Environment-driven methods and systems for predicting the growth of edible fungi

Abstract

This invention provides an environment-driven method and system for predicting the growth of edible fungi, belonging to the field of smart agriculture technology. The method includes: constructing environmental feature vectors corresponding to each time point based on environmental data within the prediction period; inputting the environmental feature vectors into a dynamic parameter mapping model trained based on actual growth parameter labels to obtain dynamic growth parameters corresponding to each time point; calling a dynamic biological growth equation that reconstructs static constant terms into dynamic variable parameters; assigning the dynamic growth parameters to the dynamic biological growth equation to calculate the predicted biomass growth value at each time point. This application utilizes machine learning to construct a mapping relationship between the environment and mechanistic parameters, and uses dynamic parameters to drive the mechanistic equation, thereby integrating the flexibility of data-driven approaches with the interpretability of mechanistic models, achieving high-precision prediction of edible fungi biomass under complex dynamic environments.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, and in particular to an environment-driven method and system for predicting the growth of edible fungi. Background Technology

[0002] In the industrialized production of edible fungi, accurate prediction of the growth process is the foundation for achieving intelligent environmental control and optimized harvesting plans.

[0003] Existing technologies typically employ prediction methods based on classical mechanistic models, such as classical growth mechanism equations. The working principle of this method is as follows: historical growth data of edible fungi is collected, and parameters such as the maximum growth rate and carrying capacity in the equation are determined through curve fitting. Subsequently, a mathematical model is constructed using these fixed parameters to describe and predict the growth trajectory of edible fungi.

[0004] However, the aforementioned existing technologies suffer from low prediction accuracy. Because the growth parameters in existing models are static constants obtained by fitting historical data, they assume the growth process occurs under relatively constant environmental conditions. When the actual production environment changes dynamically, these fixed parameters cannot reflect the real-time impact of environmental factors on the growth rate and upper growth limit, leading to a significant deviation between the model's predicted edible fungus biomass and the actual growth situation. Summary of the Invention

[0005] This invention provides an environment-driven method and system for predicting the growth of edible fungi, which addresses the shortcomings of existing technologies and enables high-precision prediction of edible fungi biomass under complex dynamic environments.

[0006] This invention provides an environment-driven method for predicting the growth of edible fungi, comprising the following steps:

[0007] Based on the environmental data within the time period to be predicted, construct the environmental feature vector corresponding to each time point;

[0008] The environmental feature vector is input into the dynamic parameter mapping model to obtain the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model. The dynamic parameter mapping model is trained based on the environmental feature vector samples and the corresponding actual growth parameter labels.

[0009] The preset dynamic biological growth equation is invoked. The dynamic biological growth equation is obtained by reconstructing the static constant terms in the classical growth mechanism equation into dynamic variable parameters that change with time and environment.

[0010] The dynamic growth parameters are assigned to the dynamic variable parameters in the dynamic biological growth equation to calculate the predicted biomass growth values ​​of edible fungi at each time point.

[0011] According to the present invention, an environment-driven method for predicting the growth of edible fungi is provided, wherein the dynamic growth parameters include growth rate parameters and growth potential parameters, and the dynamic parameter mapping model includes a first regressor sub-model and a second regressor sub-model.

[0012] The step of inputting the environmental feature vector into the dynamic parameter mapping model and obtaining the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model includes:

[0013] The environmental feature vector is input into the first regression sub-model to obtain the growth rate parameters corresponding to each time step output by the first regression sub-model.

[0014] The environmental feature vector is input into the second regression sub-model to obtain the growth potential coefficients corresponding to each time step output by the second regression sub-model.

[0015] The growth potential parameter at the corresponding time point is calculated by multiplying the preset theoretical maximum biomass constant with the growth potential coefficient.

[0016] According to the environmental-driven edible fungi growth prediction method provided by the present invention, the actual growth parameter labels include growth rate parameter labels and growth potential coefficient labels;

[0017] The first regression sub-model and the second regression sub-model were trained through the following steps:

[0018] Acquire historical observation data on edible fungi biomass and corresponding historical environmental data during the historical growth cycle;

[0019] Construct environmental feature vector samples based on the historical environmental data;

[0020] For each sampling time in the historical observation data, the relative growth rate of biomass at the current sampling time is used as the label of the growth rate parameter.

[0021] The ratio of the historical cumulative biomass at the current sampling time to the theoretical maximum biomass constant is used as the label for the growth potential coefficient.

[0022] Using the environmental feature vector samples as input, and the growth rate parameter label and the growth potential coefficient label as output targets, the first regression sub-model and the second regression sub-model are trained using a machine learning regression algorithm.

[0023] According to the present invention, an environment-driven method for predicting the growth of edible fungi includes the following steps: using the environmental feature vector sample as input and the growth rate parameter label and the growth potential coefficient label as output targets, training a first regression sub-model and a second regression sub-model using a machine learning regression algorithm, respectively.

[0024] The environmental feature vector samples are input into the first regression sub-model to obtain the predicted growth rate parameters, and the environmental feature vector samples are input into the second regression sub-model to obtain the predicted growth potential coefficients.

[0025] The estimated biomass is calculated by calling the dynamic biological growth equation and based on the predicted growth rate parameter and the predicted growth potential coefficient.

[0026] A hybrid loss function is constructed, comprising a first parameter error term, a second parameter error term, and a third parameter error term. The first parameter error term characterizes the difference between the predicted growth rate parameter and the growth rate parameter label. The second parameter error term characterizes the difference between the predicted growth potential coefficient and the growth potential coefficient label. The third parameter error term characterizes the difference between the estimated biomass and the actual biomass in the historical observation data.

[0027] The mixed loss function is minimized using the gradient descent algorithm, and the model parameters of the first regression sub-model and the second regression sub-model are synchronously iteratively updated until the model converges.

[0028] According to the present invention, an environment-driven method for predicting the growth of edible fungi includes constructing environmental feature vectors corresponding to each time period based on environmental data within the time period to be predicted, comprising:

[0029] For any moment within the time period to be predicted, determine the current time window corresponding to that moment;

[0030] Obtain environmental data for the current time window and N preceding time windows; the environmental data includes temperature, humidity, carbon dioxide concentration, and light intensity.

[0031] Calculate the statistical characteristic values ​​of the environmental data; the statistical characteristic values ​​include at least one of the mean, standard deviation, maximum value, minimum value, or cumulative value.

[0032] The statistical feature values ​​are combined to obtain the environmental feature vector.

[0033] According to the environmentally driven method for predicting the growth of edible fungi provided by the present invention, the dynamic biological growth equation is expressed by the following mathematical formula:

[0034] ;

[0035] ;

[0036] ;

[0037] in, Let be the biomass at time t; The parameter representing the growth potential at time t; The growth rate parameter at time t; The time parameter for reaching the maximum growth rate; Let be the normalized temperature value at time t. Let be the normalized humidity value at time t. Let be the normalized carbon dioxide concentration at time t. Let be the normalized illumination intensity value at time t; This represents the interaction term between temperature and humidity at time t; This represents the interaction term between temperature and carbon dioxide at time t; This represents the interaction term between temperature and illumination at time t; This represents the interaction term between humidity and carbon dioxide at time t; This represents the interaction term between humidity and light intensity at time t; This represents the interaction term between light intensity and carbon dioxide at time t; The intercept term of the first regression sub-model is... The coefficients to be fitted for the first regression sub-model; This is the intercept term of the second regression sub-model. represents the coefficients to be fitted in the second regression sub-model; v is the shape parameter.

