroom cultivation through a closed-loop perception-analysis-execution mechanism. Furthermore, the controller's deployment should be adaptable to the staggered planting operation mode of flue-cured tobacco and mushrooms—that is, during the tobacco curing season (August to October), tobacco farmers complete the tobacco curing process, and the controller provides manual assistance in controlling the curing process; during the mushroom growing season (October to July of the following year), after the tobacco season ends, the controller fully activates intelligent regulation of mushroom cultivation, enabling the staggered reuse of the same facility in different crop seasons, with the two production systems operating independently and without interference. Summary of the Invention

[0006] To address the core bottlenecks in existing mushroom environmental control technologies, such as the inability to intelligently identify mushroom growth stages in real time, insufficient precision in the coordinated regulation of multi-dimensional environmental parameters, and low utilization rate of existing dedicated curing facilities after the tobacco season due to long-term idleness, this invention provides an integrated controller and control method for flue-cured tobacco and mushrooms. By intelligently identifying mushroom growth stages and coordinating multi-objective game parameters, a closed-loop control architecture of perception-analysis-execution is formed. Under the constraint of staggered planting of tobacco and mushrooms, this invention achieves the core objectives of compatibility and reuse of the same facility for artificial curing during the tobacco season and intelligent planting during the mushroom season at the mechanistic level, and significantly improves the biological efficiency of mushrooms and the annualized utilization rate of the facility.

[0007] The technical solution of this invention is as follows: An integrated controller for flue-cured tobacco and mushroom cultivation includes a multimodal environment sensing module, a mushroom growth stage intelligent identification module, a parameter coordination module, a collaborative execution control module, and an adaptive feedback learning module. The multimodal environment sensing module collects real-time data on temperature, humidity, light intensity, and other parameters within the mushroom cultivation space using a distributed sensor array. Concentration and soil moisture content data are used to generate an environmental state vector after time-series alignment and outlier filtering. The intelligent mushroom growth stage recognition module receives the environmental state vector and mushroom morphology image data, identifies the current growth stage of the mushroom based on a lightweight convolutional neural network, and fuses the recognition result with the environmental state vector to generate a mushroom growth operation state feature map. The parameter coordination module receives the mushroom growth operation state feature map and constructs a multi-objective game coordination model based on the differentiated requirements of the mushroom for four-dimensional environmental parameters at the current growth stage. It obtains the coordinated environmental parameter combination that maximizes the overall suitability of the mushroom by solving for the Nash equilibrium point. The collaborative execution control module receives the coordinated environmental parameter combination and outputs control signals to the temperature control actuator, humidity control actuator, light control actuator, and ventilation actuator. The adaptive feedback learning module collects the actual environmental parameters and mushroom growth response data after execution, calculates the deviation from the target parameters, and updates the recognition model parameters and coordination model parameters through an online learning mechanism to form a closed-loop optimization. The controller is adapted to staggered planting scenarios for flue-cured tobacco and mushrooms. During the flue-cured tobacco curing season from August to October each year, tobacco farmers manually complete the curing process according to the process requirements, and the controller manually controls the process. During the mushroom growing season from October to July of the following year after the flue-cured tobacco season, the controller undertakes the intelligent regulation of mushroom planting throughout the entire process. The operation of the two periods is independent and does not interfere with each other.

[0008] This invention also provides a control method based on the aforementioned controller, which operates during the mushroom growing season under a staggered planting scenario of flue-cured tobacco and mushrooms. During the flue-cured tobacco curing season, tobacco farmers manually complete the curing process using traditional methods, without intervention from the aforementioned method. During the mushroom growing season, the method includes: Step S1, a multimodal environment perception step, collecting multidimensional environmental parameter data within the mushroom planting space and generating an environmental state vector; Step S2, an intelligent mushroom growth stage identification step, identifying the current growth stage of the mushrooms based on the environmental state vector and mushroom morphology image data, and generating a mushroom growth operation state feature map; Step S3, a parameter coordination step, constructing a multi-objective game coordination model and solving for the Nash equilibrium point to obtain a coordinated combination of environmental parameters; Step S4, a collaborative execution control step, outputting control signals to each actuator to adjust the environmental parameters of the planting space; and Step S5, an adaptive feedback learning step, updating the model parameters based on the deviation value and returning to Step S1 to execute closed-loop control.

[0009] The beneficial effects of this invention are as follows: First, the intelligent identification module for mushroom growth stages extracts features from mushroom morphological image data based on a lightweight convolutional neural network, and fuses the identification results with environmental state vectors across modes to generate a mushroom growth operation state feature map. This enables the controller to perceive in real time the different growth stages of the mushroom, such as the mycelial culture stage, primordium formation stage, fruiting body growth stage, or harvesting stage. The mechanism is that through the dual effect of depthwise separable convolution and attention fusion, the model retains the fine-grained discrimination ability of morphological images and introduces the contextual information of environmental parameters. Compared with the existing stage judgment methods based on fixed timetables or human experience, the identification accuracy in the 4-stage mushroom classification task can reach more than 96.5%.

[0010] Second, through the parameter coordination module, the overall suitability function of mushrooms is used as the game payoff, along with temperature, humidity, light, and... A multi-objective game coordination model is constructed using the four dimensions of concentration as the game strategy space. By solving for the Nash equilibrium point, the combination of coordination environment parameters is obtained, enabling the controller to find the equilibrium point that maximizes overall fitness when the demands of the four dimensions are mutually constrained. The mechanism is that the Nash equilibrium ensures that under the parameter combination, individual adjustment of any dimension cannot further improve the overall fitness. Compared with traditional single-dimensional threshold control, multi-dimensional game coordination keeps the overall fitness index stably above 0.85.

[0011] Third, the adaptive feedback learning module fine-tunes the parameters of the recognition and coordination models online based on the mushroom growth response reward signal. Combined with the two-layer learning architecture of edge online fine-tuning and cloud offline retraining, the controller can continuously adapt to the actual environmental differences of different mushroom varieties and different planting batches. The mechanism is that online fine-tuning locally corrects the model's fitting deviation to the current environment with a small learning rate, while cloud retraining improves the model's generalization ability with large-scale data. The two complement each other. Compared with the control strategy that is fixed after one-time training, the comprehensive suitability index can be improved from the initial 0.82 to 0.91 within 30 days of operation.

