Mushroom residue fermentation batching control method and batching device based on AI dynamic proportioning

Abstract

The invention discloses a mushroom residue fermentation batching control method based on AI dynamic proportioning and a batching device, and relates to the technical field of biological fermentation. Comprising the following steps: S1, multi-modal data acquisition: acquiring cellulose / lignin spatial distribution, decay characteristics and humidity gradient through a near-infrared spectrometer, a metal oxide semiconductor gas sensor array and a microwave resonance moisture sensor; s2, parameter determination: determining optimal parameters according to a target threshold value, the three-dimensional characteristics of the original mushroom residue stockpile and a time sequence result; and S3, adjusting and controlling: according to the optimal solution, adjusting the fermentation temperature, the auxiliary material addition and the aeration rate in real time. According to the method, the spatial features and the time series data are combined, digital twin modeling of the fermentation process is achieved, meanwhile, the optimal parameters are determined through the constructed decomposition degree-conversion rate-energy consumption three-dimensional target space in combination with the K-means clustering algorithm and B spline curvature optimization, and the fermentation parameter adjustment response time is shortened.

Description

Technical Field

[0001] The present invention relates to the field of biological fermentation technology, and in particular to a mushroom residue fermentation ingredient control method and ingredient proportioning device based on AI dynamic proportioning. Background Art

[0002] With the accelerated pace of agricultural modernization in my country, the edible fungus industry, as a key component of efficient ecological agriculture, has experienced rapid growth in recent years. However, large-scale production also generates a significant amount of waste—mushroom residue (also known as mushroom bag waste). According to incomplete statistics, the annual amount of mushroom residue generated in my country exceeds 10 million tons. This waste is rich in organic matter, cellulose, lignin, and small amounts of nutrients such as nitrogen, phosphorus, and potassium, making it a valuable resource for recycling.

[0003] Currently, the main treatment methods for mushroom residue include direct return to farmland, incineration for power generation, feed conversion, and composting. Composting is an environmentally friendly and resource-efficient method, transforming mushroom residue into a stable, harmless organic fertilizer suitable for farmland application. However, in practice, due to factors such as the high carbon-nitrogen ratio (C / N>30:1), simple structure, and unstable moisture content of mushroom residue, composting alone often suffers from slow fermentation start-up, high ammonia release, and long maturation cycles, seriously affecting its resource utilization efficiency and product quality.

[0004] Chinese invention patent publication number CN120108534A discloses a dynamic ingredient adjustment method, system, and storage medium based on online near-infrared detection. The method includes the following steps: obtaining the required nutrient content of the formulated feed based on the feed formula and the standard nutrient content of each feed raw material; detecting the actual nutrient content of each feed raw material in real time using a near-infrared sensor; optimizing and adjusting the actual ratio of each raw material in the feed formula based on the actual nutrient content of each raw material and the required nutrient content of the formulated feed; and obtaining the corresponding amount of raw materials for feed formulation based on the actual ratio of each raw material. This invention, based on near-infrared sensing technology, obtains the actual content of each nutrient in the feed raw materials in real time and adjusts the amount of each raw material in the feed formula in a timely manner. This ensures that the nutrient content of the formulated feed is consistent with the feed formula, avoiding raw material waste while ensuring that animals receive sufficient nutrients for growth.

[0005] In the process of resource utilization of mushroom residue, it can be used to produce a variety of products such as organic fertilizer, animal feed and biofuel. However, in the production of organic fertilizer, it focuses on the maturity of the material and the nutrient release characteristics. In the preparation of feed base materials, it focuses on the improvement of protein content and the degradation of toxins. When used for biofuels, it focuses on cellulose conversion rate and calorific value. Therefore, in different application scenarios, there are obvious differences between the physical and chemical properties of fermentation products. However, most existing batching systems use fixed ratios or manual adjustments based on experience, which cannot be dynamically optimized according to the target use, making it difficult for the same equipment to flexibly adapt to multiple application requirements. Summary of the Invention

[0006] The purpose of the present invention is to provide a mushroom residue fermentation ingredient control method and ingredient proportioning device based on AI dynamic proportioning to solve the problems raised in the above background technology.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for controlling the proportioning of mushroom residue fermentation ingredients based on AI dynamic proportioning, comprising: S1: Multimodal data acquisition: Acquisition of cellulose / lignin spatial distribution, spoilage characteristics, and moisture gradients using near-infrared spectrometers, metal oxide semiconductor gas sensor arrays, and microwave resonant moisture sensors; S2: Parameter determination: Based on the target threshold, the three-dimensional characteristics of the original mushroom residue pile and the time series results, the optimal parameters are determined, including: S2.1: Data preprocessing: Based on the spatial distribution, decay characteristics, and moisture gradient of cellulose / lignin, the 3D-CNN model and LSTM network model are used to determine the three-dimensional characteristics and time series results of the original mushroom residue pile; S2.2: Determine target weights: Based on the spatial distribution of the fermentation tanks, set the categories of the K-means clustering algorithm and obtain candidate solutions for each category. Simultaneously, set target performance using a usage-performance indicator mapping library. Based on the candidate solutions and the target performance, determine a comprehensive score for the candidate solutions. S2.3: Construct a three-dimensional target space: Use the normalized maturity prediction value as the X-axis, the normalized conversion rate value as the Y-axis, and the normalized energy consumption as the Z-axis to construct a three-dimensional target space; S2.4: Determine the optimal solution: Based on the three-dimensional target space and the comprehensive score, obtain the comprehensive score of the frontier solution, and determine the three largest comprehensive scores of the frontier solution. Simultaneously, obtain the curvature value of the comprehensive scores of the frontier solutions through B-spline fitting. The frontier solution corresponding to the largest curvature value is the optimal solution. S3: Adjustment and control: According to the optimal solution, the fermentation temperature, auxiliary material addition and aeration volume are adjusted in real time.