[0038] The environmentally driven method for predicting the growth of edible fungi provided by the present invention further includes:

[0039] Acquire environmental data sequences corresponding to various preset environmental control strategies;

[0040] For each of the aforementioned environmental regulation strategies, the environmental data sequence is processed using the dynamic parameter mapping model and the dynamic biological growth equation to obtain the corresponding biomass growth prediction trajectory.

[0041] Based on the predicted biomass growth trajectory, a target regulatory strategy is selected from a variety of environmental regulation strategies.

[0042] This invention also provides an environment-driven edible fungus growth prediction system, comprising the following modules:

[0043] The feature construction module is used to construct environmental feature vectors for each time period based on the environmental data within the time period to be predicted.

[0044] The parameter acquisition module is used to input the environmental feature vector into the dynamic parameter mapping model and obtain the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model. The dynamic parameter mapping model is trained based on the environmental feature vector samples and the corresponding actual growth parameter labels.

[0045] The prediction calculation module is used to call the preset dynamic biological growth equation, which is obtained by reconstructing the static constant terms in the classical growth mechanism equation into dynamic variable parameters that change with time and environment.

[0046] The prediction calculation module is also used to assign the dynamic growth parameters to the dynamic variable parameters in the dynamic biological growth equation, and calculate the predicted value of biomass growth of edible fungi at each time point.

[0047] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the environment-driven edible fungus growth prediction method as described above.

[0048] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the environment-driven edible fungus growth prediction method as described above.

[0049] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the environment-driven edible fungus growth prediction method as described above.

[0050] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0051] By constructing environmental feature vectors corresponding to each moment based on environmental data within the predicted time period, and inputting these environmental feature vectors into a dynamic parameter mapping model trained based on environmental feature vector samples and corresponding actual growth parameter labels, the precise quantification of the nonlinear mapping relationship between environmental fluctuations and biological internal growth parameters is achieved. By obtaining the dynamic growth parameters corresponding to each moment output by the dynamic parameter mapping model, and calling the dynamic biological growth equation obtained by reconstructing the static constant term in the classical growth mechanism equation into dynamic variable parameters that change with time and environment, the limitation of traditional mechanism models that can only describe fixed growth trajectories under ideal steady-state conditions is broken, giving the growth equation a structural basis for responding to external environmental fluctuations. By assigning the dynamic growth parameters to the dynamic variable parameters in the dynamic biological growth equation and calculating the predicted biomass growth values ​​of edible fungi at each moment, the accuracy and robustness of predicting edible fungi biomass growth under non-steady-state environmental conditions are significantly improved. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0053] Figure 1 This is one of the flowcharts of the environment-driven edible fungus growth prediction method provided by the present invention.

[0054] Figure 2 This is the second flowchart of the environment-driven edible fungus growth prediction method provided by the present invention.

[0055] Figure 3 This is the third flowchart of the environment-driven edible fungus growth prediction method provided by the present invention.

[0056] Figure 4 This is the fourth flowchart of the environment-driven edible fungus growth prediction method provided by the present invention.

[0057] Figure 5 This is the fifth flowchart of the environment-driven edible fungus growth prediction method provided by the present invention.

[0058] Figure 6 This is a schematic diagram comparing the growth prediction trajectories of the various models provided by this invention.

[0059] Figure 7 This is a schematic diagram of the structure of the environment-driven edible fungus growth prediction system provided by the present invention.

[0060] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0062] It should be noted that in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The terms "upper," "lower," etc., indicating orientation or positional relationships according to the accompanying drawings, are only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0063] The terms "first," "second," etc., used in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0064] The following is combined Figures 1 to 8 This invention describes the environment-driven edible fungi growth prediction method, system, electronic device, storage medium, and computer program product provided by the present invention.

[0065] This application provides an environment-driven method for predicting the growth of edible fungi. The execution subject of this method is an electronic device capable of data processing and model calculation, such as a smart agricultural control terminal, an agricultural IoT server, or a greenhouse control computer with edge computing capabilities. In this embodiment, a smart agricultural control terminal is used as the execution subject. This embodiment uses the factory cultivation of *Pleurotus ostreatus* as a specific application scenario, but the technical solution of this application is also applicable to the growth prediction of other edible fungi varieties such as *Lentinula edodes* and *Pleurotus ostreatus*.

[0066] Reference Figure 1 , Figure 1 This is one of the flowcharts illustrating the environment-driven edible fungus growth prediction method provided by this invention. For example... Figure 1 As shown, the environment-driven method for predicting the growth of edible fungi specifically includes steps 101 to 104:

[0067] Step 101: Based on the environmental data within the time period to be predicted, construct the environmental feature vector corresponding to each time point;

[0068] Step 102: Input the environmental feature vector into the dynamic parameter mapping model and obtain the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model. The dynamic parameter mapping model is trained based on the environmental feature vector samples and the corresponding actual growth parameter labels.

[0069] Step 103: Call the preset dynamic biological growth equation. The dynamic biological growth equation is obtained by reconstructing the static constant terms in the classical growth mechanism equation into dynamic variable parameters that change with time and environment.

[0070] Step 104: Assign the dynamic growth parameters to the dynamic variable parameters in the dynamic biological growth equation, and calculate the predicted biomass growth value of edible fungi at each time point.

[0071] Specifically, the intelligent agricultural control terminal first performs the step of constructing environmental feature vectors corresponding to each time period based on the environmental data within the period to be predicted.

[0072] The intelligent agricultural control terminal acquires environmental data for the predicted time period through a sensor network connected to the edible fungus growth environment. This environmental data includes, but is not limited to, temperature, humidity, carbon dioxide concentration, and light intensity.

[0073] The intelligent agricultural control terminal arranges the aforementioned environmental data according to a time series, and for each moment within the period to be predicted, extracts environmental data from that moment and the period preceding it. Through numerical normalization and statistical feature extraction, it constructs an environmental feature vector that characterizes the current environmental state and its historical cumulative effects. The environmental feature vector is a multi-dimensional numerical matrix that includes not only the current instantaneous environmental values ​​but also statistics reflecting environmental fluctuation trends.

[0074] After constructing the environmental feature vector, the intelligent agricultural control terminal performs the steps of inputting the environmental feature vector into the dynamic parameter mapping model and obtaining the dynamic growth parameters corresponding to each time point output by the dynamic parameter mapping model.

[0075] The dynamic parameter mapping model is a machine learning model pre-stored in the smart agriculture control terminal. It is trained based on environmental feature vector samples and their corresponding actual growth parameter labels.

[0076] The intelligent agricultural control terminal inputs the current environmental feature vector into a dynamic parameter mapping model. This model, through internally learned nonlinear mapping relationships, transforms the input environmental feature vector into a set of numerical values ​​describing the growth characteristics of organisms—the dynamic growth parameters. These dynamic growth parameters are not fixed constants but rather variables that fluctuate in real time with changes in the input environmental feature vector. Dynamic growth parameters include, but are not limited to, growth rate parameters characterizing the instantaneous growth rate of edible fungi under the current environment, and growth potential parameters characterizing the theoretical upper limit of growth that edible fungi can achieve under the current environment.

[0077] Subsequently, the intelligent agricultural control terminal executes the step of calling the preset dynamic biological growth equation.

[0078] The dynamic biological growth equation is obtained by reconstructing the static constant terms in the classical growth mechanism equation into dynamic variable parameters that change with time and environment. The intelligent agricultural control terminal pre-stores the framework of the growth equation based on biological mechanisms.

