An edible mushroom shelter temperature intelligent control method and system based on artificial intelligence

Abstract

This invention discloses an intelligent temperature control method and system for edible fungi containerized production based on artificial intelligence, belonging to the field of smart agriculture technology. The method includes: collecting environmental and growth status data within the container; identifying the growth stages of edible fungi based on a multimodal deep learning classifier; generating a temperature reference trajectory by invoking corresponding temperature demand strategies; constructing a digital twin model of edible fungi growth dynamics to predict future temperature demand changes; establishing a state-space model coupled with temperature, humidity, and CO2 using a model predictive control (MPC) algorithm to calculate the optimal temperature control command; and executing the control command to coordinate equipment operation. This invention achieves metabolic heat prediction through a edible fungi mycelial temperature response model and optimizes energy consumption strategies through reinforcement learning, solving the problems of low control accuracy, high energy consumption, and poor variety adaptability in existing technologies, making it suitable for industrialized production of various edible fungi.

Description

Technical Field

[0001] This invention belongs to the field of smart agriculture and edible fungi cultivation technology, specifically relating to an artificial intelligence-based intelligent temperature control method and system for edible fungi container, which is particularly suitable for precise temperature management in industrialized edible fungi production. Background Technology

[0002] As an important part of my country's agriculture, the edible fungi industry has an annual output value exceeding 300 billion yuan. With the rapid development of factory cultivation technology, edible fungi modular containers, with their significant advantages such as modular design, high mobility, and high environmental controllability, have become a key carrier for modern edible fungi production.

[0003] In the environmental control system of edible mushroom cultivation facilities, temperature regulation is a core environmental factor affecting the growth and development of edible fungi, directly impacting mycelial growth rate, primordium differentiation efficiency, fruiting body morphology, and final yield and quality. Different edible fungi varieties (such as shiitake, oyster, enoki, and king oyster mushrooms) have significantly different temperature requirements, and the same variety requires differentiated temperature management strategies at different growth stages (including mycelial growth, primordium differentiation, fruiting body development, and harvesting). For example, shiitake, as a fluctuating-temperature fruiting edible fungus, requires a diurnal temperature difference of 8-10°C for primordium differentiation; while enoki mushrooms, being a constant-temperature fruiting edible fungus, may develop deformed mushrooms if temperature fluctuations exceed ±3°C.

[0004] Currently, temperature control technology in edible mushroom cultivation cabins mainly employs traditional PID control, timed segmented control, and manual experience-based adjustment. However, these technologies have significant limitations. For example, the temperature control system for edible mushroom cultivation disclosed in CN104142699A still uses traditional PID control, with temperature control accuracy typically maintained within ±2~3℃. This fails to meet the stringent ±0.5℃ temperature stability requirements of high-end edible mushroom varieties, highlighting the problem of insufficient control precision.

[0005] Regarding personalized adaptation to varieties and growth stages, existing technologies generally adopt universal temperature control schemes. Although CN115731514A discloses a greenhouse digital twin system, this system does not design a dedicated model for the growth characteristics of edible fungi, and cannot implement precise control based on the biological characteristics and specific growth status of different edible fungi varieties. The switching of growth stages mainly relies on manual judgment, which is inaccurate and inefficient.

[0006] In terms of predictive regulation capabilities, existing control methods are reactive and lack foresight. For example, EP4483706A1 discloses a plant growth device based on digital twins, but it does not involve metabolic heat modeling of edible fungi, making it impossible to effectively predict the peak of metabolic heat release from edible fungi and changes in the external environment, resulting in large control overshoot and long adjustment time. Especially during the rapid mycelial growth period, edible fungi release a large amount of metabolic heat. Taking the mycelial growth period of enoki mushrooms as an example, its metabolic heat release power can reach 100-200W / cubic meter. If this cannot be predicted in advance, it can easily lead to temperature runaway.

[0007] In multivariate coupled control, temperature is strongly coupled with environmental parameters such as humidity and CO2 concentration, making it difficult for existing single-variable control methods to achieve coordinated control of multiple parameters. Although some studies have attempted to introduce advanced control algorithms, the reinforcement learning-based HVAC control method disclosed in CN120029058A, without being combined with the edible fungus growth kinetics model, still cannot effectively handle multivariate coupling relationships and frequently causes secondary problems such as "temperature control leading to abnormal humidity".

[0008] In terms of energy efficiency, temperature control accounts for 40% to 60% of the total energy consumption of edible mushroom cabins. Due to insufficient control efficiency and failure to fully consider peak and off-peak electricity price differences, the overall economic benefits are affected. Existing technical solutions fail to organically combine advanced algorithms such as reinforcement learning with energy consumption optimization strategies, thus failing to achieve intelligent energy consumption management.

[0009] In summary, existing technologies have failed to provide a complete solution that organically combines advanced technologies such as edible fungi growth kinetic models, deep learning-based growth stage identification, model predictive control, and reinforcement learning-based energy consumption optimization. Therefore, there is an urgent need in this field for an intelligent temperature control system for edible fungi container systems that can achieve high-precision temperature control, adapt to the needs of different varieties and growth stages, possess predictive regulation capabilities, effectively handle multivariate coupling relationships, and optimize energy consumption. Summary of the Invention

[0010] The technical problem to be solved by this invention is: how to achieve precise, intelligent, and adaptive temperature control in edible mushroom container, while meeting the differentiated temperature requirements of different edible mushroom varieties and different growth stages, improve control accuracy, reduce energy consumption, reduce human intervention, and achieve high-quality, high-yield, and low-consumption edible mushroom production.

[0011] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an intelligent temperature control method for edible mushroom mobile cabins based on artificial intelligence, comprising the following steps: Collect environmental and growth status data within the edible mushroom container; The growth status data is input into a multimodal deep learning classifier to identify the current growth stage of the edible fungus; Based on the growth stage, the corresponding temperature requirement strategy is invoked to generate a temperature reference trajectory, which includes the target temperature setpoint. Construct a digital twin model of edible fungi growth dynamics to predict temperature demand changes within a set time period based on the environmental data and the growth stage. Based on the Model Predictive Control (MPC) algorithm, a state-space model coupling three variables of temperature, humidity, and CO2 is established. According to the temperature reference trajectory and the predicted results of the temperature demand change, temperature control accuracy constraints are set, and the optimal temperature control command is solved in a rolling manner within the prediction time domain. The temperature control command is executed to coordinate the operation of the temperature control equipment inside the cabin.

