Four-flow coupling full life cycle evaluation method for planting and breeding waste resource treatment based on graph neural network

Abstract

The invention relates to the technical field of livestock and poultry manure and crop straw resourceful treatment, and particularly provides a four-flow coupling full life cycle evaluation method for planting and breeding waste resource treatment based on a graph neural network, and the method comprises the steps: obtaining four-flow real-time working condition data representing a system state, and constructing a heterogeneous process map; constructing a topology generation network, and generating a group of candidate process topology adjacency matrixes; calculating to obtain a total topology expected value, a path robustness score and a reconstruction conversion cost corresponding to each candidate process topology adjacent matrix; and obtaining a final decision score, selecting the candidate process topology adjacency matrix with the highest final decision score as the optimal process topology, and outputting a self-adaptive execution strategy corresponding to the optimal process topology. According to the method, the limitation that only parameter optimization can be carried out in a traditional fixed technological process is broken through, and dynamic reconstruction of the whole technological topological structure is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of resource treatment of agricultural and livestock waste, and specifically proposes a four-stream coupling full life cycle evaluation method for agricultural and livestock waste resource treatment based on graph neural network. Background Art

[0002] In the current field of agricultural waste resource processing, converting organic waste such as livestock and poultry manure and crop straw into energy or high-value chemicals is a core goal. To achieve this goal, it is usually necessary to combine multiple treatment units, such as anaerobic digestion, biological desulfurization, methane reforming, and methanol synthesis, to form a complete process flow. The operational performance of this process needs to be comprehensively evaluated from four dimensions: material flow, energy flow, environmental flow, and economic flow. It involves multiple complex indicators such as material conversion efficiency, energy transfer efficiency, environmental load, and operational benefits. Traditional process design often relies on a one-dimensional, fixed, and static process topology. That is, the combination and connection method of the treatment units are rarely changed once they are determined. In existing technologies, optimization methods mainly focus on parameter tuning based on established process flows, such as adjusting internal operating parameters such as temperature, pressure, or flow rate of a reaction unit. However, this type of approach has a fundamental limitation: it cannot reconfigure the process topology itself based on changes in external conditions. In real-world production, the system's input conditions are dynamic and changeable, including fluctuations in raw material properties, changes in market signals, and drift in internal equipment conditions. When these conditions change significantly, the original fixed process may no longer be the optimal choice and may even become inefficient or uneconomical. Existing technologies generally lack an intelligent mechanism that can respond to real-time data-driven, autonomously generate, and evaluate new process pathways. They are unable to deeply correlate and uniformly model multi-dimensional input data, such as raw materials, markets, and operating conditions, with the topological structure of the process flow. This results in the system being unable to fundamentally change the process pathway to avoid negative impacts or seize market opportunities when faced with situations such as high-nitrogen raw material input or soaring market prices for specific products. For example, such as activating ammonia capture units to cope with high-nitrogen raw materials, or switching to a biomass gasification pathway to quickly respond to methanol market demand. This approach, which is limited to parameter optimization, greatly limits the system's potential to maximize value over the entire life cycle in complex and dynamic environments.

[0003] Therefore, how to provide a system that can generate, evaluate and select processing processes at the topological level based on real-time, multi-dimensional input data, and achieve the transition from parameter optimization to process reconstruction, so that the system can dynamically adapt and always approach the global optimal operating state, is a technical problem that technical personnel in this field urgently need to solve.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a four-stream coupled full life cycle evaluation method for agricultural and livestock waste resource treatment based on graph neural networks to solve the problems raised in the above background technology.

[0006] The technical solutions of the present invention are as follows: A four-stream coupled life cycle assessment method for agricultural and livestock waste resource treatment based on graph neural networks, including: Step 1: Acquire multi-dimensional real-time operating data representing the system status, including raw material characteristic vectors, market signal vectors, and internal operating condition vectors, and construct a heterogeneous process map with preset processing units as nodes and potential process paths as edges. The quantitative values ​​of the material flow, energy flow, environmental flow and economic flow of the whole life cycle calculated based on the multi-dimensional real-time working condition data are set as the heterogeneous process map The 4D weight tensor of each edge in ; Step 2: Construct a topology generation network to map the heterogeneous process The multi-dimensional real-time working condition data is input into the topology generation network to generate a set of candidate process topology adjacency matrices ; Step 3: Build a multi-agent reinforcement learning system and map each candidate process topology adjacency matrix Instantiate to collaboratively learn the optimal operating parameter combination, and use the optimal operating parameter combination and the preset 4F-LCA function to calculate the adjacency matrix of each candidate process topology The corresponding topological total expected value ; Step 4: Calculate the topological adjacency matrix of each candidate process Path robustness score and restructuring switching costs ; Combined with the total expected value of the topology , path robustness score and restructuring switching costs , calculate the final decision score , and select the final decision score The highest candidate process topology adjacency matrix As the optimal process topology , to output the optimal process topology Corresponding adaptive execution strategy.

