Planting and breeding combined type agricultural waste anaerobic fermentation and biogas slurry precise recycling system

Abstract

The invention relates to a planting and breeding combined type agricultural waste anaerobic fermentation and biogas slurry accurate recycling system, in particular to the field of program control systems, and aims to realize millisecond-level accurate regulation and control of working conditions of a single fermentation tank through instant data fusion and autonomous decision of edge nodes, ensure stability of anaerobic fermentation core parameters and improve the utilization rate of biogas slurry. By means of low-delay cooperation between edge nodes, mutual interference between tank groups is effectively stabilized, overall operation robustness is improved, a cloud end continuously explores a global optimal control strategy through reinforcement learning and generates new parameters through safety verification based on a digital twinborn model constructed by total historical and external data, and finally, the overall operation robustness is improved. Through a cloud-side collaborative progressive verification and distribution mechanism, the optimized strategy is smoothly deployed to the whole network in a lossless manner, an autonomous closed loop from local real-time control to global continuous optimization is formed, and the gas production efficiency, the operation stability and the resource utilization rate are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of program control systems, and more specifically, to a system for the precise reuse of anaerobic fermentation and biogas slurry in agricultural waste that integrates crop and livestock farming. Background Technology

[0002] With the development of large-scale and intensive agriculture, integrated crop-livestock farming has become an effective way to treat agricultural waste such as livestock and poultry manure and crop straw. In this model, anaerobic fermentation technology, as the core component, can convert various organic wastes into biogas and biogas slurry, achieving energy recovery and nutrient return to the fields. A typical system includes multiple distributed anaerobic fermentation tanks to process waste materials from different sources and with different characteristics. The fermentation process is essentially a series of complex microbial biochemical reactions, and its stability and efficiency highly depend on the precise control of environmental parameters within the tank, especially temperature, pH value, volatile fatty acid concentration, and... Parameters such as material residence time are often within a narrow optimal range. For example, mesophilic fermentation usually requires a stable temperature between 35 and 40 degrees Celsius. However, actual production scenarios are complex and variable. External environmental factors such as diurnal temperature differences, seasonal climate changes, and rainwater infiltration can directly interfere with the thermal balance of the fermenter. Internal factors such as fluctuations in feed composition, uneven feed rates, and dynamic evolution of microbial communities can continuously disrupt the biochemical balance within the system. Therefore, to achieve continuous and efficient biogas production and stable biogas slurry output, it is necessary to conduct continuous and precise monitoring and real-time control of the entire fermentation process.

[0003] Currently, monitoring and control in this field mainly rely on programmable logic controllers (PLCs) or data acquisition and monitoring control system architectures, combined with IoT sensing technology for data acquisition. A typical implementation involves transmitting real-time data such as temperature, pressure, flow rate, and pH value collected by various sensors distributed throughout the fermenter and pipelines to a centralized cloud service platform or central server via wired or wireless networks. In the cloud, calculations and analyses are performed using a pre-defined rule base or classic proportional-integral-derivative (PI-DE) control algorithm, generating control commands that are then sent to on-site actuators such as heating, stirring, and feeding mechanisms. This cloud-based centralized processing solution has inherent drawbacks. First, there is a significant delay in data transmission and control loops; the time from data upload and cloud processing to command issuance often involves a lag of seconds or even minutes, which cannot meet the demands of the fermentation process. The real-time requirement for rapid response to disturbances is a significant concern in widely distributed farms, where unstable network conditions can exacerbate the risk of control failure. Secondly, the architecture aggregates large amounts of raw data to the cloud, resulting in a heavy processing load and long adjustment and optimization cycles for the control logic. This makes it difficult to adapt to changes in raw material characteristics or process adjustments, and it lacks self-learning and adaptive capabilities. Finally, simple threshold alarms or fixed-parameter control strategies cannot cope with complex fermentation processes involving multi-parameter coupling and strong nonlinearity. Especially when dealing with feed shocks or drastic environmental changes, the control is often slow, easily leading to acidification of the fermentation system, decreased gas production efficiency, or even process failure. Therefore, existing technologies urgently need a new monitoring and control system capable of millisecond-level response and edge intelligence and cloud-based collaborative optimization to ensure the stable, efficient, and adaptive operation of anaerobic fermentation processes. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a system for the precise reuse of anaerobic fermentation and biogas slurry in agricultural waste that integrates crop and livestock farming. The system utilizes an edge control module, an edge collaboration module, a cloud optimization module, and a cloud-edge collaborative update module to solve the problems mentioned in the background.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: it specifically includes: an edge control module, an edge collaboration module, a cloud optimization module, and a cloud-edge collaborative update module, wherein; Edge control module: Deployed on the edge nodes corresponding to each fermenter, it responds to real-time process data collected by the fermenter's sensors, fuses the real-time process data with the dynamic control parameter set currently applied by the edge control module, generates a process stability index, and generates and executes the first control command locally based on the comparison result between the process stability index and the dynamic control parameter set to adjust the actuators of the fermenter; the edge control module also shares local operating status data including the process stability index through a low-latency edge network; Edge collaboration module: Deployed on edge nodes, it responds to the first control command generated by the edge control module and the shared local operating status data. It obtains the local operating status data shared by at least one other edge node through the low-latency edge network, and coordinates the first control command it generates based on the distributed collaboration algorithm to generate and execute the second control command. The cloud optimization module is used to receive and aggregate historical operating data, including local operating status data, and external environmental data uploaded by each edge node. It builds and runs a digital twin model of the anaerobic fermentation process, and iteratively optimizes the dynamic control parameter set in the digital twin model through a reinforcement learning optimization engine to generate a new dynamic control parameter set. It then performs virtual deployment verification of the new dynamic control parameters to obtain the verification confidence level. The cloud-edge collaborative update module is used to generate a new set of dynamic control parameters in the cloud optimization module and complete the virtual deployment verification. Based on the verification confidence, it selects target edge nodes and controls the edge control modules on the target edge nodes to enter a dual-track operation mode. This mode compares the new control logic based on the new set of dynamic control parameters with the old control logic based on the current set of dynamic control parameters applied by the edge control modules. Based on the comparison feedback, the new set of dynamic control parameters is progressively distributed to all edge nodes in the network to update the set of dynamic control parameters applied by each edge control module. In a preferred embodiment, the process of fusing real-time process data with a dynamic control parameter set to generate a process stability index in the edge control module is specifically as follows: The edge control module performs standardized preprocessing on the real-time process data collected from the sensors. The real-time process data includes temperature, pH, stirrer current and feed flow rate, forming a standardized feature vector. At the same time, a fixed-length first-in-first-out (FIFO) data buffer is maintained to store the standardized feature vectors within the most recent time window, so that the FIFO data buffer stores multiple standardized feature vectors arranged in chronological order, forming a standardized feature vector sequence. Based on the standardized feature vector sequence stored in the first-in-first-out data buffer, a set of derived feature vectors is calculated; wherein, the derived feature vectors include the gradient consistency within the window and the curvature of the feature trajectory; The calculation process for gradient consistency within the window is performed for any parameter among temperature, pH, stirrer current, or feed flow rate. Specifically, it involves: first, calculating the difference between adjacent values ​​of the parameter within the time window corresponding to the first-in-first-out data buffer to form a difference sequence; then, calculating the dispersion of the difference sequence to obtain a dispersion measure; finally, substituting the dispersion measure into a negative exponential function with the natural constant as the base for mapping, and the result is the gradient consistency within the window for the parameter. The calculation process of the curvature of the feature trajectory is as follows: Based on the standardized feature vector sequence stored in the first-in-first-out data buffer, the approximate curvature of the path formed by the standardized feature vector sequence in the multidimensional feature space composed of each parameter dimension is calculated. The approximate curvature is obtained by calculating the magnitude of the first-order difference vector and the magnitude of the second-order difference vector of the standardized feature vector sequence, and combining the magnitude of the first-order difference vector. The edge control module combines the standardized feature vector with the derived feature vector to form a comprehensive feature vector; The edge control module calls its currently applied dynamic control parameter set, which contains a feature weight vector and a reference state vector. The edge control module subtracts the comprehensive feature vector from the reference state vector element by element to obtain a difference vector. It then multiplies the feature weight vector from the difference vector element by element to obtain a weighted deviation vector. The module calculates the magnitude of the weighted deviation vector to obtain a scalar distance value. It multiplies a scaling factor by the difference between the scalar distance value and a real offset, and uses the resulting product as the exponent of an exponential function with the natural constant as the base. The result of this exponential function calculation is incremented by one and the reciprocal is taken to obtain a continuous value between zero and one, which is defined as the process stability index.

[0006] In a preferred embodiment, the process of generating and executing a first control command locally based on the process stability index to adjust the operation of the actuator in the fermenter specifically includes: The edge control module reads an exponential action mapping table contained in the dynamic control parameter set of its current application. The exponential action mapping table divides the range of the process stability index into multiple fuzzy intervals with overlapping regions and associates each fuzzy interval with a preset control strategy for adjusting the actuator of the fermenter. The edge control module queries the index action mapping table based on the calculated process stability index, matches the fuzzy interval to which the process stability index belongs and the corresponding control strategy, and generates the corresponding first control command based on the matched control strategy. The first control command includes the command type and specific parameters for adjusting the power of the heater, the speed of the stirrer, or the opening of the feed valve.

[0007] In a preferred embodiment, while generating the corresponding first control command based on the matched control strategy, the edge control module encapsulates a structured local operating status data, which includes: the current time of generating the first control command, the calculated process stability index, the temperature and pH in the real-time process data, and the command type and specific parameters contained in the first control command; the edge control module sends the encapsulated local operating status data out through the low-latency edge network.