[0012] According to a first aspect of the present invention, the multimodal deep learning classifier uses a convolutional neural network to extract visual features of the mushroom bag image, uses a long short-term memory network to extract temporal features of the number of days of inoculation and the accumulated temperature, and outputs the growth stage identification result after fusing the visual features and the temporal features through an attention mechanism.

[0013] According to a first aspect of the present invention, the digital twin model of edible fungi growth kinetics uses an edible fungi mycelial temperature response model to calculate the mycelial growth rate and calculates the metabolic heat release based on the mycelial growth rate; the expression of the edible fungi mycelial temperature response model is:

[0014] in, The mycelial growth rate is the specific growth rate. The maximum specific growth rate at the optimum temperature. For mycelial growth activation energy, The gas constant is To cultivate the absolute temperature of the environment, The optimal temperature for mycelial growth. The relative humidity response factor is expressed as: ,in Relative humidity, The optimal relative humidity is 85%–95%. Humidity sensitivity coefficient; The CO2 concentration response factor is expressed as follows: ,in CO2 concentration This represents the CO2 inhibition coefficient. The metabolic heat release is calculated using the following formula:

[0015] in, The power of metabolic heat release. The heat production coefficient of mycelial growth metabolism. This refers to bacterial biomass.

[0016] According to a first aspect of the present invention, the state-space model uses temperature, humidity, and CO2 concentration as state variables, air conditioning cooling power, ventilation volume, and heating power as control input variables, and temperature measurement, humidity measurement, and CO2 concentration measurement as output variables; the model predictive control (MPC) algorithm takes optimal temperature control accuracy as the objective function, and adjusts the cooling power in advance based on the metabolic heat release through feedforward compensation.

[0017] According to a first aspect of the present invention, the method further includes: perceiving the current electricity price and system operating status through a reinforcement learning agent, and dynamically adjusting the temperature control command to reduce energy consumption costs while satisfying the temperature control accuracy constraint.

[0018] According to a first aspect of the invention, the method further includes: performing anomaly detection on the environmental data; when anomaly of sensor data is detected, replacing the abnormal data with the average data of redundant sensors or the moving average of historical data, and switching to a degraded control mode to maintain the basic functions of the temperature control system.

[0019] According to a first aspect of the present invention, the temperature demand strategy includes temperature setpoints, allowable fluctuation ranges, and temperature change stimulation parameters for different edible fungi varieties at various growth stages; for temperature-change fruiting edible fungi, the temperature reference trajectory includes diurnal temperature difference setpoints and temperature change stimulation timing.

[0020] According to a first aspect of the present invention, the method further includes: storing historical environmental data, growth status data and actual control effects in a database; periodically calibrating the parameters of the digital twin model of edible fungi growth kinetics online based on measured data; and introducing an error compensation term to correct the prediction result when the deviation between the model prediction value and the measured value exceeds a set threshold.

[0021] According to a first aspect of the present invention, the method further includes: acquiring actual temperature data after the execution of a control command, calculating the deviation between the actual temperature and the target temperature setpoint, adjusting the weight matrix parameters of the objective function when the deviation exceeds a set threshold, and correcting the temperature deviation in real time using a proportional-integral-derivative (PID) controller.

[0022] A second aspect of the present invention provides an intelligent temperature control system for edible mushroom mobile cabins based on artificial intelligence, comprising: The data acquisition module is used to collect environmental and growth status data inside the edible mushroom container. The growth stage identification module is used to input the growth status data into a multimodal deep learning classifier to identify the current growth stage of the edible fungus. The temperature strategy management module is used to invoke the corresponding temperature requirement strategy according to the growth stage and generate a temperature reference trajectory, which includes the target temperature setpoint. The digital twin prediction module is used to construct a digital twin model of edible fungi growth dynamics and predict temperature demand changes within a set time period based on the environmental data and the growth stage. The model predictive control module is used to establish a state-space model of temperature-humidity-CO2 three-variable coupling based on the MPC algorithm, and to set temperature control accuracy constraints and calculate the optimal temperature control command based on the temperature reference trajectory and the predicted results of temperature demand changes. The control execution module is used to execute the temperature control commands and coordinate the operation of the temperature control equipment inside the cabin; The energy consumption optimization module is used to perceive the current electricity price and system operating status through a reinforcement learning agent, and dynamically adjust the temperature control command to reduce energy consumption costs while meeting the temperature control accuracy constraints. An anomaly detection and fault tolerance module is used to detect anomalies in the environmental data and to activate a fault tolerance control strategy when anomalies in sensor data are detected.

[0023] Compared with existing technologies, the present invention provides an intelligent temperature control method and system for edible fungi container based on artificial intelligence, which has the following beneficial effects: (1) Improved control precision: Through model predictive control algorithm and metabolic heat prediction, precise temperature control is achieved to meet the temperature stability requirements of high-end edible fungi varieties.

[0024] (2) Improved yield and quality: By accurately matching the temperature requirements of varieties and growth stages, the yield of edible fungi and the rate of first-grade products are increased, and the proportion of deformed mushrooms is reduced.

[0025] (3) Energy consumption reduction: Based on predictive control and reinforcement learning energy consumption optimization strategies, reduce overall energy consumption and electricity costs.

[0026] (4) Wide adaptability of varieties: It supports temperature management of a variety of edible fungi, covering two major categories of edible fungi: constant temperature fruiting type and variable temperature fruiting type.

[0027] (5) High reliability: It has anomaly detection and fault tolerance mechanism to adapt to abnormal working conditions such as sensor failure and network interruption, and ensure production safety. Attached Figure Description

[0028] Figure 1A flowchart illustrating an artificial intelligence-based intelligent temperature control method for edible mushroom mobile cabins provided in this application; Figure 2 This is a schematic diagram of the structure of an artificial intelligence-based intelligent temperature control system for edible fungi container provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the working principle of the MPC controller provided in an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0030] This invention provides an artificial intelligence-based intelligent temperature control method and system for edible mushroom cultivation cabins. Its core lies in constructing a closed-loop control system integrating multimodal perception, intelligent cognition, model prediction, multivariate decision-making, optimized execution, and autonomous fault tolerance. This system completely changes the traditional extensive control mode of edible mushroom cultivation environments, which relies on fixed procedures or manual experience, achieving high-precision, adaptive, collaborative, and intelligent management of temperature, a key factor.

[0031] First, please refer to Figure 2 This demonstrates the physical and logical architecture of the system described in this invention. The system consists of a sensing and execution layer deployed within the edible mushroom container, an edge control layer responsible for real-time computation, and a cloud / server platform layer for in-depth analysis and strategy management.