[0079] In this embodiment, through experimental comparative analysis, the Richards model was selected as the basis for the classical growth mechanism equation. The classical Richards model includes the maximum growth rate constant *r* and the carrying capacity constant *K*. The dynamic biological growth equation invoked by the intelligent agricultural control terminal is a deconstructed form of the above classical equation, that is, the static constant *r* in the original equation is reconstructed into a growth rate parameter that varies with time and environment, and the static constant *K* is reconstructed into a growth potential parameter that varies with time and environment. Both the growth potential parameter and the growth rate parameter are related to environmental data such as temperature, humidity, carbon dioxide concentration, and light intensity. The mathematical expression of the reconstructed dynamic biological growth equation is shown below:

[0080] The dynamic biological growth equation is expressed by the following mathematical formula:

[0081] ;

[0082] ;

[0083] ;

[0084] in, Let be the biomass at time t; The parameter representing the growth potential at time t; The growth rate parameter at time t; The time parameter for reaching the maximum growth rate; Let be the normalized temperature value at time t. Let be the normalized humidity value at time t. Let be the normalized carbon dioxide concentration at time t. Let be the normalized illumination intensity value at time t; This represents the interaction term between temperature and humidity at time t, characterizing the combined effect of the temperature and humidity coupling on growth; This represents the interaction term between temperature and carbon dioxide at time t, characterizing the combined effect of the temperature-carbon dioxide coupling on growth. This represents the interaction term between temperature and light at time t, characterizing the combined effect of the temperature-light coupling effect on growth; This represents the interaction term between humidity and carbon dioxide at time t, characterizing the combined effect of the coupling effect of humidity and carbon dioxide on growth. This represents the interaction term between humidity and light at time t, characterizing the combined effect of the coupling effect of humidity and light on growth; This represents the interaction term between light and carbon dioxide at time t, characterizing the combined effect of the coupling effect of light and carbon dioxide on growth. The intercept term of the first regression sub-model represents the base growth rate when all environmental factors are normalized to 0 (i.e., the baseline environmental state). where are the coefficients to be fitted in the first regression sub-model; The main effect coefficients represent the effects of changes in four independent factors—temperature, humidity, carbon dioxide, and light—on the growth rate. The direct impact on weight; The interaction coefficient represents the effect of the corresponding pairwise environmental factor interaction terms on the growth rate. The combined influence weight. This is the intercept term of the second regression sub-model. where are the coefficients to be fitted in the second regression sub-model; where The main effect coefficients represent the effects of the four single environmental factors on growth potential. The direct impact on weight; The interaction coefficient represents the upper limit of growth potential for each pair of environmental factor interaction terms. The weights are determined jointly. v is a shape parameter used to adjust the position of the inflection point of the growth curve, so that it does not have to be strictly symmetrical at half of the theoretical maximum biomass, thus enabling more flexible fitting of the asymmetric true growth trajectory of different edible fungi varieties.

[0085] It should be noted that the normalized temperature value mentioned in the above formula is obtained by normalizing the temperature data at each time point; similarly, the normalized humidity value, normalized carbon dioxide concentration value, and normalized light intensity value are obtained by normalizing the humidity data, carbon dioxide concentration data, and light intensity data at each time point, respectively. The normalization method used in this embodiment can be the max-min normalization method, the Z-score normalization method, etc., and the specific normalization method is not specifically limited in this embodiment.

[0086] When constructing the first and second regression sub-models, the intelligent agricultural control terminal performs an interaction term screening step. Based on the statistical significance test results of historical environmental data and biomass growth data, the intelligent agricultural control terminal progressively introduces environmental factor interaction terms. Alternatively, it calculates the correlation coefficients between various combinations of environmental factors and historical growth parameters through Pearson correlation analysis, and extracts target environmental factor combinations with correlation coefficients greater than a preset correlation threshold as key interaction terms. The intelligent agricultural control terminal retains the screened key interaction terms in the first and second regression sub-models for coefficient fitting calculations.

[0087] Finally, the intelligent agricultural control terminal performs the step of assigning the dynamic growth parameters to the dynamic variable parameters in the dynamic biological growth equation, and calculating the predicted value of the biomass growth of edible fungi at each time point.

[0088] The intelligent agricultural control terminal will output the current dynamic growth parameters, i.e., growth potential parameters, from the dynamic parameter mapping model. and growth rate parameters Substitute these values ​​into the dynamic biological growth equation above, replacing the corresponding dynamic variable parameters. The intelligent agricultural control terminal then uses numerical calculations to obtain the predicted biomass growth value of the edible fungi at the current moment. The predicted biomass growth value is specifically expressed as the numerical values ​​of morphological indicators such as cap diameter, stipe length, stipe width, or fresh weight.

[0089] The intelligent agricultural control terminal repeats the above calculation process sequentially for each moment within the time period to be predicted, thereby generating a continuous biomass growth prediction curve that can respond to dynamic changes in the environment.

[0090] The environment-driven edible fungi growth prediction method provided in this embodiment combines the powerful nonlinear fitting capabilities of machine learning models with classical mechanistic models to simulate the biological process of "environment influencing growth parameters, and growth parameters determining growth outcomes." This method overcomes the shortcomings of traditional mechanistic models, where fixed parameters prevent adaptation to dynamic environmental fluctuations. It also avoids the problems of uninterpretable predictions and poor extrapolation capabilities caused by the lack of biological constraints in purely data-driven models. This method can dynamically adjust key parameters within the growth equation according to real-time environmental changes, thus providing high-precision growth prediction results even under drastic fluctuations or non-steady-state conditions, offering reliable data support for the refined environmental control of edible fungi.

[0091] This embodiment provides a further detailed explanation of the steps for constructing environmental feature vectors in the above embodiments. (Refer to...) Figure 2 , Figure 2 This is the second flowchart illustrating the environment-driven edible fungus growth prediction method provided by this invention. Figure 2 As shown, the intelligent agricultural control terminal performs the following steps to construct environmental feature vectors for each time period based on environmental data within the predicted time period:

[0092] Step 201: For any moment within the period to be predicted, determine the current time window corresponding to that moment;

[0093] Step 202: Obtain environmental data for the current time window and N preceding time windows; environmental data includes temperature, humidity, carbon dioxide concentration, and light intensity;

[0094] Step 203: Calculate the statistical characteristic values ​​of the environmental data; the statistical characteristic values ​​include at least one of the following: mean, standard deviation, maximum value, minimum value, or cumulative value;

[0095] Step 204: Combine the statistical feature values ​​to obtain the environmental feature vector.

[0096] Specifically, firstly, for any moment within the period to be predicted, the intelligent agricultural control terminal performs the step of determining the current time window corresponding to that moment.

[0097] The intelligent agricultural control terminal employs a time sliding window mechanism to process continuous time-series data. It divides the continuous time axis into a series of discrete intervals of fixed length, each interval constituting a time window. In this embodiment, the intelligent agricultural control terminal sets the length of the time window to 24 hours, i.e., dividing it in "days". For any specific time t within the period to be predicted, the intelligent agricultural control terminal identifies which "day" interval t falls into and marks that interval as the current time window.

[0098] Next, the intelligent agricultural control terminal performs the step of acquiring environmental data within the current time window and N preceding time windows.

[0099] Environmental data includes temperature (T), humidity (H), carbon dioxide concentration (C), and light intensity (L). The intelligent agricultural control terminal not only reads environmental monitoring data within the current time window (day t), but also retrospectively reads environmental monitoring data from N consecutive time windows preceding the current time window (i.e., day t-1, day t-2, ..., day tN). Here, N is a preset positive integer used to define the retrospective range of historical environmental effects. In this embodiment, the preferred value of N is 2, meaning that the intelligent agricultural control terminal will simultaneously consider the impact of the environmental conditions of the current day, yesterday, and the day before yesterday on the current growth status.

[0100] Subsequently, the intelligent agricultural control terminal performs the step of calculating the statistical characteristic values ​​of the computing environment data.