[0008] In a preferred embodiment, the process of obtaining local operating status data shared by at least one other edge node through a low-latency edge network and coordinating based on a distributed coordination algorithm in the edge collaboration module specifically includes: Upon triggering a response, the edge collaboration module retrieves local operational status data shared by its neighboring edge nodes from the low-latency edge network. For each neighboring node, the edge collaboration module calculates a dynamic coupling coefficient. The calculation process is as follows: First, it calculates the ratio of the square of the physical distance between the current node and the corresponding fermenter of the neighboring node to the square of the distance scale parameter, obtaining a distance component. Next, it calculates the square of the absolute value of the difference between the process stability index of the current node and the process stability index of the neighboring nodes, and compares it with the square of the stability difference scale parameter, obtaining a stability difference component. Then, it adds the distance component and the stability difference component to obtain a combined value. After that, it multiplies the overall attenuation coefficient by the combined value and takes a negative number to obtain an exponential value. Finally, it sets the exponent of the exponential function with the natural constant as the base as the exponential value, calculates the value of the exponential function, and the result is the dynamic coupling coefficient. Based on the calculated dynamic coupling coefficients, the edge collaboration module further calculates two collaboration metrics: The first coordination indicator is the regional average disturbance trend, which is calculated as follows: First, for each neighboring node, the specific parameter change for adjusting the heater power contained in the first control command of that neighboring node is multiplied by the corresponding dynamic coupling coefficient to obtain a weighted power change; then, the weighted power changes calculated for all neighboring nodes are summed to obtain a total weighted power change; simultaneously, the dynamic coupling coefficients corresponding to all neighboring nodes are summed, and a positive constant is added to the summation result to obtain the total coupling coefficient; finally, the total weighted power change is divided by the total coupling coefficient, and the resulting quotient is defined as the regional average disturbance trend. The second coordination metric is local state gradient consistency, which is calculated as follows: First, for each neighboring node, the following steps are performed: Calculate the difference between the temperature of the current node and the temperature of the neighboring node, and divide this difference by a preset temperature normalization factor to obtain a temperature normalized difference; calculate the difference between the pH of the current node and the pH of the neighboring node, and divide this difference by a preset pH normalization factor to obtain a pH normalized difference; add the absolute value of the temperature normalized difference to the absolute value of the pH normalized difference to obtain a... The system calculates a combined absolute difference value; multiplies this value by the corresponding dynamic coupling coefficient to obtain a weighted difference value; then, it sums the weighted difference values ​​calculated for all neighboring nodes to obtain a total weighted difference; simultaneously, it sums the dynamic coupling coefficients corresponding to all neighboring nodes and adds a positive constant to the summation to obtain a total coupling coefficient; next, it divides the total weighted difference by the total coupling coefficient to obtain an average weighted difference value; finally, it subtracts the average weighted difference value from the first value, and the result is defined as the local state gradient consistency.

[0009] In a preferred embodiment, the process of coordinating the first control command generated by itself based on the distributed cooperative algorithm to generate and execute the second control command specifically involves: The edge collaboration module is based on the calculated consistency between the regional average perturbation trend and the local state gradient, and defines a decision correction quantity for adjusting the heater power; The edge collaboration module constructs a collaborative potential energy function, the value of which is determined by the first, second and third terms. The first term is the fidelity term, and its value is equal to a first weighting coefficient multiplied by the square of the deviation of the decision correction amount from the specific parameter change used to adjust the heater power in the first control command. The second term is the consistency term, which is equal to a second weighting coefficient multiplied by the square of the difference between the adjustment amount to be decided and the regional average perturbation trend. The third term is the incentive term, whose value is equal to the product of a third weight coefficient, local state gradient consistency, and the adjustment amount to be decided. The final value of the cooperative potential energy function is equal to the sum of the first and second terms, minus the third term. The edge coordination module multiplies the first weighting coefficient by the specific parameter change used to adjust the heater power in the first control command to obtain the first product; multiplies the second weighting coefficient by the regional average disturbance trend to obtain the second product; multiplies the third weighting coefficient by the local state gradient consistency and then divides by two to obtain the third product; adds the first, second, and third products to obtain the numerator sum; adds the first and second weighting coefficients to obtain the denominator sum; and divides the numerator sum by the denominator sum, and the resulting quotient is defined as the optimal coordinated action. The edge collaboration module generates a second control command based on the optimal collaborative action. This command will override or adjust the first control command and be delivered to the actuator.

[0010] In a preferred embodiment, the process of receiving and aggregating historical operating data and external environmental data uploaded by each edge node, and constructing and running a digital twin model of the anaerobic fermentation process in the cloud optimization module specifically includes: The cloud optimization module performs time alignment and cleaning on the received historical operation data. The historical operation data includes local operation status data periodically uploaded by each edge node. The cloud optimization module connects to external data sources to obtain external environmental data, which includes meteorological forecast information reflecting atmospheric conditions and raw material characteristic parameters reflecting the composition of waste. The cloud-based optimization module performs the following steps to fuse data and extract features: First, align the timestamps of historical operating data with those of external environmental data. Next, for key process variables in the historical operating data—process stability index, temperature, and pH—calculate their statistical characteristics at multiple preset time scales, including the weighted moving average, the slope of the linear trend within a local time window, and peak characteristics within a selected time period. Then, for each feature in the external environmental data, calculate its mutual information value with the key process variables, and form a correlation weight vector based on the calculated mutual information values. Finally, combine all the calculated statistical characteristics at different time scales with the correlation weight vector to form a fused feature vector. The digital twin model is constructed using a spatiotemporal graph convolutional network architecture enhanced with an attention mechanism. In this architecture, each fermenter and its associated process pipeline are modeled as a graph node, and the physical connections or material and energy exchange relationships between fermenters or between fermenters and pipelines are modeled as edges connecting these graph nodes, thus forming a graph structure that describes the topology of the entire anaerobic fermentation system. The spatiotemporal graph convolutional network architecture uses the edge connections of the graph structure to perform spatial convolution operations on the node features composed of fused feature vectors to learn the mutual influence between multiple fermenters; at the same time, this spatiotemporal graph convolutional network architecture learns the dynamic evolution of the internal state of a single fermenter over time through its internal temporal processing unit. The attention calculation mechanism embedded in the spatiotemporal graph convolutional network architecture dynamically calculates the importance weights of different input features to the current prediction target in each operation, and dynamically calculates the importance weights of the mutual influence between any two nodes connected by an edge in the graph structure. The digital twin model takes a fused feature vector as input and outputs predicted values ​​for the gas production rate, volatile fatty acid concentration, and process stability index of the fermentation system at future moments, as well as the average process stability index of all fermenters used to calculate the average operating stability of the entire system.

[0011] In a preferred embodiment, the process of iteratively optimizing the dynamic control parameter set in the digital twin model using a reinforcement learning optimization engine to generate a new dynamic control parameter set specifically involves: In the optimization framework, the cloud optimization module is defined as an intelligent agent, the digital twin model is defined as the environment, the intelligent agent performs actions in the environment to adjust the values ​​of various parameters in the dynamic control parameter set, and the state of the environment feedback is the key indicators output by the digital twin model, including gas production rate, average process stability index of the whole system, and energy consumption level. The cloud optimization module sets a multi-objective optimization task. The objective of this task is to find the set of dynamic control parameters that can maximize the long-term comprehensive performance of the system in the environment simulated by the digital twin model, under the premise of meeting the preset process safety constraints. Long-term comprehensive performance is quantified by a multi-objective weighted reward function. The value of this reward function is equal to the sum of a gas production reward term to incentivize increased gas production, a stability reward term to incentivize improved average process stability index of the entire system, and an energy efficiency penalty term to penalize increased energy consumption, each multiplied by its corresponding positive weight coefficient. Process safety constraints are monitored in real time during the simulation of the digital twin model. These constraints include the requirement that the concentration of volatile fatty acids derived from the simulation must not exceed a set inhibition threshold, and that temperature fluctuations during the simulation must not exceed a set range limit. If any constraint is violated in the simulation, a preset penalty value corresponding to the constraint violation is subtracted from the sum of the reward functions. The reinforcement learning optimization engine employs a policy gradient algorithm with constraints. It uses a digital twin model as the simulation environment and the set of dynamic control parameters to be optimized as the action policy to be optimized. The dynamic control parameter set includes the feature weight vector and baseline state vector used in the edge control module to calculate the process stability index, and the first weight coefficient, second weight coefficient, and third weight coefficient used in the edge collaboration module to construct the collaborative potential function. By executing a large number of virtual experiments in the simulation environment, feedback data on the reward function and constraint violations are collected, and the parameters of the action policy are iteratively updated using this feedback data. After multiple rounds of iterative optimization, the reinforcement learning optimization engine outputs a new set of dynamic control parameters. After the reinforcement learning optimization engine outputs a new set of dynamic control parameters, the cloud optimization module performs the following steps to complete the security verification and distribution of the parameter set: The new dynamic control parameter set is imported into the digital twin model; within the digital twin model, various typical operating conditions and preset extreme operating conditions are set, and simulations are performed using the new dynamic control parameter set; based on the simulation results, the confidence level of the expected performance improvement brought about by applying the new dynamic control parameter set is calculated; throughout the simulation, the concentration of volatile fatty acids simulated by the digital twin model is monitored to see if it exceeds the inhibition threshold, and the simulated temperature fluctuation exceeds the range limit; only when the calculated confidence level of the expected performance improvement is higher than a preset confidence threshold, and the simulation under all set operating conditions does not trigger any process safety constraints, is the new dynamic control parameter set marked as having passed virtual deployment verification. The cloud optimization module then outputs the new dynamic control parameter set marked as having passed verification to the cloud-edge collaborative update module.