[0008] Furthermore, the spatial distribution of cellulose / lignin, decay characteristics and moisture gradients are obtained, including: S1.1: Near-infrared spectral scanning: A near-infrared spectral probe equipped with an automatic pan-tilt platform is used to scan the mushroom residue, simultaneously acquiring the azimuth and elevation angles of each spectral point. Based on the laser ranging data, the three-dimensional rectangular coordinates of the spectral point are determined, and a spatial distribution model is constructed. The grid data in the spatial distribution model is normalized, and a heat map is generated based on the RGB values ​​of different colors. S1.2: VOCs fingerprint detection: The gas in the fermentation tank is monitored by a metal oxide semiconductor gas sensor array, and a fingerprint database is constructed based on the monitored gas content; S1.3: Moisture Gradient Monitoring: Microwave resonant moisture sensors are installed at different areas on the sidewall of the fermenter to obtain the vertical moisture gradient difference and the middle layer moisture abnormality value in the fermenter. The vertical moisture gradient difference and the middle layer moisture abnormality value are compared with the preset gradient threshold range and the preset abnormality threshold, respectively. The fermentation status is determined based on the comparison results, specifically: When the vertical moisture gradient difference is within the preset gradient threshold range and the middle layer moisture abnormality value is not less than the preset abnormality threshold, the fermentation state is normal fermentation; otherwise, the fermentation state is abnormal fermentation.

[0009] Furthermore, the lignin content of the grid data in the thermal map is obtained, and the lignin content is compared with a preset content threshold. According to the comparison result, a marked voxel grid is determined, specifically: When the lignin content is greater than a preset content threshold, the corresponding voxel grid is marked; otherwise, the corresponding voxel grid is not marked; At the same time, the interconnected marked voxel grids are connected to determine the volume of the connected area, and the volume of the connected area is compared with a preset volume threshold. Based on the comparison result, the priority processing area is determined, specifically: When the volume of the connected area is greater than a preset volume threshold, the corresponding connected area is a priority processing area; otherwise, the corresponding connected area is not a priority processing area.

[0010] Furthermore, the three-dimensional characteristics and time series results of the original mushroom residue pile are determined, including: S2.1.1: Determine 3D structural features: Resample the priority regions using a trilinear interpolation algorithm and obtain normalized cellulose and lignin contents based on the resampled voxel grid. Simultaneously, output the 3D structural features using a 3D-CNN model. S2.1.2: Time series prediction: Based on the fingerprint database and fermentation status, the normalized gas content and normalized moisture content are obtained, and the predicted value of maturity and the remaining cellulose are obtained through the input and output of the LSTM network model.

[0011] Furthermore, according to the Z value of the three-dimensional target space, the data in the three-dimensional target space is divided into multiple layers of data, and the comprehensive score of the frontier solution of each layer of data is specifically: ; in: Comprehensive score for the frontier solution, is the scoring weight coefficient, For the comprehensive rating, For the frontier level, is the level attenuation factor.

[0012] Furthermore, the optimal total amount of auxiliary materials is determined based on the optimal auxiliary material ratio and the fermentation tank volume in the optimal solution, and the optimal total amount of auxiliary materials is divided into multiple small portions of auxiliary materials, and the small portions of auxiliary materials are added through a screw conveyor and a spray gun.

[0013] Furthermore, the aeration volume of each air path is divided according to the optimal aeration volume and air path composition in the optimal solution. At the same time, the average flow rate of the aeration valve is adjusted according to the flow rate, opening time and total duration of the aeration valve, specifically: ; in: is the period average flow rate, is the pulse frequency, is the maximum flow rate of the valve, For the single opening duration, is the total cycle duration.

[0014] Furthermore, when the fermentation temperature, auxiliary material addition and aeration volume are adjusted in real time, the probability of toxin accumulation inside the fermentation tank is obtained, and the cellulase and fermentation time are adjusted in real time according to the toxin accumulation probability, including: S3.1: Determine the probability of toxin accumulation: Discretize the fermentation tank into multiple discrete temperature intervals, discrete gas intervals, and discrete moisture intervals based on the temperature, gas concentration, and moisture content of the fermentation tank. Determine the probability of toxin accumulation based on the toxin probabilities corresponding to the discrete temperature intervals, discrete gas intervals, and discrete moisture intervals, specifically: ; in: is the final toxin accumulation probability, is the base conditional probability, is the weight of the i-th influencing factor, is the total number of influencing factors, is the index of influencing factors; S3.2: Determine the risk level: Compare the toxin accumulation probability with the preset probability threshold range and determine the risk level based on the comparison result; specifically: When the cumulative probability of toxins is less than the lower limit of the preset probability threshold range, the corresponding risk level is low risk; when the cumulative probability of toxins is within the preset probability threshold range, the corresponding risk level is medium risk; when the cumulative probability of toxins is greater than the upper limit of the preset probability threshold range, the corresponding risk level is high risk; S3.3: Determine first aid measures: When the risk level is medium risk, increase the cellulase content and increase the stirring speed. When the risk level is high risk, extend the fermentation time and inject lime water into the fermentation tank. At the same time, obtain the corresponding toxin accumulation probability in real time, and repeat steps S3.1-S3.3 according to the toxin accumulation probability until the risk level is low risk.