[0101] Statistical characteristics include at least one of the following: mean, standard deviation, maximum value, minimum value, or cumulative value. For each acquired time window (including the current time window and previous time windows), the intelligent agricultural control terminal calculates the statistical indicators of each environmental factor within that window.

[0102] For example, for temperature data T, the smart agriculture control terminal calculates the average temperature in the current time window (T(t)), the average temperature in the first preceding time window (T(t-1)), and the average temperature in the second preceding time window (T(t-2)).

[0103] For humidity data H, the smart agriculture control terminal calculates the average humidity (H(t)) in the current time window, the average humidity (H(t-1)) in the first preceding time window, and the average humidity (H(t-2)) in the second preceding time window.

[0104] For carbon dioxide concentration data C, the smart agriculture control terminal calculates the average carbon dioxide concentration (C(t)) in the current time window, the average carbon dioxide concentration (C(t-1)) in the first preceding time window, and the average carbon dioxide concentration (C(t-2)) in the second preceding time window.

[0105] For light intensity data L, the smart agriculture control terminal calculates the average light intensity (L(t)) in the current time window, the average light intensity (L(t-1)) in the first preceding time window, and the average light intensity (L(t-2)) in the second preceding time window.

[0106] Meanwhile, the intelligent agricultural control terminal also calculates the standard deviation characteristics that reflect the severity of environmental fluctuations, as well as the cumulative characteristics that reflect the potential for photosynthesis, such as cumulative light intensity or effective accumulated temperature.

[0107] Finally, the intelligent agricultural control terminal performs the step of combining statistical feature values ​​to obtain environmental feature vectors.

[0108] The intelligent agricultural control terminal concatenates all the statistical feature values ​​calculated above into a high-dimensional numerical vector, namely the environmental feature vector, according to a predetermined order. Taking an environmental feature vector that includes average temperature, average humidity, and average light intensity as an example, It can be represented in the following form:

[0109] .

[0110] This implementation constructs a high-dimensional environmental feature vector that incorporates the current state and historical cumulative effects by introducing a time window segmentation and multi-timescale feature extraction mechanism. This technical solution fully considers the lag and cumulative effects of biological growth; that is, the current growth state depends not only on the current environment but also on the significant influence of the environment experienced over a past period. By fusing multiple statistical features such as mean, standard deviation, and cumulative amount, this method can comprehensively capture the overall level of the environment, the intensity of fluctuations, and the cumulative energy input, thereby significantly improving the analytical accuracy of subsequent dynamic parameter mapping models in interpreting the nonlinear relationship between complex environments and growth parameters.

[0111] Reference Figure 3 , Figure 3 This is the third flowchart illustrating the environment-driven edible fungus growth prediction method provided by this invention. Figure 3As shown, this embodiment further details the composition of the dynamic growth parameters and the specific structure and calculation process of the dynamic parameter mapping model in the above embodiments, specifically including the following steps:

[0112] Step 301: Input the environmental feature vector into the first regression sub-model to obtain the growth rate parameters corresponding to each time step output by the first regression sub-model;

[0113] Step 302: Input the environmental feature vector into the second regression sub-model to obtain the growth potential coefficients corresponding to each time step output by the second regression sub-model;

[0114] Step 303: Calculate the growth potential parameters at the corresponding time point based on the product of the preset theoretical maximum biomass constant and the growth potential coefficient.

[0115] In this embodiment, the dynamic growth parameters specifically include the growth rate parameter. and growth potential parameters To achieve accurate prediction of these two parameters with different biological significance, the dynamic parameter mapping model used in the intelligent agricultural control terminal employs a dual-model architecture, specifically comprising two independently trained machine learning models: a first regression sub-model and a second regression sub-model. Both sub-models can be constructed using algorithms capable of handling high-dimensional nonlinear features, such as Gradient Boosting Tree (XGBoost), Random Forest, or Deep Neural Network (DNN). In this embodiment, both the first and second regression sub-models are constructed using the XGBoost algorithm.

[0116] The intelligent agricultural control terminal executes the steps of inputting environmental feature vectors into the dynamic parameter mapping model and obtaining the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model. Specifically, this is achieved through the following two parallel computation paths.

[0117] The first path is used to predict the growth rate. The intelligent agricultural control terminal executes the steps of inputting the environmental feature vector into the first regression sub-model to obtain the growth rate parameters corresponding to each time step output by the first regression sub-model.

[0118] The intelligent agricultural control terminal will use the environmental feature vector constructed in the aforementioned steps. The input data is fed into the first, already trained regression sub-model (denoted as model 1). In the first regression sub-model, based on the mapping relationship between the environment and growth rate learned internally, a scalar value is output, which is the growth rate parameter at the current moment. Its mathematical expression is as follows: .

[0119] The second path is used to predict growth potential. The intelligent agricultural control terminal first executes the step of inputting the environmental feature vector into the second regression sub-model to obtain the growth potential coefficients corresponding to each time step output by the second regression sub-model.

[0120] Intelligent agricultural control terminals also incorporate environmental feature vectors. Input into the already trained second regression sub-model (denoted as model). It is important to note that the direct output of the second regression sub-model is not an absolute biomass value, but a normalized coefficient, namely the growth potential coefficient. This coefficient typically ranges from 0 to 1, representing the percentage of edible fungi that can reach their theoretical maximum growth limit under current environmental stress. Its mathematical expression is as follows: .

[0121] After obtaining the growth potential coefficient, the intelligent agricultural control terminal then performs the step of calculating the growth potential parameters at the corresponding time by multiplying the preset theoretical maximum biomass constant by the growth potential coefficient.

[0122] The intelligent agricultural control terminal stores a preset theoretical maximum biomass constant. The constant This is a theoretical upper limit value derived from historical growth data of the edible fungus variety under ideal conditions (for example, the theoretical maximum cap diameter of a certain variety of *Pleurotus ostreatus* is 180mm). The intelligent agricultural control terminal outputs the growth potential coefficient from the second regression sub-model. With the theoretical maximum biomass constant Multiply them to calculate the growth potential parameter at the current moment. The calculation formula is as follows: .

[0123] This implementation introduces a dual-regressor sub-model architecture to decouple the predictions of the two core biological characteristics, growth rate and growth potential. This design avoids the potential interference that can occur when a single model simultaneously fits two very different parameters. In particular, by predicting the normalized "growth potential coefficient" rather than directly predicting the absolute "carrying capacity," the dimensional influence of individual differences between different batches of mushroom spawn is effectively eliminated. This allows the model to focus more on the relative inhibitory or promoting effects of the environment on growth potential, significantly improving the model's generalization ability and predictive stability when faced with unseen environmental conditions.

[0124] This embodiment provides a more detailed explanation of the training process of the first and second regression sub-models in the above embodiments, particularly the method for constructing the actual growth parameter labels. To enable the machine learning model to learn the correct patterns, a high-quality training dataset containing the correspondence between "environmental features" and "parameter labels" needs to be constructed. In this embodiment, the actual growth parameter labels are specifically divided into two categories: growth rate parameter labels used to train the first regression sub-model, and growth potential coefficient labels used to train the second regression sub-model.

[0125] Reference Figure 4 , Figure 4 This is the fourth flowchart illustrating the environment-driven edible fungus growth prediction method provided by this invention. Figure 4 As shown, the intelligent agricultural control terminal or the model training server connected to it performs the following model training steps:

[0126] Step 401: Obtain historical observation data of edible fungi biomass and corresponding historical environmental data within the historical growth cycle;

[0127] Step 402: Construct environmental feature vector samples based on historical environmental data;

[0128] Step 403: For each sampling time in the historical observation data, use the relative growth rate of biomass at the current sampling time as the growth rate parameter label;

[0129] Step 404: Use the ratio of historical cumulative biomass to the theoretical maximum biomass constant at the current sampling time as the growth potential coefficient label;

[0130] Step 405: Using environmental feature vector samples as input and growth rate parameter labels and growth potential coefficient labels as output targets, train the first regression sub-model and the second regression sub-model using machine learning regression algorithms respectively.