[0032] The sensing execution layer is the interface between the system and the physical world. It includes: Environmental sensor network: Inside the cabin, multiple temperature sensors (such as high-precision PT100 platinum resistance thermometers with a measurement accuracy of ±0.1℃), humidity sensors (such as capacitive polymer film sensors with an accuracy of ±2%RH), and CO2 concentration sensors (such as non-dispersive infrared (NDIR) sensors with an accuracy of ±30ppm) are evenly deployed according to the airflow organization and the layout of the bacterial beds. These sensors are networked in a wired or wireless manner and synchronously collect temperature (T), relative humidity (H), and CO2 concentration (C) values ​​at different locations inside the cabin at a fixed sampling period (e.g., 10 seconds to 1 minute).

[0033] The growth status acquisition unit mainly includes a high-definition industrial camera, fixed above the mushroom bed, which takes top-down or side-view RGB images of the mushroom bags at regular intervals (e.g., hourly). Simultaneously, this system interfaces with the production management information system to automatically obtain the inoculation date of the current batch of edible fungi, thereby calculating the inoculation days. Furthermore, the system calculates the cumulative accumulated temperature (ATT) in real time based on historical temperature data; the calculation formula is as follows: ,in The average daily temperature This is the biological zero temperature for mycelial growth of this species (e.g., 5°C for most mesophilic species).

[0034] Actuators include variable frequency drive (VFD) refrigeration / heating air conditioning units, adjustable speed ventilation fans, and auxiliary electric heaters. These devices receive command signals from the controller and adjust their operating status accordingly.

[0035] The edge control layer typically consists of industrial-grade computing gateways or industrial control systems, responsible for running real-time-critical control algorithms and data preprocessing. The cloud / server platform layer provides stronger computing power for running complex tasks such as deep learning models, digital twin simulations, and reinforcement learning training. Data is reliably transmitted between layers via industrial IoT protocols (such as MQTT and OPCUA).

[0036] Figure 1 The complete control method flowchart based on this architecture is shown, and the specific implementation details of each step are as follows: Step S101: Collect environmental data and growth status data inside the edible mushroom container. This step is performed by the data acquisition module. The module continuously reads raw data from the sensor network and information system. The raw data needs to be preprocessed to ensure quality: Data cleaning: For sensor data, use moving average filtering or median filtering to eliminate transient pulse interference.

[0037] Data Alignment: Due to slight differences in sampling times between different sensors, the system uses the control period (e.g., 1 minute) as the time base and performs timestamp alignment and interpolation on all sampled values ​​within that period to form an environmental state vector at the same moment. .

[0038] Image preprocessing: Automatic correction is performed on the acquired mushroom bag images, including brightness equalization and color correction. Image segmentation algorithms (such as threshold-based segmentation or lightweight U-Net) are used to extract the region of interest where the mushroom bag is located and remove background interference.

[0039] Data encapsulation: The processed environmental data, growth images, inoculation days, and accumulated temperature are encapsulated into structured data packets and sent to subsequent modules.

[0040] Step S102: Input the growth status data into a multimodal deep learning classifier to identify the current growth stage of the edible fungus.

[0041] This step is performed by the growth stage identification module. Its goal is to accurately determine the physiological stage of the edible fungus, which is a prerequisite for personalized regulation. The core of this module is a multimodal deep learning classifier, which is specifically implemented as an end-to-end neural network model. It receives preprocessed images and time-series data and outputs the classification results of the growth stage (e.g., mycelium growth stage, primordia differentiation stage, fruiting body growth stage, and harvesting stage).

[0042] Step S103: Invoke the corresponding temperature requirement strategy according to the growth stage to generate a temperature reference trajectory.

[0043] This step is performed by the temperature strategy management module. This module has a built-in configurable "variety-growth stage" strategy knowledge base. Upon receiving the identification result from step S102, the module combines the user-preset or automatically identified edible mushroom varieties (such as enoki mushrooms and shiitake mushrooms) with the knowledge base. The knowledge base predefines the following parameters for each "variety-stage" data pair: Target temperature setpoint For example, the mycelium growth period of enoki mushrooms is 20.0℃, and the primordium differentiation period is 13.0℃.

[0044] Allowable temperature fluctuation range For example, requiring control within ±0.3℃.

[0045] Temperature-dependent stimulation parameters (for temperature-dependent fruiting varieties such as shiitake mushrooms): including daytime target temperature. Nighttime target temperature Temperature change start time (e.g., a few days after inoculation or triggered based on growth stage), temperature change cycle (e.g., 24 hours), etc.

[0046] Based on these parameters, the module generates a temperature reference trajectory for a future period of time (such as the next 24 hours). ,in It is a discrete time series. For the isothermal phase, It is a constant value; for the temperature variation stage, A curve that changes according to a set period and amplitude.

[0047] Step S104: Construct a digital twin model of edible fungi growth dynamics, and predict the temperature demand changes within a set time period in the future based on the environmental data and the growth stage.

[0048] This step is performed by the digital twin prediction module. Specifically, the digital twin prediction module constructs a growth kinetic model that reflects the coupling relationship between the mycelial growth state of edible fungi and environmental factors. This model is not limited to a single thermodynamic calculation, but simulates the physiological activity trend of edible fungi within a set time period (i.e., the prediction time domain) based on the currently identified growth stage and real-time collected environmental data (including temperature, humidity, CO2 concentration, etc.).

[0049] The growth kinetics model calculates mycelial growth rate or metabolic activity index to extrapolate future heat production trends or temperature sensitivity changes, thereby outputting the temperature demand changes.

[0050] In some embodiments, the growth kinetics model uses the following logic for prediction: The model calculates the relative growth rate based on the deviation of the current temperature from the optimal temperature for the current growth stage, combined with water activity and CO2 concentration. When it is predicted that the mycelium is about to enter a rapid growth phase (such as the peak period of mycelial growth), the model outputs a temperature demand change signal, indicating that the temperature setpoint needs to be adjusted to a lower temperature range or that the cooling reserve needs to be increased.

[0051] Alternatively, this module runs a digital twin model of edible fungi growth dynamics. Its core function is to simulate the life activities of the mycelium and predict its metabolic heat release, which is the main source of variation in internal heat load. The model collects real-time environmental data through the aforementioned steps. Using the current growth stage as input, the system calculates the metabolic heat release power for a predicted time period (e.g., the next 4 hours) through internal kinetic equations. The predicted sequence. This predicted value will serve as a key feedforward signal, informing the control system in advance of the changing trend of thermal disturbances.

