Intelligent monitoring method for dandelion and camellia nitidissima fermentation process based on Internet of Things

Abstract

The invention relates to the technical field of monitoring control, and provides an intelligent monitoring method for a dandelion and golden camellia fermentation process based on the Internet of Things, which comprises the following steps: deploying and self-calibrating a sensor array, extracting and analyzing various data features, and judging a fermentation stage and generating a regulation and control instruction according to the data features. According to the method, a precise fermentation monitoring and control method logic is formed, through omnibearing data acquisition and deep analysis, the change of each parameter in the fermentation process can be precisely grasped, the hypha growth rate, the flora metabolism intensity, the generation of a target product and other key indexes are monitored in real time, the fermentation conditions are adjusted in time, and the fermentation efficiency is improved. And the fermentation process is stabilized in an optimal state, so that the accuracy of fermentation production is greatly improved, the fluctuation of product quality is reduced, and the consistency and stability of the quality of each batch of dandelion and golden camellia are ensured.

Description

Technical Field

[0001] The present application relates to the field of monitoring and control technology, and in particular to an intelligent monitoring method for a dandelion golden camellia fermentation process based on the Internet of Things. Background Art

[0002] During the fermentation and production process of Taraxacum chrysantha, traditional monitoring methods are unable to fully acquire all kinds of data in the fermentation bin, and traditional methods rely solely on manual observation or simple image analysis, which is inefficient and has large errors, and cannot provide a reliable basis for optimizing the fermentation process. Currently, the judgment of the fermentation stage mainly relies on manual experience, and there is a lack of scientific and accurate judgment methods. This makes it impossible to adjust the fermentation strategy in time according to actual conditions during the fermentation process, which can easily lead to excessive or insufficient fermentation, affecting product quality and output. When regulating the fermentation process, the generation of control instructions lacks intelligence and precision, resulting in poor control effects and increased production costs. Summary of the invention

[0003] In view of the shortcomings of the prior art, the present application provides an intelligent monitoring method for the fermentation process of dandelion golden camellia based on the Internet of Things, the method comprising: deploying a sensor array to respectively obtain environmental data, spectral data and visual data in the fermentation bin, and self-calibrating the sensor array;

[0004] Extract the features of environmental data, spectral data and visual data to obtain the temperature and humidity time series characteristics, gas release rate, spectral characteristics and mycelium morphology characteristics, generate the mycelium growth rate based on the temperature and humidity time series characteristics, spectral characteristics and mycelium morphology characteristics, determine the bacterial community metabolic intensity based on the gas release rate and mycelium morphology characteristics, and predict the generation rate of the target product based on the spectral characteristics;

[0005] Generate fermentation stage labels and confidence levels based on mycelium growth rate and bacterial metabolic intensity, and generate control instruction sets based on reinforcement learning in combination with the target product production rate;

[0006] Execute the control instruction set, monitor the environmental data to adjust the control instruction set and abnormal protection, and synchronize it to the cloud knowledge base.

[0007] As an optional embodiment, the hyphae morphological characteristics include hyphae density and hyphae coverage area;

[0008] The extraction logic of mycelium morphological features includes:

[0009] The hyphae morphology image is preprocessed, and the hyphae region in the hyphae morphology image is segmented according to the grayscale threshold;

[0010] The segmented hyphae regions are skeletonized by a thinning algorithm to obtain a hyphae skeleton;

[0011] Extract the topological structure of hyphae skeleton, including the number of branch points, the number of endpoints and the average length of hyphae branches;

[0012] The hyphae density and hyphae coverage area were determined according to the topological structure of the hyphae skeleton.

[0013] As an optional implementation, the logic for generating the mycelium growth rate includes:

[0014] The temperature and humidity time series characteristics, spectral characteristics and hyphae morphological characteristics at different time points were integrated and time-stamped and aligned;

[0015] Select characteristic points on the mycelium skeleton and record the temperature and humidity time series characteristics and spectral characteristics of each characteristic point;

[0016] The displacement of the feature point between two adjacent frames of hyphae morphology images is calculated to determine the local growth rate of the feature point in the time period, and the changes in the temperature and humidity time series characteristics and spectral characteristics in the time period are analyzed to determine the weight of the feature point;

[0017] The mycelium growth rate is obtained by taking the weighted average of the local growth rates of all feature points.

[0018] As an optional implementation, the logic for determining the bacterial flora metabolic intensity includes:

[0019] Monitor the dynamic changes of carbon dioxide concentration and volatile metabolite concentration, and analyze the gas release rate;

[0020] The gas release rate and hyphae morphological characteristics were correlated and analyzed;

[0021] The metabolic intensity of the bacterial community was obtained by weighted quantification based on the release rate of carbon dioxide, the release rate of volatile metabolites and the change rate of mycelium coverage area.

[0022] As an optional implementation, the prediction logic of the target product generation rate includes:

[0023] Screening the bands of spectral characteristics related to the target product;

[0024] The area of ​​the band is mapped to the substrate consumption rate, and the production rate of the output target product is predicted through a neural network in combination with environmental data;

[0025] Monitor the prediction error of the neural network to determine whether to retrain the neural network.

[0026] As an optional implementation, the logic for generating the fermentation stage label and confidence level includes:

[0027] The mycelium growth rate, bacterial community metabolic intensity and historical stage labels are integrated to obtain integrated data, and cluster analysis is performed on the integrated data based on a clustering algorithm to obtain clustering results;

[0028] Based on fuzzy logic, the mycelium growth rate and bacterial metabolic intensity were divided into fuzzy linguistic variables, and the fermentation stage labels of different fuzzy linguistic variable combinations were determined based on the clustering results.

[0029] The confidence of each fermentation stage label is calculated, and the fermentation stage label is corrected according to the confidence.

[0030] As an optional implementation, the generation logic of the control instruction set includes:

[0031] Construct a hierarchical reinforcement learning architecture, which includes a target determination phase and a regulation generation phase;

[0032] Determine the reward function based on the target product generation rate, energy consumption and pollution risk;

[0033] Monitor the operating data during the fermentation process and dynamically adjust the control generation stage and reward function.

[0034] As an optional embodiment, the target determination stage is used to determine the regulation target according to the fermentation stage label, confidence and the generation rate of the target product;

[0035] The control generation phase is used to determine the state space and action space;

[0036] The state space includes temperature, humidity, carbon dioxide concentration, mycelium density, bacterial metabolic intensity, and the production rate of target products;

[0037] The action space includes heating power, ventilation valve opening and humidification amount.

[0038] As an optional embodiment, the environmental data includes temperature, humidity, carbon dioxide concentration and volatile metabolite concentration, the spectral data is used to analyze the metabolic intensity of the bacterial community and the production rate of the target product, and the visual data includes mycelium morphology images to capture the mycelium growth morphology.

[0039] The deployment logic of the sensor array includes:

[0040] Divide the fermentation tank into regional blocks according to its structural functions;

[0041] In each area block, a sensor array is deployed in a layered and three-dimensional manner. The sensor array includes temperature and humidity sensors, gas sensors, spectral sensors, and visual sensors.

[0042] The sensor array is synchronously stimulated to obtain environmental data, spectral data and visual data in the fermentation chamber respectively.

[0043] As an optional implementation, the self-calibration logic of the sensor array includes:

[0044] Determine self-calibration conditions according to changes in time and data, and configure standard data;

[0045] During the self-calibration process, the parameters of the sensor array are adjusted according to the difference between the acquired data and the standard data;

[0046] After adjustment, verify the effect of self-calibration to provide feedback on whether to re-calibrate.