[0131] Specifically, the first step involves acquiring historical observation data on edible fungi biomass and corresponding historical environmental data throughout the historical growth cycle. The intelligent agricultural control terminal retrieves recorded data from multiple complete growth cycles from the historical database. The historical observation data includes biomass indicators (such as cap diameter, fresh weight, etc.) obtained through manual measurement or machine vision recognition at fixed time intervals (e.g., daily), denoted as a sequence. The corresponding historical environmental data includes time-series data such as temperature, humidity, and light intensity recorded synchronously with the growth process.

[0132] Next, the step of constructing environmental feature vector samples based on historical environmental data is performed. This step is consistent with the feature extraction method in the aforementioned embodiments. For each historical sampling time... tUsing historical environmental data from this moment and its preceding moments, statistical features such as mean and cumulative values ​​are calculated to generate corresponding environmental feature vector samples. .

[0133] Subsequently, the intelligent agricultural control terminal performs the steps of calculating and assigning values ​​to the actual growth parameter labels.

[0134] For growth rate parameter labeling: For each sampling moment in the historical observation data, the intelligent agricultural control terminal executes the step of using the relative growth rate of biomass at the current sampling moment as the growth rate parameter label.

[0135] Intelligent agricultural control terminals calculate local relative growth rates using biomass observations from consecutive time points. A specific calculation method can employ the finite difference method. For example, for time points... t Calculate its observed local relative growth rate This is then used as the target value (Label) for training the first regression sub-model. A feasible calculation formula is as follows:

[0136] ;

[0137] in, For the current moment t Accumulated biomass observations achieved, For the previous moment t -1 represents the cumulative biomass observations that have been achieved.

[0138] Alternatively, when the data is sufficiently dense, the growth rate can be inversely calculated using the derivative form of the Richards model. r value.

[0139] For the growth potential coefficient label: The intelligent agricultural control terminal executes the step of using the ratio of the historical cumulative biomass to the theoretical maximum biomass constant at the current sampling time as the growth potential coefficient label.

[0140] Intelligent agricultural control terminal reads the current time t Accumulated biomass observations (e.g., the current cap diameter), and obtain the theoretical maximum biomass constant for this variety. The intelligent agricultural control terminal calculates the ratio of the two to obtain the observed growth potential coefficient. This value is then used as the target value for training the second regression sub-model. The calculation formula is as follows:

[0141] ;

[0142] The physical meaning of this label lies in the fact that it reflects the time... tUnder the influence of environmental history, the proportion of the theoretical maximum growth that edible fungi have actually achieved indirectly represents the growth state that the environment at that time allowed.

[0143] Finally, the intelligent agricultural control terminal performs the steps of training the first regression sub-model and the second regression sub-model using machine learning regression algorithms.

[0144] The intelligent agricultural control terminal constructs two independent training datasets: the first dataset consists of... The second dataset consists of... The intelligent agricultural control terminal performs supervised learning training on the first and second regression sub-models based on the first and second datasets, respectively.

[0145] The training process iteratively optimizes the model parameters by minimizing the loss function between the predicted and labeled values, ultimately resulting in a model that can accurately map environmental features to growth rate parameters. First regression sub-model And the ability to accurately map environmental characteristics to growth potential coefficients. Second regression sub-model .

[0146] In one specific implementation, the first and second regression sub-models are trained using the following method:

[0147] The environmental feature vector samples are input into the first regression sub-model to obtain the predicted growth rate parameters, and the environmental feature vector samples are input into the second regression sub-model to obtain the predicted growth potential coefficients.

[0148] The estimated biomass is calculated by calling the dynamic biological growth equation and based on the predicted growth rate parameter and the predicted growth potential coefficient.

[0149] A hybrid loss function is constructed, comprising a first parameter error term, a second parameter error term, and a third parameter error term. The first parameter error term characterizes the difference between the predicted growth rate parameter and the growth rate parameter label; the second parameter error term characterizes the difference between the predicted growth potential coefficient and the growth potential coefficient label; and the third parameter error term characterizes the difference between the estimated biomass and the actual biomass in historical observation data.

[0150] The gradient descent algorithm is used to minimize the mixed loss function, and the model parameters of the first and second regression sub-models are synchronously iteratively updated until the model converges.

[0151] Specifically, first, the forward propagation prediction step is executed. The intelligent agricultural control terminal will use the constructed environmental feature vector samples... Simultaneously, the inputs are fed into the first and second regression sub-models to be trained. The first regression sub-model outputs the current predicted growth rate parameters. The second regression sub-model outputs the current predicted growth potential coefficient. .

[0152] Next, a mechanism-based estimation step is performed. The intelligent agricultural control terminal invokes a preset dynamic biological growth equation. The intelligent agricultural control terminal utilizes the currently output predicted growth rate parameters. and based on Calculated growth potential parameters Substituting these values ​​into the equation, we can obtain the estimated biomass at the current moment. This step is equivalent to simulating the real prediction process during training.

[0153] Subsequently, the intelligent agricultural control terminal performs the crucial step of constructing a hybrid loss function. To balance the accuracy of intermediate parameters and the accuracy of the final result, the intelligent agricultural control terminal constructs a hybrid loss function containing three error terms. .

[0154] The first part is the error term of the first parameter. , used to characterize the parameters for predicting growth rate The growth rate parameter labels obtained from the aforementioned steps The difference between them (e.g., using mean squared error, MSE). The second part is the second parameter error term. Used to characterize the predictive growth potential coefficient With growth potential coefficient label The differences between them. The third part is the error term for the third parameter. Used to characterize the estimated biomass calculated from the mechanistic equation. Compared with the actual biomass in historical observation data The differences between them.

[0155] Hybrid loss function It is the weighted sum of these three terms, and its mathematical expression is as follows: in, α , β , γ These are preset weighting coefficients used to balance the focus on the accuracy of intermediate parameters and the accuracy of final output.

[0156] Finally, the intelligent agricultural control terminal performs the step of minimizing the mixed loss function using the gradient descent algorithm. The intelligent agricultural control terminal then calculates the total loss... Using the backpropagation algorithm (for neural network models) or a corresponding gradient calculation method, the gradient of the loss function with respect to the internal parameters of the first and second regression sub-models is calculated. The weight parameters of these two sub-models are then iteratively updated synchronously using the gradient descent algorithm until the mixed loss function converges to a preset threshold or reaches the maximum number of iterations.

[0157] This implementation achieves end-to-end joint training by constructing a hybrid loss function that includes both parameter error and result error. This method not only requires that the intermediate parameters (r and K) predicted by the model conform to the statistical regularity of the data (approximating the label), but also strictly constrains that the final result (biomass) calculated by inputting these parameter combinations into the mechanistic equation must be consistent with actual observations. Introducing a third parameter error term essentially embeds the biological mechanistic equation as a differentiable layer into the loss function, making the trained model more tolerant of parameter prediction errors and better guaranteeing the accuracy of the final biomass prediction. This effectively avoids the prediction failure problem caused by the exponential amplification of small parameter errors by the mechanistic equation.

[0158] After training the first and second regression sub-models using machine learning regression algorithms, the intelligent agricultural control terminal also performs model evaluation and verification steps to ensure that the prediction accuracy of the constructed model meets the needs of practical applications.