[0012] Fourth, the deployment of the controller is strictly adapted to the staggered planting mode of flue-cured tobacco and mushrooms. During the flue-cured tobacco curing season, farmers manually cure the tobacco using traditional methods, while the controller manually controls the curing process. This allows the invention to be deployed without changing the farmers' existing curing techniques. The mechanism is that the controller's intelligent logic is only activated during the mushroom growing season. During the flue-cured tobacco season, only basic environmental data is collected for facility monitoring. After manual control, control commands are sent to the curing actuators. Compared to the scheme of mandatory intelligent full-process takeover, this invention has significant advantages in terms of farmer acceptance and process compatibility, and increases the annual utilization rate of the same facility by more than 50% compared to the single flue-cured tobacco mode. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the system architecture of an integrated controller for flue-cured tobacco and mushrooms provided in an embodiment of the present invention.

[0014] Figure 2 This is a flowchart of a control method for an integrated controller for flue-cured tobacco and mushrooms provided in an embodiment of the present invention. Detailed Implementation

[0015] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0016] like Figure 1 As shown in the figure, an integrated controller for flue-cured tobacco and mushrooms provided in this embodiment of the invention includes a multimodal environment perception module 1, a mushroom growth stage intelligent identification module 2, a parameter coordination module 3, a collaborative execution control module 4, and an adaptive feedback learning module 5. These five functional modules form a deeply coupled closed-loop collaborative architecture: the environmental state vector collected by the multimodal environment perception module 1 serves as the key input to the mushroom growth stage intelligent identification module 2; the mushroom growth operation state feature map output by the mushroom growth stage intelligent identification module 2 drives the model construction of the parameter coordination module 3; the coordinated environmental parameter combination output by the parameter coordination module 3 serves as the execution basis for the collaborative execution control module 4; and the adaptive feedback learning module 5 transmits the execution effect back to the mushroom growth stage intelligent identification module 2 and the parameter coordination module 3, realizing online iterative updates of the model parameters. This closed-loop structure combining forward driving and reverse feedback enables the entire controller to continuously self-optimize during the mushroom growing season, adapting to changes in environmental conditions for different mushroom varieties and different planting batches.

[0017] It should be noted that this controller only starts working during the mushroom growing season after the tobacco curing season ends. During the tobacco curing season (August to October each year), the tobacco curing process is manually regulated by tobacco farmers according to the traditional process of yellowing stage, color fixing stage, and dry rib stage. The controller is manually controlled to control the curing process. The deployment of this controller does not affect the normal curing operation of the same facility during the tobacco curing season. The operation of the two crop seasons is independent of each other in time, does not overlap, and does not interfere with each other.

[0018] In one embodiment of the present invention, the multimodal environmental sensing module 1 includes a distributed sensor array located within the planting space, the array comprising a temperature sensor, a humidity sensor, a light sensor, and... The system comprises a concentration sensor and a soil moisture sensor. Preferably, the temperature sensor is a high-precision digital temperature sensor with a measurement range covering -10℃ to 60℃, a resolution of 0.1℃, and a measurement accuracy of ±0.3℃. The humidity sensor is a capacitive relative humidity sensor with a measurement range of 0%RH to 100%RH and an accuracy of ±1.5%RH. The light sensor is a silicon photodiode type sensor with a range covering 0 lux to 200,000 lux, suitable for various stages of mushroom cultivation from light-shielded cultivation to diffused light irradiation. The concentration sensor uses the non-dispersive infrared absorption principle, with a measurement range of 0 ppm to 5000 ppm and an accuracy of ±30 ppm. The soil moisture sensor uses the frequency domain reflectometer principle, with a measurement range of 0% to 100% and an accuracy of ±2%.

[0019] Regarding the spatial layout of the sensor nodes, in one embodiment of the present invention, the spacing between adjacent sensor nodes is set to 1.5m to 3m to control deployment costs while ensuring spatial sampling density. For a typical greenhouse with an area of ​​200m², a grid deployment with a node spacing of 2m would require approximately 50 sensor nodes. The data acquisition cycle for each node is configured differently based on the rate of change of environmental parameters: the acquisition cycle for temperature and humidity data is set to 10s, and the acquisition cycle for light intensity data is set to 30s. The collection period for concentration data was set to 15 seconds, and the collection period for soil moisture content data was set to 60 seconds.

[0020] Preferably, the multimodal environment sensing module 1 further includes a data preprocessing unit, which performs three levels of preprocessing operations on the raw data reported by each sensor node. The first is time-series alignment processing. Since different types of sensors have different acquisition cycles, the data preprocessing unit uses the shortest acquisition cycle as the time reference and unifies the data from all sensor channels onto the same time axis through linear interpolation. The second is outlier filtering processing, which employs an adaptive filtering method based on the midpoint of a sliding window. When a sensor's reading at time t... Values ​​that meet the following criteria are considered outliers and are replaced with the window value: , in: This is the actual measured reading of the sensor at time t, with the unit consistent with the physical quantity measured by the corresponding sensor (such as temperature ℃, humidity %RH, etc.). For a moment At that time The median of the readings within the sliding window; The width of the sliding window (a scalar, with a value of 10 sampling points); The sensitivity coefficient (scalar, dimensionless, with a value of 3.0). This represents the median absolute deviation of the readings within the window from their median. This criterion is based on robust statistical theory and has a strong ability to suppress occasional anomalies.

[0021] Thirdly, environmental state vector generation involves assembling the aligned and filtered 5D sensor data into a 5D environmental state vector at each time step. , in: For a moment The environment state vector (5-dimensional column vector); For a moment The temperature value, in °C; This is the relative humidity value, expressed in %RH. This is the light intensity value, measured in lux. for Concentration value, in ppm; This represents the soil moisture content, expressed as a percentage. This environmental state vector serves as the core input data for the subsequent intelligent identification module 2 for mushroom growth stages.