[0012] In a preferred embodiment, the process of filtering target edge nodes based on verification confidence, controlling the target edge nodes to enter a dual-track operation mode, and progressively distributing the new dynamic control parameter set based on the comparison feedback results in the cloud-edge collaborative update module is as follows: After receiving the new dynamic control parameter set, which has passed virtual deployment verification, from the cloud optimization module, the cloud-edge collaborative update module first performs target edge node screening: for all edge nodes in the low-latency edge network, it evaluates them from three dimensions: node static attributes, node dynamic status, and node environment similarity. The evaluation criteria for the node static attribute dimension include the physical scale of the corresponding fermenter, the hardware computing power score of the edge node, and the network connection stability index. The evaluation criteria for the node dynamic status dimension directly use the process stability index recorded in the latest reported local operating status data of each edge node. The evaluation criteria for the node environment similarity dimension are obtained by calculating the degree of matching between the key parameters of the current operating environment of each node and the typical operating environment parameters on which the new dynamic control parameter set is based during optimization. The cloud-edge collaborative update module sets a weight coefficient for each of the three evaluation dimensions. For each edge node, the adaptation score is calculated as follows: the static attribute evaluation result of the node is converted into a static attribute normalized score, the process stability index of the node is used as the dynamic state score, and the environmental matching degree of the node is used as the environmental similarity score; the static attribute normalized score, dynamic state score and environmental similarity score are multiplied by the corresponding weight coefficients, and the three products are added together. The sum is the adaptation score of the edge node. The cloud-edge collaborative update module sorts all edge nodes from high to low according to their adaptation scores, selects the edge nodes with the highest adaptation scores and a preset number, and determines these selected edge nodes as target edge nodes. The cloud-edge collaborative update module controls the edge control modules on all target edge nodes to enter a dual-track operation mode. In this mode, the edge control module maintains two sets of control logic in parallel: one is a new control logic based on the new dynamic control parameter set, and the other is the old control logic based on the current dynamic control parameter set applied by the node. The cloud-edge collaborative update module sets an initial traffic scheduling ratio for all target edge nodes. This traffic scheduling ratio is defined as the percentage of control commands actually generated and issued to the actuator by the new control logic during real-time process data processing. This ratio is centrally controlled and managed by the cloud-edge collaborative update module and is called the traffic splitting threshold. When processing the received real-time process data, each target edge node dynamically determines whether to generate the actual control command using new or old control logic based on the diversion threshold. Regardless of which set of control logic ultimately generates the actual control command, the edge control module will synchronously run another set of control logic and completely record and mark all intermediate states calculated by the two sets of logic based on the same real-time process data, the generated draft control command, the final actual control command issued, and the real-time process data newly collected by the sensor after the actual control command is issued to the actuator and executed. The completely recorded and marked data constitutes the original data for comparison. The cloud-edge collaborative update module continuously collects comparative feedback data generated by each target edge node in dual-track operation mode. The comparative feedback data is extracted from the original data with complete records and tags.

[0013] In a preferred embodiment, based on the comparative feedback data, the cloud-edge collaborative update module periodically calculates a comprehensive improvement index for a new control logic. The specific process is as follows: Within a calculation cycle, for each target edge node, the same key performance indicator values ​​are extracted from the comparative feedback data of the new control logic and the old control logic. These key performance indicators include gas production per unit time and the process stability index. The difference between the key performance indicator values ​​of the new control logic and the corresponding values ​​of the old control logic at that node is calculated, and this difference is divided by the old control logic value to obtain the relative improvement rate of each key performance indicator at that node. The relative improvement rates of each key performance indicator calculated for all target edge nodes are summed. From the summation result, two penalty values ​​are subtracted. The first penalty value is equal to a preset first penalty factor multiplied by the variance of the average computing resource utilization rate of the target edge node group when running the new control logic within the calculation cycle. The second penalty value is equal to a preset second penalty factor multiplied by the total number of times the new control logic experiences anomalies or violates process safety constraints during the calculation cycle. The result obtained after the above summation and subtraction operations is defined as the comprehensive improvement index of the new control logic for that cycle. The cloud-edge collaborative update module presets a success threshold and a number of observation periods; the module decides to upgrade the promotion level only when the calculated comprehensive improvement index is consistently higher than the success threshold in multiple consecutive complete periods equal to the number of observation periods, and the total number of times the new control logic encounters anomalies or violates constraints during operation is zero. If, within any period, the calculated comprehensive improvement index is negative, or the total number of abnormalities or constraint violations during the operation of the new control logic surges and exceeds a preset alarm threshold, the cloud-edge collaborative update module triggers an automatic rollback mechanism, controlling the relevant target edge nodes to switch back to the old control logic operation, and feeding back the corresponding abnormal data to the cloud optimization module. This process iterates until the new dynamic control parameter set has passed long-term verification with high traffic ratios on all target edge nodes, and the calculated comprehensive improvement index is stable at a level higher than the success threshold. At this point, the cloud-edge collaborative update module decides to execute a steady-state synchronization across the entire network. After the network-wide steady-state synchronization command is executed, the edge control modules of all edge nodes in the system apply the new dynamic control parameter set, and the system then enters a new stable operating state. The cloud-edge collaborative update module summarizes and encapsulates the data generated during the entire deployment process, from target node selection, dual-track operation, gradual promotion to full network synchronization, including the comprehensive improvement index calculated in each cycle, the automatic rollback events that occurred, and the final overall system performance improvement effect, to form a strategy deployment performance report; the cloud-edge collaborative update module sends the strategy deployment performance report to the cloud optimization module.

[0014] The beneficial effects of this invention are as follows: Through real-time data fusion and autonomous decision-making of edge nodes, millisecond-level precise control of the operating conditions of individual fermenters is achieved, ensuring the stability of core parameters of anaerobic fermentation. With the help of low-latency collaboration between edge nodes, mutual interference between tank groups is effectively mitigated, improving the overall operational robustness. The digital twin model built on the cloud based on full historical and external data continuously discovers the globally optimal control strategy through reinforcement learning and securely verifies the generation of new parameters. Finally, through the progressive verification and distribution mechanism of cloud-edge collaboration, the optimized strategy is deployed to the entire network without loss and smoothly, forming an autonomous closed loop from local real-time control to global continuous optimization, which significantly improves gas production efficiency, operational stability and resource utilization. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0018] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application. Example 1

[0019] This embodiment provides, for example Figure 1-2 The system shown is an integrated crop-livestock farming system for anaerobic fermentation and precise reuse of biogas slurry from agricultural waste. Specifically, it includes: an edge control module, an edge collaboration module, a cloud optimization module, and a cloud-edge collaborative update module. Edge control module: Deployed on the edge nodes corresponding to each fermenter, it responds to real-time process data collected by the fermenter's sensors, fuses the real-time process data with the dynamic control parameter set currently applied by the edge control module, generates a process stability index, and generates and executes the first control command locally based on the comparison result between the process stability index and the dynamic control parameter set to adjust the actuators of the fermenter; the edge control module also shares local operating status data including the process stability index through a low-latency edge network; Edge coordination module: Deployed on edge nodes, it responds to the first control command generated by the edge control module and the shared local operating status data. It obtains the local operating status data shared by at least one other edge node through the low-latency edge network, and coordinates the first control command it generates based on the distributed coordination algorithm to generate and execute the second control command to suppress the coordination disturbance between tank groups. The cloud optimization module is used to receive and aggregate historical operating data, including local operating status data, and external environmental data uploaded by each edge node. It builds and runs a digital twin model of the anaerobic fermentation process, and iteratively optimizes the dynamic control parameter set in the digital twin model through a reinforcement learning optimization engine to generate a new dynamic control parameter set. It then performs virtual deployment verification of the new dynamic control parameters to obtain the verification confidence level. The cloud-edge collaborative update module is used to generate a new set of dynamic control parameters in the cloud optimization module and complete virtual deployment verification. Based on the verification confidence level, it selects target edge nodes and controls the edge control modules on the target edge nodes to enter a dual-track operation mode. This mode compares the new control logic based on the new set of dynamic control parameters with the old control logic based on the current set of dynamic control parameters applied by the edge control modules. Based on the comparison feedback, the new set of dynamic control parameters is progressively distributed to all edge nodes in the network to update the set of dynamic control parameters applied by each edge control module.