[0015] Furthermore, the cellulase dosage and fermentation extension time are determined based on the difference between the toxin accumulation probability and the lower limit threshold of the preset probability threshold range, specifically: ; in: is the cellulase dosage, As the basic dosage, is the dose-probability coefficient, is the intervention threshold, is the final toxin accumulation probability, To extend the fermentation time, is the risk-time conversion coefficient, Extend the duration for the maximum baseline.

[0016] A mushroom residue fermentation ingredient dosing device based on AI dynamic proportioning uses any of the above-mentioned mushroom residue fermentation ingredient control methods based on AI dynamic proportioning.

[0017] Compared with the prior art, the present invention has the following beneficial effects: First, the present invention extracts spatial features through a 3D-CNN model, acquires time series data through an LSTM network model, and combines the spatial features and time series data to achieve digital twin modeling of the fermentation process. At the same time, by constructing a three-dimensional target space of maturity, conversion rate, and energy consumption, combined with a K-means clustering algorithm and B-spline curvature optimization, the optimal parameters are determined, thereby shortening the response time for fermentation parameter adjustment. Second, the present invention uses a near-infrared spectrometer, a metal oxide semiconductor gas sensor array, and a microwave resonant moisture sensor to collaboratively acquire the spatial distribution of cellulose / lignin, VOCs fingerprint characteristics, and three-dimensional moisture gradient data of mushroom residue in real time. This not only builds a complete material feature database but also addresses the one-sided nature of traditional single-sensor monitoring data. Thirdly, the present invention can improve the cellulose conversion rate and reduce the unit processing energy consumption by normalized energy consumption analysis and frontier solution optimization algorithm while ensuring that the maturity reaches the standard; Fourthly, the present invention determines the risk level through a discretized toxin accumulation probability model, and automatically increases the amount of cellulase added, increases the stirring speed, prolongs the fermentation time, and injects lime water, thereby reducing the fermentation failure rate; Fifth: The present invention can freely switch between production modes such as organic fertilizer, feed and biofuel through the purpose-performance index mapping library, thereby not only improving equipment utilization and product qualification rate, but also improving system production capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the process of controlling the fermentation ingredients of mushroom residue in the present invention; Figure 2 It is the temperature control curve diagram of the present invention; Figure 3 A comparison chart of aeration flow stability in the present invention; Figure 4 This is a comparison chart of indicator data in the present invention; Figure 5 This is a diagram showing the dynamic change of toxin probability and automatic intervention changes in the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] In the process of resource utilization of mushroom residue, it can be used to produce a variety of products such as organic fertilizer, animal feed and biofuel. However, in the production of organic fertilizer, the focus is on the maturity and nutrient release characteristics of the material. In the preparation of feed base, the focus is on increasing the protein content and the toxin degradation effect. When used for biofuel, the focus is on cellulose conversion rate and calorific value. Therefore, there are obvious differences between the physical and chemical properties of fermentation products in different application scenarios. Most existing ingredient systems use fixed ratios or manual adjustment methods based on experience, which cannot be dynamically optimized according to the target use, making it difficult for the same equipment to flexibly adapt to multiple application requirements. This application uses a near-infrared spectrometer, a gas sensor array and a moisture sensor to collect the spatial distribution, corruption characteristics and humidity gradient data of mushroom residue in real time. At the same time, through a 3D-CNN model and an LSTM model, the three-dimensional characteristics of the material and the fermentation time series changes are analyzed, and a three-dimensional target space with maturity, conversion rate and energy consumption as dimensions is constructed. At the same time, the optimal fermentation parameters are determined through the K-means clustering algorithm and the B-spline fitting algorithm. By obtaining the toxin accumulation probability in real time, the cellulase addition and fermentation time are dynamically adjusted, thereby realizing intelligent optimization control of the fermentation process.

[0021] Example 1 refer to Figures 1-4 This embodiment provides a method for controlling the proportioning of mushroom residue fermentation ingredients based on AI dynamic proportioning, and the method includes the following steps: Step S1: Multimodal data acquisition. This involves scanning the mushroom residue using a near-infrared spectrometer to determine the spatial distribution of cellulose and lignin within the residue. Simultaneously, a metal oxide semiconductor gas sensor array and a microwave resonant moisture sensor are used to determine the decay characteristics and humidity gradient within the fermentation tank. The details are as follows: Step S1.1: Near-infrared spectral scanning. A high-resolution near-infrared spectral probe (e.g., an InGaAs detector) with a wavelength range of 900-2500nm is installed above the fermentation tank feed port. An automatic rotating platform is installed below the high-resolution near-infrared spectral probe. The automatic rotating platform is a two-dimensional platform driven by a stepper motor, capable of both horizontal and vertical rotation. This rotation drives the high-resolution near-infrared spectral probe, enabling multi-angle scanning of the mushroom residue on the conveyor belt.

[0022] Furthermore, a high-resolution near-infrared spectrometer probe is used to perform a spiral scan of the raw material surface at a linear speed of 10 cm / s, collecting five spectral points per square centimeter. A pan-tilt encoder simultaneously acquires the azimuth and elevation angles of each spectral point. Based on the laser ranging data, the corresponding three-dimensional rectangular coordinates of each spectral point are determined. Specifically, a Kriging interpolation algorithm is used to generate a corresponding spatial distribution model using the three-dimensional rectangular coordinates corresponding to each spectral point.