[0007] Preferably, in step 1, the step of obtaining the multi-dimensional real-time operating condition data includes: Obtaining the raw material characteristic vector by online sensor measurement, wherein the raw material characteristic vector includes carbon-nitrogen ratio, moisture content and lignin content; obtaining the market signal vector through an external interface, wherein the market signal vector includes a methanol market price, a carbon credit price, and a grid electricity price; The internal operating condition vector is acquired through sensor collection of each processing unit, and the internal operating condition vector includes temperature, pressure and flow.

[0008] Preferably, in step 1, the four-dimensional weight tensor is set The steps include: The transmission coefficient that characterizes the conversion efficiency of matter between two nodes is set as the four-dimensional weight tensor Material flow weight in ; The value representing the energy transfer efficiency or loss on the path is set as the four-dimensional weight tensor The energy flow weight in ; The equivalent value representing the environmental load generated by the path is set as the four-dimensional weight tensor Environmental flow weights in; The numerical value representing the operational benefit associated with the path is set as the four-dimensional weight tensor The economic flow weight in .

[0009] Preferably, in step 2, the topology generation network includes a graph variational autoencoder and a generative decoder; The graph variational autoencoder is used to transform the heterogeneous process graph The multi-dimensional real-time working condition data is encoded into a topological distribution latent space middle; The generative decoder is used to distribute the latent space from the topology and generate a set of candidate process topology adjacency matrices based on the multi-dimensional real-time working condition data. .

[0010] Preferably, the step 2 generates the set of candidate process topology adjacency matrix After that, it also includes: For each candidate process topology adjacency matrix Perform structural and physical constraint checks and discard candidate process topology adjacency matrices that fail the checks; Construct a value proxy model and use the value proxy model to analyze the topological adjacency matrix of the candidate processes that have passed the verification. Calculate its heuristic value score HVS; Set a heuristic value score threshold , and select the heuristic value score HVS higher than the heuristic value score threshold The candidate process topology adjacency matrix , used for processing in subsequent steps.

[0011] Preferably, in step 3, the 4F-LCA function is set to be composed of the weighted sum of four sub-functions: material flow, energy flow, environmental flow and economic flow; The weight factors corresponding to the four sub-functions are introduced to balance economic benefits, environmental impact, energy self-sufficiency rate and material conversion rate; The weight factor is based on the market signal vector Make real-time adjustments to achieve the total desired value of the topology Dynamic calculation.

[0012] Preferably, in step 3, each candidate process topology adjacency matrix The instantiation steps include: In the multi-agent reinforcement learning system, the candidate process topology adjacency matrix Each functional node in is instantiated as an independent agent; The goal of the independent agents is to find their respective optimal operating parameters through collaborative learning; The optimal operating parameters are used to maximize the adjacency matrix of the candidate process topology The total expected value of the topology calculated based on the 4F-LCA function .

[0013] Preferably, in step 4, the path robustness score is calculated The cost of the reconstruction conversion The steps include: Construct a Monte Carlo simulation method to calculate the path robustness score by forward propagating the input uncertainty in the multi-dimensional real-time working condition data ; Set up a comprehensive loss assessment model, by evaluating the adjacency matrix of switching from the current operating topology to the candidate process topology The energy, material and time losses generated during the reconstruction are calculated to obtain the reconstruction conversion cost. .

[0014] Preferably, in step 4, the final decision score is calculated The steps include: Set a small regularization constant ; Restructuring switching costs Cost normalization factor After dimensionless processing, the regularization constant Add them together to get a denominator term; The total expected value of the topology and the path robustness score Multiply them together to get a numerator term; Divide the numerator by the denominator to calculate the final decision score .

[0015] Preferably, in step 4, the adaptive execution strategy is a hierarchical dynamic reconstruction strategy, and the execution of the strategy includes: Set a low-cost reconstruction threshold ; The optimal process topology Corresponding reconstruction conversion cost With the low-cost reconstruction threshold Make comparisons; When the reconstruction conversion cost Below the low-cost reconstruction threshold When , a first-level reconstruction is triggered to perform smooth switching; When the reconstruction conversion cost Above the low-cost reconstruction threshold , triggers the secondary reconstruction to start the planned reconstruction and generate an optimal transition operation sequence.