[0159] The intelligent agricultural control terminal first divides the constructed dataset into a training set and a test set according to a preset ratio. In this embodiment, following the time series order, the first 70% of the data is used as the training set for model parameter learning, and the last 30% of the data is used as an independent test set for model performance evaluation.

[0160] For the first regression sub-model (growth rate prediction model) The intelligent agricultural control terminal inputs environmental feature vector samples from the test set into the trained first regression sub-model to obtain a predicted sequence of growth rates. Subsequently, the intelligent agricultural control terminal calculates the predicted sequence against the corresponding growth rate parameter labels (i.e., the actual observed calculated values) in the test set. The statistical indicators of error between the two are: root mean square error (RMSE) and coefficient of determination (R²). For intelligent agricultural control terminals, the root mean square error (RMSE) and coefficient of determination (R²) are selected. 2 The root mean square error (RMSE) is used as the core evaluation metric. It quantifies the average deviation between the predicted growth rate and the actual observed rate; a smaller value indicates higher model accuracy. The coefficient of determination (COD) characterizes the model's ability to explain the variability of growth rates; a value closer to 1 indicates a better fit. The intelligent agricultural control terminal determines that when the R-value of the first regression sub-model on the test set... 2 If the RMSE is greater than a preset threshold (e.g., 0.85) and less than a preset tolerance, the model is considered to have passed training.

[0161] Similarly, for the second regression sub-model (growth potential coefficient prediction model) The intelligent agricultural control terminal inputs test set data into the model to obtain a sequence of predicted growth potential coefficients. The intelligent agricultural control terminal then calculates the correlation between this predicted value sequence and the growth potential coefficient labels in the test set. RMSE and R between ) 2 Indicators. Due to the growth potential coefficient It is a value between 0 and 1. The smart agriculture control terminal will focus on whether its RMSE index is at an extremely low level (e.g., less than 0.05) to ensure the accuracy of the upper limit of growth prediction.

[0162] Furthermore, to verify the overall performance of the combined two sub-models, the intelligent agricultural control terminal also performed a full-link verification step. The intelligent agricultural control terminal substituted the predicted dynamic growth rate parameters and growth potential parameters from the test set into the reconstructed dynamic biological growth equation to calculate the final biomass prediction trajectory. The intelligent agricultural control terminal compared this prediction trajectory with the actual biomass observation data recorded in the test set, and recalculated the overall biomass RMSE and R0. 2 Only after both the sub-model evaluation and the overall end-to-end evaluation have passed verification will the smart agriculture control terminal recognize the currently trained first and second regression sub-models as the final usable dynamic parameter mapping models and deploy them in actual production forecasting.

[0163] This embodiment provides a detailed explanation of the application process of optimizing environmental control decisions using the aforementioned prediction method. (Refer to...) Figure 5 , Figure 5 This is the fifth flowchart illustrating the environment-driven edible fungus growth prediction method provided by this invention. Figure 5 As shown, the intelligent agricultural control terminal can not only passively predict growth, but also actively assist in decision-making. Specifically, it performs the following steps to select the optimal environmental control strategy:

[0164] Step 501: Obtain environmental data sequences corresponding to various preset environmental control strategies;

[0165] Step 502: For each environmental regulation strategy, the environmental data sequence is processed by a dynamic parameter mapping model and a dynamic biological growth equation to obtain the corresponding biomass growth prediction trajectory.

[0166] Step 503: Based on the predicted biomass growth trajectory, select the target regulation strategy from a variety of environmental regulation strategies.

[0167] Specifically, firstly, the intelligent agricultural control terminal performs the step of acquiring environmental data sequences corresponding to various preset environmental control strategies.

[0168] Environmental control strategies refer to a combination of control targets set for the greenhouse environment over a future period. The intelligent agricultural control terminal stores or receives multiple alternative control schemes input by the user. For example, strategy A is set to "constant temperature and high humidity mode," with corresponding environmental data sequences showing a temperature maintained at 18℃±0.5℃ and relative humidity maintained at 90% for the next 7 days; strategy B is set to "variable temperature stimulation mode," with corresponding environmental data sequences showing a diurnal temperature range of 5℃ and relative humidity maintained at 85%. Based on these set strategy parameters, the intelligent agricultural control terminal generates virtual environmental factor time-series data for each corresponding future period.

[0169] Next, the intelligent agricultural control terminal executes the environmental regulation strategy for each type of environmental control strategy, and processes the environmental data sequence through a dynamic parameter mapping model and a dynamic biological growth equation to obtain the corresponding biomass growth prediction trajectory.

[0170] The intelligent agricultural control terminal sequentially inputs the virtual environment data sequences corresponding to strategy A and strategy B. and The input is fed into the dynamic parameter mapping model.

[0171] First, the dynamic parameter sequences under strategy A and strategy B are calculated using a dynamic parameter mapping model. and Subsequently, iterative calculations were performed using the dynamic biological growth equation to output the predicted biomass growth trajectory corresponding to strategy A. Biomass growth prediction trajectory corresponding to strategy B These trajectories visually demonstrate how edible fungi will grow in the future if appropriate regulatory strategies are implemented.

[0172] Finally, the intelligent agricultural control terminal performs the step of selecting the target regulation strategy from a variety of environmental regulation strategies based on the predicted biomass growth trajectory.

[0173] The intelligent agricultural control terminal performs quantitative evaluation on each of the generated prediction trajectories. The evaluation index can be set as "expected final yield" (i.e., the biomass value at the end of the prediction cycle) or "growth rate" (i.e., the time required to reach the maturity harvest standard).

[0174] For example, if the production goal is to pursue maximum output, intelligent agricultural control terminals are more suitable. and If strategy A has a higher expected output, then strategy A will be selected as the target control strategy.

[0175] If the production goal is to bring the produce to market before the holiday and achieve the fastest possible maturity, the intelligent agricultural control terminal compares the time points at which both strategies reach the harvest threshold and selects the strategy with the shorter time. The intelligent agricultural control terminal then sends the selected target control strategy to the greenhouse environmental control system, which directly drives equipment such as fans, wet curtains, and heaters to perform corresponding environmental adjustment actions.

[0176] This implementation method allows the intelligent agricultural control terminal to quickly simulate the expected consequences of various management measures without conducting actual destructive experiments by inputting environmental strategies under different assumptions into the model. This significantly reduces production risks and improves resource utilization efficiency.

[0177] To further verify the effectiveness and superiority of the environment-driven edible fungus growth prediction method proposed in this application in practical applications, a set of comparative experiments was designed in this embodiment. Independent fungal log samples (numbered 062 / 11) that were not involved in model training were selected as the verification objects to evaluate the generalization ability and prediction accuracy of different models when faced with novel samples and dynamic environments.

[0178] This implementation method selects three different prediction models for comparison:

[0179] The first model is the Environment-Driven-Dynamic Growth Model (ED-DGM) proposed in this application. The intelligent agricultural control terminal will use the trained first regression sub-model. Second regression sub-model The process involves solidification, inputting actual environmental sequence data of the verification object during its complete growth cycle, dynamically generating parameters and calculating the growth trajectory according to the method described in the aforementioned embodiments.

[0180] The second model is the classical static Logistic model (i.e., the classical growth mechanism equation). This model uses the traditional Logistic equation, where the parameters r and K are fixed constants obtained by fitting historical average data and are not dynamically adjusted with changes in the environment.

[0181] The third type of model is the pure neural network model (ANN). This model uses a backpropagation (BP) neural network architecture. Its input features are consistent with those of the ED-DGM model (i.e., the same environmental feature vectors), but its output directly corresponds to the predicted value of biomass growth. It does not include the constraints of biological mechanism equations and parameter mapping processes, and is a typical "end-to-end" black box model.