[0020] In this embodiment, it is specifically necessary to explain that in the edge control module, the process of fusing real-time process data with dynamic control parameter sets to generate a process stability index is as follows: The edge control module performs standardized preprocessing on the real-time process data collected from the sensors. The real-time process data includes temperature, pH, stirrer current and feed flow rate. The preprocessing is used to eliminate the influence of different physical dimensions and form a standardized feature vector. Standardized preprocessing can be done by Z-score standardization or Min-Max normalization to map the values ​​of each parameter to a unified dimension range, which is convenient for subsequent fusion calculation. At the same time, a fixed-length first-in-first-out (FIFO) data buffer is maintained to store the standardized feature vectors within the most recent time window. This ensures that the FIFO data buffer stores multiple standardized feature vectors arranged in chronological order, forming a standardized feature vector sequence. The fixed length corresponds to, for example, a 60-second time window. Data is collected once per second, so the buffer length is 60, which is used to provide sufficient short-term historical data to calculate dynamic features. Based on the standardized feature vector sequence stored in the first-in-first-out data buffer, a set of derived feature vectors that can characterize the short-term dynamic behavior of the process are calculated; wherein, the derived feature vectors include gradient consistency within the window and feature trajectory curvature; The calculation process for gradient consistency within the window is performed on any parameter among temperature, pH, agitator current, or feed flow rate. Specifically, it involves: First, calculating the difference between adjacent values ​​of the parameter within the time window corresponding to the first-in-first-out (FIFO) data buffer to form a difference sequence; then, calculating the dispersion of this difference sequence to obtain a dispersion measure; finally, substituting the dispersion measure into a negative exponential function with the natural constant as the base for mapping, and obtaining the result is the gradient consistency within the window for the parameter. The closer the value is to one, the more stable the recent trend of the parameter's change. The dispersion is preferably calculated using the mean absolute deviation, and the attenuation coefficient in the negative exponential function is used to adjust the strength of sensitivity to fluctuations. Its value can be set according to the stability requirements of the specific process. The calculation process of the curvature of the feature trajectory is as follows: Based on the standardized feature vector sequence stored in the first-in-first-out data buffer, the approximate curvature value of the path formed by the standardized feature vector sequence in the multidimensional feature space composed of various parameter dimensions is calculated. The approximate curvature value is obtained by calculating the magnitude of the first-order difference vector and the magnitude of the second-order difference vector of the standardized feature vector sequence, and combining the magnitude of the first-order difference vector. This calculation method can effectively capture the abrupt change in the direction of change of the process state in the multidimensional space, and is an effective indicator of the agility of dynamic behavior. The edge control module combines the standardized feature vector with the derived feature vector to form a comprehensive feature vector; The edge control module calls its currently applied dynamic control parameter set, which contains a feature weight vector and a reference state vector. The edge control module subtracts the integrated feature vector from the reference state vector element-wise to obtain a difference vector. It then multiplies the feature weight vector and the difference vector element-wise to obtain a weighted deviation vector. The module calculates the magnitude of the weighted deviation vector to obtain a scalar distance value representing the degree of integrated deviation. A scaling factor greater than zero is multiplied by the difference between the scalar distance value and a real offset, and the resulting product is used as the exponent of an exponential function with the natural constant as its base. The calculation of this exponential function... The result, when added to one and the reciprocal taken, yields a continuous value between zero and one. This value is defined as the process stability index, where one represents complete stability and zero represents extreme instability. The process stability index, calculated using the aforementioned nonlinear saturation function, can accurately and continuously quantify the instantaneous stability of the fermentation process. The closer the value is to 1, the more stable it is, providing an intuitive and reliable scalar basis for subsequent control decisions. Preferably, the stability threshold of the process stability index is set to 0.8. When the index value is greater than or equal to 0.8, the fermentation process is determined to be in a stable state; when it is less than 0.8, it indicates that the process has an instability risk and requires intervention. The specific process of generating and executing the first control command locally based on the process stability index to adjust the operation of the fermenter's actuators is as follows: The edge control module reads an exponential action mapping table from the dynamic control parameter set of its current application. This table divides the range of the process stability index into multiple fuzzy intervals with overlapping regions, and associates each interval with a preset control strategy for adjusting the actuators of the fermenter. For example, the interval [0, 0.6) can be defined as a "high-risk" interval, associated with a strategy to significantly increase heating power and stirring rate; the interval [0.6, 0.8) can be defined as a "caution" interval, associated with a strategy to slightly increase temperature or fine-tune stirring; and the interval [0.8, 1.0] can be defined as a "stable" interval, associated with a strategy to maintain current parameters or perform energy-saving fine-tuning. This fuzzy interval division with overlapping regions helps avoid frequent oscillations of control commands near the threshold, improving the smoothness of system control. The edge control module queries the index action mapping table based on the calculated process stability index, matches the fuzzy interval to which the process stability index belongs and the corresponding control strategy, and generates the corresponding first control command based on the matched control strategy. The first control command includes the command type and specific parameters for adjusting the power of the heater, the speed of the stirrer, or the opening of the feed valve. While generating the corresponding first control command based on the matched control strategy, the edge control module encapsulates a structured local operating status data. This local operating status data includes: the current time of generating the first control command, the calculated process stability index, the temperature and pH from the real-time process data, and the command type and specific parameters contained in the first control command. This structured encapsulation of the local operating status data ensures the integrity and uniformity of key information. The process stability index and original key parameters (temperature, pH) provide core inputs for assessing the status of itself and neighboring tanks, while the summary of the first control command reveals the node's control intention. These three elements together constitute the data foundation for edge collaborative decision-making. The edge control module then sends the encapsulated local operating status data out through a low-latency edge network. The transmission method can be periodic active broadcasting or responding to data requests from other edge nodes, thereby establishing a real-time, lightweight state synchronization mechanism among distributed edge nodes, laying the foundation for information flow for subsequent group collaborative optimization.

[0021] In this embodiment, it is specifically necessary to explain that in the edge collaboration module, the process of obtaining local operating status data shared by at least one other edge node through a low-latency edge network and coordinating based on a distributed collaboration algorithm is as follows: After the response is triggered, the edge collaboration module obtains the local operating status data shared by its neighboring edge nodes from the low-latency edge network. For each neighboring node, the edge collaboration module calculates a dynamic coupling coefficient to quantify the potential influence of the neighboring node on the current node. The dynamic coupling coefficient reflects both spatial proximity and state similarity. The larger the value, the more significant the influence of the neighboring node on the current node. The calculation of the dynamic coupling coefficient utilizes both the physical distance between the current node and the corresponding fermenters of neighboring nodes, as well as the degree of difference in their process stability indices. The parameters used to calculate the dynamic coupling coefficient include a distance scale parameter, a stability difference scale parameter, and an overall decay coefficient. These parameters are contained in a set of dynamic control parameters generated and distributed by the cloud optimization module. The distance scale parameter normalizes the physical distance, the stability difference scale parameter normalizes the difference in process stability indices, and the overall decay coefficient controls the rate at which the coupling strength decays with the overall difference. The specific calculation process is as follows: First, calculate the distance between the current node and the corresponding fermenters of neighboring nodes. The distance component is obtained by comparing the square of the physical distance between fermentation tanks with the square of the distance scale parameter. Next, the square of the absolute value of the difference between the process stability index of the current node and the process stability index of the neighboring nodes is calculated and compared with the square of the stability difference scale parameter to obtain a stability difference component. Then, the distance component and the stability difference component are added to obtain a combined value. After that, the overall attenuation coefficient is multiplied by the combined value and then the result is negative to obtain an exponential value. Finally, the exponent of the exponential function with the natural constant as the base is set as the exponential value, and the value of the exponential function is calculated. The result is the dynamic coupling coefficient. Based on the calculated dynamic coupling coefficients, the edge collaboration module further calculates two collaboration metrics: The first coordination indicator is the regional average disturbance trend, which is calculated as follows: First, for each neighboring node, the specific parameter change for adjusting heater power included in the first control command of that neighboring node is multiplied by the corresponding dynamic coupling coefficient to obtain a weighted power change. Then, the weighted power changes calculated for all neighboring nodes are summed to obtain a total weighted power change. At the same time, the dynamic coupling coefficients corresponding to all neighboring nodes are summed, and a positive constant is added to the summation result to prevent division by zero, resulting in a total coupling coefficient. Finally, the total weighted power change is divided by the total coupling coefficient, and the quotient is defined as the regional average disturbance trend. This calculation method makes neighboring nodes with stronger coupling between the control intention and the current node have a greater weight in influencing the regional trend. The second coordination metric is local state gradient consistency, which is calculated as follows: First, for each neighboring node, the following steps are performed: Calculate the difference between the temperature of the current node and the temperature of the neighboring node, and divide this difference by a preset temperature normalization factor to obtain a temperature normalized difference; calculate the difference between the pH of the current node and the pH of the neighboring node, and divide this difference by a preset pH normalization factor to obtain a pH normalized difference; add the absolute value of the temperature normalized difference to the absolute value of the pH normalized difference to obtain a combined absolute difference value; multiply the combined absolute difference value by the corresponding dynamic coupling coefficient to obtain a weighted difference value; then, for the... The weighted difference values ​​calculated from neighboring nodes are summed to obtain a total weighted difference. Simultaneously, the dynamic coupling coefficients corresponding to all neighboring nodes are summed, with a positive constant added to the summation to prevent division by zero, resulting in a total coupling coefficient. Next, the total weighted difference is divided by the total coupling coefficient to obtain an average weighted difference value. Finally, the average weighted difference value is subtracted from the number one, and the result is defined as the local state gradient consistency. The closer this index value is to one, the more consistent the current node is with its neighboring nodes in key state parameters. Temperature and pH normalization factors are used to eliminate the influence of different dimensions, enabling comparable superposition of temperature and pH differences. The process of coordinating the first control command generated by itself based on the distributed cooperative algorithm to generate and execute the second control command is as follows: The edge collaboration module is based on the calculated consistency between the regional average perturbation trend and the local state gradient, and defines a decision correction quantity for adjusting the heater power; The edge collaboration module constructs a collaborative potential energy function, the value of which is determined by the first, second and third terms. The first term is the fidelity term, the value of which is equal to a first weighting coefficient multiplied by the square of the deviation of the correction amount to be decided from the specific parameter change used to adjust the heater power in the first control command. This term is used to ensure that the collaborative correction does not deviate excessively from the local optimal command calculated based on the state of the tank, and to maintain the performance of the single tank. The second term is the consistency term, which is equal to a second weighting coefficient multiplied by the square of the difference between the decision correction amount and the regional average disturbance trend. This term is used to drive the current node's actions to align with the overall trend of the local region, thereby achieving load balancing and disturbance synergy suppression. The third term is the incentive term, which is equal to the product of a third weight coefficient, local state gradient consistency, and the amount of adjustment to be decided. This term is directional. When the local state gradient consistency is high, its effect tends to incentivize the adoption of control actions in the same direction as the regional trend, thereby achieving synergistic effects through state consistency. The final value of the collaborative potential energy function is equal to the sum of the first and second values, minus the third value. The first, second, and third weighting coefficients are preset positive numbers, included in the set of dynamic control parameters generated and distributed by the cloud optimization module. The relative magnitudes of these three weighting coefficients determine the priority of the three objectives of fidelity, consistency, and incentive in the final decision. The edge coordination module multiplies the first weighting coefficient by the specific parameter change used to adjust the heater power in the first control command to obtain the first product; it multiplies the second weighting coefficient by the regional average disturbance trend to obtain the second product; it multiplies the third weighting coefficient by the local state gradient consistency and then divides by two to obtain the third product; it adds the first, second, and third products to obtain the numerator sum; it adds the first and second weighting coefficients to obtain the denominator sum; and it divides the numerator sum by the denominator sum, and the resulting quotient is defined as the optimal coordinated action. This calculation process is the condition that the decision correction quantity should satisfy when the coordinated potential energy function reaches its mathematical minimum. The optimal coordinated action can be calculated directly and efficiently through this analytical solution, avoiding a complex iterative optimization process and meeting the real-time requirements of the edge side. The edge collaboration module generates a second control command based on the optimal collaborative action. This command will override or adjust the first control command and be delivered to the actuator to suppress collaborative disturbances between tank groups. The generation of the second control command marks that the edge collaboration module has completed real-time decision-making based on local group information, realizing collaborative control from single-point intelligence to group intelligence.