[0016] Compared with the prior art, the present invention has at least the following beneficial effects: 1. This approach transcends the limitations of traditional fixed process flows, which are limited to parameter optimization, by enabling dynamic reconstruction of the entire process topology. By autonomously generating new process pathways, the system can fundamentally change the combination and connection of processing units to adapt to real-time changes in feedstock characteristics, market signals, and internal operating conditions, significantly improving the environmental adaptability and overall optimization potential of the entire system.

[0017] 2. This approach establishes a scientific and comprehensive intelligent decision-making framework, integrating quantitative modeling of material, energy, environmental, and economic flows. The final decision is based not only on the overall expected value of the pathway but also weighs the stability risks in actual operation against the conversion costs of switching to the new process, ensuring that the selected process pathway strikes a balance between theoretical optimality, practical feasibility, and economic rationality.

[0018] 3. This method achieves efficient target focus within a vast array of possibilities through a generative network and phased screening mechanism. The system first creatively generates multiple candidate process topologies. Then, using physical constraint verification and a value proxy model, it rapidly eliminates a large number of infeasible or low-potential options. This significantly reduces the computational load of subsequent refined evaluations and improves decision-making efficiency and speed.

[0019] 4. This approach establishes a safe and efficient bridge from optimal topology calculation to actual physical execution. Through a hierarchical dynamic reconfiguration strategy, it differentiates process switching costs. For low-cost changes, smooth transitions are implemented, while for high-cost changes, planned reconfiguration is initiated and an optimal transition plan is generated to minimize losses. This ensures that the optimal solution at the computational level can be safely and economically implemented in production, forming a complete closed-loop adaptive control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will be further explained below in conjunction with the accompanying drawings and Examples: Figure 1 This is a flowchart of a four-stream coupled full life cycle evaluation method for agricultural and livestock waste resource treatment based on graph neural networks in the present invention. DETAILED DESCRIPTION

[0021] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0022] Example 1 A four-stream coupled life cycle assessment method for agricultural and livestock waste resource treatment based on graph neural networks includes the following steps: Step 1: Obtain multi-dimensional real-time operating data that characterizes the system status. The multi-dimensional real-time operating data includes raw material characteristic vectors, market signal vectors, and internal operating condition vectors, and construct a heterogeneous process map that includes preset processing units as nodes and potential process paths as edges. ; Set the quantitative values ​​of material flow, energy flow, environmental flow and economic flow calculated based on multi-dimensional real-time working condition data as heterogeneous process maps The 4D weight tensor of each edge in ; Step 2: Build a topology generation network to map heterogeneous processes Input multi-dimensional real-time working condition data into the topology generation network to generate a set of candidate process topology adjacency matrices ; Step 3: Build a multi-agent reinforcement learning system and map each candidate process topology adjacency matrix Instantiate to collaboratively learn the optimal operating parameter combination, and use the optimal operating parameter combination and the preset 4F-LCA function to calculate the adjacency matrix of each candidate process topology The corresponding topological total expected value ; Step 4: Calculate the topological adjacency matrix of each candidate process Path robustness score and restructuring switching costs ; Combined topological total expected value , path robustness score and restructuring switching costs , calculate the final decision score , and select the final decision score The highest candidate process topology adjacency matrix As the optimal process topology , to output the optimal process topology Corresponding adaptive execution strategy.