[0182] The intelligent agricultural control terminal runs the three models described above, generating corresponding growth prediction curves based on the same set of input environment sequences. The prediction results are then compared with the measured growth data (including length, width, and height) of the validation object (mushroom stick 062 / 11). To quantify the prediction performance, this implementation method selects the root mean square error (RMSE) and the coefficient of determination (R²). 2 () as an evaluation indicator.

[0183] The experimental results are shown in the table below:

[0184] Table 1. Comparison of the growth prediction performance of three models for mycelium 062 / 11 (unit: mm, with length as the growth index)

[0185]

[0186] Analysis of the above experimental data shows that:

[0187] In terms of length prediction, the ED-DGM model of this application has an RMSE of only 8.7mm, significantly lower than the 22.3mm of the classic static Logistic model and the 12.5mm of the pure neural network model; at the same time, its R... 2 The value of 0.94 indicates that the predicted curve closely matches the actual growth trajectory.

[0188] In width and height prediction, the ED-DGM model also demonstrated an overwhelming advantage. Especially in width prediction, the classic static model's R² was significantly higher. 2 With an R² value of only 0.45, it is clear that it is largely ineffective in capturing the changing patterns of width growth, while the ED-DGM model has a much higher R² value. 2 The result reached 0.91, demonstrating that the dynamic parameter mechanism has good adaptability to different morphological indicators.

[0189] To more intuitively demonstrate the advantages of the ED-DGM model proposed in this application in capturing dynamic environmental changes, this implementation method further combines... Figure 6 A detailed comparative analysis of the growth prediction trajectories of each model was conducted. Figure 6 This is a schematic diagram comparing the growth prediction trajectories of the various models provided by this invention. Figure 6 The study demonstrates the fitting of predicted curves generated by the ED-DGM model, the classic static Logistic model, and the pure neural network model to the measured data points for the length growth index of mushroom stick number 056 / 04 under the same dynamically changing environment.

[0190] like Figure 6As shown, the horizontal axis represents growth days, and the vertical axis represents stipe length, in millimeters (mm). The blue dots in the figure represent measured data, indicating the actual growth data obtained each day. The solid red line represents the predicted trajectory of the Environment-Driven Dynamic Growth Model (ED-DGM) proposed in this application; the dashed green line represents the predicted trajectory of the Classical Logistic model, i.e., the classical growth mechanism equation; and the dotted purple line represents the predicted trajectory of the Artificial Neural Network Model (ANN Model).

[0191] From an overall perspective, during the actual growth process, the curve formed by the blue dots is not a smooth upward curve from the slow growth stage to the rapid growth stage, and then to the maturity stage, but rather exhibits obvious fluctuations. Especially between the 3rd and 4th days (the green background area in the diagram, corresponding to the rapid growth stage), due to drastic fluctuations in external environmental factors (such as sudden changes in temperature and humidity), the actual growth rate shows a significant lag and inflection point (i.e.,...). Figure 6 The growth slowdown (marked as "Growth slowdown due to environmental fluctuations") resulted in the measured length on day 4 (approximately 82 mm) being significantly lower than expected.

[0192] The three models showed significant differences in their responses to this specific environmental fluctuation event:

[0193] The classic static Logistic model (green dashed line), because its internal parameters r and K are fixed as constants, can only describe an idealized, monotonically increasing S-shaped curve. Therefore, it completely ignores the environmental degradation from day 3 to day 4, continuing to predict that growth will increase linearly at the average rate. This results in its predicted value (approximately 118 mm) on day 4 being much higher than the measured value, producing a huge positive bias. This clearly exposes the fundamental flaw of static models: their inability to respond to environmental dynamics.

[0194] Although the pure neural network model (purple dashed line) learned certain nonlinear relationships through data-driven learning, at the turning point from day 3 to day 4, its predicted trajectory (approximately 88 mm), while closer to the measured value than the static model, still failed to accurately capture the magnitude of growth stagnation. More importantly, from day 5 to day 6 (the red background area in the figure, corresponding to the maturity stage), after the environment recovered, a significant deviation again appeared between the pure neural network model's predicted trajectory (approximately 160 mm) and the final measured value (approximately 178 mm), indicating that its extrapolation prediction ability is insufficient in the absence of mechanistic constraints, and it is prone to "fit failure".

[0195] In contrast, the ED-DGM model (solid red line) of this application exhibits optimal dynamic adaptability. From day 3 to day 4, the intelligent agricultural control terminal, through the dynamic parameter mapping model, sensed environmental changes in real time and automatically lowered the dynamic growth rate parameter r(t), causing the red predicted curve to accurately exhibit a downward-curving "growth stagnation" characteristic, highly coinciding with the actual data points (blue dots). From day 5 to day 6, as the environment improved, the model automatically increased the parameter, causing the predicted curve to rapidly recover and accurately track the final measured yield (approximately 175 mm).

[0196] Figure 6 The comparative results clearly demonstrate that the ED-DGM model not only retains the biological morphological characteristics of classic growth curves but also possesses the ability to adjust its growth rhythm according to environmental fluctuations, just like real organisms. This result fully illustrates that combining data-driven dynamic parameter mapping with the structural constraints of mechanistic models can effectively achieve accurate and reliable growth prediction in real and complex dynamic environments.

[0197] Reference Figure 7 , Figure 7 This is a schematic diagram of the structure of the environment-driven edible fungus growth prediction system provided by the present invention. The system includes:

[0198] The feature construction module is used to construct environmental feature vectors for each time period based on the environmental data within the time period to be predicted.

[0199] The parameter acquisition module is used to input environmental feature vectors into the dynamic parameter mapping model and obtain the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model. The dynamic parameter mapping model is trained based on environmental feature vector samples and corresponding actual growth parameter labels.

[0200] The prediction calculation module is used to call the preset dynamic biological growth equation. The dynamic biological growth equation is obtained by reconstructing the static constant terms in the classical growth mechanism equation into dynamic variable parameters that change with time and environment.

[0201] The prediction and calculation module is also used to assign the dynamic growth parameters to the dynamic variable parameters in the dynamic biological growth equation, and calculate the predicted value of biomass growth of edible fungi at each time point.

[0202] In one possible implementation, the system further includes a model training module; the model training module is used for:

[0203] Acquire historical observation data on edible fungi biomass and corresponding historical environmental data during the historical growth cycle;

[0204] Construct environmental feature vector samples based on historical environmental data;

[0205] For each sampling time in the historical observation data, the relative growth rate of biomass at the current sampling time is used as the growth rate parameter label;

[0206] The ratio of historical cumulative biomass to the theoretical maximum biomass constant at the current sampling time is used as the growth potential coefficient label;

[0207] Using environmental feature vector samples as input and growth rate parameter labels and growth potential coefficient labels as output targets, the first regression sub-model and the second regression sub-model are trained using machine learning regression algorithms.

[0208] It should be noted that the environment-driven edible fungus growth prediction system provided by the present invention can execute the environment-driven edible fungus growth prediction method of any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0209] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 8 As shown, the electronic device may include a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute the environment-driven edible fungus growth prediction method provided in the above embodiments.

[0210] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0211] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer is able to execute the environment-driven edible fungus growth prediction method provided in the above embodiments.

[0212] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the environment-driven edible fungus growth prediction method provided in the above embodiments.