[0047] Compared with the prior art, the beneficial effects of the present application are: from the deployment and self-calibration of the sensor array to the extraction and analysis of various data features and the fermentation stage judgment and control instruction generation based on this, a precise fermentation monitoring and control method logic is formed. By acquiring data from all directions and conducting in-depth analysis, it is possible to accurately grasp the changes in various parameters during the fermentation process, monitor key indicators such as mycelium growth rate, bacterial metabolic intensity, and target product generation in real time, adjust the fermentation conditions in time, and stabilize the fermentation process in the optimal state, which greatly improves the accuracy of fermentation production, reduces product quality fluctuations, and ensures the consistency and stability of the quality of each batch of Dandelion Golden Camellia.

[0048] By deploying sensor arrays and self-calibrating, comprehensive and accurate environmental data, spectral data, and visual data can be obtained in the fermentation chamber. This lays a solid foundation for the subsequent precise extraction of various data features, ensuring the accuracy of data such as temperature and humidity time series characteristics, gas release rate, spectral characteristics, and mycelium morphological characteristics, thereby improving the reliability of mycelium growth rate, bacterial metabolic intensity calculation, and target product production rate prediction, providing strong data support for the scientific analysis of the fermentation process.

[0049] Fermentation stage labels and confidence levels are generated based on mycelium growth rate and bacterial metabolic intensity. Compared with traditional manual experience judgment, this method is more scientific and accurate, and can reflect the fermentation process in real time and objectively, providing an important basis for the precise control of the fermentation process. It enables operators to adjust fermentation strategies in time according to different fermentation stages, avoid excessive or insufficient fermentation, and effectively improve product quality and output.

[0050] Combined with the generation rate of the target product, a control instruction set is generated based on reinforcement learning, which fully considers multiple factors such as the generation rate of the target product, energy consumption and pollution risk. This intelligent control method can achieve precise control of the fermentation process, while ensuring the generation of the target product, reducing energy consumption, reducing pollution risks, improving production efficiency, reducing production costs, and enhancing the market competitiveness of products.

[0051] During the execution of the control instruction set, the control instruction set is adjusted and abnormal protection is performed by monitoring the environmental data, and then synchronized to the cloud knowledge base. On the one hand, the control strategy can be optimized in time according to the dynamic changes of the fermentation process to ensure that the fermentation process is always in the best state; on the other hand, the abnormal protection mechanism can quickly discover and handle abnormal situations in the fermentation process to avoid fermentation failures caused by abnormalities. The establishment of a cloud knowledge base is conducive to the accumulation of data and experience in the fermentation process, providing reference and improvement direction for subsequent fermentation production, and promoting the continuous improvement of the entire fermentation production technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0053] Figure 1 A method flow chart of the intelligent monitoring method of the fermentation process of Taraxacum chrysantha based on the Internet of Things provided in an embodiment of the present application;

[0054] Figure 2 A logic diagram for extracting mycelium morphological features of an intelligent monitoring method for the fermentation process of Taraxacum chrysantha based on the Internet of Things provided in an embodiment of the present application;

[0055] Figure 3 This is a logic diagram for generating mycelium growth rate of the intelligent monitoring method for the fermentation process of Taraxacum chrysantha based on the Internet of Things provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application are clearly and completely described below in conjunction with the drawings in the specification. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0057] Example The application scenario of the embodiment of the present application is to produce a more effective probiotic "double bacteria" dandelion golden flower tea through microbial fermentation. The base tea is made by taking young leaves, picking, washing, steaming, airing, rolling, frying and drying. Later, a series of fungus culture and tea making technologies such as sterilization, inoculation, cultivation, drying and shaping are carried out through modern microbial technology. The Amsterdam Assemblage is precisely placed, and the optimal fermentation environment is accurately controlled to produce a high amount of Eurotium cristatum (golden flower fungus). The dandelion flavonoid content is increased through microbial fermentation to produce a more effective probiotic "double bacteria" dandelion golden flower tea.

[0058] like Figure 1 As shown, a method flow chart of an intelligent monitoring method for the fermentation process of dandelion golden camellia based on the Internet of Things is provided for an embodiment of the present application, and the method comprises:

[0059] S1. Deploy the sensor array to obtain the environmental data, spectral data and visual data in the fermentation chamber, and self-calibrate the sensor array.

[0060] Environmental data include temperature, humidity, carbon dioxide concentration and volatile metabolite concentration, spectral data are used to analyze the metabolic intensity of the bacterial community and the production rate of target products, and visual data include mycelial morphology images to capture the mycelial growth morphology.

[0061] The deployment logic of the sensor array includes:

[0062] Divide the fermentation tank into regional blocks according to its structural functions;

[0063] In each area block, a sensor array is deployed in a layered and three-dimensional manner. The sensor array includes temperature and humidity sensors, gas sensors, spectral sensors, and visual sensors.

[0064] The sensor array is synchronously stimulated to obtain environmental data, spectral data and visual data in the fermentation chamber respectively.

[0065] There are differences in fermentation conditions at different locations in the fermentation warehouse. For example, the temperature, humidity and gas concentration in the area near the heating equipment or the vents vary. Dividing the area into blocks can more accurately monitor the fermentation environment in different areas and provide a basis for subsequent precise regulation. Through on-site investigation of the fermentation warehouse, combined with the internal layout of the fermentation warehouse, such as the location of the fermentation equipment and the direction of the ventilation duct, the fermentation warehouse is digitally modeled through 3D modeling software. According to the model of the fermentation warehouse, the fermentation warehouse is divided into several blocks of moderate size. For example, for a fermentation warehouse with a length of 10 meters, a width of 8 meters and a height of 5 meters, it can be divided into 8 blocks of 2.5 meters × 4 meters × 5 meters, ensuring that the environmental characteristics of each block during the fermentation process are representative. In this way, the regional and refined monitoring of the environment in the fermentation warehouse can be realized, the accuracy and pertinence of data collection can be improved, and a clear location framework can be provided for the layered and three-dimensional deployment of sensors, so that the deployment of the sensor array is more planned and the environmental data of each area can be better obtained.

[0066] The environmental conditions at different heights in the fermentation bin are also different. The layered three-dimensional deployment can fully obtain the environmental data at different height levels in the fermentation bin, and more realistically reflect the spatial changes of the fermentation environment. In each area block, a sensor plane is set every 1 meter along the height direction of the fermentation bin. On each plane, various sensors are evenly distributed at a certain interval. For example, temperature and humidity sensors are used to obtain the temperature and humidity in the fermentation bin. One is installed every 2 meters. Gas sensors include carbon dioxide sensors and volatile metabolite sensors, which are used to monitor carbon dioxide concentration and volatile metabolite concentration, respectively. Each sensor is 3 meters apart. Spectral sensors are used for subsequent analysis of bacterial metabolic intensity and target product generation rate. One is installed at the center of each area block. Visual sensors need to be equipped with lenses that can adapt to the light conditions in the fermentation bin. They are installed on an adjustable angle bracket to capture mycelium morphology images. Two are installed in each area block and taken from different angles. In this way, more comprehensive and accurate fermentation environmental data can be obtained, providing multi-dimensional information for analyzing the fermentation process, and ensuring that when the sensor array is synchronously stimulated, rich environmental data, spectral data and visual data can be obtained from different heights and positions, providing sufficient data support for subsequent data analysis and fermentation process regulation.

[0067] In order to ensure that each sensor acquires data at the same time and the time synchronization of the data is guaranteed, the subsequent data analysis is more reliable and effective; a unified trigger time is set for each sensor, and a synchronous trigger instruction is sent to the sensors in each area block, so that the temperature and humidity sensors, gas sensors, spectral sensors and visual sensors start to acquire data at the same time; thereby ensuring that the acquired environmental data, spectral data and visual data have time consistency, avoiding analysis errors caused by asynchronous data acquisition time, and providing a reliable data basis for the subsequent accurate analysis of the relationship between environmental factors and bacterial metabolism and mycelium growth during the fermentation process, which is conducive to precise fermentation regulation based on the data.

