Mushroom house carbon dioxide adjusting method based on multi-mode growth stage recognition

Abstract

The invention discloses a method for adjusting carbon dioxide in a mushroom house based on multi-modal growth stage recognition. The method comprises the following steps: S1, obtaining pre-processed mushroom house environment data and pre-processed mushroom house multi-modal image data; s2, outputting a mushroom growth stage classification label and a mushroom growth stage confidence vector; s3, constructing a multi-target fitness function according to a stage sensing result and generating a stage weight group; s4, obtaining an optimal gas regulation control strategy; s5, adjusting the carbon dioxide concentration of the mushroom house according to the optimal gas adjusting control strategy; and S6, calculating an environmental stability index, triggering the whale optimization algorithm to re-execute the steps S4 to S6 when the environmental stability index exceeds a limit or the confidence vector in the mushroom growth stage is lower than a threshold value, and circularly executing the steps S1 to S6 in the whole mushroom house production period. According to the method, the problems of premature convergence and local optimization are effectively avoided, and the environmental adaptability and the energy consumption optimization effect of a gas regulation strategy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of mushroom houses, and in particular to a carbon dioxide regulation method for a mushroom house based on multimodal growth stage recognition. Background Art

[0002] With the continuous improvement of the intelligent level of facility agriculture, a variety of sensors and image acquisition equipment have gradually been introduced into the mushroom house environmental control system. Carbon dioxide concentration is an important environmental factor affecting mushroom growth, and its precise regulation has become a key link in mushroom house management. However, the existing carbon dioxide regulation methods in mushroom houses generally have certain limitations.

[0003] At present, most mushroom houses adopt rule-based control methods based on simple threshold settings or single sensor data, which often ignore the stage-by-stage changes in environmental requirements during mushroom growth, resulting in a lack of targeted and dynamic adaptability in the adjustment strategy. At the same time, traditional methods mainly rely on single-modal input of environmental data, and have weak ability to recognize and understand the growth status of mushrooms, and cannot fully utilize the rich feature information in multimodal image data.

[0004] On the other hand, some environmental control methods that introduce intelligent optimization algorithms have the problem of being out of touch with the actual growth stage. They fail to adjust the optimization direction in real time based on the dynamic process of mushroom growth, resulting in high energy consumption or excessive environmental fluctuations during the carbon dioxide regulation process, affecting the continuity and quality of mushroom growth. In addition, in existing multimodal data processing, the robustness against modal missing or modal distortion is insufficient, and it is easy to cause overall regulation decision-making errors due to local data anomalies.

[0005] In summary, there is an urgent need for a new adaptive adjustment method that can combine the mushroom growth stage, integrate the characteristics of multimodal data and have efficient global optimization capabilities to improve the intelligence level and production efficiency of mushroom house environmental control. Summary of the Invention

[0006] One purpose of the present invention is to propose a mushroom house carbon dioxide regulation method based on multimodal growth stage identification, which effectively avoids premature convergence and local optimal problems, and improves the environmental adaptability and energy consumption optimization effect of the gas regulation strategy.

[0007] According to 1 of the embodiments of the present invention, a method for regulating carbon dioxide in a mushroom house based on multimodal growth stage identification comprises the following steps:

[0008] S1. Synchronously collect environmental data and multimodal image data from the mushroom house. The environmental data includes carbon dioxide concentration, air temperature, air humidity, and light intensity, and the multimodal image data includes visible light images and infrared thermal images. Data integrity testing, timestamp alignment, and noise suppression are performed on the environmental data and multimodal image data to obtain preprocessed environmental data and multimodal image data.

[0009] S2. Generate a modality loss mask based on the preprocessed multimodal image data of the mushroom house. Input the preprocessed multimodal image data and the modality loss mask into a deep separable sparse Transformer model, and output a mushroom growth stage classification label and a mushroom growth stage confidence vector.

[0010] S3. Combine the mushroom growth stage classification labels, mushroom growth stage confidence vectors, and preprocessed mushroom house environment data to form a stage perception result. Based on the stage perception result, construct a multi-objective fitness function and generate a stage weight group.

[0011] S4. Initialize the Whale Optimization Algorithm population, map the mushroom growth phase confidence vector to a dynamic search control factor, and perform global search and local convergence iterations using a multi-objective fitness function as the evaluation criterion. During the Whale Optimization Algorithm iteration, update the fitness evaluation in real time, output a candidate gas regulation sequence, perform temporal smoothing compensation on the candidate gas regulation sequence, and obtain the optimal gas regulation control strategy.

[0012] S5. Distribute the optimal gas regulation control strategy to the mushroom house execution equipment network, which includes variable frequency fans, solenoid valves, and a carbon dioxide injection system. This network regulates carbon dioxide concentration in the mushroom house according to the optimal gas regulation control strategy and simultaneously collects actual execution feedback data.

[0013] S6. Calculate the environmental stability index based on the actual execution feedback data and the confidence vector of the mushroom growth stage. When the environmental stability index exceeds the limit or the confidence vector of the mushroom growth stage is lower than the threshold, trigger the whale optimization algorithm to re-execute steps S4 to S6. Circulate steps S1 to S6 throughout the entire mushroom house production cycle to achieve adaptive dynamic adjustment of carbon dioxide concentration and optimal control of energy consumption in the mushroom house.

[0014] Optionally, step S1 includes the following steps:

[0015] S11. Synchronously collect environmental data and multimodal image data of the mushroom house during each synchronization cycle of the mushroom house data acquisition device. The environmental data consists of carbon dioxide concentration, air temperature, air humidity, and light intensity. The multimodal image data consists of visible light images and infrared thermal images of the mushroom house. The total number of cycles for data acquisition is N.

[0016] S12. During each synchronization cycle, record the carbon dioxide concentration, air temperature, air humidity, light intensity, visible light image, and infrared thermal image of the mushroom house corresponding to the i-th cycle, forming a complete original mushroom house environment dataset and an original mushroom house multimodal image dataset.

[0017] S13. Perform integrity checks on the carbon dioxide concentration, air temperature, air humidity, and light intensity at each sampling point in the original mushroom house environmental dataset, removing any missing or obviously abnormal records. Perform validity checks on each set of mushroom house visible light images and mushroom house infrared thermal images in the original mushroom house multimodal image dataset, removing any samples with missing or severely distorted images and retaining any data that pass the tests. This yields the integrity-tested mushroom house environmental dataset and the integrity-tested mushroom house multimodal image dataset.