[0023] Furthermore, the grid data in the spatial distribution model is normalized to obtain the corresponding normalized values. At the same time, the color corresponding to each grid data is determined based on the normalized value corresponding to each grid data and the RGB value corresponding to each color, thereby generating the corresponding heat map.

[0024] In this embodiment, the three-dimensional grid data in the thermal map is scanned and the lignin content corresponding to each grid data is determined. The obtained lignin content is compared with a preset content threshold (which is set according to the specific data and is not specifically explained in this embodiment, such as 25%), and the corresponding voxel grid is marked according to the comparison result. Specifically: When the obtained lignin content is greater than the preset content threshold (25%), the voxel grid corresponding to the lignin content is marked. Conversely, when the obtained lignin content is not greater than the preset content threshold (25%), the voxel grid corresponding to the lignin content is not marked.

[0025] Furthermore, based on the marked voxel grids, the interconnected voxel grids are connected to obtain multiple connected regions, and the volume size of each connected region is determined, specifically: ; in: is the volume of the connected region, is the number of labeled voxels in the connected region, is the volume of the voxel grid.

[0026] In this embodiment, the volume of the connected area obtained is compared with a preset volume threshold (which is set according to specific data and is not specifically explained in this embodiment, for example, 10 cm 3 ) and determine the priority treatment areas based on the comparison results. Specifically: When the volume of the connected area is greater than the preset volume threshold (e.g. 10cm 3 ), the corresponding connected area is the priority processing area. On the contrary, when the volume of the connected area is not greater than the preset volume threshold (for example, 10cm 3), the corresponding connected area is not the priority processing area.

[0027] Step S1.2: VOCs fingerprint detection. Six mounting holes are evenly spaced on the top cover of the fermenter, and each mounting hole is equipped with a metal oxide semiconductor gas sensor array to monitor ammonia, sulfide, and alcohol / aldehyde gases in the fermenter.

[0028] In this embodiment, a pulsed air pump is used to extract gas from the fermenter and transmit the extracted gas to the metal oxide semiconductor gas sensor array to obtain the corresponding gas content. It is worth noting that in this embodiment, the gas in the fermenter is extracted every 30 minutes.

[0029] Furthermore, based on the obtained gas content within the fermentation tank, the corresponding ethanol gas content, propionaldehyde gas content, ammonia gas content, hydrogen sulfide gas content, butyric acid gas content, and dimethyl disulfide gas content were determined. A fingerprint database corresponding to the different gas content levels was also constructed, as shown in Table 1 below.

[0030] Table 1: Fingerprint database table Corruption Level Gas combination Early Ethanol and propionaldehyde mid-term Ammonia and hydrogen sulfide Late stage Butyric acid and dimethyl disulfide Step S1.3: Moisture Gradient Monitoring. Three sets of microwave resonant moisture sensors are installed on the sidewalls of the fermenter, with the vertical spacing between each set set at 30 cm. Specifically, the sensors on the upper layer monitor the surface area within the fermenter, the sensors on the middle layer monitor the core of the fermenter, and the sensors on the lower layer monitor the bottom of the fermenter.

[0031] Furthermore, when monitoring the moisture content of a fermenter, the microwave resonant moisture sensor's emitter emits a microwave signal that penetrates the fermenter's interior and is then received by the corresponding receiver. The moisture content of the fermenter is determined based on the microwave signal received by the receiver.

[0032] In this embodiment, the moisture content at different locations in the upper, middle, and lower layers of the fermenter is determined based on the moisture content obtained by each group of microwave resonant moisture sensors. Furthermore, the corresponding vertical moisture gradient difference and the moisture anomaly value in the middle layer are obtained based on the moisture content at different locations in the upper, middle, and lower layers of the fermenter, specifically: ; in: is the vertical moisture gradient difference, is the abnormal value of moisture in the middle layer, is the moisture content of the lower layer, is the moisture content of the upper layer, is the moisture content of the middle layer.

[0033] Furthermore, the obtained vertical moisture gradient difference and the mid-layer moisture anomaly value are compared with a preset gradient threshold range (set according to specific data, not specifically explained in this embodiment, for example, 5%-8%) and a preset anomaly threshold (set according to specific data, not specifically explained in this embodiment, for example, 3%), and the fermentation state in the fermentation tank is determined based on the comparison results. Specifically: When the vertical moisture gradient difference is within the preset gradient threshold range (e.g., 5%-8%), and the middle layer moisture abnormality value is not less than the preset abnormality threshold (e.g., 3%), the fermentation state in the fermentation tank is normal. Otherwise, the fermentation state in the fermentation tank is abnormal.

[0034] Step S2: Parameter determination. This involves determining the three-dimensional characteristics and time series results of the original mushroom residue pile based on the priority processing areas determined in step S1.1, the fingerprint database constructed in step S1.2, and the fermentation status determined in step S1.3. Furthermore, the corresponding optimal parameters are determined based on the set target thresholds. The details are as follows: Step S2.1: Data preprocessing. This involves obtaining the corresponding three-dimensional features and time series results using the 3D-CNN model and LSTM network model based on the priority processing areas determined in step S1.1, the fingerprint database constructed in step S1.2, and the fermentation status determined in step S1.3. The details are as follows: Step S2.1.1: Determine 3D structural features. Specifically, based on the priority regions identified in the thermal map in Step S1.1, resample them to a 1 cm³ standard voxel grid using a trilinear interpolation algorithm. Each voxel includes cellulose content, lignin content, and temperature values. Furthermore, based on the resampled voxel grid, determine the maximum and minimum measured cellulose and lignin contents. Based on these maximum and minimum measured values, the cellulose and lignin contents of each voxel are normalized to obtain the corresponding normalized cellulose and lignin contents.