[0022] In one embodiment of the present invention, the intelligent identification module 2 for mushroom growth stages is used to receive the environmental state vector output by the multimodal environmental perception module 1. The system also uses mushroom morphology image data acquired through an image acquisition unit to identify the current growth stage of the mushrooms in real time. The image acquisition unit consists of at least one set of industrial-grade cameras, aimed at the mushroom cultivation area, with a frame acquisition frequency of one frame every 10 minutes and an image resolution of no less than 1280×960 pixels.

[0023] Regarding the division of mushroom growth stages, this invention divides the entire growth cycle of mushrooms into four stages: mycelial culture stage, primordium formation stage, fruiting body growth stage, and harvesting stage. The mycelial culture stage is characterized by a white, fluffy mycelium covering the surface of the culture medium; the primordium formation stage is characterized by the appearance of rice-grain-sized protruding primordia on the surface of the culture medium; the fruiting body growth stage is characterized by the gradual unfolding of the cap and elongation of the stipe; and the harvesting stage is characterized by the cap being fully unfolded or slightly rolled outwards.

[0024] Preferably, the lightweight convolutional neural network in the mushroom growth stage intelligent recognition module 2 adopts a depthwise separable convolutional architecture. In one embodiment of the present invention, the feature extraction backbone of this network adopts an improved MobileNetV3-Small structure, with an input image size of 224×224×3 pixels. The backbone network contains 11 bottleneck modules, each of which adopts a combination structure of depthwise separable convolution and pointwise convolution, wherein the depthwise convolution kernel size is 3×3 or 5×5, and the channel expansion ratio is 4 to 6. The last bottleneck module of the backbone network outputs a 576-dimensional feature vector. Above the backbone network, the present invention sets a classification head: this classification head is a mushroom growth stage classification head, containing a fully connected layer and a 4-class Softmax output layer, corresponding to the four stages of mycelial culture, primordia formation, fruiting body growth, and harvesting. The total number of parameters of the entire network is controlled within 4.2M, and the single-frame inference time on the AI ​​edge computing unit equipped with an NPU acceleration chip does not exceed 35ms.

[0025] Furthermore, the intelligent recognition module 2 for mushroom growth stages also includes an attention fusion layer, which is used to integrate the environmental state vector. Cross-modal fusion with image features. Specifically, environment state vectors. First, it is mapped to an environment embedding vector of the same dimension as the image features through a two-layer fully connected network. The dimension is 576. Subsequently, the attention fusion layer calculates the cross-modal attention weights. : , in: The attention weight vector (576 dimensions) has a value range of... ; The 576-dimensional image feature vector output by the backbone network; The environment embedding vector (576 dimensions); This indicates a vector concatenation operation, resulting in a vector with a dimension of 1152 after concatenation. This is the attention weight matrix, with dimensions of 576×1152; This is the bias vector, with a dimension of 576; The Sigmoid activation function is defined as follows: Its output range is .

[0026] Characteristic diagram of the growth status of fused mushrooms Calculated using the following formula: , in: For a moment A feature map of the mushroom growth operation status (576-dimensional vector); This represents the element-wise multiplication operation; , , The definition is the same as above. This cross-modal fusion mechanism enables the identification of growth stages to not only rely on morphological image information, but also utilize contextual information from environmental parameters to improve the accuracy and robustness of identification. Preferably, in the classification task of mushroom 4 stages, this identification model achieves an average classification accuracy of 96.5% on a validation set containing 6000 mushroom morphologically labeled images.

[0027] In one embodiment of the present invention, the parameter coordination module 3 is used to receive the mushroom growth operation status feature map output by the mushroom growth stage intelligent recognition module 2. And based on the mushrooms' requirements for temperature, humidity, light, and [other factors] at their current growth stage. Differential concentration requirements are identified, and a multi-objective game-theoretic coordination model is constructed to solve for the optimal combination of coordinated environmental parameters. This module is the core technical component of this invention for achieving refined control of environmental parameters throughout the entire growth cycle of mushrooms.

[0028] Regarding the specific manifestations of environmental parameter differences between different stages of mushroom growth, taking the mycelial culture period and the fruiting body growth period as examples: the optimal temperature range for the mycelial culture period is 24℃ to 26℃, the optimal relative humidity is 65%RH to 75%RH, and complete darkness is required, i.e., light intensity approaches 0 lux. The concentration can be relatively high, ranging from 1500 ppm to 2500 ppm; the optimal temperature range for fruiting body growth is 15℃ to 22℃, the optimal relative humidity is 85%RH to 95%RH, and the optimal light intensity is 200 lux to 800 lux. The concentration ranged from 800 ppm to 1500 ppm. It can be seen that mushrooms at different growth stages are sensitive to temperature, humidity, light, and... Significant differences exist in the required concentrations. Specifically, temperature needs to be lowered as the primordia formation stage transitions to the fruiting body growth stage, humidity needs to be increased in sync with the development of the fruiting body, and light intensity needs to transition from complete darkness to weak diffused light. The concentration needs to be gradually reduced as the fruiting body grows.

[0029] To address the aforementioned problem of coordinating and controlling environmental parameters across various dimensions at different growth stages of mushrooms, this invention models the coordination of environmental parameters throughout the entire mushroom growth cycle as a game with a mushroom comprehensive suitability function as the payoff, and temperature, humidity, light, and [other parameters as needed]. This is a multi-objective game problem with concentration as the four dimensions of the game strategy space. A fitness function for mushrooms in the temperature dimension is defined. Taking temperature as an example, the temperature suitability function for mushrooms is defined as follows: , in: This is the actual controlled temperature value, in °C. This represents the optimal central temperature for the mushrooms at their current growth stage, expressed in °C. The value is determined based on the mushroom growth and operational status characteristic diagram. The dynamic determination of the growth stages of fungi was obtained from the analysis, for example, the fruiting body growth period was set at 18.5℃; This is a temperature tolerance width parameter for mushrooms, expressed in °C. It characterizes the degree tolerable for temperatures deviating from the optimum value during the current growth stage of the mushroom. The value is 2.8 °C during the fruiting body growth period. hour, The maximum value of 1.0 is taken; as the temperature deviates from the optimum value, the fitness decreases Gaussian. For the mycelial culture period, the optimal temperature center value is... The value is 25℃, which is the temperature tolerance width parameter. The value is 2.0℃; for the primordium formation period, The value is set to 15℃. The value is taken as 1.5℃; for the harvesting period, The value is set to 18℃. The value is 3.0℃. Humidity, light, and The fitness functions for the three dimensions of concentration have the exact same Gaussian form, differing only in the central value of the parameter and the tolerance width.