[0018] S14. Synchronize the integrity-checked Mushroom House environment dataset and the integrity-checked Mushroom House multimodal image dataset according to their timestamps, ensuring that the time difference between each piece of Mushroom House environment data and its corresponding Mushroom House multimodal image data is less than a set time difference threshold. This yields the timestamp-aligned Mushroom House environment dataset and the timestamp-aligned Mushroom House multimodal image dataset.

[0019] S15. Apply sliding window filtering to each piece of carbon dioxide concentration, air temperature, air humidity, and light intensity in the mushroom house environment dataset after timestamp alignment, and obtain the preprocessed mushroom house environment dataset E. proc The effective period number of the preprocessed mushroom house environment data set is recorded as N s .

[0020] S16. Apply Gaussian filtering to each visible light image of the mushroom house in the timestamp-aligned multimodal image dataset, and apply median filtering to each infrared thermal image of the mushroom house, to construct a preprocessed multimodal image dataset I of the mushroom house. proc , the effective number of cycles of the preprocessed mushroom house multimodal image dataset is also N s .

[0021] Optionally, step S2 includes the following steps:

[0022] S21. Based on the preprocessed mushroom house multimodal image dataset I proc The actual modal validity in each acquisition cycle is determined by setting the dynamic modal weight factor Construct the modality missing mask matrix Mask i , used to implement modal feature reconstruction and sparse attention dynamic adjustment in the deep separable sparse Transformer model:

[0023]

[0024] in, represents the weight ratio of the visible light image data of the mushroom house in the i-th cycle during attention fusion, represents the weight ratio of the infrared thermal image data of the mushroom house in the i-th cycle during attention fusion. When the modality is complete, both are 0.5. When the visible light image data of the mushroom house is missing or severely distorted, the value is adjusted to 0.1. When the infrared thermal image data of the mushroom house is missing or severely distorted, the value is adjusted to 0.9.

[0025] S22. Designing a modality-complementary sparse attention mechanism within the deep separable sparse Transformer model to optimize the Mushroom House multimodal image dataset I proc To solve the mid-modal missing problem, we introduce a cross-modal complementary coefficient when calculating the self-attention weight of each modality. Cross-modal complementarity coefficient Used to compensate for missing or weak modal features through modal features:

[0026]

[0027] in, represents the cross-modal complementary sensitivity factor, which is used to control the sensitivity of the difference between the carbon dioxide concentration in the mushroom house and the target concentration to the modal compensation degree. target Indicates the target carbon dioxide concentration at the current mushroom growth stage, C i represents the actual measured carbon dioxide concentration of the mushroom house during the i-th cycle, ensuring that when the gap between the actual carbon dioxide concentration in the mushroom house and the target concentration increases, the compensation strength of the effective mode for the missing mode is adaptively improved;

[0028] S23. Modal missing mask matrix Mask i and cross-modal complementarity coefficients Self-attention weight matrix working together in the deep separable sparse Transformer model Get the multimodal sparse self-attention weight matrix that integrates the multimodal image data of the mushroom house and the environment data of the mushroom house:

[0029]

[0030] in, represents the initial self-attention weight matrix without sparse adjustment;

[0031] S24. Use the multimodal sparse self-attention weight matrix to fuse the spatial and temporal features of the mushroom house multimodal image data and output the mushroom growth stage classification label P i and the mushroom growth stage confidence vector γ i , Mushroom Growth Stage Classification Label P i The value range is {P1, P2, P3, P4, P5}, which correspond to the hyphae laying period, primordium emergence period, fruiting body differentiation period, rapid expansion development period and pre-harvest period, respectively. The confidence vector γ of the mushroom growth stage i Used to indicate the credibility of the deep separable sparse Transformer model's classification results for the mushroom growth stage in the i-th cycle:

[0032]

[0033] in, Indicates that the classification label of the mushroom growth stage in the i-th cycle is P j confidence level.

[0034] Optionally, step S3 includes the following steps:

[0035] S31. Record the mushroom growth stage classification labels and mushroom growth stage confidence vectors output by the deep separable sparse Transformer model for each acquisition cycle, along with the preprocessed carbon dioxide concentration, preprocessed air temperature, preprocessed air humidity, and preprocessed light intensity of the mushroom room, in chronological order to form a stage perception result set S.

[0036] S32. For each collection cycle in the stage perception result set S, according to the mushroom growth stage P i The target environmental parameters are used to calculate the carbon dioxide concentration deviation of the mushroom house. Air temperature deviation E T,i , air humidity deviation E H,i and light intensity deviation E L,i ;

[0037] S33. Extracting mushroom growth stage confidence vector γ i In the current stage P i The corresponding confidence Multiply the confidence by the preset basic weight coefficient to get the carbon dioxide concentration weight of the mushroom house. Air temperature weight w T,i , air humidity weight w H,i, light intensity weight w L,i ;

[0038] S34. Set the carbon dioxide concentration deviation in the mushroom house Weighted with carbon dioxide concentration Multiply the air temperature deviation E T,i With air temperature weight w T,i Multiply the air humidity deviation E H,i With air humidity weight w H,i Multiply the light intensity deviation E L,i and the light intensity weight w L,i Multiply them, and then add the four product results to get the multi-objective fitness function F of the current acquisition cycle i , multi-objective fitness function F i The smaller the value, the closer the mushroom house environment is to the target state.

[0039] Optionally, the carbon dioxide concentration deviation of the mushroom house is the absolute difference between the pretreatment CO2 concentration and the corresponding target CO2 concentration;

[0040] The air temperature deviation E T,i is the absolute difference between the pre-treated air temperature and the corresponding target air temperature;

[0041] The air humidity deviation E H,i is the absolute difference between the pre-treated air humidity and the corresponding target air humidity;

[0042] The light intensity deviation E L,i is the absolute difference between the pre-processed light intensity and the corresponding target light intensity.

[0043] Optionally, step S4 includes the following steps:

[0044] S41. Extract the mushroom growth stage confidence of each collection cycle based on the stage perception result set Set mushroom growth stage confidence Mapping to dynamic exploration radius factor Used to control the global exploration range of the whale optimization algorithm. In the mushroom growth stage, the higher the confidence, the smaller the exploration radius, which tends to be local development.