[0022] In this embodiment, it is specifically necessary to explain the process in the cloud optimization module of receiving and aggregating historical operating data and external environmental data uploaded by each edge node, and constructing and running a digital twin model of the anaerobic fermentation process as follows: The cloud optimization module performs time alignment and cleaning on the received historical operation data. The historical operation data includes local operation status data periodically uploaded by each edge node. The cloud optimization module connects to external data sources to obtain external environmental data. The external environmental data includes meteorological forecast information reflecting atmospheric conditions (ambient temperature, solar intensity, precipitation) and raw material characteristic parameters reflecting waste composition (such as carbon-nitrogen ratio of feed, organic load rate). The cloud-based optimization module performs the following steps to fuse data and extract features: First, the timestamps of historical operating data and external environmental data are aligned. Next, for key process variables in the historical operating data—process stability index, temperature, and pH—statistical characteristics are calculated at multiple preset time scales: one hour, day, and week. These characteristics include weighted moving averages, linear trend slopes within local time windows, and peak characteristics within selected time periods. Weighted moving averages assign higher weights to recent data to more sensitively reflect the latest trends. Linear trend slopes quantify the rate of change of process variables and are a key indicator for predicting short-term state evolution. Then, for each feature in the external environmental data, the mutual information value between it and the key process variables is calculated. Based on the calculated mutual information values, a correlation weight vector is formed, which quantifies the nonlinear correlation strength between various external environmental features and internal process variables. Mutual information effectively captures nonlinear relationships and is more suitable for complex biochemical processes than traditional correlation coefficients. Finally, all the calculated statistical characteristics at different time scales are combined with the correlation weight vector to form a fused feature vector, which integrates multi-scale information on the process's internal dynamics and external disturbances. The digital twin model is constructed using a spatiotemporal graph convolutional network architecture enhanced with an attention mechanism. In this architecture, each fermenter and its associated process pipeline are modeled as a graph node, and the physical connections or material and energy exchange relationships between fermenters or between fermenters and pipelines are modeled as edges connecting these graph nodes. This forms a graph structure that describes the topology of the entire anaerobic fermentation system. This graph structure can naturally express the spatial and process relationships between the components in the system. The spatiotemporal graph convolutional network architecture uses the edge connections of a graph structure to perform spatial convolution operations on node features composed of fused feature vectors to learn the mutual influence between multiple fermenters. For example, how temperature fluctuations in one tank affect adjacent tanks through heat exchange. At the same time, this spatiotemporal graph convolutional network architecture learns the dynamic evolution of the internal state of a single fermenter over time through its internal temporal processing unit, thereby achieving temporal prediction of processes such as gas production and acidification. The attention calculation mechanism embedded in the spatiotemporal graph convolutional network architecture dynamically calculates the importance weights of different input features to the current prediction target in each operation, and dynamically calculates the importance weights of the mutual influence between any two nodes connected by an edge in the graph structure. This enables the digital twin model to adaptively focus on key features and key coupling paths under the current conditions, improving prediction accuracy and interpretability. The digital twin model also incorporates a mathematical model describing the mechanistic dynamics of microbial growth and substrate consumption as prior knowledge into the computation process of the spatiotemporal graph convolutional network architecture. This allows the predictions of the digital twin model to simultaneously follow data-driven principles and biochemical reaction mechanisms. This hybrid modeling approach can ensure the physical rationality of predictions based on mechanisms and improve the generalization ability of the model when dealing with small samples or drastic changes in operating conditions. The digital twin model takes the fused feature vector as input and outputs predicted values ​​for the gas production rate, volatile fatty acid concentration and process stability index of the fermentation system at future time points, as well as the average process stability index of all fermenters used to calculate the average operating stability of the entire system. The process of iteratively optimizing the dynamic control parameter set in the digital twin model using a reinforcement learning optimization engine to generate a new dynamic control parameter set is as follows: In the optimization framework, the cloud optimization module is defined as an intelligent agent, the digital twin model is defined as the environment, the intelligent agent performs actions in the environment to adjust the values ​​of various parameters in the dynamic control parameter set, and the state of the environment feedback is the key indicators output by the digital twin model, including gas production rate, average process stability index of the whole system, and energy consumption level. The cloud optimization module sets a multi-objective optimization task. The objective of this task is to find the set of dynamic control parameters that can maximize the long-term comprehensive performance of the system in the environment simulated by the digital twin model, under the premise of meeting the preset process safety constraints. Long-term overall performance is quantified using a multi-objective weighted reward function. This function is equal to the sum of three factors: a gas production reward to incentivize increased gas production, a stability reward to incentivize improved average process stability index, and an energy efficiency penalty to penalize increased energy consumption. Each factor is multiplied by its corresponding positive weighting coefficient. The gas production reward is positively correlated with the cumulative gas production over the prediction period, the stability reward is positively correlated with the average process stability index over the prediction period, and the energy efficiency penalty is positively correlated with the total heating and stirring energy consumption over the prediction period. The three weighting coefficients are used to balance the cumulative gas production and energy consumption. The relative importance of the three objectives of yield, stability, and energy consumption can be set through domain knowledge or Pareto front analysis. Process safety constraints are monitored in real time during the simulation of the digital twin model. These constraints include ensuring that the concentration of volatile fatty acids derived from the simulation does not exceed a set inhibition threshold, and that temperature fluctuations during the simulation do not exceed a set range limit. If any constraint is violated in the simulation, a preset penalty value corresponding to the constraint violation is subtracted from the sum of the reward functions. The inhibition threshold is used to prevent acidification of the fermentation system, and the range limit is used to ensure that the microbial activity is within the high-efficiency range. The reinforcement learning optimization engine employs a policy gradient algorithm with constraints. It uses a digital twin model as the simulation environment and the set of dynamic control parameters to be optimized as the action policy. The dynamic control parameter set includes the feature weight vector and baseline state vector used in the edge control module to calculate the process stability index, and the first, second, and third weight coefficients used in the edge collaboration module to construct the collaborative potential function. By executing numerous virtual trials in the simulation environment, feedback data on the reward function and constraint violations are collected, and these feedback data are used to iteratively update the parameters of the action policy. The policy gradient algorithm with constraints can satisfy all safety constraints with a high probability while exploring the optimal policy, ensuring the safety of the optimization results. In each policy update, the algorithm not only pursues the growth of the reward function value, but also ensures that the updated policy can meet all process safety constraints by introducing a constraint violation penalty mechanism. After multiple rounds of iterative optimization, the reinforcement learning optimization engine outputs a new set of dynamic control parameters that is both performance-enhanced and safe and reliable. After the reinforcement learning optimization engine outputs a new set of dynamic control parameters, the cloud optimization module performs the following steps to complete the security verification and distribution of this parameter set: Importing the new dynamic control parameter set into the digital twin model; setting various typical operating conditions and preset extreme operating conditions within the digital twin model, and performing simulations using the new dynamic control parameter set; calculating the confidence level of the expected performance improvement brought about by applying the new dynamic control parameter set based on the simulation results; the confidence level can be calculated based on the statistical significance test of multiple simulation results and historical benchmark performance, for example, requiring the improvement to be statistically significant (P-value less than 0.05); monitoring the performance improvement throughout the simulation process. Whether the concentration of volatile fatty acids simulated by the twin model exceeds the inhibition threshold, and whether the simulated temperature fluctuation exceeds the range limit; an exemplary value for the inhibition threshold is 5,000 mg / L, and an exemplary value for the temperature fluctuation range limit is ±2 degrees Celsius of the reference temperature; only when the calculated confidence level of the expected performance improvement is higher than a preset confidence level threshold (e.g., 95%), and the simulation run under all set operating conditions does not trigger any process safety constraints, is the new dynamic control parameter set marked as having passed virtual deployment verification. The cloud optimization module will output the new dynamic control parameter set marked as having passed verification to the cloud-edge collaborative update module.