[0030] Temperature, humidity, light and After calculating the fitness functions for the four dimensions of concentration separately, the overall fitness function of mushrooms is aggregated through a weighted product: , in: This is the overall suitability function for mushrooms, with an output range of [range missing]. Dimensionless; Let be the vector of coordinated environment parameters to be solved. For temperature, For humidity, Light intensity, for Concentration; d is the parameter dimension index, which takes temperature as an example. ,humidity ,illumination and concentration ; parameter vector Components in dimension d; Let be the fitness function of mushrooms on dimension d; Let the importance weights (scalar, dimensionless) of the fungus to the d-th dimension environmental parameter at the current growth stage satisfy the following conditions: Its specific value is determined by the mushroom growth operation status characteristic map. It is dynamically generated through a fully connected mapping network.

[0031] The Nash equilibrium solution of a game is defined as a parameter vector that simultaneously satisfies the following conditions. For any other parameter vector Suitability functions for each dimension None of them can be strictly improved simultaneously, and the overall suitability is... To achieve the maximum, in one embodiment of the invention, an alternating direction multiplier method is used for iterative solution. Specifically, in the k-th iteration, the parameters of the other three dimensions are fixed sequentially, and the fitness function of a single dimension is optimized to update the parameter components of that dimension. This process is repeated iteratively until convergence. The convergence criterion is that the Euclidean norm of the difference between the parameter vectors of two adjacent iterations is less than a threshold. , The value is 0.01.

[0032] Furthermore, since the aforementioned game problem may have multiple Nash equilibria, this invention introduces a Pareto front search strategy to select a Pareto optimal solution set from all Nash equilibria, and then selects the solution from this set that satisfies the overall fitness index. The maximum equilibrium solution is taken as the final output combination of coordinated environment parameters. The comprehensive fitness index is defined as: , That is, the comprehensive suitability index directly uses the mushroom comprehensive suitability function as the evaluation index. Preferably, it is the combination of coordinated environmental parameters output after Pareto front search. Make The overall suitability of the operation should be maintained at a high level, with a value of not less than 0.85.

[0033] In one embodiment of the present invention, the cooperative execution control module 4 is used to combine the coordination environment parameters output by the parameter coordination module 3. This is converted into specific control signals for the physical actuators. The actuator network managed by this module includes temperature control actuators, humidity control actuators, light control actuators, and ventilation actuators.

[0034] Preferably, the collaborative execution control module 4 adopts a zoned fine-grained control strategy. In terms of physical spatial layout, the present invention divides the mushroom cultivation area into several cultivation rack zones according to the mushroom cultivation layout. Each zone is independently equipped with low-power LED diffused lights, and the light intensity can be adjusted within the range of 0 lux to 2000 lux. A shading curtain is also provided to achieve complete light avoidance to meet the needs of the mycelium cultivation period. Temperature and humidity parameters are finely controlled through airflow organization within the space and an independent humidification system: a temperature controller actuator is used to adjust the overall temperature of the mushroom cultivation space to coordinated parameters. Different target temperatures are switched at different growth stages (approximately 25℃ during mycelial culture, approximately 15℃ during primordia formation, and approximately 18.5℃ during fruiting body growth). The humidity control actuator uses an ultrasonic atomizing humidifier combined with a ventilation actuator to precisely maintain the humidity in the growing space within a high humidity range of 85%RH to 95%RH during the fruiting body growth period. It should be noted that during the tobacco curing season, this facility is manually controlled by tobacco farmers using traditional curing techniques, and the controller issues instructions via human intervention; the above-mentioned zoned refined control strategy is only used during the mushroom growing season.

[0035] Regarding the generation of control signals, the cooperative execution control module 4 will coordinate the combination of environmental parameters. With the current environment state vector The difference is used as the control deviation, and an incremental PID control algorithm is used to calculate the output increment of each actuator. Specifically, for the d-th dimension environmental parameter, its control signal increment is... The calculation formula is: , Where: k is the current control cycle number (dimensionless integer); Let d be the control deviation of the d-th dimension parameter in the k-th control cycle. To coordinate the target value, This is the current actual measurement value; This is the proportional gain coefficient. This is the integral gain coefficient. These are the differential gain coefficients (all three are scalars, and their units are related to the corresponding physical quantities). Their initial values ​​are determined using the Ziegler-Nichols method based on the actuator's response characteristics, for example, in a temperature control actuator. The initial value is 2.5. The initial value is 0.08. The initial value is 0.6. Control cycle. The setting is 30 seconds, meaning that the control output of each actuator is updated every 30 seconds based on the latest environmental data.

[0036] Furthermore, the collaborative execution control module 4 is also equipped with an actuator conflict arbitration mechanism. When the heating command of the temperature control actuator and the heat dissipation command of the ventilation actuator are triggered simultaneously in the same control cycle, the arbitration mechanism assigns priority according to the absolute value of the deviation. The command with the larger absolute value of the deviation is executed first, and the command with the smaller deviation is executed in the next control cycle, thereby avoiding energy waste and action cancellation between actuators.

[0037] In one embodiment of the present invention, the adaptive feedback learning module 5 is used to construct an online learning closed loop for the controller, enabling the entire system to continuously self-optimize during long-term operation. This module collects actual environmental parameter data and mushroom growth response data after the collaborative execution control module 4 has executed. The mushroom growth response data includes mycelial extension rate, primordium density, fruiting body diameter growth rate, and yield per unit time.

[0038] The core mechanism of the adaptive feedback learning module 5 is an online parameter update strategy based on temporal difference. The mushroom growth response reward signal is defined at time t. for: , in: Let be the reward signal at time t (scalar, dimensionless). The normalized growth increment of mushrooms at time t is calculated by weighted summation of mycelial extension rate and fruiting body diameter growth rate. Normalization yields (scalar, dimensionless). The normalized energy consumption cost at time t is calculated by dividing the power consumption of all actuators. Normalization yields (scalar, dimensionless). The reward weight for mushroom growth is set to 0.7 by default. The energy consumption penalty weight has a default value of 0.3, and the sum of the two is 1.0, which represents the controller's trade-off preference between promoting mushroom growth and reducing energy consumption.