[0045] S42. Initialize the whale optimization algorithm population. The position vector of each individual in the population is defined as the mushroom room gas regulation control parameter triplet in, is the carbon dioxide injection rate of the jth individual in the kth generation, T vent,k,j is the ventilation duration, V valve,k,j is the air valve opening ratio;

[0046] S43. In each iteration of the whale optimization algorithm, the dynamic exploration control factor δ i Adjust individual updates and define the capture control vector A k,i With the guidance vector C k,i for:

[0047] A k,i =δ i (2r k -1),C k,i =2r k ;

[0048] Among them, r k is a random vector;

[0049] When rounding up the control vector |A k,i When |<1, the individual moves towards the optimal individual of the current generation. Move closer:

[0050]

[0051] When rounding up the control vector |A k,i When |≥1, the spiral bubble network convergence strategy is executed and the growth stage guidance angle is introduced

[0052]

[0053] in, Indicates the mushroom growth stage P i The corresponding convergence direction guide angle, l is the spiral random variable, b0 is the initial spiral convergence coefficient, X k,j is the gas regulation control parameter vector of the jth individual in the kth generation, including the current carbon dioxide injection rate, ventilation duration and air valve opening ratio, which is used to describe the specific control variables of the mushroom house gas regulation scheme. k+1,j is the updated gas regulation control parameter vector of the jth individual in the k+1th generation, based on the kth generation population state and the capture control vector A k,i and the guide vector C k,i Evolution obtained;

[0054] S44. After completing the population position update of each generation, based on the multi-objective fitness function F i Evaluate the fitness of each individual and introduce the environmental stability coefficient σ i :

[0055]

[0056] Among them, α is the stability sensitivity factor, C i is the current actual carbon dioxide concentration, is the target carbon dioxide concentration for the stage;

[0057] The multi-objective fitness function is modified based on the environmental stability coefficient to obtain the environmental corrected fitness:

[0058] F′ i =F i ·σ i ;

[0059] And the environmentally corrected fitness F′ i The smallest individual updates the optimal individual

[0060] S45. Reaching the maximum iteration number G in the whale optimization algorithm max After that, the complete gas adjustment candidate sequence Q is obtained candidate ,The gas regulation candidate sequence consists of the optimal gas regulation control parameters of each ,acquisition period;

[0061] S46. Adjusting the candidate sequence Q for gas candidate , introduce the inertia correction smoothing mechanism based on the confidence-driven mushroom growth stage, and define the inertia smoothing factor ω for each cycle i , and apply the inertia correction formula to generate the final optimal gas regulation control strategy Q optimal :

[0062] Q optimal,i =ω i Q candidate,i +(1-ω i )Q optimal,i-1 ;

[0063] Among them, Q optimal,i-1 The gas regulation control strategy that has been smoothed in the previous cycle;

[0064] S47. Output the optimal gas regulation control strategy Q after inertia correction and smoothing. optimal , for the mushroom room execution equipment network to perform corresponding carbon dioxide injection, ventilation and air valve opening adjustment operations.

[0065] Optionally, the optimal gas regulation control strategy Q optimal,i According to the following standard output, the carbon dioxide injection rate, ventilation duration and air valve opening of the mushroom house are determined respectively:

[0066] When mushroom growth stage classification label P i It is the mycelium laying period P1 and the carbon dioxide concentration in the mushroom house is lower than the target concentration When the carbon dioxide injection rate Take the corresponding component in the gas regulation candidate sequence as an upper limit of 10%, and the ventilation duration T vent,i Keep the original value, the air valve opening Vvalve,i 5% reduction;

[0067] When mushroom growth stage classification label P i It is the primordium emergence stage P2 or the fruiting body differentiation stage P3, and the mushroom house is dioxygenated.

[0068] When the carbon concentration is close to the target concentration ±50ppm, the carbon dioxide injection rate Use the inertia correction smoothing value, ventilation duration T vent,i Moderately increase 10%, the air valve opening V valve,i Keep smooth value;

[0069] When mushroom growth stage classification label P i It is the rapid expansion and development stage P4 or the early harvest stage P5, and the carbon dioxide concentration in the mushroom house is higher than the target concentration When the carbon dioxide injection rate is above Directly decrease by 15%, ventilation duration T vent,i Increase by 20%, air valve opening V valve,i Synchronization increased by 10%.

[0070] The beneficial effects of the present invention are:

[0071] (1) This paper proposes a deep separable sparse Transformer model, which combines a dynamic modality loss mask with a cross-modal complementary sparse attention mechanism to specifically optimize the modality loss and noise interference problems in the multimodal image data of the mushroom house. By dynamically generating a modality weight matrix and introducing a cross-modal complementary coefficient based on the carbon dioxide concentration deviation, it adaptively compensates for the missing or weakened modal features, achieving high-reliability identification of mushroom growth stages under incomplete or distorted data conditions, significantly improving the accuracy of growth stage discrimination.

[0072] (2) In terms of optimization control strategy, the present invention dynamically maps the mushroom growth stage confidence vector to the exploration control factor in the whale optimization algorithm, and proposes a dynamic exploration radius adjustment mechanism driven by stage confidence. It can dynamically adjust the global exploration and local development weights of the whale optimization population according to the judgment credibility of the current growth status of the mushroom, thereby effectively avoiding premature convergence and local optimal problems, and improving the environmental adaptability and energy consumption optimization effect of the gas regulation strategy.

[0073] (3) The present invention introduces an inertia correction smoothing mechanism, and dynamically smoothes the gas regulation candidate sequence output by the whale optimization through the time series inertia compensation formula driven by the confidence of the growth stage, ensuring the continuity and stability of the regulation strategy during the periodic change process. It can automatically and smoothly transition the control parameters according to the stage judgment confidence, and effectively suppress the drastic jump of the regulation strategy during the execution process. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0075] Figure 1 This is a flow chart of a mushroom house carbon dioxide regulation method based on multimodal growth stage recognition proposed by the present invention. DETAILED DESCRIPTION

[0076] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0077] refer to Figure 1 1. A method for regulating carbon dioxide in a mushroom house based on multimodal growth stage recognition, comprising the following steps:

[0078] S1. Synchronously collect environmental data and multimodal image data from the mushroom house. The environmental data includes carbon dioxide concentration, air temperature, air humidity, and light intensity, and the multimodal image data includes visible light images and infrared thermal images. Data integrity testing, timestamp alignment, and noise suppression are performed on the environmental data and multimodal image data to obtain preprocessed environmental data and multimodal image data.