[0068] The self-calibration logic for the sensor array includes:

[0069] Determine self-calibration conditions according to changes in time and data, and configure standard data;

[0070] During the self-calibration process, the parameters of the sensor array are adjusted according to the difference between the acquired data and the standard data;

[0071] After adjustment, verify the effect of self-calibration to provide feedback on whether to re-calibrate.

[0072] During long-term use, the sensor will be affected by environmental factors (such as temperature and humidity changes) or its own aging, resulting in deviations in the acquired data. Setting self-calibration conditions and configuring standard data can timely detect and correct these deviations to ensure the accuracy of the data acquired by the sensor; the self-calibration conditions are set to perform regular calibration every 24 hours. At the same time, when the data acquired by the sensor fluctuates abnormally or exceeds the preset threshold range, self-calibration is triggered immediately. Abnormal fluctuations refer to, for example, temperature data changes exceeding ±5°C within 10 minutes, and there is no similar change in temperature sensors in other area blocks; and conditions greater than the preset threshold range refer to, for example, carbon dioxide concentration exceeding ±20% of the normal fermentation range; standard data is obtained through High-precision standard sources are measured and acquired in a laboratory environment. For example, a constant temperature and humidity chamber is used as the standard data for temperature and humidity. The temperature accuracy can reach ±0.1°C and the humidity accuracy can reach ±2%RH. A standard gas generator is used to configure standard gases of known concentrations of carbon dioxide and volatile metabolites for calibrating gas sensors. The standard data of spectral sensors are obtained by measuring standard samples with known spectral characteristics. These standard data are stored in a database for self-calibration. This enables timely detection and correction of sensor acquisition deviations, ensuring the accuracy and reliability of sensor data, and providing an accurate comparison benchmark for subsequent difference calculations between the acquired data and the standard data, making adjustment of sensor array parameters more targeted.

[0073] By comparing the difference between the data obtained by the sensor and the standard data, the sensor parameters are adjusted so that the result of the data obtained by the sensor is closer to the true value, ensuring the accuracy of the data obtained by the sensor; during the self-calibration process, the sensor obtains data again and compares the obtained data with the standard data. For the temperature and humidity sensor, if the difference between the obtained temperature value and the standard temperature value is greater than the allowable error range, such as ±0.5℃, the temperature compensation coefficient of the temperature and humidity sensor is adjusted according to the size of the difference. For the gas sensor, if the difference between the obtained carbon dioxide concentration and the standard concentration is too large, the gas sensor is calibrated by adjusting the sensitivity parameters. The spectrum sensor adjusts the wavelength calibration parameters and intensity calibration parameters according to the difference between the obtained spectrum data and the standard spectrum; thereby effectively improving the measurement accuracy of the sensor and ensuring the accuracy of the data. After the parameters are adjusted, the sensor obtains more accurate data, providing reliable data for verifying the self-calibration effect, which is convenient for judging whether the self-calibration is successful.

[0074] Determine whether the self-calibration has achieved the expected effect. If not, re-perform self-calibration to ensure that the sensor is always in the best working state. After the sensor completes parameter adjustment, obtain data again and compare it with the standard data. Calculate the error between the adjusted data and the standard data. If the error is within the allowable error range, the self-calibration is determined to be successful. If the error is still greater than the allowable error range, re-execute the self-calibration process. At the same time, record the results of each self-calibration in the database, including the calibration time, error data before and after calibration, etc., so as to analyze the performance change trend of the sensor later. This ensures the effectiveness of the sensor self-calibration and ensures that the sensor continuously and stably provides accurate data. If the self-calibration is successful, the sensor can continue to work normally and provide reliable data for fermentation process monitoring. If the self-calibration fails and re-self-calibration is performed, fermentation control errors caused by inaccurate sensor data can be avoided to ensure the smooth progress of the fermentation process.

[0075] S2. Extract the features of environmental data, spectral data and visual data to obtain the temperature and humidity time series characteristics, gas release rate, spectral characteristics and mycelium morphology characteristics, generate the mycelium growth rate according to the temperature and humidity time series characteristics, spectral characteristics and mycelium morphology characteristics, determine the bacterial community metabolic intensity according to the gas release rate and mycelium morphology characteristics, and predict the production rate of the target product based on the spectral characteristics.

[0076] Mycelial morphological characteristics include mycelial density and mycelial coverage area;

[0077] like Figure 2 As shown in the figure, the extraction logic of mycelium morphological features includes:

[0078] The hyphae morphology image is preprocessed, and the hyphae region in the hyphae morphology image is segmented according to the grayscale threshold;

[0079] The segmented hyphae regions are skeletonized by a thinning algorithm to obtain a hyphae skeleton;

[0080] Extract the topological structure of hyphae skeleton, including the number of branch points, the number of endpoints and the average length of hyphae branches;

[0081] The hyphae density and hyphae coverage area were determined according to the topological structure of the hyphae skeleton.

[0082] The original mycelium morphology image may have problems such as noise and uneven illumination, which will affect the accuracy of subsequent feature extraction. Preprocessing removes noise and segments the mycelium area according to the grayscale threshold, which can highlight the mycelium morphology and facilitate subsequent analysis. The mycelium morphology image is denoised by the Gaussian filtering algorithm. The Gaussian filtering algorithm effectively smoothes the image and reduces noise interference by weighted averaging the neighborhood of each pixel in the mycelium morphology image. Then, the appropriate grayscale threshold is automatically calculated by the Otsu algorithm. The Otsu algorithm is based on the grayscale histogram of the image and determines the optimal grayscale threshold by maximizing the inter-class variance. The part of the mycelium morphology image with a grayscale value greater than or equal to the grayscale threshold is identified as the mycelium area, and the part less than the grayscale threshold is removed as the background, thereby achieving accurate segmentation of the mycelium area. The image quality is improved, the mycelium area is accurately separated, and a clear image basis is provided for subsequent feature extraction. The segmented mycelium area provides an accurate object for skeletonization processing, ensuring that the subsequent refinement algorithm can accurately act on the mycelium and avoid background interference.

[0083] The appearance of mycelium is relatively complex. The mycelium skeleton obtained through skeletonization processing can simplify the shape of mycelium, more clearly show the growth structure and branching direction of mycelium, and facilitate the extraction of topological structure features. Through the morphological-based refinement algorithm, by continuously deleting edge pixels and retaining skeleton pixels, the mycelium area is gradually refined into a skeleton with a single pixel width. During the processing, the edge pixels of the mycelium area are iteratively deleted according to specific deletion rules until they can no longer be deleted, and finally a complete mycelium skeleton is obtained; thereby simplifying the shape of mycelium, highlighting the growth structure and branching characteristics of mycelium, and facilitating the extraction of topological structure. The generated mycelium skeleton is the basis for extracting topological structure features. An accurate mycelium skeleton can ensure that the extracted topological features such as the number of branch points and the number of endpoints are accurate and reliable.

[0084] The topological structure characteristics of the hyphae skeleton can reflect the growth pattern and complexity of the hyphae, and are of great significance for analyzing the growth status and metabolism of the hyphae. By traversing each pixel point of the hyphae skeleton, the branch points and endpoints are identified by judging the pixel situation in the neighborhood of the pixel point. For example, when there are three or more pixels connected to a pixel point in its neighborhood, the pixel point is identified as a branch point; when there is only one pixel connected to a pixel point in its neighborhood, the pixel point is identified as an endpoint. The average length of the hyphae branch can be obtained by tracing the path length from the branch point to the endpoint and averaging the lengths of all branches. In this way, the key topological features that can reflect the growth status of the hyphae are obtained, providing a quantitative basis for in-depth analysis of the hyphae growth. The extracted topological structure characteristics are an important basis for determining the hyphae density and hyphae coverage area, which directly affects the subsequent comprehensive evaluation of the hyphae growth status.