[0039] Based on the aforementioned reward signals, the adaptive feedback learning module 5 updates the weight parameters of the attention fusion layer in the mushroom growth stage intelligent recognition module 2 using an online learning method based on exponential moving average. And the fitness function parameters in parameter coordination module 3. Specifically, when the accumulated reward signal is lower than a preset threshold... At this time, gradient fine-tuning of the model parameters is triggered, with the fine-tuning learning rate set to 0.01 times the initial training learning rate to ensure the stability of the update. Preferably, the adaptive feedback learning module 5 also uploads historical running data to the cloud server, where the cloud uses a larger dataset for offline retraining. After training, the updated model parameters are pushed down to the AI ​​edge computing unit, realizing a two-layer learning architecture of edge online fine-tuning plus cloud offline retraining.

[0040] In one embodiment of the present invention, the above five functional modules run on an AI edge computing unit, which is equipped with a neural network acceleration chip for locally performing inference computation of a lightweight convolutional neural network and iterative solution of the parameter coordination model. The AI ​​edge computing unit connects to sensor nodes and a cloud server via an IoT communication module. Preferably, the sensor nodes and the AI ​​edge computing unit use 4G or Wi-Fi communication protocols, with a communication distance covering 3km, suitable for low-power long-distance transmission in mushroom cultivation facilities, with an uplink data rate of 0.3kbps to 50kbps. The AI ​​edge computing unit and the cloud server use the MQTT protocol to interact via a 4G or Wi-Fi network for uploading historical operating data, receiving model update packages, and remote management commands.

[0041] like Figure 2 As shown, this embodiment of the invention also provides a control method based on the above-mentioned integrated controller for flue-cured tobacco and mushrooms. This control method corresponds one-to-one with the five functional modules in the above system embodiment, forming a closed-loop control process from steps S1 to S5. It should be noted that the control method is only activated during the mushroom growth season; during the flue-cured tobacco curing season, this controller and the method do not intervene in the flue-cured tobacco curing process, which is completed manually by tobacco farmers using traditional curing techniques.

[0042] Step S1, Multimodal Environment Sensing Step. In this step, the distributed sensor array of the multimodal environment sensing module 1 collects temperature data, humidity data, and light intensity data in the planting space in real time according to their respective acquisition cycles. Concentration data and soil moisture content data were collected. After acquisition, the data preprocessing unit performed time-series alignment on the multi-source heterogeneous sensor data, unifying the time axis through linear interpolation using the shortest acquisition cycle as the time base. Subsequently, outlier filtering based on the median of the sliding window was performed, with the window width set to 10 sampling points and the sensitivity coefficient set to 3.0. Readings exceeding the tolerance were replaced with the median of the window. After the above preprocessing, the data from each channel were assembled into a 5-dimensional environmental state vector. The data is then output to step S2. The execution frequency of this step is consistent with the shortest acquisition cycle to ensure that subsequent processing modules can obtain the latest environmental data.

[0043] Step S2, Intelligent Recognition of Mushroom Growth Stage. In this step, the image acquisition unit acquires morphological images of the mushroom area at a frequency of one frame every 10 minutes, and crops the images to a standard input size of 224×224×3 pixels. The lightweight convolutional neural network MobileNetV3-Small backbone extracts features from the input image, generating a 576-dimensional image feature vector. Meanwhile, the environment state vector Mapped to a 576-dimensional context embedding vector via a two-layer fully connected network. The attention fusion layer calculates the fusion weights according to the aforementioned cross-modal attention formula. And generate a fused mushroom growth operation status feature map. Classification head pairs The process involves classifying the mushrooms and outputting a label indicating their current growth stage. In this step, a change in growth stage is only confirmed when the classification results are consistent across three consecutive frames, to avoid frequent changes in control strategies due to occasional misidentification. The confirmed growth stage label and the mushroom growth status feature map are then displayed. Output them together to step S3.

[0044] Step S3, parameter coordination step. In this step, the parameter coordination module 3 first extracts the optimal temperature, optimal humidity, optimal light intensity, and optimal humidity for the mushroom at the current growth stage from a preset mushroom growth stage parameter lookup table based on the mushroom growth stage labels output in step S2. The center value of concentration and tolerance width parameters were then established. Subsequently, a system was constructed to measure the effects of temperature, humidity, light, and [other factors] on the mushroom. The four-dimensional suitability function of concentration is used to construct a multi-objective game coordination model for game payoffs. The model employs an alternating direction multiplier method for iterative solution: in each iteration, the parameters of the other three dimensions are fixed sequentially, and the suitability function of a single dimension is optimized to update the parameter components of that dimension. This process is repeated until the change in the parameter vector is less than the convergence threshold of 0.01. If multiple Nash equilibria are obtained through game solving, a comprehensive suitability index is selected using a Pareto front search. The maximum equilibrium solution. The final output is the combination of coordinated environmental parameters. Includes target temperature value, target humidity value, target light value, and target The concentration value is output to step S4.

[0045] Preferably, the mushroom growth stage parameter lookup table described in step S3 pre-sets the environmental parameter requirements for mushrooms at each growth stage. The optimal temperature center value for the mycelial culture period is 25℃, and the tolerance width is 2℃. This stage requires complete darkness, i.e., light intensity of 0 lux. The concentration can be relatively high, with a central value of 2000 ppm; during the primordium formation stage, a sudden drop in temperature is required, with the optimal central temperature dropping to 15℃ and a tolerance range of only 1.5℃. Simultaneously, weak diffused light is needed, with a central light intensity of 500 lux. During the fruiting body growth stage, the optimal central temperature rises back to 18.5℃, the optimal central humidity is 90%RH with a tolerance range of 3%RH, and the optimal central light intensity is 600 lux. The concentration center value was reduced to 1200 ppm; the parameter requirements during the harvest period were close to those during the fruiting body growth period, with a center value of 18℃ for temperature and 85%RH for humidity. All the above parameter data are derived from agronomic standards and planting experience in the field of edible mushroom cultivation, ensuring that the input parameters of the multi-objective game coordination model have sufficient agronomic basis.