[0023] In this embodiment, it is specifically necessary to explain the process in the cloud-edge collaborative update module, which involves filtering target edge nodes based on verification confidence, controlling the target edge nodes to enter dual-track operation mode, and progressively distributing the new dynamic control parameter set based on the comparison feedback results. After receiving the new dynamic control parameter set, which has passed virtual deployment verification, from the cloud optimization module, the cloud-edge collaborative update module first performs target edge node screening: for all edge nodes in the low-latency edge network, it evaluates them from three dimensions: node static attributes, node dynamic status, and node environment similarity. The evaluation criteria for the node static attribute dimension include the physical scale of the corresponding fermenter, the hardware computing power score of the edge node, and the network connection stability index. The evaluation criteria for the node dynamic status dimension directly use the process stability index recorded in the latest reported local operating status data of each edge node. The evaluation criteria for the node environment similarity dimension are obtained by calculating the degree of matching between the key parameters of the current operating environment of each node and the typical operating environment parameters on which the new dynamic control parameter set is based during optimization. The cloud-edge collaborative update module sets a weight coefficient greater than zero for each of the three evaluation dimensions, and the sum of the three weight coefficients is one. For each edge node, the adaptation score is calculated as follows: the static attribute evaluation result of the node is converted into a static attribute normalized score, the process stability index of the node is used as the dynamic state score, and the environmental matching degree of the node is used as the environmental similarity score; the static attribute normalized score, dynamic state score and environmental similarity score are multiplied by the corresponding weight coefficients, and the three products are added together. The sum is the adaptation score of the edge node. The cloud-edge collaborative update module sorts all edge nodes from high to low according to their adaptation scores, selects the edge nodes with the highest adaptation scores and a preset number, and determines these selected edge nodes as target edge nodes. The cloud-edge collaborative update module controls the edge control modules on all target edge nodes to enter a dual-track operation mode. In this mode, the edge control module maintains two sets of control logic in parallel: one is a new control logic based on the new dynamic control parameter set, and the other is the old control logic based on the current dynamic control parameter set applied by the node. The cloud-edge collaborative update module uniformly sets an initial traffic scheduling ratio for all target edge nodes. This traffic scheduling ratio is defined as the percentage of control commands actually generated and issued to the actuator by the new control logic during real-time process data processing. This ratio is centrally controlled and managed by the cloud-edge collaborative update module and is called the traffic splitting threshold. For example, the initial traffic scheduling ratio can be set to five percent to conduct preliminary verification of the new logic with the smallest possible range. When processing the received real-time process data, each target edge node dynamically determines whether to use the new or old control logic to generate the actual control command based on the diversion threshold. Regardless of which control logic ultimately generates the actual control command, the edge control module will simultaneously run another set of control logic and completely record and mark all intermediate states calculated by the two sets of logic based on the same real-time process data, the generated draft control command, the final actual control command issued, and the real-time process data newly collected by the sensor after the actual control command is issued to the actuator and executed. The completely recorded and marked data constitutes the raw data for comparison. This dual-track synchronous operation and full-process recording mechanism ensures fair comparison between the new and old logics under exactly the same input conditions, providing a reliable data foundation for subsequent accurate evaluation. The cloud-edge collaborative update module continuously collects comparative feedback data generated by each target edge node in dual-track operation mode. The comparative feedback data is extracted from the original data with complete records and tags. Based on comparative feedback data, the cloud-edge collaborative update module periodically calculates the comprehensive improvement index of a new control logic. The specific process is as follows: Within a calculation cycle, for each target edge node, the same key performance indicator values ​​are extracted from the comparative feedback data of the new and old control logics. These key performance indicators include at least gas production per unit time and the process stability index. The difference between the key performance indicator values ​​of the new and old control logics at that node is calculated, and this difference is divided by the old control logic value to obtain the relative improvement rate of each key performance indicator at that node. The relative improvement rates of each key performance indicator calculated for all target edge nodes are summed. From the summation result, two penalty items are subtracted. The first penalty item is equal to a preset first penalty factor multiplied by the calculation cycle. The variance of the average computing resource (including CPU and memory) utilization rate of the target edge node group when running new control logic is calculated. The second penalty is equal to a preset second penalty factor multiplied by the total number of times the new control logic runs abnormally or violates process safety constraints within the calculation cycle. The result obtained after the above summation and subtraction operations is defined as the comprehensive improvement index of the new control logic in that cycle. The degree of computing resource fluctuation can be quantified by calculating the variance of the average CPU utilization rate of the target edge node group. The values ​​of the first penalty factor and the second penalty factor are used to balance performance improvement with system stability and security. For example, they can be set to 0.1 and 10.0 respectively. The cloud-edge collaborative update module has a preset success threshold and a number of observation periods. For example, the success threshold can be set to 0.05 (i.e., the overall improvement index is required to be consistently higher than 5%), and the number of observation periods can be set to 3. The module decides to upgrade the promotion level only when the calculated overall improvement index is consistently higher than the success threshold in multiple consecutive complete periods equal to the number of observation periods, and the total number of times the new control logic runs abnormally or violates constraints is zero. Upgrading the promotion level manifests in two ways: First, increasing the traffic scheduling ratio, i.e., adjusting the diversion threshold value, allowing more real-time requests to be processed by the new logic; each upgrade can be set to a fixed step size, such as an increase of 10%. Second, distributing the new dynamic control parameter set to the next batch of edge nodes with slightly lower adaptation scores and controlling them to enter a dual-track operation mode. If, within any period, the calculated comprehensive improvement index is negative or the total number of anomalies or constraint violations during the operation of the new control logic surges and exceeds a preset alarm threshold; an example value for the alarm threshold is 3 times per calculation period, then the cloud-edge collaborative update module triggers an automatic rollback mechanism, controlling the relevant target edge nodes to fully switch back to the old control logic operation, and feeding back the corresponding abnormal data to the cloud optimization module; this process iterates cyclically. The process continues until the new dynamic control parameter set has passed long-term verification with high traffic ratios on all target edge nodes, and the calculated comprehensive improvement index remains stable above the success threshold. At this point, the cloud-edge collaborative update module decides to execute a network-wide steady-state synchronization. An example value for "high traffic ratio" is 95%, and "long-term verification" means that the promotion conditions are met for multiple consecutive observation periods (e.g., 10). The network-wide steady-state synchronization process is as follows: instructions are sent to all remaining edge nodes in the network, except for the target edge nodes that have completed verification, instructing the edge control modules on these remaining edge nodes to directly replace their locally stored original dynamic control parameter sets with the new dynamic control parameter set. Since sufficient verification has been carried out on the most representative nodes in the early stage, the operational risk of this network-wide synchronization is reduced to an extremely low level. After the network-wide steady-state synchronization command is executed, the edge control modules of all edge nodes in the system apply the new dynamic control parameter set, and the system then enters a new stable operating state. The cloud-edge collaborative update module summarizes and encapsulates the data generated during the entire deployment process—from target node selection, dual-track operation, gradual promotion to full network synchronization—including the comprehensive improvement index calculated in each cycle, automatic rollback events, and the final overall system performance improvement, into a strategy deployment performance report. This report is then sent to the cloud optimization module. This report provides the cloud optimization module with practical deployment experience to initiate the next round of dynamic control parameter set optimization, thus forming a complete autonomous closed loop from optimization, verification, deployment to feedback. This feedback loop mechanism enables the system to continuously utilize actual production data to drive the evolution of control strategies, achieving continuous performance improvement and adaptive optimization.

[0024] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0025] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0026] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0027] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0028] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0029] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0030] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A system for the precise reuse of anaerobic fermentation and biogas slurry in agricultural waste through integrated crop-livestock farming, characterized in that, Specifically, it includes: The module comprises an edge control module, an edge collaboration module, a cloud optimization module, and a cloud-edge collaboration update module; Edge control module: Deployed on the edge nodes corresponding to each fermenter, it responds to real-time process data collected by the fermenter's sensors, fuses the real-time process data with the dynamic control parameter set currently applied by the edge control module, generates a process stability index, and generates and executes the first control command locally based on the comparison result between the process stability index and the dynamic control parameter set to adjust the actuators of the fermenter; the edge control module also shares local operating status data including the process stability index through a low-latency edge network; Edge collaboration module: Deployed on edge nodes, it responds to the first control command generated by the edge control module and the shared local operating status data. It obtains the local operating status data shared by at least one other edge node through the low-latency edge network, and coordinates the first control command it generates based on the distributed collaboration algorithm to generate and execute the second control command. The cloud optimization module is used to receive and aggregate historical operating data, including local operating status data, and external environmental data uploaded by each edge node. It builds and runs a digital twin model of the anaerobic fermentation process, and iteratively optimizes the dynamic control parameter set in the digital twin model through a reinforcement learning optimization engine to generate a new dynamic control parameter set. It then performs virtual deployment verification of the new dynamic control parameters to obtain the verification confidence level. The cloud-edge collaborative update module is used to generate a new set of dynamic control parameters in the cloud optimization module and complete virtual deployment verification. Based on the verification confidence level, it selects target edge nodes and controls the edge control modules on the target edge nodes to enter a dual-track operation mode. This mode compares the new control logic based on the new set of dynamic control parameters with the old control logic based on the current set of dynamic control parameters applied by the edge control modules. Based on the comparison feedback, the new set of dynamic control parameters is progressively distributed to all edge nodes in the network to update the set of dynamic control parameters applied by each edge control module.