[0079] S2. Generate a modality loss mask based on the preprocessed multimodal image data of the mushroom house. Input the preprocessed multimodal image data and the modality loss mask into a deep separable sparse Transformer model, and output a mushroom growth stage classification label and a mushroom growth stage confidence vector.

[0080] S3. Combine the mushroom growth stage classification labels, mushroom growth stage confidence vectors, and preprocessed mushroom house environment data to form a stage perception result. Based on the stage perception result, construct a multi-objective fitness function and generate a stage weight group.

[0081] S4. Initialize the Whale Optimization Algorithm population, map the mushroom growth phase confidence vector to a dynamic search control factor, and perform global search and local convergence iterations using a multi-objective fitness function as the evaluation criterion. During the Whale Optimization Algorithm iteration, update the fitness evaluation in real time, output a candidate gas regulation sequence, perform temporal smoothing compensation on the candidate gas regulation sequence, and obtain the optimal gas regulation control strategy.

[0082] S5. Distribute the optimal gas regulation control strategy to the mushroom house execution equipment network, which includes variable frequency fans, solenoid valves, and a carbon dioxide injection system. This network regulates carbon dioxide concentration in the mushroom house according to the optimal gas regulation control strategy and simultaneously collects actual execution feedback data.

[0083] S6. Calculate the environmental stability index based on the actual execution feedback data and the confidence vector of the mushroom growth stage. When the environmental stability index exceeds the limit or the confidence vector of the mushroom growth stage is lower than the threshold, trigger the whale optimization algorithm to re-execute steps S4 to S6. Circulate steps S1 to S6 throughout the entire mushroom house production cycle to achieve adaptive dynamic adjustment of carbon dioxide concentration and optimal control of energy consumption in the mushroom house.

[0084] In this embodiment, step S1 includes the following steps:

[0085] S11. Synchronously collect environmental data and multimodal image data of the mushroom house during each synchronization cycle of the mushroom house data acquisition device. The environmental data consists of carbon dioxide concentration, air temperature, air humidity, and light intensity. The multimodal image data consists of visible light images and infrared thermal images of the mushroom house. The total number of cycles for data acquisition is N.

[0086] S12. During each synchronization cycle, record the carbon dioxide concentration, air temperature, air humidity, light intensity, visible light image, and infrared thermal image of the mushroom house corresponding to the i-th cycle, forming a complete original mushroom house environment dataset and an original mushroom house multimodal image dataset.

[0087] S13. Perform integrity checks on the carbon dioxide concentration, air temperature, air humidity, and light intensity at each sampling point in the original mushroom house environmental dataset, removing any missing or obviously abnormal records. Perform validity checks on each set of mushroom house visible light images and mushroom house infrared thermal images in the original mushroom house multimodal image dataset, removing any samples with missing or severely distorted images and retaining any data that pass the tests. This yields the integrity-tested mushroom house environmental dataset and the integrity-tested mushroom house multimodal image dataset.

[0088] S14. Synchronize the integrity-checked Mushroom House environment dataset and the integrity-checked Mushroom House multimodal image dataset according to their timestamps, ensuring that the time difference between each piece of Mushroom House environment data and its corresponding Mushroom House multimodal image data is less than a set time difference threshold. This yields the timestamp-aligned Mushroom House environment dataset and the timestamp-aligned Mushroom House multimodal image dataset.

[0089] S15. Apply sliding window filtering to each piece of carbon dioxide concentration, air temperature, air humidity, and light intensity in the mushroom house environment dataset after timestamp alignment, and obtain the preprocessed mushroom house environment dataset E. proc The effective period number of the preprocessed mushroom house environment data set is recorded as N s .

[0090] S16. Apply Gaussian filtering to each visible light image of the mushroom house in the timestamp-aligned multimodal image dataset, and apply median filtering to each infrared thermal image of the mushroom house, to construct a preprocessed multimodal image dataset I of the mushroom house. proc , the effective number of cycles of the preprocessed mushroom house multimodal image dataset is also N s .

[0091] In this embodiment, step S2 includes the following steps:

[0092] S21. Based on the preprocessed mushroom house multimodal image dataset I proc The actual modal validity in each acquisition cycle is determined by setting the dynamic modal weight factor Construct the modality missing mask matrix Mask i , used to implement modal feature reconstruction and sparse attention dynamic adjustment in the deep separable sparse Transformer model:

[0093]

[0094] in, represents the weight ratio of the visible light image data of the mushroom house in the i-th cycle during attention fusion, represents the weight ratio of the infrared thermal image data of the mushroom house in the i-th cycle during attention fusion. When the modality is complete, both are 0.5. When the visible light image data of the mushroom house is missing or severely distorted, the value is adjusted to 0.1. When the infrared thermal image data of the mushroom house is missing or severely distorted, the value is adjusted to 0.9.

[0095] S22. Designing a modality-complementary sparse attention mechanism within the deep separable sparse Transformer model to optimize the Mushroom House multimodal image dataset I proc To solve the mid-modal missing problem, we introduce a cross-modal complementary coefficient when calculating the self-attention weight of each modality. Cross-modal complementarity coefficient Used to compensate for missing or weak modal features through modal features:

[0096]

[0097] in, represents the cross-modal complementary sensitivity factor, which is used to control the sensitivity of the difference between the carbon dioxide concentration in the mushroom house and the target concentration to the modal compensation degree. target Indicates the target carbon dioxide concentration at the current mushroom growth stage, C i represents the actual measured carbon dioxide concentration of the mushroom house during the i-th cycle, ensuring that when the gap between the actual carbon dioxide concentration in the mushroom house and the target concentration increases, the compensation strength of the effective mode for the missing mode is adaptively improved;

[0098] S23. Modal missing mask matrix Mask i and cross-modal complementarity coefficients Self-attention weight matrix working together in the deep separable sparse Transformer model Get the multimodal sparse self-attention weight matrix that integrates the multimodal image data of the mushroom house and the environment data of the mushroom house:

[0099]

[0100] in, represents the initial self-attention weight matrix without sparse adjustment;

[0101] S24. Use the multimodal sparse self-attention weight matrix to fuse the spatial and temporal features of the mushroom house multimodal image data and output the mushroom growth stage classification label P i and the mushroom growth stage confidence vector γ i , Mushroom Growth Stage Classification Label P i The value range is {P1, P2, P3, P4, P5}, which correspond to the hyphae laying period, primordium emergence period, fruiting body differentiation period, rapid expansion development period and pre-harvest period, respectively. The confidence vector γ of the mushroom growth stage i Used to indicate the credibility of the deep separable sparse Transformer model's classification results for the mushroom growth stage in the i-th cycle:

[0102]

[0103] in, Indicates that the classification label of the mushroom growth stage in the i-th cycle is P j confidence level.