[0046] Regarding the data flow in this step, the characteristic diagram of the mushroom growth operation status is shown. This model not only provides growth stage labels for parameter lookup but also dynamically generates an importance weight vector for four-dimensional environmental parameters of the mushroom through a fully connected mapping network. This weight vector reflects the differences in the mushroom's sensitivity to various environmental factors at different growth stages: for example, during the mycelial growth stage, the sensitivity to temperature is significantly higher than that to light, at which point temperature has a larger weight; while during the primordia formation stage, the mushroom is extremely sensitive to temperature changes, and temperature has a dominant weight; during the fruiting body growth stage, the sensitivity to humidity and light increases simultaneously, and the corresponding weights are adjusted accordingly. This dynamic weighting mechanism enables the multi-objective game coordination model to adaptively adjust the parameter coordination strategy according to the current physiological state of the mushroom, providing more refined optimization calculations on the parameter dimensions with the most intense conflicts.

[0047] Step S4, Cooperative Execution Control Step. In this step, the cooperative execution control module 4 coordinates the combination of environmental parameters. The current environment state vector provided by the multimodal environment perception module 1 Compare and calculate the control deviations for each dimension. This applies to temperature, humidity, and... For continuously adjustable parameters such as concentration, an incremental PID control algorithm is used to generate control signal increments for each actuator. For lighting parameters, different light intensity setpoints are issued to the lighting actuators based on the differentiated needs of different growth stages of the mushrooms. Preferably, the collaborative execution control module 4 also performs conflict detection on the commands of each actuator before execution. When heating commands and heat exhaust ventilation commands are detected to be triggered simultaneously within the same control cycle, their execution priorities are determined according to the absolute value of the deviation, avoiding energy waste. After the control signal is output, each actuator adjusts the environmental parameters of the mushroom cultivation space according to the commands.

[0048] Furthermore, the implementation of the zoning control strategy in this step in the temperature dimension is as follows: the collaborative execution control module 4 coordinates the combination of environmental parameters. Target temperature value Based on the distribution characteristics of different mushroom cultivation racks, a unified target temperature setting is set for each zone. By controlling the opening angles of the air inlets and outlets at both ends of the space, a uniform airflow is created within the space to maintain the target temperature. The opening angle of the air inlets is precisely controlled by a stepper motor, with an adjustment range of 0° to 90° and an angle resolution of 1°. Regarding humidity, the mushroom cultivation space is equipped with an ultrasonic humidifier array, consisting of four ultrasonic humidifiers, each with a humidification capacity of 3 kg / h. This array can precisely maintain the humidity in the cultivation space within a high humidity range of 85%RH to 95%RH during the fruiting body growth period. Concentration control is achieved through ventilation actuators: when the mushroom cultivation space... When the concentration exceeds the upper limit threshold of the current growth stage, the vents automatically open for ventilation; when When the concentration is below the lower limit threshold, it can be supplemented by a directional gas supply device.

[0049] Step S5, Adaptive Feedback Learning Step. In this step, the adaptive feedback learning module 5 collects execution result data after each control cycle, including the actual environmental parameters and mushroom growth response indicators after execution. The control deviation vector is calculated by comparing the actual environmental parameters with the coordinated target parameters. Simultaneously, the mushroom growth response reward signal is calculated. The effectiveness of the current control strategy on promoting mushroom growth and controlling energy consumption is comprehensively evaluated. When the average reward signal within the sliding window falls below a preset threshold... At this point, online model parameter fine-tuning is triggered: the attention fusion layer weights of the mushroom growth stage intelligent recognition module 2 and the fitness function parameters of the parameter coordination module 3 are updated using the fine-tuning learning rate. After fine-tuning, the updated parameters take effect immediately, and the control flow returns to step S1 to continue the next round of closed-loop control.

[0050] Preferably, the sliding window length in this step is set to 48 control cycles, corresponding to a 24-hour observation duration. A preset reward threshold is also included. The value is set to 0.6. When the average reward signal over 48 cycles is lower than 0.6, it indicates that the overall effect of the current control strategy is poor, and parameter fine-tuning is required. The fine-tuning process uses a stochastic gradient descent optimizer, and the fine-tuning learning rate is set to... Each fine-tuning iteration is limited to no more than 10 steps to avoid system oscillations caused by excessive updates. The fine-tuning targets two levels: the first level is the weight matrix of the attention fusion layer in the mushroom growth stage intelligent recognition module 2. and bias By fine-tuning, attention fusion can better adapt to the image and environmental data distribution characteristics under the current planting environment; the second level is the temperature tolerance width parameter for each growth stage in parameter coordination module 3. The importance weight w is used to fine-tune the coordination model so that it can more accurately reflect the actual sensitivity of the mushroom variety to various environmental factors under the current environmental conditions.

[0051] Furthermore, the adaptive feedback learning module 5 also establishes an abnormal state detection mechanism. When the control deviation of any environmental parameter continuously exceeds the safety threshold within 12 consecutive control cycles, the module automatically triggers an alarm and switches to a safety mode: suspending the output of the game coordination algorithm, switching to a preset conservative parameter scheme to maintain basic environmental stability, and simultaneously sending an alarm notification to the management personnel. The safety thresholds are set according to the physiological limits of mushrooms: the temperature deviation safety threshold is ±5℃, and the humidity deviation safety threshold is ±15%RH. The safe threshold for concentration deviation is ±500 ppm. Once the abnormal condition is resolved, the system automatically resumes parameter coordination control mode.

[0052] Preferably, the adaptive feedback learning module 5 uploads the accumulated daily running data to the cloud server in batches every 24 hours. After accumulating no less than 7 days of data, the cloud initiates an offline retraining task. After training is completed, the updated model weight file is pushed to the AI ​​edge computing unit for local model replacement. This two-layer learning architecture, combining edge online fine-tuning with cloud offline retraining, ensures the controller's real-time adaptability in daily operation while also achieving better model generalization ability through training on a larger dataset in the cloud.