[0085] Mycelium density and mycelium coverage area are important indicators for measuring the growth scale and vitality of mycelium, and are of great value for evaluating the growth status of mycelium and the fermentation effect during the fermentation process. The mycelium density is determined by calculating the ratio of the number of pixels of the mycelium skeleton to the total area of ​​the mycelium morphological image. For the mycelium coverage area, the segmented mycelium area is first binarized, and then the number of pixels in the mycelium area in the binarized image is counted. Then, based on the conversion relationship between the pixels of the mycelium morphological image and the actual area, the number of pixels is converted into the actual coverage area. This quantifies the growth scale and vitality of mycelium and provides intuitive quantitative indicators for evaluating the fermentation effect. Mycelium density and mycelium coverage area, as important components of mycelium morphological characteristics, provide key data support for determining the metabolic intensity of the bacterial community and analyzing the mycelium growth rate.

[0086] like Figure 3 As shown in the figure, the generation logic of mycelium growth rate includes:

[0087] The temperature and humidity time series characteristics, spectral characteristics and hyphae morphological characteristics at different time points were integrated and time-stamped and aligned;

[0088] Select characteristic points on the mycelium skeleton and record the temperature and humidity time series characteristics and spectral characteristics of each characteristic point;

[0089] The displacement of the feature point between two adjacent frames of hyphae morphology images is calculated to determine the local growth rate of the feature point in the time period, and the changes in the temperature and humidity time series characteristics and spectral characteristics in the time period are analyzed to determine the weight of the feature point;

[0090] The mycelium growth rate is obtained by taking the weighted average of the local growth rates of all feature points.

[0091] The temperature and humidity time series characteristics, spectral characteristics and mycelium morphological characteristics reflect the fermentation environment and mycelium growth from different angles. Integrating them and aligning the time marks can comprehensively consider the impact of various factors on mycelium growth and ensure the accuracy of subsequent analysis. The temperature and humidity time series data, spectral data and mycelium morphological characteristic data obtained at different time points are stored in chronological order. Through timestamp matching, it is ensured that different types of data at the same time point can be accurately associated. For example, each time data is obtained, a timestamp accurate to the second is recorded. When the data is fused, the corresponding data is integrated according to the timestamp. This provides a comprehensive and time-synchronized data foundation, creating conditions for accurately analyzing the relationship between mycelium growth rate and various factors. The integrated data makes it convenient to select feature points on the mycelium skeleton, and can simultaneously obtain multiple aspects of information corresponding to the feature point, providing rich data support for subsequent analysis.

[0092] Selecting representative feature points on the mycelium skeleton and recording their related temperature, humidity and spectral characteristics will help analyze the local growth of mycelium under different environmental conditions, and then determine the weight of each factor affecting mycelium growth; evenly selecting multiple feature points on the mycelium skeleton, the interval can be determined according to the mycelium density and image resolution, for example, one feature point is selected every 10 pixels, for each selected feature point, find the temperature and humidity time series characteristics and spectral characteristics of the corresponding time point, and associate this information with the feature point for storage; thereby obtaining multi-feature information of feature points at different positions, providing specific data samples for analyzing the local growth rate and the weight of each factor, which directly affects the accuracy of subsequent mycelium growth rate calculations.

[0093] The local growth rate is obtained by calculating the displacement of the feature points between two adjacent frames of images, and the weights are determined by analyzing the changes in temperature, humidity and spectral characteristics. This can comprehensively consider the impact of environmental factors on the local growth of mycelium, making the calculated mycelium growth rate more in line with the actual situation. By comparing the coordinates of the feature points in two adjacent mycelium morphological images, the displacement of the feature points is calculated using the Euclidean distance formula. For example, the coordinates of the feature points in the first frame of the image are , the coordinates in the second frame image are , then the displacement of the feature point in two adjacent frames is , combined with the time interval between the acquisition of two frames of images, the moving distance of the feature point in unit time, that is, the local growth rate, is obtained. For example, a feature point moves 10 pixels between images collected in two consecutive hours. It is known that the conversion ratio of image pixels to actual length is 1 pixel = 0.1 mm, and the time interval is 2 hours. The local growth rate at the feature point is 0.5 mm / h.

[0094] As for weight determination, for the temperature and humidity time series characteristics, the temperature and humidity value of each feature point at the first time point is taken as the benchmark, and the temperature and humidity values ​​of the subsequent time points are divided by the benchmark value to obtain a dimensionless temperature and humidity feature sequence, and the spectral characteristics are processed synchronously to eliminate the dimension differences of the data, so that different types of data can be analyzed at the same scale, improving the accuracy and reliability of grey correlation analysis; the local growth rate data sequence is set as the reference sequence, and the temperature and humidity feature sequence and the spectral feature sequence are set as the comparison sequence. The absolute difference between the reference sequence and the comparison sequence of each feature point at each time point is calculated, and then the minimum difference and the maximum difference of the two levels are determined. Finally, the correlation coefficient is calculated according to the grey correlation coefficient formula to quantify the degree of correlation between each factor and the local growth rate of each feature point at different time points, providing specific data support for the subsequent comprehensive calculation of the correlation degree.

[0095] The grey correlation coefficient of each feature point is arithmetic averaged to obtain the correlation between the temperature and humidity time series characteristics and the spectral characteristics of the feature point and the local growth rate. For example, for a certain feature point, the correlation between its temperature and humidity time series characteristics and the local growth rate and the correlation between its spectral characteristics and the local growth rate are calculated, so as to comprehensively consider the correlation at each time point and obtain a quantitative index that can reflect the degree of influence of each factor on the local growth of mycelium, providing a clear basis for weight determination; the higher the correlation, the greater the influence of the factor on the local growth of mycelium, so giving it a higher weight can more reasonably comprehensively consider the influence of various factors on the mycelium growth rate. , so that the calculated mycelium growth rate is more in line with the actual situation; for each feature point, the correlation between the temperature and humidity time series characteristics and the spectral characteristics is normalized to obtain the weight of each factor, so as to reasonably allocate the weight of each factor, so that the influence of each factor can be fully considered when calculating the mycelium growth rate, and the accuracy and scientificity of the mycelium growth rate calculation can be improved; thereby, the environmental factors are comprehensively considered, the local growth rate is accurately calculated, and the weight of each factor is determined, so that the calculation of the mycelium growth rate is more scientific. The obtained local growth rate and the weight of each factor are the key data for calculating the overall mycelium growth rate, which directly determines the accuracy of the final mycelium growth rate.

[0096] The mycelium growth rate is obtained by weighted averaging the local growth rates of all feature points, which can comprehensively reflect the overall growth of the mycelium and provide a quantitative comprehensive indicator for evaluating the growth status of the mycelium during the fermentation process. The mycelium growth rate is calculated using the weighted average formula based on the local growth rates of each feature point determined previously and the corresponding weights. This provides a quantitative indicator that can comprehensively reflect the overall growth of the mycelium, facilitates intuitive evaluation of the mycelium growth status, and provides an important basis for determining the metabolic intensity of the flora and adjusting the fermentation conditions.

[0097] The logic for determining the metabolic intensity of the bacterial community includes:

[0098] Monitor the dynamic changes of carbon dioxide concentration and volatile metabolite concentration, and analyze the gas release rate;

[0099] The gas release rate and hyphae morphological characteristics were correlated and analyzed;

[0100] The metabolic intensity of the bacterial community was obtained by weighted quantification based on the release rate of carbon dioxide, the release rate of volatile metabolites and the change rate of mycelium coverage area.