[0104] In this embodiment, step S3 includes the following steps:

[0105] S31. Record the mushroom growth stage classification labels and mushroom growth stage confidence vectors output by the deep separable sparse Transformer model for each acquisition cycle, along with the preprocessed carbon dioxide concentration, preprocessed air temperature, preprocessed air humidity, and preprocessed light intensity of the mushroom room, in chronological order to form a stage perception result set S.

[0106] S32. For each collection cycle in the stage perception result set S, according to the mushroom growth stage P i The target environmental parameters are used to calculate the carbon dioxide concentration deviation of the mushroom house. Air temperature deviation E T,i , air humidity deviation E H,i and light intensity deviation E L,i ;

[0107] S33. Extracting mushroom growth stage confidence vector γ i In the current stage P i The corresponding confidence Multiply the confidence by the preset basic weight coefficient to get the carbon dioxide concentration weight of the mushroom house. Air temperature weight w T,i , air humidity weight w H,i , light intensity weight w L,i ;

[0108] S34. Set the carbon dioxide concentration deviation in the mushroom house Weighted with carbon dioxide concentration Multiply the air temperature deviation E T,i With air temperature weight w T,i Multiply the air humidity deviation E H,i With air humidity weight w H,i Multiply the light intensity deviation E L,i and the light intensity weight w L,i Multiply them, and then add the four product results to get the multi-objective fitness function F of the current acquisition cycle i , multi-objective fitness function F i The smaller the value, the closer the mushroom house environment is to the target state.

[0109] In this embodiment, the carbon dioxide concentration deviation of the mushroom house is the absolute difference between the pretreatment CO2 concentration and the corresponding target CO2 concentration;

[0110] Air temperature deviation E T,i is the absolute difference between the pre-treated air temperature and the corresponding target air temperature;

[0111] Air humidity deviation E H,iis the absolute difference between the pre-treated air humidity and the corresponding target air humidity;

[0112] Light intensity deviation E L,i is the absolute difference between the pre-processed light intensity and the corresponding target light intensity.

[0113] In this embodiment, step S4 includes the following steps:

[0114] S41. Extract the mushroom growth stage confidence of each collection cycle based on the stage perception result set Set mushroom growth stage confidence Mapping to dynamic exploration radius factor Used to control the global exploration range of the whale optimization algorithm. In the mushroom growth stage, the higher the confidence, the smaller the exploration radius, which tends to be local development.

[0115] S42. Initialize the whale optimization algorithm population. The position vector of each individual in the population is defined as the mushroom room gas regulation control parameter triplet in, is the carbon dioxide injection rate of the jth individual in the kth generation, T vent,k,j is the ventilation duration, V valve,k,j is the air valve opening ratio;

[0116] S43. In each iteration of the whale optimization algorithm, the dynamic exploration control factor δ i Adjust individual updates and define the capture control vector A k,i With the guidance vector C k,i for:

[0117] A k,i =δ i (2r k -1),C k,i =2r k ;

[0118] Among them, r k is a random vector;

[0119] When rounding up the control vector |A k,i When |<1, the individual moves towards the optimal individual of the current generation. Move closer:

[0120]

[0121] When rounding up the control vector |A k,i When |≥1, the spiral bubble network convergence strategy is executed and the growth stage guidance angle is introduced

[0122]

[0123] in, Indicates the mushroom growth stage P i The corresponding convergence direction guide angle, l is the spiral random variable, b0 is the initial spiral convergence coefficient, X k,j is the gas regulation control parameter vector of the jth individual in the kth generation, including the current carbon dioxide injection rate, ventilation duration and air valve opening ratio, which is used to describe the specific control variables of the mushroom house gas regulation scheme. k+1,j is the updated gas regulation control parameter vector of the jth individual in the k+1th generation, based on the kth generation population state and the capture control vector A k,i and the guide vector C k,i Evolution obtained;

[0124] S44. After completing the population position update of each generation, based on the multi-objective fitness function F i Evaluate the fitness of each individual and introduce the environmental stability coefficient σ i :

[0125]

[0126] Among them, α is the stability sensitivity factor, C i is the current actual carbon dioxide concentration, is the target carbon dioxide concentration for the stage;

[0127] The multi-objective fitness function is modified based on the environmental stability coefficient to obtain the environmental corrected fitness:

[0128] F′ i =F i ·σ i ;

[0129] And the environmentally corrected fitness F′ i The smallest individual updates the optimal individual

[0130] S45. Reaching the maximum iteration number G in the whale optimization algorithm max After that, the complete gas adjustment candidate sequence Q is obtained candidate ,The gas regulation candidate sequence consists of the optimal gas regulation control parameters of each ,acquisition period;

[0131] S46. Adjusting the candidate sequence Q for gas candidate , introduce the inertia correction smoothing mechanism based on the confidence-driven mushroom growth stage, and define the inertia smoothing factor ω for each cycle i , and apply the inertia correction formula to generate the final optimal gas regulation control strategy Q optimal :

[0132] Q optimal,i =ωi Q candidate,i +(1-ω i )Q optimal,i-1 ;

[0133] Among them, Q optimal,i-1 The gas regulation control strategy that has been smoothed in the previous cycle;

[0134] S47. Output the optimal gas regulation control strategy Q after inertia correction and smoothing. optimal , for the mushroom room execution equipment network to perform corresponding carbon dioxide injection, ventilation and air valve opening adjustment operations.

[0135] In this embodiment, the optimal gas regulation control strategy Q optimal,i According to the following standard output, the carbon dioxide injection rate, ventilation duration and air valve opening of the mushroom house are determined respectively:

[0136] When mushroom growth stage classification label P i It is the mycelium laying period P1 and the carbon dioxide concentration in the mushroom house is lower than the target concentration When the carbon dioxide injection rate Take the corresponding component in the gas regulation candidate sequence as an upper limit of 10%, and the ventilation duration T vent,i Keep the original value, the air valve opening V valve,i 5% reduction;

[0137] When mushroom growth stage classification label P i It is the primordium emergence stage P2 or the fruiting body differentiation stage P3, and the mushroom house is dioxygenated.