2. The integrated crop-livestock anaerobic fermentation and biogas slurry precision reuse system for agricultural waste, as described in claim 1, is characterized in that: In the edge control module, the process of fusing real-time process data with dynamic control parameter sets to generate a process stability index is as follows: The edge control module performs standardized preprocessing on the real-time process data collected from the sensors. The real-time process data includes temperature, pH, stirrer current and feed flow rate, forming a standardized feature vector. At the same time, a fixed-length first-in-first-out (FIFO) data buffer is maintained to store the standardized feature vectors within the most recent time window, so that the FIFO data buffer stores multiple standardized feature vectors arranged in chronological order, forming a standardized feature vector sequence. Based on the standardized feature vector sequence stored in the first-in-first-out data buffer, a set of derived feature vectors is calculated; wherein, the derived feature vectors include the gradient consistency within the window and the curvature of the feature trajectory; The calculation process for gradient consistency within the window is performed for any parameter among temperature, pH, stirrer current, or feed flow rate. Specifically, it involves: first, calculating the difference between adjacent values ​​of the parameter within the time window corresponding to the first-in-first-out data buffer to form a difference sequence; then, calculating the dispersion of the difference sequence to obtain a dispersion measure; finally, substituting the dispersion measure into a negative exponential function with the natural constant as the base for mapping, and the result is the gradient consistency within the window for the parameter. The calculation process of the curvature of the feature trajectory is as follows: Based on the standardized feature vector sequence stored in the first-in-first-out data buffer, the approximate curvature of the path formed by the standardized feature vector sequence in the multidimensional feature space composed of each parameter dimension is calculated. The approximate curvature is obtained by calculating the magnitude of the first-order difference vector and the magnitude of the second-order difference vector of the standardized feature vector sequence, and combining the magnitude of the first-order difference vector. The edge control module combines the standardized feature vector with the derived feature vector to form a comprehensive feature vector; The edge control module calls its currently applied dynamic control parameter set, which contains a feature weight vector and a reference state vector. The edge control module subtracts the comprehensive feature vector from the reference state vector element by element to obtain a difference vector. It then multiplies the feature weight vector from the difference vector element by element to obtain a weighted deviation vector. The module calculates the magnitude of the weighted deviation vector to obtain a scalar distance value. It multiplies a scaling factor by the difference between the scalar distance value and a real offset, and uses the resulting product as the exponent of an exponential function with the natural constant as the base. The result of this exponential function calculation is incremented by one and the reciprocal is taken to obtain a continuous value between zero and one, which is defined as the process stability index.

3. The system for precise reuse of anaerobic fermentation and biogas slurry in agricultural waste integrated with crop-livestock farming as described in claim 2, characterized in that: The specific process of generating and executing the first control command locally based on the process stability index to adjust the operation of the fermenter's actuators is as follows: The edge control module reads an exponential action mapping table contained in the dynamic control parameter set of its current application. The exponential action mapping table divides the range of the process stability index into multiple fuzzy intervals with overlapping regions and associates each fuzzy interval with a preset control strategy for adjusting the actuator of the fermenter. The edge control module queries the index action mapping table based on the calculated process stability index, matches the fuzzy interval to which the process stability index belongs and the corresponding control strategy, and generates the corresponding first control command based on the matched control strategy. The first control command includes the command type and specific parameters for adjusting the power of the heater, the speed of the stirrer, or the opening of the feed valve.

4. The integrated crop-livestock anaerobic fermentation and biogas slurry precision reuse system for agricultural waste, as described in claim 3, is characterized in that: While generating the corresponding first control command based on the matched control strategy, the edge control module encapsulates a structured local operating status data, which includes: the current time of generating the first control command, the calculated process stability index, the temperature and pH in the real-time process data, and the command type and specific parameters contained in the first control command. The edge control module then sends the encapsulated local operating status data out through the low-latency edge network.

5. The integrated crop-livestock anaerobic fermentation and biogas slurry precision reuse system for agricultural waste, as described in claim 4, is characterized in that: In the edge collaboration module, the process of obtaining local operating status data shared by at least one other edge node through a low-latency edge network and coordinating based on a distributed collaboration algorithm is as follows: Upon triggering a response, the edge collaboration module retrieves local operational status data shared by its neighboring edge nodes from the low-latency edge network. For each neighboring node, the edge collaboration module calculates a dynamic coupling coefficient. The calculation process is as follows: First, it calculates the ratio of the square of the physical distance between the current node and the corresponding fermenter of the neighboring node to the square of the distance scale parameter, obtaining a distance component. Next, it calculates the square of the absolute value of the difference between the process stability index of the current node and the process stability index of the neighboring nodes, and compares it with the square of the stability difference scale parameter, obtaining a stability difference component. Then, it adds the distance component and the stability difference component to obtain a combined value. After that, it multiplies the overall attenuation coefficient by the combined value and takes a negative number to obtain an exponential value. Finally, it sets the exponent of the exponential function with the natural constant as the base as the exponential value, calculates the value of the exponential function, and the result is the dynamic coupling coefficient. Based on the calculated dynamic coupling coefficients, the edge collaboration module further calculates two collaboration metrics: The first coordination indicator is the regional average disturbance trend, which is calculated as follows: First, for each neighboring node, the specific parameter change for adjusting the heater power contained in the first control command of that neighboring node is multiplied by the corresponding dynamic coupling coefficient to obtain a weighted power change; then, the weighted power changes calculated for all neighboring nodes are summed to obtain a total weighted power change; simultaneously, the dynamic coupling coefficients corresponding to all neighboring nodes are summed, and a positive constant is added to the summation result to obtain the total coupling coefficient; finally, the total weighted power change is divided by the total coupling coefficient, and the resulting quotient is defined as the regional average disturbance trend. The second coordination index is the consistency of local state gradients. The calculation process is as follows: First, for each neighboring node, perform the following steps: calculate the difference between the temperature of the current node and the temperature of the neighboring node, and divide the difference by a preset temperature normalization factor to obtain a temperature normalization difference. Calculate the difference between the pH of the current node and the pH of its neighboring nodes, and divide this difference by a preset pH normalization factor to obtain a pH normalized difference value. Add the absolute value of the temperature normalized difference to the absolute value of the pH normalized difference to obtain a combined absolute difference value. Multiply the combined absolute difference value by the corresponding dynamic coupling coefficient to obtain a weighted difference value. Then, sum the weighted difference values ​​calculated for all neighboring nodes to obtain a weighted difference sum. At the same time, sum the dynamic coupling coefficients corresponding to all neighboring nodes, and add a positive constant to the summation result to obtain a coupling coefficient sum. Next, divide the weighted difference sum by the coupling coefficient sum to obtain an average weighted difference value. Finally, subtract the average weighted difference value from the number one, and the result is defined as the local state gradient consistency.

6. The integrated crop-livestock anaerobic fermentation and biogas slurry precision reuse system for agricultural waste, as described in claim 5, is characterized in that: The process of coordinating the first control command generated by itself based on the distributed cooperative algorithm to generate and execute the second control command is as follows: The edge collaboration module is based on the calculated consistency between the regional average perturbation trend and the local state gradient, and defines a decision correction quantity for adjusting the heater power; The edge collaboration module constructs a collaborative potential energy function, the value of which is determined by the first, second and third terms. The first term is the fidelity term, and its value is equal to a first weighting coefficient multiplied by the square of the deviation of the decision correction amount from the specific parameter change used to adjust the heater power in the first control command. The second term is the consistency term, which is equal to a second weighting coefficient multiplied by the square of the difference between the adjustment amount to be decided and the regional average perturbation trend. The third term is the incentive term, whose value is equal to the product of a third weight coefficient, local state gradient consistency, and the adjustment amount to be decided. The final value of the cooperative potential energy function is equal to the sum of the first and second terms, minus the third term. The edge coordination module multiplies the first weighting coefficient by the specific parameter change used to adjust the heater power in the first control command to obtain the first product; multiplies the second weighting coefficient by the regional average disturbance trend to obtain the second product; multiplies the third weighting coefficient by the local state gradient consistency and then divides by two to obtain the third product; adds the first, second, and third products to obtain the numerator sum; adds the first and second weighting coefficients to obtain the denominator sum; and divides the numerator sum by the denominator sum, and the resulting quotient is defined as the optimal coordinated action. The edge collaboration module generates a second control command based on the optimal collaborative action. This command will override or adjust the first control command and be delivered to the actuator.

7. The integrated crop-livestock anaerobic fermentation and biogas slurry precision reuse system for agricultural waste, as described in claim 6, is characterized in that: The cloud optimization module receives and aggregates historical operational data and external environmental data uploaded by each edge node, and constructs and runs a digital twin model of the anaerobic fermentation process, specifically as follows: The cloud optimization module performs time alignment and cleaning on the received historical operation data. The historical operation data includes local operation status data periodically uploaded by each edge node. The cloud optimization module connects to external data sources to obtain external environmental data, which includes meteorological forecast information reflecting atmospheric conditions and raw material characteristic parameters reflecting the composition of waste. The cloud-based optimization module performs the following steps to fuse data and extract features: First, align the timestamps of historical operating data with those of external environmental data. Then, for key process variables such as process stability index, temperature, and pH in historical operating data, calculate their statistical characteristics at multiple preset time scales in the recent hour, day, and week. The statistical characteristics include weighted moving average, linear trend slope within a local time window, and peak characteristics within a selected time period. Then, for each feature in the external environment data, the mutual information value between it and the key process variables is calculated, and the correlation weight vector is formed based on the calculated mutual information value. Finally, all the calculated statistical features at different time scales are combined with the correlation weight vector to form a fused feature vector. The digital twin model is constructed using a spatiotemporal graph convolutional network architecture enhanced with an attention mechanism. In this architecture, each fermenter and its associated process pipeline are modeled as a graph node, and the physical connections or material and energy exchange relationships between fermenters or between fermenters and pipelines are modeled as edges connecting these graph nodes, thus forming a graph structure that describes the topology of the entire anaerobic fermentation system. The spatiotemporal graph convolutional network architecture uses the edge connections of the graph structure to perform spatial convolution operations on the node features composed of fused feature vectors to learn the mutual influence between multiple fermenters; at the same time, this spatiotemporal graph convolutional network architecture learns the dynamic evolution of the internal state of a single fermenter over time through its internal temporal processing unit. The attention calculation mechanism embedded in the spatiotemporal graph convolutional network architecture dynamically calculates the importance weights of different input features to the current prediction target in each operation, and dynamically calculates the importance weights of the mutual influence between any two nodes connected by an edge in the graph structure. The digital twin model takes a fused feature vector as input and outputs predicted values ​​for the gas production rate, volatile fatty acid concentration, and process stability index of the fermentation system at future moments, as well as the average process stability index of all fermenters used to calculate the average operating stability of the entire system.