[0052] Step S105: Based on the Model Predictive Control (MPC) algorithm, establish a state-space model with three coupled variables: temperature, humidity, and CO2. According to the temperature reference trajectory and the predicted results of the temperature demand change, set temperature control accuracy constraints and solve for the optimal temperature control command in the prediction time domain.

[0053] This step is executed by the model predictive control module. This is the core component for achieving high-precision, coordinated control. This module incorporates a state-space predictive model describing the dynamic coupling relationship between temperature, humidity, and CO2 concentration within the control chamber. In each control cycle (e.g., time k), the module performs the following operations: Model state update: The latest sensor measurements are used as the current state.

[0054] Optimize problem construction: to track temperature reference trajectory With humidity and CO2 stability as the primary objective, and considering equipment energy consumption and stability, a constrained optimization problem in the finite time domain is constructed. The decision variable of the optimization problem is the sequence of control commands to the actuators (air conditioner, fan, heater) over multiple future control cycles.

[0055] Feedforward compensation: The metabolic heat sequence predicted in step S104 is used to... As a prediction model with known perturbation input.

[0056] Rolling solution: Solving this optimization problem yields a set of optimal future control command sequences. .

[0057] Step S106: Execute the temperature control command to coordinate the operation of the temperature control equipment inside the cabin.

[0058] This step is performed by the control execution module. This module receives the optimal control instruction sequence calculated in step S105, but only executes the first instruction in the sequence. For example, specific cooling power percentages and fan speed setpoints are sent to the corresponding frequency converters or drives. After the actuators operate, the environmental state of the cabin changes. In the next control cycle (at time k+1), the system starts again from step S101, repeating the "identification-prediction-optimization-execution" process based on new environmental measurement data. This periodic, iterative optimization based on the latest feedback, known as "rolling optimization," forms a closed-loop control, enabling the system to continuously adapt to dynamic changes.

[0059] Deep implementation of the growth stage identification module 1. Data preparation and annotation: Model training requires a large amount of labeled data. Through collaboration with large-scale edible mushroom production bases, a dataset covering multiple varieties, different growth stages, and various environmental conditions was constructed. The dataset includes: Image samples: Hundreds of thousands of images of mushroom bags labeled with growth stages.

[0060] Time-series samples: vaccination days and cumulative temperature data recorded synchronously with the images.

[0061] Data labeling was done by experienced cultivation engineers to ensure accurate labeling.

[0062] Specifically, 500,000 images of mushroom bags covering three varieties—enoki mushrooms, shiitake mushrooms, and oyster mushrooms—were collected and labeled with "mycelium growth period," "primordia differentiation period," "fruiting body growth period," and "harvest period," and were matched with the corresponding inoculation days and cumulative temperature data.

[0063] 2. Specific implementation of network structure: The classifier adopts a two-branch fusion structure.

[0064] Visual Feature Extraction Branch: Input is the preprocessed ROI image of the mushroom bag (e.g., scaled to 224x224 pixels). A convolutional neural network pre-trained on ImageNet, such as ResNet-18 or EfficientNet-B0, is used as the backbone. Its last fully connected layer is removed, and the output of the last convolutional layer is subjected to global average pooling to obtain a fixed-length visual feature vector. (e.g., 512-dimensional). This vector encodes deep semantic information of the image, such as texture (characterizing hyphal density), shape (characterizing the outline of primordia or fruiting bodies), and color features.

[0065] Temporal Feature Extraction Branch: The input is data within a short time window, such as the [number of days since inoculation, cumulative accumulated temperature] sequence of the most recent 7 days. This is treated as a two-dimensional time series. A Long Short-Term Memory (LSTM) network is used for processing. LSTM units can remember long-term dependencies, such as recognizing specific patterns in accumulated temperature accumulation during the initiation period. The LSTM hidden state at the last time step is extracted as the temporal feature vector. (e.g., 128 dimensions).

[0066] Attention fusion layer: This is key to achieving effective multimodal learning. (Simple splicing) and It may be impossible to balance the importance of both in different situations. Therefore, this implementation employs an attention mechanism. Specifically, the concatenated features [ , Input a small fully connected network, followed by a Softmax function, to generate two attention weights. and + =1 The features after weighted fusion are: This allows the network to adaptively focus on more reliable feature sources, such as relying more on temporal data when the image is blurred.

[0067] Classification output layer: This layer will fuse features. Input a fully connected layer with the number of neurons equal to the number of classes in the growth phase (e.g., 4 classes), followed by a Softmax activation function, and output the probability of each class. The category with the highest probability is selected as the final identification result, and its maximum probability value is used as the identification confidence.

[0068] Specifically, the visual branch uses a ResNet-18 pre-trained on ImageNet, with its fully connected layers removed, outputting a 512-dimensional feature vector. The temporal branch uses a single-layer LSTM with 128 hidden units, processing a temporal input of length 7 [number of days since vaccination, cumulative temperature]. The fusion layer concatenates the two feature vectors and then generates attention weights through a fully connected layer with 128 neurons and a Softmax function.

[0069] 3. Model Training and Deployment: The system is trained on a prepared dataset using the cross-entropy loss function and the Adam optimizer. To prevent overfitting, data augmentation (such as image rotation and color jitter) and Dropout techniques are employed. After training, the model is deployed to a cloud server or edge device. During inference, the system runs the classifier periodically (e.g., hourly) and outputs the current growth stage. When confidence falls below a preset threshold (e.g., 0.85), the system logs a "low-confidence recognition" message and may selectively notify a human reviewer to ensure the reliability of the system's decisions.

[0070] Detailed Implementation of Metabolic Heat Prediction Based on Digital Twin and Digital Twin Prediction Module In this embodiment, the digital twin prediction model is a hybrid model that combines a physiological and ecological mechanism model with data-driven approaches.

[0071] 1. Specific implementation of the mycelial growth kinetic model: The core of the model is the mycelial growth rate. The computational equation quantifies the combined effects of environmental factors on mycelial growth:

[0072] Each parameter and factor is explained in detail below: : Maximum specific growth rate at the optimum temperature. This is a variety-specific parameter, in units of... For example, oyster mushrooms It may be approximately 0.20 The value was lower for slower-growing varieties such as morels. This value was obtained through fitting growth experiments under controlled laboratory conditions.

[0073] Mycelial growth activation energy, in units of It reflects the sensitivity of mycelial growth and metabolism to temperature. The higher the value, the more sensitive the growth rate is to temperature changes. A typical range is 35- between.

[0074] Ideal gas constant, with a value of .

[0075] Ambient absolute temperature, in units of It is obtained by adding 273.15 to the sensor measurement value.