[0138] When the carbon concentration is close to the target concentration ±50ppm, the carbon dioxide injection rate Use the inertia correction smoothing value, ventilation duration T vent,i Moderately increase 10%, the air valve opening V valve,i Keep smooth value;

[0139] When mushroom growth stage classification label P i It is the rapid expansion and development stage P4 or the early harvest stage P5, and the carbon dioxide concentration in the mushroom house is higher than the target concentration When the carbon dioxide injection rate is above Directly decrease by 15%, ventilation duration T vent,i Increase by 20%, air valve opening V valve,i Synchronization increased by 10%.

[0140] Example 1:

[0141] Located in a large facility agriculture park, Company A has commissioned a newly built mushroom cultivation temperature control center with a total area of 2,800 square meters. The center is divided into six independent mushroom house units, each equipped with an environmental monitoring and gas regulation system to control key environmental parameters such as carbon dioxide concentration, temperature, humidity, and light intensity. The project considered adopting an intelligent and adaptive environmental regulation solution at the early design stage to overcome the previous problems of mushroom house environmental control, such as slow response, insensitivity to stage changes, and high energy consumption.

[0142] In the application scenario, the project team deployed the system of the present invention for the first time. Carbon dioxide concentration, temperature, humidity and light sensors were installed in the mushroom house, and visible light cameras and infrared thermal imagers were installed on the top and side walls of each unit respectively. An infrastructure covering the entire cycle and multimodal data collection capabilities was built. The system is set to synchronously collect environmental data and multimodal image data every 30 minutes, with uninterrupted collection throughout the year.

[0143] In actual operation, within each synchronization cycle, the system first completes data integrity detection, eliminates data loss and image distortion samples caused by sensor fluctuations or acquisition anomalies, and after timestamp alignment, uses sliding window filtering to remove high-frequency noise for environmental data, applies Gaussian filtering to visible light images, and applies median filtering to infrared thermal images, effectively improving data stability and consistency of image features.

[0144] Subsequently, the system dynamically generates a modality missing mask based on the collected preprocessed image data and inputs it into the deep separable sparse Transformer model. Through the cross-modal complementary sparse attention mechanism, when there is local blur in the infrared image or insufficient exposure of the visible light image, the model can automatically adjust the weight ratio and compensate for the missing modality information, thereby maintaining the accuracy of the mushroom growth stage classification. Taking an inspection in early December 2024 as an example, the infrared thermal imager in Unit 4 of the mushroom house had noise in part of the screen due to the low temperature at night, but the system still correctly identified that the period was in the fruiting body differentiation stage P3, and the classification confidence was as high as 0.91, which was nearly 13% higher than the confidence of traditional methods.

[0145] Based on the stage recognition results, the system combines the mushroom growth stage confidence vector with the environmental data to construct a multi-objective fitness function. During the whale optimization algorithm stage, the system dynamically adjusts the global search radius in real time based on the confidence. When the recognition confidence is high (such as a confidence of 0.9 or above), the system narrows the search range, performs local optimization, and quickly converges to the optimal gas regulation parameters; when the confidence is lower than 0.6, the search radius is automatically increased to enhance the exploration capability and avoid falling into the local optimum. During the mycelium laying period in mushroom house No. 5, the system detected a slight drift in the sensor, causing the stage recognition confidence to drop to 0.58. The optimization algorithm adaptively adjusted the parameter search radius and successfully avoided the problem of long-term deviation in carbon dioxide concentration.

[0146] The adjustment parameters are further processed through an inertia correction and smoothing mechanism, ensuring continuous and smooth adjustment within each cycle. Taking pre-harvest data as an example, the traditional method directly sets the fan start time and CO2 injection rate based on the optimal point, resulting in large fluctuations in CO2 concentration within a short period of time (maximum fluctuation of ±80ppm). However, after applying inertia correction, the proposed method controls the CO2 concentration fluctuation range to within ±35ppm, significantly improving environmental stability.

[0147] After a complete mushroom production cycle, the project team conducted a comparative evaluation of mushroom house units using the method of the present invention and the traditional rule-based control method. The results are summarized as follows:

[0148] Table 1 Comparative evaluation data of mushroom house units between the method of the present invention and the traditional rule control method

[0149]

[0150]

[0151] In terms of stage recognition accuracy, the traditional threshold + dense Transformer method has a significantly increased misjudgment rate in night data with a modal missing rate of more than 20%, while the method of the present invention still maintains an identification accuracy of more than 95% under the condition of a missing rate of 30%. In terms of energy consumption, the present invention optimizes the gas regulation strategy by dynamically adapting to stage changes and environmental changes, and achieves an overall energy saving effect of more than 11%. Most importantly, by stabilizing the control of the mushroom growth environment, the yield per unit area is ultimately increased by about 7%, which brings significant economic benefits to actual production.

[0152] In summary, this Example 1 fully verifies that the present invention can provide a practical and effective technical improvement solution to the problems of insufficient perception in the environmental adjustment stage, fragile multimodal fusion, discontinuous adjustment strategy, and high energy consumption in the existing technology, and has achieved excellent application effects in real scenarios, and has extremely high promotion and application value.

[0153] This paper proposes a deep separable sparse Transformer model, which combines a dynamic modality loss mask with a cross-modal complementary sparse attention mechanism to specifically optimize the modality loss and noise interference problems in the multimodal image data of mushroom houses. By dynamically generating a modality weight matrix and introducing a cross-modal complementary coefficient based on carbon dioxide concentration deviation, it adaptively compensates for the missing or weakened modal features, achieving high-reliability identification of mushroom growth stages even under conditions of incomplete or distorted data, significantly improving the accuracy of growth stage discrimination.

[0154] In terms of optimization control strategy, the present invention dynamically maps the mushroom growth stage confidence vector into the exploration control factor in the whale optimization algorithm, and proposes a dynamic exploration radius adjustment mechanism driven by stage confidence. It can dynamically adjust the global exploration and local development weights of the whale optimization population according to the judgment credibility of the current growth status of the mushroom, thereby effectively avoiding premature convergence and local optimal problems, and improving the environmental adaptability and energy consumption optimization effect of the gas regulation strategy.