[0035] Furthermore, three data channels in the 3D-CNN model are set based on the normalized cellulose content, normalized lignin content, and the temperature corresponding to the infrared thermal imaging. Specifically, the 3D-CNN model outputs 128 corresponding three-dimensional structural features. The first 32 dimensions represent the spatial distribution of cellulose, the middle 64 dimensions represent the lignin-temperature coupling, and the last 32 dimensions represent the structural heterogeneity.

[0036] Step S2.1.2: Time Series Prediction. Based on the fingerprint database constructed in step S1.2 and the fermentation state determined in step S1.3, the gas content obtained by the metal oxide semiconductor gas sensor array and the moisture content obtained by the microwave resonant moisture sensor are normalized to obtain the corresponding normalized gas content and normalized moisture content.

[0037] Furthermore, the normalized gas content and the normalized moisture content are used as inputs of the LSTM network model, and the corresponding predicted maturity value and the amount of cellulose residue are obtained as outputs.

[0038] Step S2.2: Determine target weights. Based on the spatial distribution of the fermentation tanks, the raw mushroom residue pile is divided into five categories: the core area, the tank wall area, the middle and lower layers of the pile, the surface area, and the remaining area. Furthermore, based on the 128-dimensional three-dimensional structural features obtained in step S2.1.1, a K-means clustering algorithm is used to obtain candidate solutions corresponding to each category.

[0039] Furthermore, through the purpose-performance index mapping library that is set, the loading threshold values ​​of organic fertilizer, feed and biofuel are set. Specifically, when setting the organic fertilizer loading threshold value, its corresponding degree of maturity is not less than 80%, and the proportion of available nitrogen is not less than 30%. When setting the feed loading threshold value, its corresponding crude protein content is not less than 18%, and the aflatoxin degradation rate is not less than 95%. When setting the biofuel loading threshold value, its corresponding cellulose conversion rate is not less than 75%, and the calorific value is not less than 16MJ / kg. In the present embodiment, the target purpose is set through the HMI interface, that is, according to the purpose-performance index mapping library that is set, the performance size corresponding to the target purpose is determined.

[0040] Specifically, a comprehensive score is determined for each candidate solution based on the performance corresponding to the target application and the candidate solutions for each partition category. It is worth noting that there are multiple candidate solutions for each partition category. In other words, the comprehensive scores of each candidate solution are compared to determine the maximum comprehensive score. The candidate solution with the maximum comprehensive score is the optimal solution for that partition category.

[0041] In this embodiment, the formula for obtaining the comprehensive score is specifically: ; in: For the comprehensive rating, is the maturity weight, is the maturity deviation, is the conversion rate weight, is the cellulose conversion rate, is the energy consumption weight, is the energy consumption coefficient.

[0042] Step S2.3: Construct a three-dimensional target space. Specifically, based on the predicted maturity values ​​obtained by the LSTM network model in step S2.1.2, obtain the corresponding maximum and minimum predicted maturity values. Each predicted maturity value is normalized based on the maximum and minimum predicted maturity values ​​to obtain a normalized predicted maturity value.

[0043] Furthermore, based on the spatial distribution characteristics of cellulose obtained by the 3D-CNN model in step S2.1.1, the hourly change of the corresponding spatial distribution characteristics of cellulose is obtained, and the corresponding normalized conversion rate value is obtained by the inverse Sigmoid transformation algorithm, specifically: ; in: is the normalized value of conversion rate, is the original value of cellulose conversion rate, is the smoothing factor.

[0044] Furthermore, a power sensor (such as a smart meter) is used to obtain the specific energy consumption of the fermentation tank during the fermentation process. Specifically, within a preset time period, all specific energy consumptions of the fermentation tank during the fermentation of the same raw material are obtained, and the minimum specific energy consumption within the preset time period is determined. The actual total energy consumption obtained is then normalized based on the determined minimum specific energy consumption to obtain a normalized energy consumption.

[0045] In this embodiment, the obtained normalized maturity prediction value is used as the X-axis, the obtained normalized conversion rate value is used as the Y-axis, and the obtained normalized energy consumption is used as the Z-axis to construct a corresponding three-dimensional target space.

[0046] Step S2.4: Determine the optimal solution. That is, based on the three-dimensional target space and Z value constructed in step S2.3, divide the data in the three-dimensional target space into multiple layers of data, and obtain the corresponding comprehensive score of the frontier solution in each layer of data, specifically: ; in: Comprehensive score for the frontier solution, is the scoring weight coefficient, For the comprehensive rating, For the frontier level, is the level attenuation factor.

[0047] Furthermore, the comprehensive scores of all frontier solutions are compared to determine the three largest comprehensive scores. At the same time, based on the data size corresponding to the comprehensive scores of the frontier solutions, the corresponding curvature values ​​are obtained through B-spline fitting, specifically: ; in: is the curvature value, is the second-order derivative of the frontier solution, is the first derivative of the frontier solution.

[0048] In this embodiment, the corresponding curvature value is obtained based on the frontier solution corresponding to the comprehensive score of the three largest frontier solutions determined, and the three curvature values ​​are compared to determine the maximum curvature value. The frontier solution corresponding to the maximum curvature value is the optimal solution.