[0076] The optimal absolute temperature for mycelial growth is also a varietal parameter.

[0077] Humidity response factor. Adopting a Gaussian response form: .in, The measured relative humidity (%) The optimal humidity is typically 85%-95%. This is the humidity sensitivity coefficient, which determines the tolerance of the growth rate to humidity deviations. When Indicates no inhibition; the further the deviation, The closer it is to 0, the stronger the inhibition.

[0078] : Concentration response factor. An inhibition model is used. .in, CO2 concentration (ppm) This is the CO2 inhibition coefficient. When... When =0 (fresh air), =1. With The increase in CO2 and the decrease in the factor simulate the inhibitory effect of high concentration CO2 on mycelial growth.

[0079] 2. Specific implementation of metabolic heat release calculation: In calculation Then, combined with bacterial biomass (Unit: kg) can be used to predict metabolic heat release power. :

[0080] : The heat production coefficient of mycelial growth metabolism, in units of Its physical meaning is the heat released for every 1 kilogram of mycelial biomass produced. This coefficient needs to be determined through calorimetry experiments, and there are significant differences between different varieties.

[0081] : Bacterial biomass. In modular cell production, it is difficult to weigh directly online. This implementation uses two estimation methods: (1) Mechanism estimation: Based on the initial inoculation strain weight and culture medium dry weight, combined with the above growth model, the mechanism was estimated... The cumulative biomass is calculated by integrating the integrals. (2) Data-driven estimation: This involves establishing a regression model from visual features (such as mycelial coverage and image texture) to biomass, and using periodically captured images for indirect estimation. In practical systems, these two methods are often combined to improve robustness.

[0082] Taking enoki mushroom as an example, the parameters were obtained through laboratory growth experiments as follows:

[0083] The above parameters can be obtained by measuring the dry weight growth rate of mycelia under different temperature, humidity and CO2 conditions in a controlled environment chamber, and then fitting the formula using a nonlinear regression method.

[0084] 3. Specific implementation of online model calibration and error compensation: To ensure the long-term predictive accuracy of digital twin models, they must be calibrated online.

[0085] Data storage: The system continuously stores historical environmental data, control commands, and observations that indirectly reflect growth status (such as image-based biomass estimation and stage identification results) into the time series database.

[0086] Regular parameter calibration: The system initiates a calibration process every 24 hours (or after the end of a cultivation stage). Data from a past period (e.g., the past 5 days) is selected as the calibration dataset. A loss function is constructed with the objective of minimizing the deviation between the model's predicted biomass growth trend or indirect representation of metabolic heat and the observed values. Gradient descent or Bayesian optimization algorithms are used to adjust key adjustable parameters in the model (such as...). Fine-tuning is then performed. The calibration process is conducted offline in the cloud and does not affect real-time control.

[0087] Real-time error compensation: Even after calibration, short-term predictions may still be biased. Therefore, the system implements a real-time compensation mechanism. In each control cycle, the model outputs the predicted error... It also calculates a prediction error compensation term. This compensation term is derived based on the moving average of the deviations between the model-predicted temperature changes and the actual measured temperature changes over the most recent few periods. When providing to the MPC controller, what is actually used is This compensation can quickly correct systematic model mismatches or unmodeled perturbations.

[0088] The model predictive control module is the real-time control core of this invention, and its working principle is as follows: Figure 3 As shown, it abandons the traditional controller's "post-event correction" mode and instead adopts a proactive control strategy of "predict first, optimize later". This section will elaborate on the implementation details of its four key sub-processes.

[0089] 1. Establishment and Function of State-Space Prediction Model To achieve accurate predictions, a mathematical model, namely a state-space model, is first needed to describe the dynamic coupling relationship between temperature, humidity, and carbon dioxide concentration within the shelter. This model is obtained through system identification methods. Its core idea is to treat the shelter as a dynamic system, where the system's "state" consists of temperature, humidity, and CO2 concentration. The model's inputs include control commands for equipment such as air conditioners, fans, and heaters, as well as measurable external disturbances (such as outdoor temperature) and internal disturbances (i.e., the mycelial metabolic heat predicted by the digital twin module). This model can quantitatively describe how the environmental state of the shelter will change over a future period when a set of control commands and given disturbances are applied. This model is the mathematical foundation for the MPC's "prediction" function, enabling the controller to proactively assess the consequences of different control strategies.

[0090] 2. Specific steps of the rolling optimization process like Figure 3 As shown, in each control cycle (e.g., at time k once per minute), the MPC controller executes a complete "rolling optimization" loop, with the following specific steps: Measurement and Acquisition: The controller first acquires the actual measured values ​​of the current cabin environment (temperature, humidity, CO2) as the starting point for optimization. Simultaneously, it receives the future temperature reference trajectory from the temperature strategy management module and the future metabolic heat prediction sequence from the digital twin module.

[0091] Prediction: Based on the state-space model described above, the controller uses the current state as a starting point to predict the environmental state for multiple future control cycles (referred to as the "prediction time domain," such as the next 30 minutes). It simulates how the environment will evolve under different candidate control command sequences.

[0092] Optimization Solution: The controller searches for the optimal sequence among all possible candidate control sequences. The selection criteria are defined by an "objective function," which primarily aims to achieve two objectives: first, to make the predicted temperature trajectory as close as possible to the set reference trajectory to ensure control accuracy; second, to make the changes in control commands as smooth as possible to avoid frequent start-ups or drastic actions of the equipment, thereby protecting the equipment and saving energy. This process also needs to satisfy a series of "constraints," such as the power limit of the equipment, the allowable temperature fluctuation range, and the safety limits for humidity and CO2. This is a constrained optimization problem, which is completed online using an efficient solver.

[0093] Execution and Rolling: After obtaining the optimal control sequence, the controller only issues the first control command in the sequence to the executing device (e.g., setting the specific power of the air conditioner). In the next control cycle (time k+1), the controller no longer uses the second command from the previous sequence, but instead remeasures the latest environmental state and repeats the entire "prediction-optimization" process starting from this new state. This periodic, iterative optimization based on the latest feedback is the essence of "rolling optimization," enabling the system to continuously adapt to dynamic changes and possessing strong anti-interference capabilities.