[0023] The present invention provides a four-stream coupled full life cycle evaluation method for agricultural waste resource treatment based on graph neural network; multi-dimensional real-time working condition data is acquired to characterize the system state, wherein the raw material characteristic vector Characterize the physical and chemical properties of input materials, market signal vector Reflects external economic environment information, internal working condition vector Describe the system's internal operating state parameters; heterogeneous process maps is constructed, which includes nodes representing processing units and edges representing potential process paths; a four-dimensional weight tensor Set as a heterogeneous process map The attributes of each edge in the process unify the transmission characteristics of material flow, energy flow, environmental flow and economic flow. Material flow quantifies material conversion efficiency, energy flow quantifies energy consumption and energy recovery, environmental flow quantifies environmental load impact, and economic flow quantifies cost-effectiveness indicators. A topology generation network is constructed and used to input heterogeneous process maps. and multi-dimensional real-time process data to generate a set of candidate process topology adjacency matrices ; A multi-agent reinforcement learning system is constructed, and each candidate process topology adjacency matrix is instantiated for collaborative learning of optimal operating parameters; using the optimal operating parameters and a preset 4F-LCA function, each candidate process topology adjacency matrix The corresponding topological total expected value Calculated; path robustness score and restructuring switching costs For each candidate process topology adjacency matrix Calculated; final decision score By combining the topological total expected value , path robustness score and restructuring switching costs Calculate the final decision score The highest candidate process topology adjacency matrix Selected as the optimal process topology , and outputs an adaptive execution strategy; This method achieves the transition from parameter optimization to topology reconstruction by constructing a complete decision space driven by real-time multidimensional data, enabling the system to autonomously create and evaluate new process flows based on the dynamic changes of the external environment and internal state, ultimately forming a closed-loop adaptive control system of evaluation, generation, decision-making, and execution.

[0024] Example 2 In step 1, the step of obtaining the multi-dimensional real-time operating condition data includes: The raw material characteristic vector is obtained by online sensor measurement, and the raw material characteristic vector includes carbon-nitrogen ratio, moisture content and volatile solid content; Obtaining a market signal vector through an external interface, the market signal vector includes methanol market price, carbon credit price, and grid electricity price; The internal working condition vector is acquired through sensor collection of each processing unit, and the internal working condition vector includes temperature, pressure and flow.

[0025] In step 1, set the four-dimensional weight tensor The steps include: The transmission coefficient that characterizes the conversion efficiency of matter between two nodes is set as a four-dimensional weight tensor Material flow weight in ; The value representing the energy transfer efficiency or loss on the path is set as a four-dimensional weight tensor The energy flow weight in ; The equivalent value of the environmental load generated by the characterization path is set as a four-dimensional weight tensor Environmental flow weights in; The numerical value representing the operational benefit of the path association is set as a four-dimensional weight tensor The economic flow weight in .

[0026] This embodiment is a further explanation of embodiment 1; the acquisition of multi-dimensional real-time working condition data is refined; the raw material characteristic vector Obtained through online sensor measurement, including carbon-nitrogen ratio , moisture content and lignin content ; Market signal vector Obtained through external interface, including methanol market price , carbon credit prices and grid electricity prices ; Internal working condition vector Collected by sensors of each processing unit, including temperature ,pressure and traffic ; Four-dimensional weight tensor The setting steps of material flow weight are also clarified; It is set as the transmission coefficient that characterizes the efficiency of material conversion between two nodes; energy flow weight It is set as a value that characterizes the energy transfer efficiency or loss on the path; environmental flow weight is set as the equivalent value of the environmental load generated by the pathway; the economic flow weight is set as a numerical value that characterizes the operational benefits associated with the path; For example, for an anaerobic fermentation unit, its material flow weight can be quantified by the following empirical formula: ; in: is the dimensionless process intrinsic coefficient calibrated by historical data fitting or experiment, is the moisture content of the raw materials, is the carbon-nitrogen ratio, is the fermentation temperature; the exponential term in the formula is used to describe the effect of temperature on fermentation efficiency, where the constant 35 is the optimal fermentation temperature, and the constant 10 is the temperature sensitivity coefficient, whose dimension is consistent with the temperature to ensure that the exponent is a dimensionless value; The value representing the energy transfer efficiency or loss on the path is set as the energy flow weight in the four-dimensional weight tensor. For example, the energy flow weight of a path can be expressed as the net energy consumption per unit material processing, and its calculation formula is: ; in: and are the input power and recovery power of the process, is the material flow rate. The energy value can be further combined with the grid electricity price for economic quantification; The equivalent value representing the environmental load generated by the pathway is set as the environmental flow weight in the four-dimensional weight tensor. For example, the environmental flow weight of a pathway can focus on greenhouse gas emissions and be estimated using the following formula: ; in: 、 is the gas emission, GWP is the global warming potential, The price of carbon credits is introduced as a unit conversion factor to convert the emission units from kilograms to tons to match the units of carbon credit prices and ensure dimensional consistency; The numerical value representing the operational benefits associated with the path is set as the economic flow weight in the four-dimensional weight tensor; for example, for a process path that produces methanol, its economic flow weight can be expressed as: ; in: is the methanol mass yield per unit mass of raw material, dimensionless; is the material flow rate, is the market price of methanol, It is the comprehensive operating cost per unit time including energy consumption, manpower, etc. By accurately and multi-dimensionally defining the input data and mapping it to the four-dimensional weight tensor of the graph, a quantitative and comprehensive decision-making basis is provided for subsequent models, ensuring that the evaluation of the process path can dynamically reflect the real-time changes in raw materials, market and equipment conditions.