[0213] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0214] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0215] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An environment-driven method for predicting the growth of edible fungi, characterized in that, include: Based on the environmental data within the time period to be predicted, construct the environmental feature vector corresponding to each time point; The environmental feature vector is input into the dynamic parameter mapping model to obtain the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model. The dynamic parameter mapping model is trained based on the environmental feature vector samples and the corresponding actual growth parameter labels. The dynamic growth parameters include growth rate parameters and growth potential parameters. The dynamic parameter mapping model includes a first regression sub-model and a second regression sub-model. The step of inputting the environmental feature vector into the dynamic parameter mapping model and obtaining the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model includes: The environmental feature vector is input into the first regression sub-model to obtain the growth rate parameters corresponding to each time step output by the first regression sub-model. The environmental feature vector is input into the second regression sub-model to obtain the growth potential coefficients corresponding to each time step output by the second regression sub-model. The growth potential parameter at the corresponding time is calculated by multiplying the preset theoretical maximum biomass constant with the growth potential coefficient. A preset dynamic biological growth equation is invoked. This dynamic biological growth equation is obtained by reconstructing the static constant terms in the classical growth mechanism equation into dynamic variable parameters that change with time and environment. The dynamic biological growth equation is expressed by the following mathematical formula: ； ； ； in, Let be the biomass at time t; The parameter representing the growth potential at time t; The growth rate parameter at time t; The time parameter for reaching the maximum growth rate; Let be the normalized temperature value at time t. Let be the normalized humidity value at time t. Let be the normalized carbon dioxide concentration at time t. Let be the normalized illumination intensity value at time t; This represents the interaction term between temperature and humidity at time t; This represents the interaction term between temperature and carbon dioxide at time t; This represents the interaction term between temperature and illumination at time t; This represents the interaction term between humidity and carbon dioxide at time t; This represents the interaction term between humidity and light intensity at time t; This represents the interaction term between light intensity and carbon dioxide at time t; The intercept term of the first regression sub-model is... The coefficients to be fitted for the first regression sub-model; This is the intercept term of the second regression sub-model. represents the coefficients to be fitted in the second regression sub-model; v represents the shape parameter; The dynamic growth parameters are assigned to the dynamic variable parameters in the dynamic biological growth equation to calculate the predicted biomass growth values ​​of edible fungi at each time point.

2. The environment-driven edible fungus growth prediction method according to claim 1, characterized in that, The actual growth parameter labels include growth rate parameter labels and growth potential coefficient labels; The first regression sub-model and the second regression sub-model were trained through the following steps: Acquire historical observation data on edible fungi biomass and corresponding historical environmental data during the historical growth cycle; Construct environmental feature vector samples based on the historical environmental data; For each sampling time in the historical observation data, the relative growth rate of biomass at the current sampling time is used as the label of the growth rate parameter. The ratio of the historical cumulative biomass at the current sampling time to the theoretical maximum biomass constant is used as the label for the growth potential coefficient. Using the environmental feature vector samples as input, and the growth rate parameter label and the growth potential coefficient label as output targets, the first regression sub-model and the second regression sub-model are trained using a machine learning regression algorithm.

3. The environment-driven edible fungus growth prediction method according to claim 2, characterized in that, The process of using the environmental feature vector samples as input and the growth rate parameter label and the growth potential coefficient label as output targets, and training the first regression sub-model and the second regression sub-model respectively using a machine learning regression algorithm, includes: The environmental feature vector samples are input into the first regression sub-model to obtain the predicted growth rate parameters, and the environmental feature vector samples are input into the second regression sub-model to obtain the predicted growth potential coefficients. The estimated biomass is calculated by calling the dynamic biological growth equation and based on the predicted growth rate parameter and the predicted growth potential coefficient. A hybrid loss function is constructed, comprising a first parameter error term, a second parameter error term, and a third parameter error term. The first parameter error term characterizes the difference between the predicted growth rate parameter and the growth rate parameter label. The second parameter error term characterizes the difference between the predicted growth potential coefficient and the growth potential coefficient label. The third parameter error term characterizes the difference between the estimated biomass and the actual biomass in the historical observation data. The mixed loss function is minimized using the gradient descent algorithm, and the model parameters of the first regression sub-model and the second regression sub-model are synchronously iteratively updated until the model converges.

4. The environment-driven edible fungus growth prediction method according to claim 1, characterized in that, The step of constructing environmental feature vectors for each time period based on environmental data within the period to be predicted includes: For any moment within the time period to be predicted, determine the current time window corresponding to that moment; Obtain environmental data for the current time window and N preceding time windows; the environmental data includes temperature, humidity, carbon dioxide concentration, and light intensity. Calculate the statistical characteristic values ​​of the environmental data; the statistical characteristic values ​​include at least one of the mean, standard deviation, maximum value, minimum value, or cumulative value. The statistical feature values ​​are combined to obtain the environmental feature vector.

5. The environment-driven edible fungus growth prediction method according to claim 1, characterized in that, Also includes: Acquire environmental data sequences corresponding to various preset environmental control strategies; For each of the aforementioned environmental regulation strategies, the environmental data sequence is processed using the dynamic parameter mapping model and the dynamic biological growth equation to obtain the corresponding biomass growth prediction trajectory. Based on the predicted biomass growth trajectory, a target regulatory strategy is selected from a variety of environmental regulation strategies.

6. An environment-driven edible fungus growth prediction system, characterized in that, include: The feature construction module is used to construct environmental feature vectors for each time period based on the environmental data within the time period to be predicted. The parameter acquisition module is used to input the environmental feature vector into a dynamic parameter mapping model to obtain the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model. The dynamic parameter mapping model is trained based on environmental feature vector samples and corresponding actual growth parameter labels. The dynamic growth parameters include growth rate parameters and growth potential parameters. The dynamic parameter mapping model includes a first regression sub-model and a second regression sub-model. The step of inputting the environmental feature vector into the dynamic parameter mapping model to obtain the dynamic growth parameters corresponding to each time step output by the dynamic parameter mapping model includes: inputting the environmental feature vector into the first regression sub-model to obtain the growth rate parameters corresponding to each time step output by the first regression sub-model; inputting the environmental feature vector into the second regression sub-model to obtain the growth potential coefficients corresponding to each time step output by the second regression sub-model; and calculating the growth potential parameters at the corresponding time step based on the product of a preset theoretical maximum biomass constant and the growth potential coefficients. The prediction and calculation module is used to invoke a preset dynamic biological growth equation. This dynamic biological growth equation is obtained by reconstructing the static constant terms in the classical growth mechanism equation into dynamic variable parameters that change with time and the environment. The dynamic biological growth equation is expressed by the following mathematical formula: ； ； ； in, Let be the biomass at time t; The parameter representing the growth potential at time t; The growth rate parameter at time t; The time parameter for reaching the maximum growth rate; Let be the normalized temperature value at time t. Let be the normalized humidity value at time t. Let be the normalized carbon dioxide concentration at time t. Let be the normalized illumination intensity value at time t; This represents the interaction term between temperature and humidity at time t; This represents the interaction term between temperature and carbon dioxide at time t; This represents the interaction term between temperature and illumination at time t; This represents the interaction term between humidity and carbon dioxide at time t; This represents the interaction term between humidity and light intensity at time t; This represents the interaction term between light intensity and carbon dioxide at time t; The intercept term of the first regression sub-model is... The coefficients to be fitted for the first regression sub-model; This is the intercept term of the second regression sub-model. represents the coefficients to be fitted in the second regression sub-model; v represents the shape parameter; The prediction calculation module is also used to assign the dynamic growth parameters to the dynamic variable parameters in the dynamic biological growth equation, and calculate the predicted value of biomass growth of edible fungi at each time point.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the environment-driven edible fungus growth prediction method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the environment-driven edible fungus growth prediction method as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the environment-driven edible fungus growth prediction method as described in any one of claims 1 to 5.

Images