[0094] 3. Integrated Implementation of Energy Consumption Optimization Strategies In this embodiment, a higher-level reinforcement learning agent collaborates with the MPC optimizer. This agent does not directly output control commands but indirectly guides the system to achieve energy savings by dynamically adjusting the parameters of the MPC optimization problem. The agent continuously senses external information (such as whether the current electricity price is peak, flat, or off-peak) and internal states (such as system load and predicted heat demand). During peak electricity price periods, the agent sends commands to the MPC to increase the penalty weight for changes in equipment power. This makes the MPC more inclined to choose a smooth, energy-saving control strategy, reducing high-power operation time. Conversely, during off-peak electricity price periods, the agent may reduce this penalty weight and appropriately relax the restrictions on the rate of temperature change, allowing the MPC to more actively utilize inexpensive electricity for cooling, "storing" cold energy for the enclosure structure and mushroom bags to offset some of the heat load during subsequent peak periods. In this way, operating costs are optimized while ensuring temperature control accuracy.

[0095] Specifically, in this embodiment, the specific operating logic of the reinforcement learning agent is as follows: State space: contains the current time step. Current temperature inside the makeshift hospital and target temperature The deviation.

[0096] Action Space: The action 'a' output by the agent is a continuous value, used to adjust the energy consumption weight coefficient in the objective function of the MPC controller. .

[0097] Reward function: The design is as follows:

[0098] in, This indicates how much electricity the air conditioner or heating device consumed at time t. It is generally a fixed constant.

[0099] 4. Implementation of monitoring and real-time correction mechanisms To address sudden disturbances (such as personnel entry / exit or equipment malfunctions) that the model cannot fully predict, this embodiment incorporates a parallel, fast safety loop. This system continuously monitors the deviation between the actual temperature and the target setpoint. When the deviation consistently exceeds a certain safety threshold, it indicates that the MPC main loop may be insufficiently responsive. At this point, the system immediately activates a parallel, tunable proportional-integral-derivative (PID) controller. This PID controller quickly calculates a correction based on the real-time deviation and directly adds it to the control command output by the MPC, rapidly fine-tuning the temperature to ensure it quickly returns to the set range. Simultaneously, the system can adaptively and temporarily increase the weight of "temperature tracking accuracy" in the MPC optimization objective, prompting the MPC to more actively eliminate deviations in subsequent optimizations. Once the deviation stabilizes, the PID corrector automatically exits, and the system fully reverts to the MPC-dominated optimization mode. This mechanism ensures that the system possesses both the long-term optimization performance of the MPC and rapid robustness in response to unexpected situations.

[0100] Detailed implementation of the anomaly detection and fault tolerance module This module serves as a "safety net" to ensure the long-term stable operation of the system, achieving end-to-end fault tolerance from data anomaly identification to seamless switching of control strategies.

[0101] 1. Multi-level anomaly detection method The system employs a combination of rule validation, statistical analysis, and machine learning to monitor sensor data in real time. Physical rule verification: Set an absolutely reasonable physical range for each sensor (e.g., the temperature cannot be lower than -5°C or higher than 50°C), and immediately alarm if the data exceeds the limit.

[0102] Spatial consistency analysis: Leveraging the advantages of redundant sensor placement, it calculates the differences between sensor readings at different locations at the same time. If a sensor's reading consistently deviates significantly from the average of its neighboring sensors, the sensor's data is deemed suspicious.

[0103] Timing consistency prediction: A simple prediction model (e.g., based on historical trends) is built for the data stream of each sensor to predict its next sampled value. If the actual measured value deviates significantly from the predicted value multiple times in a row, it indicates that the sensor may be experiencing drift or sudden failure.

[0104] Unsupervised detection using machine learning: Deploying the Isolation Forest algorithm. The model is trained using historical normal data, allowing it to learn normal correlation patterns between multi-sensor data. When running online, the algorithm can identify "outliers" that deviate from the overall pattern, such as an abnormal spike in CO2 readings when temperature and humidity are normal.

[0105] 2. Hierarchical fault-tolerant control strategy Once an anomaly is detected, the system initiates a tiered response based on the scope of the fault's impact: Level 1 Fault Tolerance (Data Layer): For a single sensor failure, the system automatically performs data replacement. If there are redundant sensors, the weighted average of their reliable data is used to replace the faulty data; if there is no redundancy, the moving average of historical data at that point is used. The control algorithm continues to run using the replaced data, and the system logs and prompts for maintenance.

[0106] Level 2 Fault Tolerance (Strategy Layer): If multiple associated sensors in a certain area malfunction simultaneously, making the state of that area unreliable, the MPC module will adaptively adjust its internal model, temporarily reducing or ignoring the weight of state variables in the faulty area, and primarily relying on measurements from other reliable areas for optimized control, while simultaneously increasing the alarm level.

[0107] Level 3 Fault Tolerance (System Level): In the event of a severe fault, such as a core controller communication interruption or the failure of more than half of the critical sensors, the anomaly detection module determines that the system health is severely compromised. It will force the system to switch control modes, downgrading from the optimized MPC mode to a pre-configured standby mode. The standby mode can be robust PID control based on a few reliable sensors, or timed control operating strictly according to a preset schedule. Its sole purpose is to prevent drastic environmental degradation and buy time for manual maintenance. Simultaneously, the system will trigger the highest level of multi-channel alarms.

[0108] Detailed implementation of the temperature strategy management module This module is a "translator" that transforms agronomic knowledge into executable targets for the controller, and its core is a structured policy knowledge base.

[0109] 1. Construction of the strategy knowledge base The knowledge base is stored in the form of a database or configuration file, predefining complete temperature control parameters for each combination of "edible fungi variety - growth stage". For example, for "enoki mushroom - mycelium growth stage", the strategy explicitly records a target temperature of 20.0°C with an allowable fluctuation range of ±0.3°C. For "shiitake mushroom - primordia differentiation stage", the strategy is more detailed, including not only daytime target temperature (e.g., 18.0°C) and nighttime target temperature (e.g., 10.0°C), but also specifying the temperature variation cycle (24 hours), the day-night time division point, and the total number of days the temperature variation stimulus needs to last. These parameters are a digital representation of the cultivation experts' experience.

[0110] 2. Dynamic generation of temperature reference trajectory This module is triggered when the growth stage identification module identifies a new stage. It retrieves the corresponding strategy from the knowledge base based on the current variety and stage label. Then, it generates a temperature reference trajectory for a future period based on the strategy.

[0111] For stages requiring constant temperature (such as the mycelium growth period of enoki mushrooms), the trajectory is a flat straight line.

[0112] For stages requiring temperature-dependent stimulation (such as the primordia differentiation period of shiitake mushrooms), the trajectory is a periodic sawtooth wave that switches periodically between the target daytime and nighttime temperatures.