[0155] The present invention introduces an inertia correction smoothing mechanism. Through the time series inertia compensation formula driven by the confidence of the growth stage, the gas regulation candidate sequence output by the whale optimization is dynamically smoothed to ensure the continuity and stability of the regulation strategy during periodic changes. It can automatically and smoothly transition the control parameters according to the stage judgment confidence, effectively suppressing the drastic jumps of the regulation strategy during execution.

[0156] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for regulating carbon dioxide in a mushroom house based on multimodal growth stage recognition, characterized in that: The steps include: S1. Synchronously collect and preprocess the mushroom house environment data and the mushroom house multimodal image data to obtain preprocessed mushroom house environment data and preprocessed mushroom house multimodal image data; S2. Generate a modality loss mask based on the preprocessed multimodal image data of the mushroom house. Input the preprocessed multimodal image data and the modality loss mask into a deep separable sparse Transformer model, and output a mushroom growth stage classification label and a mushroom growth stage confidence vector. S3. Combine the mushroom growth stage classification labels, mushroom growth stage confidence vectors, and preprocessed mushroom house environment data to form a stage perception result. Based on the stage perception result, construct a multi-objective fitness function and generate a stage weight group. S4. Perform global search and local convergence iterations using a multi-objective fitness function as the evaluation criterion. During the iteration of the whale optimization algorithm, the fitness evaluation is updated in real time, and a candidate gas regulation sequence is output. Time-series smoothing compensation is performed on the candidate gas regulation sequence to obtain the optimal gas regulation control strategy. S5. Send the optimal gas regulation control strategy to the mushroom house execution device network, and implement the mushroom house carbon dioxide concentration adjustment according to the optimal gas regulation control strategy; S6. Calculate the environmental stability index. When the environmental stability index exceeds the limit or the mushroom growth stage confidence vector is lower than the threshold, the whale optimization algorithm is triggered to re-execute steps S4 to S6, and steps S1 to S6 are cyclically executed during the entire mushroom house production cycle.

2. A mushroom house carbon dioxide regulation method based on multimodal growth stage recognition according to claim 1, characterized in that: The step S1 comprises the following steps: S11. Synchronously collect environmental data and multimodal image data of the mushroom house during each synchronization cycle of the mushroom house data acquisition device. The environmental data includes carbon dioxide concentration, air temperature, air humidity, and light intensity. The multimodal image data includes visible light images and infrared thermal images of the mushroom house. S12. During each synchronization cycle, record the carbon dioxide concentration, air temperature, air humidity, light intensity, visible light image, and infrared thermal image of the mushroom house corresponding to the i-th cycle, forming a complete original mushroom house environment dataset and an original mushroom house multimodal image dataset. S13. Perform integrity checks on each sampling point in the original Mushroom House environment dataset, removing any missing or obviously abnormal records. Perform validity checks on the original Mushroom House multimodal image dataset, removing samples with missing or severely distorted images, and retaining the qualified data. This yields the integrity-checked Mushroom House environment dataset and the integrity-checked Mushroom House multimodal image dataset. S14. Synchronize the integrity-checked Mushroom House environment dataset and the integrity-checked Mushroom House multimodal image dataset according to their timestamps, ensuring that the time difference between each piece of Mushroom House environment data and its corresponding Mushroom House multimodal image data is less than a set time difference threshold. This yields the timestamp-aligned Mushroom House environment dataset and the timestamp-aligned Mushroom House multimodal image dataset. S15. Apply sliding window filtering to the mushroom house environment dataset after timestamp alignment to obtain the preprocessed mushroom house environment dataset E. proc The effective period number of the preprocessed mushroom house environment data set is recorded as N s . S16. Apply Gaussian filtering to each visible light image of the mushroom house in the timestamp-aligned multimodal image dataset, and apply median filtering to each infrared thermal image of the mushroom house, to construct a preprocessed multimodal image dataset I of the mushroom house. proc , the effective number of cycles of the preprocessed mushroom house multimodal image dataset is also N s .

3. The method for regulating carbon dioxide in a mushroom house based on multimodal growth stage recognition according to claim 2, characterized in that: The step S2 comprises the following steps: S21. Based on the preprocessed mushroom house multimodal image dataset I proc The actual modal validity in each acquisition cycle is determined by setting the dynamic modal weight factor Construct the modality missing mask matrix Mask i , used to implement modal feature reconstruction and sparse attention dynamic adjustment in the deep separable sparse Transformer model: in, represents the weight ratio of the visible light image data of the mushroom house in the i-th cycle during attention fusion, represents the weight ratio of the infrared thermal image data of the mushroom house in the i-th cycle during attention fusion. When the modality is complete, both are 0.

5. When the visible light image data of the mushroom house is missing or severely distorted, the value is adjusted to 0.

1. When the infrared thermal image data of the mushroom house is missing or severely distorted, the value is adjusted to 0.

9. S22. Designing a modality-complementary sparse attention mechanism within the deep separable sparse Transformer model to optimize the Mushroom House multimodal image dataset I proc To solve the mid-modal missing problem, we introduce a cross-modal complementary coefficient when calculating the self-attention weight of each modality. Cross-modal complementarity coefficient Used to compensate for missing or weak modal features through modal features: in, represents the cross-modal complementary sensitivity factor, which is used to control the sensitivity of the difference between the carbon dioxide concentration in the mushroom house and the target concentration to the modal compensation degree. target Indicates the target carbon dioxide concentration at the current mushroom growth stage, C i represents the actual measured carbon dioxide concentration of the mushroom house during the i-th cycle, ensuring that when the gap between the actual carbon dioxide concentration in the mushroom house and the target concentration increases, the compensation strength of the effective mode for the missing mode is adaptively improved; S23. Modal missing mask matrix Mask i and cross-modal complementarity coefficients Self-attention weight matrix working together in the deep separable sparse Transformer model Get the multimodal sparse self-attention weight matrix that integrates the multimodal image data of the mushroom house and the environment data of the mushroom house S24. Use the multimodal sparse self-attention weight matrix to fuse the spatial and temporal features of the mushroom house multimodal image data and output the mushroom growth stage classification label P i and the mushroom growth stage confidence vector γ i , Mushroom Growth Stage Classification Label P i The value range is {P1, P2, P3, P4, P5}, which correspond to the mycelial laying period, primordium emergence period, fruiting body differentiation period, rapid expansion development period and pre-harvest period, respectively. The mushroom growth stage confidence vector is used to represent the credibility of the deep separable sparse Transformer model's discrimination results for the classification label of the mushroom growth stage in the i-th cycle.