[0101] The dynamic changes in carbon dioxide concentration and volatile metabolite concentration can directly reflect the metabolic activity of the bacterial community, and analyzing their release rate can quantify the metabolic activity of the bacterial community. Through gas sensors, such as carbon dioxide sensors and volatile metabolite sensors, the changes in gas concentration in the fermentation chamber are monitored in real time, and data is acquired at a certain time interval (for example, 10 minutes). The gas release rate is obtained by calculating the ratio of the difference between two adjacent data acquisitions and the time interval. For example, if the carbon dioxide concentration increases from 3% to 3.5% within 10 minutes, the carbon dioxide release rate is (3.5%-3%)÷10=0.05% / minute. This can reflect the intensity of the metabolic activity of the bacterial community in real time and quantitatively, providing key data for subsequent analysis. Accurate gas release rate data is the basis for correlation analysis with mycelial morphological characteristics, which affects the accuracy of determining the metabolic intensity of the bacterial community.

[0102] Microbial metabolism is closely related to mycelial growth. Correlating the gas release rate with mycelial morphological characteristics for analysis can provide a more comprehensive understanding of the actual situation of microbial metabolism and provide a basis for accurately determining the metabolic intensity of the microbial community. Through correlation analysis methods, such as Pearson correlation coefficient analysis, the correlation between the carbon dioxide release rate and the volatile metabolite release rate and mycelial morphological characteristics such as mycelial density and mycelial coverage area is calculated. Through analysis, it is found that when the mycelial coverage area increases, the carbon dioxide release rate usually increases, indicating that there is a positive correlation between the two. This reveals the intrinsic connection between microbial metabolism and mycelial growth, providing a more comprehensive perspective for quantifying the metabolic intensity of the microbial community. The results of the correlation analysis provide a reference for determining the weight of each factor in quantifying the metabolic intensity of the microbial community, which affects the accuracy of the final calculation of the metabolic intensity of the microbial community.

[0103] By weighted quantification of the carbon dioxide release rate, volatile metabolite release rate, and mycelium coverage area change rate, multiple factors can be comprehensively considered to obtain a quantitative index that can accurately reflect the metabolic intensity of the flora; the weight of each factor is determined based on the correlation results obtained from the association analysis. For example, after analysis, it is found that the carbon dioxide release rate has a greater impact on the metabolic intensity of the flora, and it is given a weight of 0.4; the volatile metabolite release rate has a weight of 0.3; and the mycelium coverage area change rate has a weight of 0.3. Then, the flora metabolic intensity is calculated by weighting; thus, a comprehensive and quantitative flora metabolic intensity index is provided, which intuitively reflects the metabolic activity of the flora during the fermentation process, and provides an important basis for judging whether the fermentation state is normal and whether the fermentation conditions need to be adjusted.

[0104] The prediction logic of the target product generation rate includes:

[0105] Screening the bands of spectral characteristics related to the target product;

[0106] The area of ​​the band is mapped to the substrate consumption rate, and the production rate of the output target product is predicted through a neural network in combination with environmental data;

[0107] Monitor the prediction error of the neural network to determine whether to retrain the neural network.

[0108] Spectral data contains a lot of information, but not all bands are related to the generation of target products. Screening out the bands of spectral characteristics related to the target products can reduce data redundancy and improve the accuracy and efficiency of prediction. The spectral data are analyzed by the least squares discriminant analysis method. This method finds out the spectral bands that have the greatest impact on the change of target product concentration by establishing a regression model between spectral data and target product concentration. For example, when analyzing the spectral data of the fermentation process of dandelion golden camellia, it was found that the 700-800nm ​​and 1200-1300nm bands are closely related to the generation of target products. This reduces the amount of data processing, highlights the key information related to the generation of target products, and improves the accuracy and efficiency of the prediction model. The screened spectral characteristic bands are the basis for subsequent mapping of substrate consumption rate and predicting target product generation rate, which directly affects the accuracy of prediction.

[0109] The area of ​​the band of spectral characteristics is mapped to the substrate consumption rate, and the generation rate of the target product is predicted by the neural network in combination with the environmental data. This can comprehensively consider the influence of spectral information and environmental factors on the generation of the target product, and realize the effective prediction of the generation rate of the target product. Through experiments, it is found that there is a linear relationship between the area of ​​a specific spectral band and the substrate consumption rate, which can be fitted using a linear regression equation. The obtained substrate consumption rate and environmental data (such as temperature, humidity and oxygen concentration, etc.) are used as inputs and input into a pre-trained neural network model. The neural network model can use a multi-layer perceptron. The model is trained with a large amount of experimental data so that it can learn the complex relationship between the input data and the generation rate of the target product, thereby predicting the generation rate of the output target product. In this way, a variety of factors are comprehensively considered to realize the effective prediction of the generation rate of the target product, providing a basis for the control and optimization of the fermentation process. The predicted target product generation rate is an important reference for judging the fermentation effect and adjusting the fermentation conditions. At the same time, the prediction error is also the basis for judging whether the neural network needs to be retrained.

[0110] As the fermentation process proceeds, the environment and fermentation conditions will change, resulting in a decrease in the prediction accuracy of the neural network. By monitoring the prediction error, the inaccuracy of the neural network model can be discovered in a timely manner, and the neural network can be retrained to ensure the accuracy of the prediction. After each prediction of the generation rate of the target product, the predicted value is compared with the actual measured value, and the prediction error is calculated. For example, the mean square error is used as an error indicator. When the prediction error is greater than a preset threshold (such as 5%), new experimental data is collected, the neural network is retrained, and the parameters of the neural network model are updated to improve the accuracy of the prediction. This ensures the accuracy of the prediction of the generation rate of the target product, so that the prediction results can better guide the control and optimization of the fermentation process. Accurate prediction results provide a reliable basis for adjusting fermentation conditions and optimizing fermentation processes, which helps to improve fermentation efficiency and product quality.

[0111] S3. Generate fermentation stage labels and confidence levels based on mycelium growth rate and bacterial metabolic intensity, and generate a control instruction set based on reinforcement learning in combination with the target product production rate.

[0112] The logic for generating fermentation stage labels and confidence levels includes:

[0113] The mycelium growth rate, bacterial community metabolic intensity and historical stage labels are integrated to obtain integrated data, and cluster analysis is performed on the integrated data based on a clustering algorithm to obtain clustering results;

[0114] Based on fuzzy logic, the mycelium growth rate and bacterial metabolic intensity were divided into fuzzy linguistic variables, and the fermentation stage labels of different fuzzy linguistic variable combinations were determined based on the clustering results.

[0115] The confidence of each fermentation stage label is calculated, and the fermentation stage label is corrected according to the confidence.

[0116] The mycelium growth rate, bacterial metabolic intensity, and historical stage labels each contain part of the information about the fermentation process. Integrating these data and performing cluster analysis can mine data features from multiple dimensions, discover potential patterns in different fermentation stages, and provide a basis for accurately dividing the fermentation stages. The mycelium growth rate and bacterial metabolic intensity acquired in real time are integrated with the historical stage label data stored in the database. For example, the mycelium growth rate and bacterial metabolic intensity are recorded every 2 hours, and the corresponding historical stage labels are associated. Cluster analysis is performed using a density-based spatial clustering algorithm. First, a suitable neighborhood radius is set according to the data characteristics (for example, eps=0.5) and the minimum number of samples (for example, minPts=5). The density-based spatial clustering algorithm will scan the data points, divide the density-connected data points into the same cluster, and identify the outliers. The integrated data will be divided into multiple clusters through the density-based spatial clustering algorithm. Each cluster represents a fermentation state with similar characteristics. Valuable information is extracted from the multi-dimensional data, and the different states of the fermentation process are preliminarily classified, laying the foundation for the subsequent accurate determination of the fermentation stage labels. The clustering results provide a reference framework for the fuzzy logic to divide the fermentation stage labels, making the application of fuzzy logic more targeted and reasonable.