[0113] When the growth stage changes (e.g., from the mycelium growth stage to the primordium differentiation stage), the module does not cause the target temperature to jump immediately. Instead, it generates a smooth transition trajectory (e.g., a linear decrease from 20°C to 13°C over 2 hours). This transition trajectory serves as the tracking target for the MPC, avoiding physiological shock to the edible fungi. At the same time, its set transition rate is also passed to the MPC controller as a constraint.

[0114] The following two typical scenarios illustrate the implementation process and technical effects of the system of the present invention.

[0115] Example 1: Intelligent Temperature Control Throughout the Factory Cultivation of Enoki Mushrooms System Configuration and Initialization: The system of this invention is deployed in a 12m×8m×3.5m enoki mushroom cultivation container. The user selects "enoki mushroom" as the cultivation variety on the front-end interface. System initialization begins, all sensors start working, and the strategy knowledge base loads enoki mushroom-specific parameters.

[0116] Precise temperature control during mycelial growth: After inoculation, images show sparse hyphae and a short inoculation period, which the system identifies as the "mycelial growth period." The strategy module invokes the strategy: target temperature 20.0°C, fluctuation range ±0.3°C. The digital twin model predicts metabolic heat based on the current mild environment. Approximately 50W. The MPC controller initially operates with the primary objective of tracking 20.0°C. Due to the low internal heat load, the MPC's output commands mainly coordinate with the inverter air conditioner to operate at low power to maintain the temperature, and intermittently activate ventilation to regulate humidity and CO2. The actual temperature is stably controlled within the range of 19.8°C to 20.2°C.

[0117] Intelligent switching and energy-saving operation during the primordia differentiation period: On day 18 of cultivation, industrial camera images show that the mycelium on the surface of the bags is dense white and completely covered, and the accumulated temperature has reached the set threshold. The multimodal classifier outputs "primordia differentiation period" with high confidence. The strategy module immediately switches the temperature reference trajectory to 13.0°C. At the same time, the digital twin model predicts that mycelial metabolism remains vigorous during this stage. The predicted value has risen to approximately 150W. When MPC tracks the new setpoint, it adjusts accordingly. The forecasts escalated the cooling commands in advance, achieving stable, overshoot-free cooling. Energy optimization played a significant role during this phase: at 11 PM, during off-peak electricity hours, the reinforcement learning agent's actions were used to reduce the energy consumption penalty coefficient of the MPC. The temperature was lowered, allowing for a faster cooling rate. After MPC solving, the system operated at higher power for cooling during the early morning hours, lowering the temperature to 12.8°C to store cold energy for the enclosure structure and the bacterial culture bags. During peak daytime electricity prices, the agent will... When the power is increased, the MPC control becomes more conservative, relying mainly on stored cold energy to maintain the temperature, which significantly reduces peak power consumption.

[0118] Example of anomaly handling: During operation, the anomaly detection module, using the isolated forest algorithm, identified a temperature sensor data stream located on the east wall as an "outlier," exhibiting a significant statistical discrepancy with adjacent sensors. The fault-tolerant module immediately determined the sensor was faulty and replaced its data with the reading from the corresponding sensor on the west wall (compensated for spatial temperature difference). MPC control remained unaffected, and a maintenance work order for the sensor was generated in the background.

[0119] Example 2: Precise and repeatable control of temperature-induced stimulation in shiitake mushrooms Temperature variation strategy configuration: The system knowledge base has pre-set shiitake mushroom strategies in the shiitake mushroom cultivation container. The strategy for the "primordia differentiation period" is as follows: target temperature of 18.0°C during the day (6:00-18:00) and target temperature of 10.0°C at night (18:00-6:00), with a temperature difference of 8°C, for 7 days.

[0120] Dynamic trajectory tracking: When the mycelium reaches physiological maturity and the system identifies the primordia differentiation phase, the temperature strategy management module generates a zigzag reference trajectory with a 24-hour cycle. This tests the dynamic performance of the controller. The MPC needs to solve an optimization problem tracking this changing setpoint in each control cycle. The feedforward role of the digital twin is crucial: the model predicts that mycelial metabolism will slow down when the temperature drops in the evening. The temperature drops slightly, so the MPC will issue cooling / ventilation commands earlier and more proactively; while during the morning warming phase, it will reduce cooling or start heating earlier. Through this "prediction + optimization" approach, the MPC achieves accurate tracking of the dynamic trajectory, with the actual temperature curve smoothly matching the set value, the day-night temperature difference remaining stable within 8.0±0.5°C, and no overshoot at the reversal point.

[0121] Technical Effects: This invention completely changes the situation where temperature stimulation of shiitake mushrooms relies on experienced farmers, resulting in inaccurate control and large batch-to-batch variations. Through automated, high-precision, and repeatable temperature control, it ensures that each batch of shiitake mushrooms forms primordia synchronously under optimal temperature stimulation, significantly improving the uniformity of primordia and the consistency of final yield and quality, providing a reliable technical guarantee for the industrialized and standardized production of high-end shiitake mushrooms.

[0122] System Module Description The system claimed in this application consists of eight modules. The specific embodiments described above have provided detailed and operable technical support for the implementation of each module, and the corresponding relationships are summarized as follows: Data acquisition module: corresponds to step S101 in the implementation method, and the sensing execution layer in the hardware architecture. Specifically, it implements temperature, humidity, and CO2 sensor arrays, an industrial camera, data acquisition circuits, and preprocessing software.

[0123] Growth Stage Recognition Module: Corresponds to the "Intelligent Recognition of Multimodal Growth Stages" section. Specifically, it is implemented as a deep learning model containing a CNN visual branch, an LSTM temporal branch, an attention fusion layer, and a Softmax classifier, deployed as a callable software service.

[0124] Temperature Strategy Management Module: This corresponds to the sections on "Dynamic Temperature Reference Trajectory Generation" and "Detailed Implementation of Temperature Strategy Management Module." Specifically, it is implemented as a software module with a built-in variety knowledge base (database / configuration file) and a trajectory generation algorithm (including logic for constant temperature, variable temperature, and smooth transition).

[0125] Digital twin prediction module: This corresponds to the sections on "Metabolic Heat Prediction Based on Digital Twins" and "Detailed Implementation of the Digital Twin Prediction Module." Specifically, it provides computational services for running mycelial growth kinetics models and online calibration algorithms, outputting metabolic heat prediction sequences.

[0126] Model Predictive Control Module: Corresponds to the sections on "Multivariable Model Predictive Control (MPC) Rolling Optimization" and "Detailed Implementation of the Model Predictive Control (MPC) Module". Specifically, it is implemented as real-time control software integrating a state-space model, a quadratic programming solver, rolling optimization logic, and a monitoring and correction loop.