[0053] To verify the technical effectiveness of this invention, a comparative experiment was conducted in a modern greenhouse with an area of ​​600m² for nine months (covering a complete tobacco curing season and a mushroom growing season). The greenhouse was divided into two 300m² areas: the experimental area was equipped with the integrated tobacco and mushroom controller described in this invention, while the control area was equipped with a traditional PID temperature controller and a timed irrigation controller. The experimental procedure strictly followed the staggered planting schedule of tobacco and mushrooms: during the tobacco curing season from August to October, both the experimental and control areas were manually controlled by tobacco farmers according to the traditional curing process of yellowing, color fixing, and drying stages. During this period, the controller of this invention did not intervene in the curing process, but only collected basic environmental data; during the mushroom growing season from October to May of the following year, the experimental area was fully controlled by the controller of this invention for the intelligent regulation of mushrooms (using the gray-white variety of oyster mushrooms), while the control area was controlled by a traditional controller according to fixed thresholds. During the experiment, multiple indicators such as the qualified rate of tobacco curing, mushroom yield and quality, environmental control accuracy, and operating costs were recorded.

[0054] Regarding the precision of environmental control during the mushroom growing season, the root mean square (RMS) deviation of temperature control in the experimental area was 0.4℃, and the RMS deviation of humidity control was 1.8%RH. The root mean square value of the concentration control deviation was 38 ppm; the corresponding indicators for the control area were 1.2℃, 4.5%RH, and 95 ppm. The control accuracy of the experimental area was improved by 67%, 60%, and 60% in the three dimensions, respectively. In the comparison of the flue-cured tobacco curing season, since both areas were manually controlled by tobacco farmers using the same traditional curing process, and the curing process was controlled by the controller of this invention through manual adjustment, the qualified rate of flue-cured tobacco curing in both areas remained at the normal level of tobacco farmers with no significant difference. This proves that the deployment of the controller of this invention has no adverse effect on the normal progress of flue-cured tobacco curing, and the two production systems are completely staggered in time and do not interfere with each other.

[0055] In terms of mushroom yield and quality, the biological efficiency of oyster mushrooms in the experimental area was 98%, an increase of 23 percentage points compared to 75% in the control area; the single-batch harvesting cycle was 28 days, a reduction of 22.2% compared to 36 days in the control area. During the 7-month mushroom growing season, the experimental area completed 4 batches of oyster mushroom production, while the control area only completed 3 batches. The total cumulative yield of mushrooms in the experimental area was approximately 65% ​​higher than that in the control area. Regarding flue-cured tobacco, since both areas were manually controlled by tobacco farmers using traditional curing techniques, the yield per mu, proportion of high-quality tobacco, total sugar content, nicotine content, and potassium-chloride ratio of flue-cured tobacco in both the experimental and control areas were at the normal levels for local tobacco farmers, with no statistically significant differences between the two areas. This further confirms the good compatibility of the controller of this invention with the traditional flue-cured tobacco curing process.

[0056] In terms of energy consumption and labor costs, the experimental area, by adopting a multi-objective game-theoretic optimization of environmental parameter combinations, avoided conflicting control actions between different devices, resulting in a 35% reduction in overall energy consumption during the mushroom growing season compared to the control area. In terms of electricity costs, the experimental area consumed an average of 1250 kWh per month during the mushroom growing season, compared to 1920 kWh in the control area, saving approximately 470 yuan per month. Due to the fully automated closed-loop control during the mushroom growing season, daily management of the experimental area only requires one technician for remote monitoring via mobile phone, compared to the traditional model requiring two personnel in the control area, reducing labor costs by approximately 50%. During the tobacco curing season, because the controller of this invention allows for manual control of the curing process, tobacco curing in both areas was carried out by tobacco farmers using traditional methods, resulting in no difference in labor costs.

[0057] Regarding the accuracy of intelligent recognition, the mushroom growth stage intelligent recognition module 2 achieved a recognition accuracy of 96.5% during the actual operation of the mushroom growth season over a 7-month period. Misidentifications were mainly concentrated at the transition moments between adjacent growth stages, and the continuous 3-frame confirmation mechanism effectively avoided erroneous switching of control strategies caused by misidentification. The adaptive feedback learning module 5 completed three rounds of online parameter fine-tuning within the first 30 days of operation, gradually increasing the comprehensive suitability index of the multi-objective game coordination model from an initial 0.82 to 0.91, verifying the effectiveness of the closed-loop adaptive learning mechanism.

[0058] In summary, the integrated controller and control method for flue-cured tobacco and mushrooms provided by this invention achieves significantly better results than traditional solutions in terms of environmental control precision, mushroom biological efficiency, yield and quality, and operating cost control during the mushroom growing season through the deep coupling and closed-loop collaboration of five major functional modules. At the same time, the deployment of the controller is fully adaptable to the staggered planting scenarios of flue-cured tobacco and mushrooms. During the flue-cured tobacco curing season, tobacco farmers manually cure the tobacco using traditional techniques, while the controller issues instructions manually. The two production systems are completely staggered in time, independent of each other, and do not interfere with each other, realizing the efficient reuse of the same facility in different crop seasons.

[0059] It is particularly noteworthy that the core innovation of this invention lies in the intelligent management of the mushroom cultivation stage. Through a closed-loop architecture formed by intelligent identification of mushroom growth stages, multi-objective game parameter coordination, collaborative execution control, and adaptive feedback learning, precise control can be achieved across all environmental dimensions at different growth stages of mushroom cultivation. This overcomes the fundamental problem that traditional fixed-threshold control strategies are ill-suited to the evolving needs of mushroom growth stages. Furthermore, the deployment of this invention remains completely independent of the tobacco curing process, ensuring that tobacco farmers can complete curing operations normally using traditional methods during the tobacco season. Without altering the tobacco production method, the intelligent operation during the mushroom growing season significantly improves the utilization rate and return on investment of tobacco curing facilities over an annualized time dimension. These technical effects are the core advantages that distinguish this invention from existing single-crop control technologies.