[0117] The changes in mycelium growth rate and bacterial metabolic intensity during the fermentation process are continuous, and it is difficult to divide the fermentation stage with precise numerical boundaries. Fuzzy logic can convert these continuously changing variables into fuzzy language variables, which is more in line with the fuzzy characteristics of stage division in the fermentation process. Combined with the clustering results, the fermentation stage labels can be determined more accurately; the fermentation stage labels include germination period, growth period, stable period and decline period, and the mycelium growth rate is divided into three fuzzy language variables of "low", "medium" and "high", and the bacterial metabolic intensity is also divided into three fuzzy language variables of "weak", "medium" and "strong". For example, when the mycelium growth rate is less than a certain threshold (such as 0.1mm / h), it is judged as "low", and in the threshold interval (0.1-0.5mm / h ) is “medium”, and greater than 0.5 mm / h is “high”. The metabolic intensity of the bacterial community is comprehensively judged according to indicators such as the carbon dioxide release rate and the volatile metabolite release rate. Combined with the clustering results, a fuzzy rule table is formulated. For example, when the mycelium growth rate is “high” and the metabolic intensity of the bacterial community is “strong”, and the clustering results show that the characteristics of this group of data are most similar to the “growth period” in the historical data, the fermentation stage label of this stage is determined to be “growth period”; thereby fully considering the fuzziness of the variables in the fermentation process, being more in line with the actual fermentation situation, and improving the accuracy and rationality of the fermentation stage division. The clear fermentation stage labels provide a basis for calculating confidence and correcting labels, and also provide key information for the subsequent determination of regulation targets.

[0118] The fermentation process is affected by many factors and there is uncertainty. Calculating confidence can quantify the reliability of the fermentation stage label judgment. When the confidence is low, the label can be corrected to improve the accuracy of the label and provide a more reliable basis for subsequent regulation. Confidence is calculated through evidence theory. For example, the fermentation stage label determined by fuzzy logic is used as evidence. The confidence of each label is calculated according to the degree of conflict and support between different evidences. Assuming that the confidence of the "growth period" label is 0.75 (less than 0.85) through calculation, the label needs to be corrected at this time. By re-examining the changing trends of mycelium growth rate and bacterial metabolic intensity, and comparing with historical data, if it is found that the bacterial metabolic intensity is in the "strong" range, but has a downward trend, combined with the clustering results that are similar to some characteristics of the "stable period", the fermentation stage label is corrected to "stable period"; thereby improving the reliability of the fermentation stage label, reducing the erroneous judgment caused by uncertainty, and making subsequent label-based regulatory decisions more scientific. The corrected fermentation stage label and confidence provide more accurate input for the target determination stage, affecting the determination of regulatory targets and the formulation of regulatory strategies.

[0119] The generation logic of the control instruction set includes:

[0120] Construct a hierarchical reinforcement learning architecture, which includes a target determination phase and a regulation generation phase;

[0121] Determine the reward function based on the target product generation rate, energy consumption and pollution risk;

[0122] Monitor the operating data during the fermentation process and dynamically adjust the control generation stage and reward function.

[0123] The fermentation process is complex, involving multiple goals and multiple control methods. The hierarchical reinforcement learning architecture can decompose complex control tasks into sub-tasks at different levels, making the learning process more efficient, facilitating the separate processing of target determination and control generation problems, and improving the generation quality of the control instruction set. Through the two-layer architecture, the upper layer is the target determination stage, and the lower layer is the control generation stage. In the target determination stage, the control target is determined according to the fermentation stage label, confidence and target product generation rate. In the control generation stage, specific control instructions are generated based on the determined control target and the state space and action space. The hierarchical architecture is implemented through programming, and the framework is built using Python's reinforcement learning library to define the input, output and processing logic of different stages. In this way, complex control problems are decomposed, the efficiency and accuracy of reinforcement learning are improved, and the generation of control instruction sets is more targeted and systematic. A structured framework is provided for the subsequent determination of reward functions, monitoring of operating data and dynamic adjustment, ensuring the orderly progress of the control instruction set generation process.

[0124] The reward function is the core of reinforcement learning. It is used to guide the intelligent agent (the control system in the application scenario of this embodiment) to learn the optimal strategy. The reward function is determined according to the generation rate, energy consumption and pollution risk of the target product. It enables the control system to take into account energy consumption and pollution control while pursuing the generation of the target product, thereby maximizing the overall benefit. The reward function is defined as ,in, is the rate of formation of the target product, For energy consumption, For pollution risk, , and is the weight coefficient, which is determined through experiments and data analysis, such as , and The higher the generation rate of the target product, the higher the reward; the higher the energy consumption and pollution risk, the lower the reward. For example, when the generation rate of the target product increases by 10%, the reward increases by 0.6; when the energy consumption increases by 10%, the reward decreases by 0.2; thereby guiding the control system to balance among multiple objectives, optimize the fermentation process, improve production efficiency, reduce costs and environmental impacts. The reward function determines the learning direction of the control system, affects the utilization of the state space and action space in the control generation stage, and the generation of the final control instruction set.

[0125] The fermentation process is dynamic, and the initially set control strategy will not be able to adapt to various changes in the process. Monitoring the operating data during the fermentation process and dynamically adjusting the control generation stage and reward function can enable the control system to respond to changes in the fermentation process in real time and maintain the effectiveness and adaptability of the control. Obtain the operating data of the fermentation process at regular intervals (such as 30 minutes), including temperature, humidity, carbon dioxide concentration, mycelium density, bacterial metabolic intensity, and target product generation rate, and analyze the changing trends of these data. If it is found that the generation rate of the target product is lower than expected for a long time and the energy consumption is too high, it indicates that the current control strategy is unreasonable. At this time, adjust the weight coefficient of the reward function, such as appropriately increasing ,reduce , to encourage the control system to pay more attention to improving the rate of target product generation, and at the same time re-evaluate the state space and action space according to data changes. For example, if it is found that temperature changes have a greater impact on the generation of target products, the quantization interval of temperature in the state space can be appropriately reduced to improve the accuracy of temperature control; so that the control system can adapt to the dynamic changes of the fermentation process, maintain a good control effect, and improve the quality and output of fermentation products. The dynamically adjusted control generation stage and reward function provide more realistic parameters and conditions for the target determination stage and the generation of control instruction sets, ensuring the accuracy and effectiveness of control instructions.

[0126] The target determination stage is used to determine the regulatory target based on the fermentation stage label, confidence level, and the production rate of the target product;

[0127] The control generation phase is used to determine the state space and action space;

[0128] The state space includes temperature, humidity, carbon dioxide concentration, mycelium density, bacterial metabolic intensity, and the production rate of target products;

[0129] The action space includes heating power, ventilation valve opening and humidification amount.

[0130] The fermentation stage label, confidence and target product generation rate reflect the current fermentation status and expected goals. The control target is determined by combining this information, which can make the control strategy targeted and meet the needs of different fermentation stages. When the fermentation stage label is "growth period" and the confidence is high (such as 0.9), and the target product generation rate is lower than expected, the control target is determined to be "increasing the target product generation rate while ensuring the stability of the bacterial community metabolic intensity". If the fermentation stage label is "stable period", but the target product generation rate begins to decline, and the confidence is 0.8, the control target can be set to "maintain the current target product generation rate, control energy consumption and pollution risks". By writing a logical judgment program, the corresponding control target is determined according to different combinations of fermentation stage labels, confidence and target product generation rates, and stored in the target library of the control system. This clarifies the specific control direction for the control system, makes the control more targeted, and improves the controllability of the fermentation process. The determined control target provides guidance for the control generation stage, affecting the selection of state space and action space and the generation of control instructions.