4. The method for regulating carbon dioxide in a mushroom house based on multimodal growth stage recognition according to claim 3, characterized in that: The step S3 comprises the following steps: S31. Record the mushroom growth stage classification labels and mushroom growth stage confidence vectors output by the deep separable sparse Transformer model for each acquisition cycle, along with the preprocessed carbon dioxide concentration, preprocessed air temperature, preprocessed air humidity, and preprocessed light intensity of the mushroom room, in chronological order to form a stage perception result set S. S32. For each collection cycle in the stage perception result set S, according to the mushroom growth stage P i The target environmental parameters are used to calculate the carbon dioxide concentration deviation of the mushroom house. Air temperature deviation E T,i , air humidity deviation E H,i and light intensity deviation E L,i ; S33. Extracting mushroom growth stage confidence vector γ i In the current stage P i The corresponding confidence Multiply the confidence by the preset basic weight coefficient to get the carbon dioxide concentration weight of the mushroom house. Air temperature weight w T,i , air humidity weight w H,i , light intensity weight w L,i ,in, Indicates that the classification label of the mushroom growth stage in the i-th cycle is P j confidence level; S34. Set the carbon dioxide concentration deviation in the mushroom house Weighted with carbon dioxide concentration Multiply the air temperature deviation E T,i With air temperature weight w T,i Multiply the air humidity deviation E H,i With air humidity weight w H,i Multiply the light intensity deviation E L,i and the light intensity weight w L,i Multiply them, and then add the four product results to get the multi-objective fitness function F of the current acquisition cycle i , multi-objective fitness function F i The smaller the value, the closer the mushroom house environment is to the target state.

5. The method for regulating carbon dioxide in a mushroom house based on multimodal growth stage recognition according to claim 4, characterized in that: The carbon dioxide concentration deviation in the mushroom house is the absolute difference between the pretreatment CO2 concentration and the corresponding target CO2 concentration; The air temperature deviation E T,i is the absolute difference between the pre-treated air temperature and the corresponding target air temperature; The air humidity deviation E H,i is the absolute difference between the pre-treated air humidity and the corresponding target air humidity; The light intensity deviation E L,i is the absolute difference between the pre-processed light intensity and the corresponding target light intensity.

6. The method for regulating carbon dioxide in a mushroom house based on multimodal growth stage recognition according to claim 4, characterized in that: The step S4 comprises the following steps: S41. Extract the mushroom growth stage confidence of each collection cycle based on the stage perception result set Set mushroom growth stage confidence Mapping to dynamic exploration radius factor Used to control the global exploration range of the whale optimization algorithm; S42. Initialize the whale optimization algorithm population. The position vector of each individual in the population is defined as the mushroom room gas regulation control parameter triplet in, is the carbon dioxide injection rate of the jth individual in the kth generation, T vent,k,j is the ventilation duration, V valve,k,j is the air valve opening ratio; S43. In each iteration of the whale optimization algorithm, the dynamic exploration control factor δ i Adjust individual updates and define the capture control vector A k,i With the guidance vector C k,i ; When rounding up the control vector |A k,i When |<1, the individual moves towards the optimal individual of the current generation. Move closer: When rounding up the control vector |A k,i When |≥1, the spiral bubble network convergence strategy is executed and the growth stage guidance angle is introduced in, Indicates the mushroom growth stage P i The corresponding convergence direction guide angle, l is the spiral random variable, b0 is the initial spiral convergence coefficient, X k,j is the gas regulation control parameter vector of the jth individual in the kth generation, X k+1,j is the updated gas regulation control parameter vector of the jth individual in the k+1th generation; S44. After completing the population position update of each generation, based on the multi-objective fitness function F i Evaluate the fitness of each individual and introduce the environmental stability coefficient σ i : Among them, α is the stability sensitivity factor, C i is the current actual carbon dioxide concentration, is the target carbon dioxide concentration for the stage; The multi-objective fitness function is modified based on the environmental stability coefficient to obtain the environmental modified fitness F′ i and use the environmentally corrected fitness F′ i The smallest individual updates the optimal individual S45. Reaching the maximum iteration number G in the whale optimization algorithm max After that, the complete gas adjustment candidate sequence Q is obtained candidate ,The gas regulation candidate sequence consists of the optimal gas regulation control parameters of each ,acquisition period; S46. Adjusting the candidate sequence Q for gas candidate , introduce the inertia correction smoothing mechanism based on the confidence-driven mushroom growth stage, and define the inertia smoothing factor ω for each cycle i , and apply the inertia correction formula to generate the final optimal gas regulation control strategy Q optimal ; S47. Output the optimal gas regulation control strategy Q after inertia correction and smoothing. optimal , for the mushroom room execution equipment network to perform corresponding carbon dioxide injection, ventilation and air valve opening adjustment operations.

7. The method for regulating carbon dioxide in a mushroom house based on multimodal growth stage recognition according to claim 6, characterized in that: The optimal gas regulation control strategy Q optimal,i According to the following standard output, the carbon dioxide injection rate, ventilation duration and air valve opening of the mushroom house are determined respectively: When mushroom growth stage classification label P i It is the mycelium laying period P1 and the carbon dioxide concentration in the mushroom house is lower than the target concentration When the carbon dioxide injection rate Take the corresponding component in the gas regulation candidate sequence as an upper limit of 10%, and the ventilation duration T vent,i Keep the original value, the air valve opening V valve,i 5% reduction; When mushroom growth stage classification label P i It is the primordium emergence stage P2 or the fruiting body differentiation stage P3, and the mushroom house is dioxygenated. When the carbon concentration is close to the target concentration ±50ppm, the carbon dioxide injection rate Use the inertia correction smoothing value, ventilation duration T vent,i Moderately increase 10%, the air valve opening V valve,i Keep smooth value; When mushroom growth stage classification label P i It is the rapid expansion and development stage P4 or the early harvest stage P5, and the carbon dioxide concentration in the mushroom house is higher than the target concentration When the carbon dioxide injection rate is above 50ppm, Directly decrease by 15%, ventilation duration T vent,i Increase by 20%, air valve opening V valve,i Synchronization increased by 10%.

Images