8. The integrated crop-livestock anaerobic fermentation and biogas slurry precision reuse system for agricultural waste, as described in claim 7, is characterized in that: The process of iteratively optimizing the dynamic control parameter set in the digital twin model using a reinforcement learning optimization engine to generate a new dynamic control parameter set is as follows: In the optimization framework, the cloud optimization module is defined as an intelligent agent, the digital twin model is defined as the environment, the intelligent agent performs actions in the environment to adjust the values ​​of various parameters in the dynamic control parameter set, and the state of the environment feedback is the key indicators output by the digital twin model, including gas production rate, average process stability index of the whole system, and energy consumption level. The cloud optimization module sets a multi-objective optimization task. The objective of this task is to find the set of dynamic control parameters that can maximize the long-term comprehensive performance of the system in the environment simulated by the digital twin model, under the premise of meeting the preset process safety constraints. Long-term comprehensive performance is quantified by a multi-objective weighted reward function. The value of this reward function is equal to the sum of a gas production reward term to incentivize increased gas production, a stability reward term to incentivize improved average process stability index of the entire system, and an energy efficiency penalty term to penalize increased energy consumption, each multiplied by its corresponding positive weight coefficient. Process safety constraints are monitored in real time during the simulation of the digital twin model. These constraints include the requirement that the concentration of volatile fatty acids derived from the simulation must not exceed a set inhibition threshold, and that temperature fluctuations during the simulation must not exceed a set range limit. If any constraint is violated in the simulation, a preset penalty value corresponding to the constraint violation is subtracted from the sum of the reward functions. The reinforcement learning optimization engine employs a policy gradient algorithm with constraints. It uses a digital twin model as the simulation environment and the set of dynamic control parameters to be optimized as the action policy to be optimized. The dynamic control parameter set includes the feature weight vector and baseline state vector used in the edge control module to calculate the process stability index, and the first weight coefficient, second weight coefficient, and third weight coefficient used in the edge collaboration module to construct the collaborative potential function. By executing a large number of virtual experiments in the simulation environment, feedback data on the reward function and constraint violations are collected, and the parameters of the action policy are iteratively updated using this feedback data. After multiple rounds of iterative optimization, the reinforcement learning optimization engine outputs a new set of dynamic control parameters. After the reinforcement learning optimization engine outputs a new set of dynamic control parameters, the cloud optimization module performs the following steps to complete the security verification and distribution of the parameter set: The new dynamic control parameter set is imported into the digital twin model; within the digital twin model, various typical operating conditions and preset extreme operating conditions are set, and simulations are performed using the new dynamic control parameter set; based on the simulation results, the confidence level of the expected performance improvement brought about by applying the new dynamic control parameter set is calculated; throughout the simulation, the concentration of volatile fatty acids simulated by the digital twin model is monitored to see if it exceeds the inhibition threshold, and the simulated temperature fluctuation exceeds the range limit; only when the calculated confidence level of the expected performance improvement is higher than a preset confidence threshold, and the simulation under all set operating conditions does not trigger any process safety constraints, is the new dynamic control parameter set marked as having passed virtual deployment verification. The cloud optimization module then outputs the new dynamic control parameter set marked as having passed verification to the cloud-edge collaborative update module.

9. The integrated crop-livestock anaerobic fermentation and biogas slurry precision reuse system for agricultural waste, as described in claim 8, is characterized in that: In the cloud-edge collaborative update module, the process of filtering target edge nodes based on verification confidence, controlling the target edge nodes to enter dual-track operation mode, and progressively distributing the new dynamic control parameter set based on the comparison feedback results is as follows: After receiving the new dynamic control parameter set, which has passed virtual deployment verification, from the cloud optimization module, the cloud-edge collaborative update module first performs target edge node screening: for all edge nodes in the low-latency edge network, it evaluates them from three dimensions: node static attributes, node dynamic status, and node environment similarity. The evaluation criteria for the node static attribute dimension include the physical scale of the corresponding fermenter, the hardware computing power score of the edge node, and the network connection stability index. The evaluation criteria for the node dynamic status dimension directly use the process stability index recorded in the latest reported local operating status data of each edge node. The evaluation criteria for the node environment similarity dimension are obtained by calculating the degree of matching between the key parameters of the current operating environment of each node and the typical operating environment parameters on which the new dynamic control parameter set is based during optimization. The cloud-edge collaborative update module sets a weight coefficient for each of the three evaluation dimensions. For each edge node, the adaptation score is calculated as follows: the static attribute evaluation result of the node is converted into a static attribute normalized score, the process stability index of the node is used as the dynamic state score, and the environmental matching degree of the node is used as the environmental similarity score; the static attribute normalized score, dynamic state score and environmental similarity score are multiplied by the corresponding weight coefficients, and the three products are added together. The sum is the adaptation score of the edge node. The cloud-edge collaborative update module sorts all edge nodes from high to low according to their adaptation scores, selects the edge nodes with the highest adaptation scores and a preset number, and determines these selected edge nodes as target edge nodes. The cloud-edge collaborative update module controls the edge control modules on all target edge nodes to enter a dual-track operation mode. In this mode, the edge control module maintains two sets of control logic in parallel: one is a new control logic based on the new dynamic control parameter set, and the other is the old control logic based on the current dynamic control parameter set applied by the node. The cloud-edge collaborative update module sets an initial traffic scheduling ratio for all target edge nodes. This traffic scheduling ratio is defined as the percentage of control commands actually generated and issued to the actuator by the new control logic during real-time process data processing. This ratio is centrally controlled and managed by the cloud-edge collaborative update module and is called the traffic splitting threshold. When processing the received real-time process data, each target edge node dynamically determines whether to generate the actual control command using new or old control logic based on the diversion threshold. Regardless of which set of control logic ultimately generates the actual control command, the edge control module will synchronously run another set of control logic and completely record and mark all intermediate states calculated by the two sets of logic based on the same real-time process data, the generated draft control command, the final actual control command issued, and the real-time process data newly collected by the sensor after the actual control command is issued to the actuator and executed. The completely recorded and marked data constitutes the original data for comparison. The cloud-edge collaborative update module continuously collects comparative feedback data generated by each target edge node in dual-track operation mode. The comparative feedback data is extracted from the original data with complete records and tags.

10. The integrated crop-livestock anaerobic fermentation and biogas slurry precision reuse system for agricultural waste, as described in claim 9, is characterized in that: Based on the aforementioned comparative feedback data, the cloud-edge collaborative update module periodically calculates a comprehensive improvement index for a new control logic. The specific process is as follows: Within a calculation cycle, for each target edge node, the same key performance index values ​​are extracted from the comparison feedback data of the new control logic and the old control logic. The key performance indexes include gas production per unit time and process stability index. Calculate the difference between the key performance indicators of the new control logic and the corresponding values ​​of the old control logic at the node, and then divide this difference by the value of the old control logic to obtain the relative improvement rate of each key performance indicator at the node. Sum the relative improvement rates of each key performance indicator calculated for all target edge nodes. Subtract two penalty amounts from the above summation result. The first penalty amount is equal to a preset first penalty factor multiplied by the variance of the average computing resource utilization rate of the target edge node group when running the new control logic in the calculation period. The second penalty amount is equal to a preset second penalty factor multiplied by the total number of times the new control logic occurs or violates process safety constraints during the operation of the new control logic in the calculation period. The result obtained after the above summation and subtraction operations is defined as the comprehensive improvement index of the new control logic in this period. The cloud-edge collaborative update module presets a success threshold and a number of observation periods; the module decides to upgrade the promotion level only when the calculated comprehensive improvement index is consistently higher than the success threshold in multiple consecutive complete periods equal to the number of observation periods, and the total number of times the new control logic encounters anomalies or violates constraints during operation is zero. If the calculated comprehensive improvement index is negative within any period, or if the total number of abnormalities or constraint violations during the operation of the new control logic surges and exceeds a preset alarm threshold, the cloud-edge collaborative update module will trigger an automatic rollback mechanism, control the relevant target edge nodes to switch back to the old control logic, and feed back the corresponding abnormal data to the cloud optimization module. This process is repeated until the new dynamic control parameter set has passed long-term verification with high traffic ratios on all target edge nodes, and the calculated comprehensive improvement index is stable at a level higher than the success threshold. At this time, the cloud-edge collaborative update module decides to execute the steady-state synchronization of the entire network. After the network-wide steady-state synchronization command is executed, the edge control modules of all edge nodes in the system apply the new dynamic control parameter set, and the system then enters a new stable operating state. The cloud-edge collaborative update module summarizes and encapsulates the data generated during the entire deployment process, from target node selection, dual-track operation, gradual promotion to full network synchronization, including the comprehensive improvement index calculated in each cycle, the automatic rollback events that occurred, and the final overall system performance improvement effect, to form a strategy deployment performance report; the cloud-edge collaborative update module sends the strategy deployment performance report to the cloud optimization module.

Images