[0049] Step S3: Adjustment and Control. This involves real-time adjustment of the temperature, auxiliary material addition, and aeration rate during the fermentation process based on the optimal solution determined in step S2.4, namely, the optimal temperature, optimal auxiliary material ratio, and optimal aeration rate. Specifically, based on the optimal temperature and the real-time temperature, the total fermentation time is divided into time periods, i.e., multiple fermentation time periods. Simultaneously, based on the fermentation time intervals and temperature rise intervals corresponding to the fermentation time periods, the temperature rise rate corresponding to each fermentation time period is determined. In other words, the corresponding temperature operation is performed within each fermentation time period based on the determined temperature rise rate.

[0050] Furthermore, based on the optimal excipient ratio and the fermentation tank volume, the optimal total amount of excipients is determined, divided into multiple small portions, and added to the fermentation tank via a screw conveyor and a spray gun. Specifically, during the addition of the small portions, the first and last additions are dissolved and then atomized and sprayed via the spray gun. The remaining excipients are then added directly to the fermentation tank via the screw conveyor.

[0051] Furthermore, based on the optimal aeration volume, the aeration volume of each air path is divided according to the composition of the air paths, that is, the aeration volume of the bottom air path is set to 70% of the total aeration volume, and the aeration volume of the side wall air path is set to 30% of the total aeration volume. At the same time, the average flow rate of the aeration valve is adjusted according to the flow rate, opening time and total duration of the aeration valve, specifically: ; in: is the period average flow rate, is the pulse frequency, is the maximum flow rate of the valve, For the single opening duration, is the total cycle duration.

[0052] refer to Figure 2 , Figure 2 is the temperature control curve in this embodiment, Figure 2 It can be seen that during the high-temperature period of 55°C-60°C, the fluctuation under AI dynamic control was within ±1°C, while the fluctuation under traditional control reached ±5°C. Furthermore, at the end of fermentation, 72 hours later, AI control still maintained 55°C, while traditional control had dropped to 45°C. This means that AI control can precisely respond to the thermodynamic requirements of the fermentation process and reduce the impact of temperature fluctuations on microbial activity.

[0053] refer to Figure 3 , Figure 3 is a comparison chart of aeration flow stability in this embodiment, Figure 3 It can be seen that the flow rate fluctuation range has been reduced from ±0.3m³ / h with traditional aeration to ±0.1m³ / h, and the average flow rate has increased from 0.5m³ / h to 0.6m³ / h, a 20% improvement. In other words, AI control can stabilize dissolved oxygen levels, promote the growth of aerobic microorganisms, and avoid localized hypoxia or excessive aeration.

[0054] refer to Figure 4 , Figure 4 The dynamic change of toxin probability and automatic intervention change diagram in this embodiment are shown by Figure 4 It can be seen that the maturity rate has increased from 75% in the traditional process to 98.5%, an increase of 31%, thus ensuring the quality of the compost product. The cellulose conversion rate has increased from 60% in the traditional process to 78.5%, an increase of 30%, thereby improving the efficiency of organic waste degradation. The unit energy consumption has decreased from 0.5kWh / kg in the traditional process to 0.35kWh / kg, a decrease of 30%, thereby reducing operating costs. The fermentation cycle has been reduced from 20 days in the traditional process to 17 days, a decrease of 15%, thereby improving production efficiency. Labor costs have been reduced from 100% in the traditional process to 15%, a decrease of 85%, thereby reducing labor dependence.

[0055] This embodiment also provides a mushroom residue fermentation batching device based on AI dynamic proportioning, which uses the above-mentioned mushroom residue fermentation batching control method based on AI dynamic proportioning.

[0056] Example 2 This embodiment provides a method for controlling the proportioning of mushroom residue fermentation ingredients based on AI-based dynamic proportioning. Its specific implementation is similar to that of Example 1, differing in that, during the real-time monitoring of the fermentation product within the fermenter, the probability of toxin accumulation within the fermenter is obtained in real time, and cellulase and fermentation time are adjusted in real time based on the toxin accumulation probability. The present invention is illustrated below with reference to specific implementations of this embodiment.

[0057] In this embodiment, the cellulase and fermentation time are adjusted in real time according to the probability of toxin accumulation, as follows: Step S3.1: Determine the probability of toxin accumulation. Specifically, the temperature, gas concentration, and moisture content within the fermenter are discretized into multiple discrete temperature intervals, discrete gas intervals, and discrete moisture intervals, depending on the fermentation product within the fermenter. Simultaneously, the toxin probability corresponding to each discrete interval is determined based on the discrete intervals.

[0058] In this embodiment, the temperature within the fermenter is measured using a high-resolution near-infrared spectrometer, the gas content within the fermenter is measured using a metal oxide semiconductor gas sensor array, and the moisture content within the fermenter is measured using a microwave resonant moisture sensor. Specifically, based on the measured temperature, a corresponding discrete temperature interval is determined, and the corresponding toxin probability for each discrete temperature interval is obtained. Based on the measured gas content, a corresponding discrete gas interval is determined, and the corresponding toxin probability for each discrete gas interval is obtained. Based on the measured moisture content, a corresponding discrete moisture interval is determined, and the corresponding toxin probability for each discrete moisture interval is obtained.

[0059] Furthermore, based on the toxin probabilities corresponding to the discrete temperature intervals, the toxin probabilities corresponding to the discrete gas intervals, and the toxin probabilities corresponding to the discrete moisture intervals, the corresponding toxin cumulative probabilities are determined as follows: ; in: is the final toxin accumulation probability, is the base conditional probability, is the weight of the i-th influencing factor, is the total number of influencing factors, Index of influencing factors.