[0127] Control execution module: corresponds to the actuator drive layer in the hardware architecture. Specifically, it consists of hardware interface circuits and driver software that convert digital control commands into inverter analog signals and relay switching signals.

[0128] Energy consumption optimization module: This corresponds to the "Integrated Implementation of Energy Consumption Optimization Strategies" section. Specifically, it is implemented as an intelligent agent based on a reinforcement learning framework, capable of sensing external electricity prices and internal states, and dynamically adjusting MPC optimization parameters through predefined interfaces.

[0129] Anomaly Detection and Fault Tolerance Module: This corresponds to the "Detailed Implementation of Anomaly Detection and Fault Tolerance Module" section. Specifically, it is implemented as a background daemon or service containing multi-layered anomaly detection algorithms, data substitution logic, and hierarchical control mode switching strategies.

[0130] The above eight modules interact with each other through standard industrial communication protocols, working together to form a complete intelligent control system.

[0131] Finally, it should be noted that the above are merely preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent temperature control in a mobile mushroom cultivation facility based on artificial intelligence, characterized in that, Includes the following steps: Collect environmental and growth status data within the edible mushroom container; The growth status data is input into a multimodal deep learning classifier to identify the current growth stage of the edible fungus; Based on the growth stage, the corresponding temperature requirement strategy is invoked to generate a temperature reference trajectory, which includes the target temperature setpoint. Construct a digital twin model of edible fungi growth dynamics to predict temperature demand changes within a set time period based on the environmental data and the growth stage. Based on the Model Predictive Control (MPC) algorithm, a state-space model with three coupled variables—temperature, humidity, and CO2—is constructed to characterize the dynamic coupling relationship between the environmental data. Based on the temperature reference trajectory and the predicted changes in temperature demand, a temperature control accuracy constraint is set, and the optimal temperature control command is solved in a rolling manner within the prediction time domain; wherein, The temperature control command is executed to coordinate the operation of the temperature control equipment inside the cabin.

2. The method according to claim 1, characterized in that, The multimodal deep learning classifier uses a convolutional neural network to extract visual features from the mushroom bag image, and a long short-term memory network to extract temporal features of inoculation days and accumulated temperature. After fusing the visual features and the temporal features through an attention mechanism, it outputs the identification result of the growth stage.

3. The method according to claim 1, characterized in that, The digital twin model of edible fungi growth kinetics uses a mycelial temperature response model to calculate the mycelial growth rate and calculates the metabolic heat release based on the mycelial growth rate; the expression for the mycelial growth rate calculated by the mycelial temperature response model is as follows: in, The mycelial growth rate is the specific growth rate. The maximum specific growth rate at the optimum temperature. For mycelial growth activation energy, The gas constant is The absolute temperature of the culture environment in the environmental data. This is the optimal temperature for mycelial growth, and its value is dynamically adjusted based on the currently identified growth stage of the edible fungus. The relative humidity response factor is expressed as: ,in The relative humidity in the environmental data. To determine the optimal relative humidity, the value is dynamically adjusted based on the currently identified growth stage of the edible fungi, ranging from 85% to 95%. Humidity sensitivity coefficient; The CO2 concentration response factor is expressed as follows: ,in The CO2 concentration in the environmental data. This represents the CO2 inhibition coefficient. The metabolic heat release is calculated using the following formula: in, The power of metabolic heat release. The heat production coefficient of mycelial growth metabolism. This refers to bacterial biomass.

4. The method according to claim 3, characterized in that, The state-space model uses temperature, humidity, and CO2 concentration as state variables, air conditioning cooling power, ventilation volume, and heating power as control input variables, and temperature measurement, humidity measurement, and CO2 concentration measurement as output variables. The Model Predictive Control (MPC) algorithm takes optimal temperature control accuracy as its objective function and introduces the metabolic heat release as a measurable disturbance into the model. It then adjusts the cooling power in advance based on the metabolic heat release through feedforward compensation.

5. The method according to claim 1, characterized in that, Also includes: By using a reinforcement learning agent to perceive the current electricity price and system operating status, and under the premise of meeting the temperature control accuracy constraints, the optimal temperature control command is economically modified and the temperature control command is dynamically adjusted.

6. The method according to claim 1, characterized in that, Also includes: The environmental data is subjected to anomaly detection. When anomaly is detected in the sensor data, the average data of redundant sensors or the moving average of historical data is used to replace the abnormal data, and the system switches to a degraded control mode to maintain the basic function of the temperature control system.

7. The method according to claim 1, characterized in that, The temperature requirement strategy includes temperature setpoints, allowable fluctuation ranges, and temperature change stimulation parameters for different edible fungi varieties at each growth stage; for temperature-change fruiting edible fungi, the temperature reference trajectory includes diurnal temperature difference setpoints and temperature change stimulation timing.

8. The method according to claim 1, characterized in that, Also includes: Historical environmental data, growth status data, and actual control effects are stored in a database. The parameters of the digital twin model of edible fungi growth kinetics are periodically calibrated online based on the measured data using the gradient descent method. When the deviation between the model's predicted value and the measured value exceeds a set threshold, an error compensation term is introduced to correct the prediction result.

9. The method according to claim 1, characterized in that, Also includes: The system collects actual temperature data after the control command is executed, calculates the deviation between the actual temperature and the target temperature setpoint, and adjusts the weight matrix parameters of the objective function when the deviation exceeds a set threshold. The system then uses a proportional-integral-derivative (PID) controller to correct the temperature deviation in real time.

10. An intelligent temperature control system for edible mushroom mobile cabins based on artificial intelligence, characterized in that, include: The data acquisition module is used to collect environmental and growth status data inside the edible mushroom container. The growth stage identification module is used to input the growth status data into a multimodal deep learning classifier to identify the current growth stage of the edible fungus. The temperature strategy management module is used to invoke the corresponding temperature requirement strategy according to the growth stage and generate a temperature reference trajectory, which includes the target temperature setpoint. The digital twin prediction module is used to construct a digital twin model of edible fungi growth dynamics and predict temperature demand changes within a set time period based on the environmental data and the growth stage. The model predictive control module is used to establish a state-space model of temperature-humidity-CO2 three-variable coupling based on the MPC algorithm, and to set temperature control accuracy constraints and calculate the optimal temperature control command based on the temperature reference trajectory and the predicted results of temperature demand changes. The control execution module is used to execute the temperature control commands and coordinate the operation of the temperature control equipment inside the cabin.

Images