[0131] Clarifying the state space and action space is the basis for reinforcement learning to generate control instructions. The state space covers the key factors that affect the fermentation process, and the action space includes operations that can be used to control the fermentation process. Reasonable determination of these two spaces can enable the control system to find the optimal control strategy within a feasible range; the state space includes temperature (setting range is 20-35℃, quantization interval is 1℃), humidity (setting range is 50%-80%, quantization interval is 5%), carbon dioxide concentration (setting range is 0.1%-0.5%, quantization interval is 0.05%), mycelium density, bacterial metabolic intensity and the generation rate of the target product; the action space includes heating power .... The setting range is 0-10kW, and the adjustment interval is 0.5kW), ventilation valve opening (the setting range is 0-100%, and the adjustment interval is 5%) and humidification amount (the setting range is 0-5L / h, and the adjustment interval is 0.5L / h), and these state and action information are stored in the control system in the form of data structure, which is convenient for reinforcement learning algorithm to process; thereby providing the control system with a clear control range and operation options, enabling reinforcement learning to search for the optimal control strategy within a reasonable range, and improving the accuracy and effectiveness of control. The determined state space and action space are the basis for the reinforcement learning algorithm to generate the control instruction set, which determines the specific form and value range of the control instruction.

[0132] S4. Execute the control instruction set, monitor the environmental data to adjust the control instruction set and abnormal protection, and synchronize to the cloud knowledge base.

[0133] The fermentation process is dynamically changing. Real-time acquisition and analysis of environmental data can timely grasp the fermentation situation, determine whether the current control instruction set is effective, and provide a basis for adjusting the instruction set. Through various sensors deployed in the fermentation warehouse, such as temperature and humidity sensors and gas sensors, environmental data are acquired every 15 minutes, including temperature, humidity, carbon dioxide concentration and volatile metabolite concentration, etc., and the changing trends of the acquired data are observed; thereby timely obtaining the latest information on the fermentation process, providing data support for precise control, and improving the monitoring ability of the fermentation process. Accurate real-time data and analysis results provide a basis for judging whether the control instruction set needs to be adjusted and how to adjust it.

[0134] Clarify the gap between the current fermentation status and the preset target, judge the control effect, and determine whether the control instruction set needs to be adjusted to ensure that the fermentation process develops in the expected direction; compare the real-time acquired and analyzed data with the preset fermentation target, which includes the appropriate fermentation temperature range, humidity range and expected value of the target product generation rate, and calculate the deviation between the actual data and the fermentation target value; thereby intuitively judging the execution effect of the control instruction set, timely discovering deviations in the fermentation process, and providing a quantitative basis for adjusting the control strategy. The evaluation results determine whether to start the adjustment process of the control instruction set. If there is a deviation, guide the subsequent adjustment steps.

[0135] According to the comparative evaluation results, the specific adjustment direction and amplitude are determined to make the control instruction set more in line with the fermentation needs and optimize the fermentation process. If the temperature is too high, the adjustment strategy is determined according to the deviation. If the temperature deviation is between +1 and +2°C, the heat entry can be reduced by reducing the ventilation valve opening by 5% and observing the temperature change. If the deviation is greater than +2°C, the heating power is reduced by 1kW and the ventilation valve opening is increased by 10%. The adjustment of parameters such as humidity and carbon dioxide concentration also follows a similar logic. According to the deviation range and change trend of different parameters, corresponding adjustment strategies are formulated. These strategies are pre-stored in the strategy library, and the specific adjustment plan is determined by calling the data in the strategy library. In this way, the control instructions are adjusted in a targeted manner to make the fermentation environment closer to the ideal state and improve the fermentation efficiency and product quality. The determined adjustment strategy provides specific content for generating a new control instruction set and is a key step in adjusting the control instruction set.

[0136] The adjustment strategy is converted into specific control instructions to achieve precise control of the fermentation equipment and ensure that the fermentation process proceeds according to the optimized plan; based on the determined adjustment strategy, a new set of control instructions is generated. For example, if the adjustment strategy is to reduce the heating power by 1kW, increase the ventilation valve opening by 10%, and increase the humidification volume by 0.5L / h, these adjustments are converted into control instructions that can be recognized by the equipment, and the instructions are transmitted to the control system of the fermentation equipment through the network. The control system adjusts the operating parameters of the heating equipment, ventilation equipment and humidification equipment according to the instructions; thereby achieving precise control of the fermentation equipment, effectively adjusting the fermentation environment, and ensuring the stability and efficiency of the fermentation process. The new set of control instructions is issued to the fermentation equipment for execution, and at the same time provides a new starting point for subsequent monitoring and re-evaluation.

[0137] Check whether the adjusted control instruction set achieves the expected effect, provide feedback for subsequent further optimization, and ensure the effectiveness and sustainability of the control; after the new control instruction set is executed for a period of time (for example, 30 minutes), obtain and analyze the environmental data again, compare the data before and after the adjustment, and evaluate whether the various parameters are close to the preset targets, such as observing whether the temperature is reduced to the target range after adjustment. If the expected effect is achieved, continue to monitor according to the current control instruction set. If the expected effect is not achieved, re-evaluate the deviation, adjust the strategy, and generate a new control instruction set again to form a closed-loop feedback adjustment mechanism; thereby continuously optimizing the control instruction set, improving the stability and controllability of the fermentation process, and ensuring the consistency of product quality. The effect verification results determine whether the control instruction set needs to be adjusted again, which affects the control direction and rhythm of the entire fermentation process.

[0138] Clarify the parameter range of normal fermentation, provide judgment criteria for timely detection of abnormal situations, and ensure that measures can be taken quickly when abnormalities occur to avoid serious impacts on the fermentation process and product quality; set reasonable abnormal thresholds for each monitoring parameter based on a large amount of experimental data and production experience, such as the upper limit of temperature is set to 35°C and the lower limit is set to 20°C; the upper limit of humidity is set to 80% and the lower limit is set to 50%; the upper limit of carbon dioxide concentration is set to 0.6% and the lower limit is set to 0.05%. Store these abnormal thresholds in the database as a basis for abnormal judgment; thereby providing quantitative standards for abnormal monitoring, improving the timeliness and accuracy of abnormal discovery, and ensuring the safety and stability of the fermentation process. The abnormal threshold is the basis of abnormal detection and determines whether the subsequent abnormal handling process is started.

[0139] Monitor various parameters in the fermentation process in real time, and promptly detect abnormal situations beyond the normal range, providing guarantees for rapid response and handling of abnormalities; in the process of data acquisition and transmission, compare the acquired environmental data with the preset abnormal threshold in real time. Once a parameter is found to be beyond the threshold range, the abnormal alarm mechanism is immediately triggered. For example, when the temperature sensor detects that the temperature reaches 36°C, a high temperature alarm signal is issued; thereby capturing abnormal situations in a timely manner, buying time for subsequent processing, reducing the harm of abnormalities to the fermentation process, and reducing production losses.

[0140] Accurately determine the cause of the abnormality and provide a basis for taking targeted solutions to ensure that the abnormality can be effectively eliminated and normal fermentation can be restored. When the abnormal alarm is triggered, automatically collect relevant data within a period of time before and after the abnormality occurs, including environmental data and equipment operation data, and analyze the collected data. If the temperature rises abnormally, it is determined that the temperature abnormality is caused by a failure of the heating equipment or poor ventilation. In this way, the cause of the abnormality can be quickly and accurately determined, the pertinence and efficiency of abnormal handling can be improved, the time for abnormal handling can be reduced, and the impact on production can be reduced.