[0060] Step S3.2: Determine the risk level. This involves comparing the final toxin accumulation probability obtained in step S3.1 with a preset probability threshold range (set based on specific data and not specifically explained in this embodiment, e.g., 15%-30%), and determining the corresponding risk level based on the comparison result. Specifically: If the final cumulative probability of toxins is less than the lower limit of the preset probability threshold range (e.g., 15%), the corresponding risk level is low, and continuous monitoring is performed. If the final cumulative probability of toxins is within the preset probability threshold range (e.g., 15%-30%), the corresponding risk level is medium, and the cellulase content and stirring speed are increased. If the final cumulative probability of toxins is greater than the upper limit of the preset probability threshold range (e.g., 30%), the corresponding risk level is high, and the fermentation time is extended, and 5% lime water is injected into the fermentation tank for adjustment.

[0061] Step S3.3: Determine first aid measures. That is, when the risk level is medium, determine the cellulase dosage based on the difference between the final toxin accumulation probability and the lower limit threshold of the preset probability threshold range, specifically: ; in: is the cellulase dosage, As the basic dosage, is the dose-probability coefficient, is the intervention threshold, is the final toxin accumulation probability.

[0062] Specifically, during the process of adding the cellulase, the cellulase is added to the priority treatment area according to the priority treatment area determined in step S1.1.

[0063] Furthermore, when the risk level is high, the fermentation time that needs to be extended is determined based on the difference between the final toxin accumulation probability and the lower limit threshold of the preset probability threshold range, specifically: ; in: To extend the fermentation time, is the risk-time conversion coefficient, Extend the duration for the maximum benchmark, is the intervention threshold, is the final toxin accumulation probability.

[0064] Specifically, in the process of determining the cellulase dosage and fermentation time, the corresponding toxin accumulation probability is obtained in real time, and based on the obtained toxin accumulation probability, steps S3.1 to S3.3 are repeated until the corresponding risk level is low risk.

[0065] refer to Figure 5 , Figure 5 The dynamic change of toxin probability and automatic intervention change diagram in this embodiment are shown by Figure 5It can be seen that the probability of toxin accumulation fluctuates and increases over time. When the probability of toxin accumulation is between 15% and 30%, the corresponding risk level is medium risk, and when it exceeds 30%, the corresponding risk level is high risk. Furthermore, when the probability of toxin accumulation exceeds 15%, intervention is triggered to prevent the risk from escalating.

[0066] Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is limited by the accompanying embodiments and their equivalents.

Claims

1. A mushroom residue fermentation ingredient control method based on AI dynamic proportioning, characterized in that: Includes: S1: Multimodal data acquisition: Acquisition of cellulose / lignin spatial distribution, spoilage characteristics, and moisture gradients using near-infrared spectrometers, metal oxide semiconductor gas sensor arrays, and microwave resonant moisture sensors; S2: Parameter determination: Based on the target threshold, the three-dimensional characteristics of the original mushroom residue pile and the time series results, the optimal parameters are determined, including: S2.1: Data preprocessing: Based on the spatial distribution, decay characteristics, and moisture gradient of cellulose / lignin, the 3D-CNN model and LSTM network model are used to determine the three-dimensional characteristics and time series results of the original mushroom residue pile; S2.2: Determine target weights: Based on the spatial distribution of the fermentation tanks, set the categories of the K-means clustering algorithm and obtain candidate solutions for each category. Simultaneously, set target performance using a usage-performance indicator mapping library. Based on the candidate solutions and the target performance, determine a comprehensive score for the candidate solutions. S2.3: Construct a three-dimensional target space: Use the normalized maturity prediction value as the X-axis, the normalized conversion rate value as the Y-axis, and the normalized energy consumption as the Z-axis to construct a three-dimensional target space; S2.4: Determine the optimal solution: Based on the three-dimensional target space and the comprehensive score, obtain the comprehensive score of the frontier solution, and determine the three largest comprehensive scores of the frontier solution. Simultaneously, obtain the curvature value of the comprehensive scores of the frontier solutions through B-spline fitting. The frontier solution corresponding to the largest curvature value is the optimal solution. S3: Adjustment and control: According to the optimal solution, the fermentation temperature, auxiliary material addition and aeration volume are adjusted in real time.

2. The method for controlling mushroom residue fermentation ingredients based on AI dynamic proportioning according to claim 1, characterized in that: Obtain cellulose / lignin spatial distribution, decay characteristics and moisture gradients, including: S1.1: Near-infrared spectral scanning: A near-infrared spectral probe equipped with an automatic pan-tilt platform is used to scan the mushroom residue, simultaneously acquiring the azimuth and elevation angles of each spectral point. Based on the laser ranging data, the three-dimensional rectangular coordinates of the spectral point are determined, and a spatial distribution model is constructed. The grid data in the spatial distribution model is normalized, and a heat map is generated based on the RGB values ​​of different colors. S1.2: VOCs fingerprint detection: The gas in the fermentation tank is monitored by a metal oxide semiconductor gas sensor array, and a fingerprint database is constructed based on the monitored gas content; S1.3: Moisture Gradient Monitoring: Microwave resonant moisture sensors are installed at different areas on the sidewall of the fermenter to obtain the vertical moisture gradient difference and the middle layer moisture abnormality value in the fermenter. The vertical moisture gradient difference and the middle layer moisture abnormality value are compared with the preset gradient threshold range and the preset abnormality threshold, respectively. The fermentation status is determined based on the comparison results, specifically: When the vertical moisture gradient difference is within the preset gradient threshold range and the middle layer moisture abnormality value is not less than the preset abnormality threshold, the fermentation state is normal fermentation; otherwise, the fermentation state is abnormal fermentation.