[0060] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. An integrated controller for flue-cured tobacco and mushroom cultivation, characterized in that, include: The multimodal environmental sensing module is used to collect real-time data on temperature, humidity, and light intensity within the planting space through a distributed sensor array. Concentration data and soil moisture content data were collected, and the collected multidimensional environmental parameter data were time-series aligned and outlier filtered to generate an environmental state vector. The intelligent mushroom growth stage recognition module is used to receive the environmental state vector and mushroom morphology image data acquired by the image acquisition unit, extract features from the mushroom morphology image data based on a lightweight convolutional neural network, identify the current growth stage of the mushroom, and fuse the recognition result with the environmental state vector to generate a mushroom growth operation status feature map. The parameter coordination module is used to receive the mushroom growth operation status feature map and, based on the mushroom's response to temperature, humidity, light, and other factors at the current growth stage, coordinate the parameters accordingly. Differentiated concentration requirements parameters are used to construct a multi-objective game coordination model. By solving the Nash equilibrium point, a combination of coordinated environmental parameters that maximizes the overall suitability for mushroom growth is obtained. The collaborative execution control module is used to receive the coordinated environmental parameter combination and output control signals to the temperature control actuator, humidity control actuator, light control actuator and ventilation actuator; The controller operates in a staggered planting scenario for flue-cured tobacco and mushrooms. During the flue-cured tobacco curing season, tobacco farmers manually assist in completing the curing process using traditional methods, and the controller manually controls the curing process. During the mushroom growing season after the flue-cured tobacco season, the controller is responsible for the intelligent environmental regulation of the mushrooms throughout the entire process.

2. The integrated controller for flue-cured tobacco and mushroom cultivation according to claim 1, characterized in that, The multimodal environmental sensing module includes a temperature sensor, a humidity sensor, and a light sensor. Concentration sensors and soil moisture sensors are included, with the temperature sensor having a measurement accuracy of ±0.3℃ and the humidity sensor having a measurement accuracy of ±1.5%RH. The concentration sensor has a measurement accuracy of ±30ppm, the spacing between adjacent sensor nodes in the distributed sensor array is 1.5m to 3m, and the data acquisition cycle is 5s to 30s.

3. The integrated controller for flue-cured tobacco and mushroom cultivation according to claim 1, characterized in that, The intelligent identification module for mushroom growth stages includes the mycelial culture period, primordium formation period, fruiting body growth period, and harvesting period. The lightweight convolutional neural network adopts a depthwise separable convolutional structure, with no more than 5M model parameters and a single-frame inference time of no more than 50ms.

4. The integrated controller for flue-cured tobacco and mushroom cultivation according to claim 1, characterized in that, In the parameter coordination module, the multi-objective game coordination model uses the mushroom comprehensive suitability function as the game payoff function, and uses temperature, humidity, light intensity, and... Concentration serves as the game strategy space, and the combination of coordinated environmental parameters ensures that the overall suitability index of the operation during the corresponding operating period is not less than 0.

85.

5. The integrated controller for flue-cured tobacco and mushroom cultivation according to claim 3, characterized in that, The lightweight convolutional neural network includes a feature extraction backbone network, a classification head, and an attention fusion layer. The feature extraction backbone network adopts the MobileNet architecture, the classification head corresponds to the classification of mushroom growth stages, and the attention fusion layer is used to perform cross-modal fusion of environmental state vectors and image features.

6. The integrated controller for flue-cured tobacco and mushroom cultivation according to claim 4, characterized in that, The multi-objective game coordination model uses the alternating direction multiplier method for iterative solution when solving for Nash equilibrium points, and introduces the Pareto front search strategy to select the equilibrium solution that maximizes the comprehensive fitness index from multiple Nash equilibrium points as the final coordination environment parameter combination.

7. The integrated controller for flue-cured tobacco and mushroom cultivation according to claim 1, characterized in that, The controller also includes an AI edge computing unit, which is equipped with a neural network acceleration chip for locally running the inference computation of the lightweight convolutional neural network and the multi-objective game coordination model. The AI ​​edge computing unit is connected to a cloud server through an IoT communication module for uploading historical running data and downloading updated model parameters. The IoT communication module supports the MQTT protocol and 4G and Wi-Fi communication protocols. The 4G and Wi-Fi communication protocols are used for long-distance data transmission between sensor nodes and the AI ​​edge computing unit, and the MQTT protocol is used for data interaction between the AI ​​edge computing unit and the cloud server.

8. The integrated controller for flue-cured tobacco and mushroom cultivation according to claim 7, characterized in that, It also includes an adaptive feedback learning module, which is used to collect actual environmental parameter data and mushroom growth response data after the collaborative execution control module is executed, calculate the deviation value between the actual environmental parameters and the target coordinated environmental parameters, and update the recognition model parameters of the mushroom growth stage intelligent recognition module and the coordination model parameters of the parameter coordination module based on the deviation value through an online learning mechanism.

9. A control method based on the integrated controller for flue-cured tobacco and mushroom cultivation according to any one of claims 1 to 8, characterized in that, The control method operates during the mushroom growing season under a staggered planting scenario of flue-cured tobacco and mushrooms. During the flue-cured tobacco curing season, tobacco farmers manually complete the curing process using traditional methods, and the control method does not intervene in the curing process. During the mushroom growing season, the control method includes the following steps: Step S1, Multimodal Environmental Sensing Step: Real-time collection of temperature, humidity, and light intensity data within the mushroom cultivation space using a distributed sensor array. Concentration data and soil moisture content data are used to perform time-series alignment and outlier filtering on the collected multidimensional environmental parameter data to generate an environmental state vector. Step S2, intelligent identification step of mushroom growth stage: Based on the environmental state vector and mushroom morphology image data, the current growth stage of the mushroom is identified by a lightweight convolutional neural network, and the identification result is fused with the environmental state vector to generate a mushroom growth operation status feature map; Step S3, parameter coordination step: Based on the mushroom growth operation state characteristic map, construct a multi-objective game coordination model with the mushroom comprehensive suitability function as the game payoff, and obtain the coordination environment parameter combination by solving the Nash equilibrium point; Step S4, Coordinated execution control step: Based on the coordinated environmental parameter combination, output control signals to each actuator to adjust the environmental parameters of the mushroom cultivation space; Step S5, Adaptive Feedback Learning Step: Collect actual environmental parameter data and mushroom growth response data after execution, calculate the deviation value between the actual environmental parameter and the target parameter, update the identification model parameter in step S2 and the coordination model parameter in step S3 through the online learning mechanism based on the deviation value, and return to execute step S1.

Images