[0141] According to the cause of the abnormality, take corresponding measures to eliminate the abnormality, restore the normal fermentation process, and ensure product quality and production safety; if the cause of the abnormality is that the temperature is too high due to a failure of the heating equipment, stop the operation of the heating equipment immediately, start the backup cooling equipment to lower the temperature, and send a fault notification to the maintenance personnel at the same time, informing them of the type and location of the fault so that timely maintenance can be carried out. If the abnormality is caused by poor ventilation, automatically increase the opening of the ventilation valve, and check whether the ventilation duct is blocked. If there is any blockage, clean it up in time; thereby effectively solving the abnormal problem, restoring the stability of the fermentation environment, reducing the impact of the abnormality on product quality, and ensuring the continuity of production. The result of the abnormal handling affects whether the fermentation process can return to normal. If the handling is successful, normal fermentation monitoring will continue. If the handling fails, it is necessary to further adjust the handling measures or suspend the fermentation for a comprehensive inspection.

[0142] Record the detailed information and handling process of the abnormality to provide data support for subsequent analysis and improvement, avoid similar abnormalities from happening again, and improve the reliability and stability of the fermentation process; after the abnormality is handled, record the time of the abnormality, abnormal parameters, abnormal causes, handling measures and handling results in detail in the cloud knowledge base, and regularly summarize and analyze the abnormal records to find out potential risk points and weak links. For example, if a batch frequently has humidity abnormalities caused by humidity sensor failure, the maintenance and replacement cycle of the humidity sensor can be optimized; thereby accumulating experience in abnormality handling, providing a basis for optimizing fermentation technology and equipment maintenance, and improving the stability and reliability of the entire fermentation process.

Claims

1. An intelligent monitoring method for the fermentation process of dandelion golden camellia based on the Internet of Things, characterized in that: include: Deploy a sensor array to obtain environmental data, spectral data, and visual data in the fermentation chamber, and self-calibrate the sensor array; Extract the features of environmental data, spectral data and visual data to obtain the temperature and humidity time series characteristics, gas release rate, spectral characteristics and mycelium morphology characteristics, generate the mycelium growth rate based on the temperature and humidity time series characteristics, spectral characteristics and mycelium morphology characteristics, determine the bacterial community metabolic intensity based on the gas release rate and mycelium morphology characteristics, and predict the generation rate of the target product based on the spectral characteristics; Generate fermentation stage labels and confidence levels based on mycelium growth rate and bacterial metabolic intensity, and generate control instruction sets based on reinforcement learning in combination with the target product production rate; Execute the control instruction set, monitor the environmental data to adjust the control instruction set and abnormal protection, and synchronize it to the cloud knowledge base.

2. The intelligent monitoring method for the fermentation process of Camellia chrysantha based on Internet of Things as claimed in claim 1, characterized in that: The hyphae morphological characteristics include hyphae density and hyphae coverage area; The extraction logic of mycelium morphological features includes: The hyphae morphology image is preprocessed, and the hyphae region in the hyphae morphology image is segmented according to the grayscale threshold; The segmented hyphae regions are skeletonized by a thinning algorithm to obtain a hyphae skeleton; Extract the topological structure of hyphae skeleton, including the number of branch points, the number of endpoints and the average length of hyphae branches; The hyphae density and hyphae coverage area were determined according to the topological structure of the hyphae skeleton.

3. The intelligent monitoring method for the fermentation process of Camellia chrysantha based on Internet of Things as claimed in claim 2, characterized in that: The generation logic of the hyphae growth rate includes: The temperature and humidity time series characteristics, spectral characteristics and hyphae morphological characteristics at different time points were integrated and time-stamped and aligned; Select characteristic points on the mycelium skeleton and record the temperature and humidity time series characteristics and spectral characteristics of each characteristic point; The displacement of the feature point between two adjacent frames of hyphae morphology images is calculated to determine the local growth rate of the feature point in the time period, and the changes in the temperature and humidity time series characteristics and spectral characteristics in the time period are analyzed to determine the weight of the feature point; The mycelium growth rate is obtained by taking the weighted average of the local growth rates of all feature points.

4. The intelligent monitoring method for the fermentation process of Camellia chrysantha based on Internet of Things as claimed in claim 3, characterized in that: The logic for determining the bacterial flora metabolic intensity includes: Monitor the dynamic changes of carbon dioxide concentration and volatile metabolite concentration, and analyze the gas release rate; The gas release rate and hyphae morphological characteristics were correlated and analyzed; The metabolic intensity of the bacterial community was obtained by weighted quantification based on the release rate of carbon dioxide, the release rate of volatile metabolites and the change rate of mycelium coverage area.

5. The intelligent monitoring method for the fermentation process of Camellia chrysantha based on Internet of Things as claimed in claim 4, characterized in that: The prediction logic of the generation rate of the target product includes: Screening the bands of spectral characteristics related to the target product; The area of ​​the band is mapped to the substrate consumption rate, and the production rate of the output target product is predicted through a neural network in combination with environmental data; Monitor the prediction error of the neural network to determine whether to retrain the neural network.

6. The intelligent monitoring method for the fermentation process of Camellia chrysantha based on Internet of Things as claimed in claim 5, characterized in that: The generation logic of the fermentation stage label and confidence includes: The mycelium growth rate, bacterial community metabolic intensity and historical stage labels are integrated to obtain integrated data, and cluster analysis is performed on the integrated data based on a clustering algorithm to obtain clustering results; Based on fuzzy logic, the mycelium growth rate and bacterial metabolic intensity were divided into fuzzy linguistic variables, and the fermentation stage labels of different fuzzy linguistic variable combinations were determined based on the clustering results. The confidence of each fermentation stage label is calculated, and the fermentation stage label is corrected according to the confidence.

7. The intelligent monitoring method for the fermentation process of Camellia chrysantha based on Internet of Things as claimed in claim 6, characterized in that: The generation logic of the control instruction set includes: Construct a hierarchical reinforcement learning architecture, which includes a target determination phase and a regulation generation phase; Determine the reward function based on the target product generation rate, energy consumption and pollution risk; Monitor the operating data during the fermentation process and dynamically adjust the control generation stage and reward function.

8. The method for intelligent monitoring of the fermentation process of Camellia chrysantha based on Internet of Things according to claim 7, characterized in that: The target determination stage is used to determine the regulation target according to the fermentation stage label, confidence and the generation rate of the target product; The control generation phase is used to determine the state space and action space; The state space includes temperature, humidity, carbon dioxide concentration, mycelium density, bacterial metabolic intensity, and the production rate of target products; The action space includes heating power, ventilation valve opening and humidification amount.

9. The method for intelligent monitoring of the fermentation process of Camellia chrysantha based on Internet of Things as claimed in claim 8, characterized in that: The environmental data include temperature, humidity, carbon dioxide concentration and volatile metabolite concentration, the spectral data are used to analyze the bacterial community metabolic intensity and the production rate of the target product, and the visual data include hyphae morphology images to capture the hyphae growth morphology; The deployment logic of the sensor array includes: Divide the fermentation tank into regional blocks according to its structural functions; In each area block, a sensor array is deployed in a layered and three-dimensional manner. The sensor array includes temperature and humidity sensors, gas sensors, spectral sensors, and visual sensors. The sensor array is synchronously stimulated to obtain environmental data, spectral data and visual data in the fermentation chamber respectively.

10. The intelligent monitoring method for the fermentation process of Camellia chrysantha based on Internet of Things according to claim 9, characterized in that: The self-calibration logic of the sensor array includes: Determine self-calibration conditions according to changes in time and data, and configure standard data; During the self-calibration process, the parameters of the sensor array are adjusted according to the difference between the acquired data and the standard data; After adjustment, verify the effect of self-calibration to provide feedback on whether to re-calibrate.

Images