3. The method for controlling mushroom residue fermentation ingredients based on AI dynamic proportioning according to claim 2, characterized in that: Obtain the lignin content of the grid data in the thermal map, and compare the lignin content with a preset content threshold. According to the comparison result, determine the marked voxel grid, specifically: When the lignin content is greater than a preset content threshold, the corresponding voxel grid is marked; Otherwise, the corresponding voxel grid is not marked; At the same time, the interconnected marked voxel grids are connected to determine the volume of the connected area, and the volume of the connected area is compared with a preset volume threshold. Based on the comparison result, the priority processing area is determined, specifically: When the volume of the connected area is greater than a preset volume threshold, the corresponding connected area is a priority processing area; otherwise, the corresponding connected area is not a priority processing area.

4. The method for controlling mushroom residue fermentation ingredients based on AI dynamic proportioning according to claim 1, characterized in that: Determine the three-dimensional characteristics and time series results of the original mushroom residue pile, including: S2.1.1: Determine 3D structural features: Resample the priority regions using a trilinear interpolation algorithm and obtain normalized cellulose and lignin contents based on the resampled voxel grid. Simultaneously, output the 3D structural features using a 3D-CNN model. S2.1.2: Time series prediction: Based on the fingerprint database and fermentation status, the normalized gas content and normalized moisture content are obtained, and the predicted value of maturity and the remaining cellulose are obtained through the input and output of the LSTM network model.

5. The method for controlling mushroom residue fermentation ingredients based on AI dynamic proportioning according to claim 1, characterized in that: According to the Z value of the three-dimensional target space, the data in the three-dimensional target space is divided into multiple layers of data. The comprehensive score of the frontier solution of each layer of data is specifically: ； in: Comprehensive score for the frontier solution, is the scoring weight coefficient, For the comprehensive rating, For the frontier level, is the level attenuation factor.

6. The method for controlling mushroom residue fermentation ingredients based on AI dynamic proportioning according to claim 1, characterized in that: According to the optimal auxiliary material ratio and the fermentation tank volume in the optimal solution, the optimal total amount of auxiliary materials is determined, and the optimal total amount of auxiliary materials is divided into multiple auxiliary material portions, and the auxiliary material portions are added through a screw conveyor and a spray gun.

7. The method for controlling mushroom residue fermentation ingredients based on AI dynamic proportioning according to claim 1, characterized in that: According to the optimal aeration volume and air path composition in the optimal solution, the aeration volume of each air path is divided. At the same time, according to the flow rate, opening time and total duration of the aeration valve, the average flow rate of the aeration valve is adjusted. Specifically, ； in: is the period average flow rate, is the pulse frequency, is the maximum flow rate of the valve, For the single opening duration, is the total cycle duration.

8. The method for controlling mushroom residue fermentation ingredients based on AI dynamic proportioning according to claim 1, characterized in that: When adjusting the fermentation temperature, auxiliary material addition and aeration volume in real time, the probability of toxin accumulation inside the fermentation tank is obtained, and cellulase and fermentation time are adjusted in real time according to the toxin accumulation probability, including: S3.1: Determine the probability of toxin accumulation: Discretize the fermentation tank into multiple discrete temperature intervals, discrete gas intervals, and discrete moisture intervals based on the temperature, gas concentration, and moisture content of the fermentation tank. Determine the probability of toxin accumulation based on the toxin probabilities corresponding to the discrete temperature intervals, discrete gas intervals, and discrete moisture intervals, specifically: ； in: is the final toxin accumulation probability, is the base conditional probability, is the weight of the i-th influencing factor, is the total number of influencing factors, is the index of influencing factors; S3.2: Determine the risk level: Compare the toxin accumulation probability with the preset probability threshold range and determine the risk level based on the comparison result; specifically: When the cumulative probability of toxins is less than the lower limit of the preset probability threshold range, the corresponding risk level is low risk; when the cumulative probability of toxins is within the preset probability threshold range, the corresponding risk level is medium risk; when the cumulative probability of toxins is greater than the upper limit of the preset probability threshold range, the corresponding risk level is high risk; S3.3: Determine first aid measures: When the risk level is medium risk, increase the cellulase content and increase the stirring speed. When the risk level is high risk, extend the fermentation time and inject lime water into the fermentation tank. At the same time, obtain the corresponding toxin accumulation probability in real time, and repeat steps S3.1-S3.3 according to the toxin accumulation probability until the risk level is low risk.

9. The method for controlling mushroom residue fermentation ingredients based on AI dynamic proportioning according to claim 8, characterized in that: According to the difference between the toxin accumulation probability and the lower limit threshold of the preset probability threshold range, the cellulase dosage and fermentation extension time are determined, specifically: ； in: is the cellulase dosage, As the basic dosage, is the dose-probability coefficient, is the intervention threshold, is the final toxin accumulation probability, To extend the fermentation time, is the risk-time conversion coefficient, Extend the duration for the maximum baseline.

10. A mushroom residue fermentation batching device based on AI dynamic proportioning, characterized in that: A mushroom residue fermentation ingredient control method based on AI dynamic proportioning as described in any one of claims 1 to 9 is used.

Images