[0027] Example 3 In step 2, the topology generation network includes a graph variational autoencoder and a generative decoder; Graph variational autoencoder is used to transform heterogeneous process graphs and multi-dimensional real-time working condition data are encoded into a topological distribution latent space middle; Generative decoder is used to distribute latent space from topology The process topology adjacency matrix of the candidate process is generated based on the multi-dimensional real-time working condition data. ; Step 2 is to generate a set of candidate process topology adjacency matrices After that, it also includes: For each candidate process topology adjacency matrix Perform structural and physical constraint checks and discard candidate process topology adjacency matrices that fail the checks; Construct a value proxy model and use it to analyze the topological adjacency matrix of the candidate processes that have passed the verification. Calculate its heuristic value score HVS; Set a heuristic value score threshold , and select the heuristic value score HVS higher than the heuristic value score threshold The candidate process topology adjacency matrix , used for processing in subsequent steps.

[0028] This embodiment is a detailed description of the topology generation and screening process in Example 1. The topology generation network is designed as a structure including a graph variational autoencoder and a generative decoder. The graph variational autoencoder converts the heterogeneous process graph into and multi-dimensional real-time working condition data are encoded into a low-dimensional, continuous topological distribution latent space The graph variational autoencoder includes an encoder network and a decoder network. The encoder network uses a three-layer graph convolutional neural network, each layer contains 128 hidden units, and the activation function uses the ReLU function. The final output is the mean of the latent space vector and variance The latent space dimension is set to 1 / 4 to 1 / 2 of the number of nodes in the original graph. The specific value depends on the scale of the graph. It is 16-dimensional when there are less than 50 nodes, 32-dimensional when there are 50-200 nodes, and 64-dimensional when there are more than 200 nodes. Variational inference uses the reparameterization technique, namely: ; in is a standard normal distribution sampling value, Represents element-wise multiplication; The generative decoder then extracts the latent space A set of candidate process topology adjacency matrices is generated based on the specific working condition data. ; The generation process is represented by the following function: ; in, is the candidate process topology adjacency matrix; is the latent space vector; is the raw material characteristic vector; is the market signal vector; is the internal working condition vector; is a generative decoder function that transforms the latent space vector With the conditional vector Mapped into an adjacency matrix through multi-layer graph convolution operations , the specific calculation process is: Fusion of latent space vectors with conditional vectors to generate feature embedding matrices for all nodes : ; in, is the total number of nodes in the graph, is the node embedding dimension, MLP is the multi-layer perceptron, Represents vector concatenation.

[0029] By performing the inner product calculation on the node embedding matrix and applying the Sigmoid activation function, the elements in the adjacency matrix (i.e. the probability of the existence of an edge) are decoded and generated: ; in: The Sigmoid activation function ensures that the value of each element in the adjacency matrix is ​​between (0, 1), representing the probability of edge generation; finally, the binary adjacency matrix is ​​determined by threshold judgment or Bernoulli sampling; The intermediate node embedding matrix generated for the decoder; The training of graph variational autoencoder uses the negative variation lower bound as the loss function, and its goal is to minimize the loss. The specific expression is: ; The first term is the reconstruction loss, which measures the similarity between the generated image and the original image; the second term is the KL divergence regularization term, which constrains the latent space distribution to be close to the prior distribution. ; The reconstruction loss uses binary cross entropy to calculate the element-by-element error of the adjacency matrix, and the analytical expression of KL divergence is: ; in, is the latent space dimension; and Represents the mean vector of the encoder output and variance vector The training adopts Adam optimizer, the learning rate is set to 0.001, the batch size is 32, and the number of training rounds is 1000; In generating the candidate process topology adjacency matrix Afterwards, a two-stage screening process was performed; first, for each Structural and physical constraint checks are performed to discard infeasible solutions that have isolated nodes or violate physical and chemical laws. Subsequently, the topologies that pass the check are evaluated by a value proxy model to calculate their heuristic value scores (HVS), which are calculated as follows: ; in, is the heuristic value score of the i-th candidate topology, It is a fast estimate of the potential four-stream value of the topology by the value proxy model. The model adopts a deep neural network structure and the input is the topological adjacency matrix And the corresponding node feature matrix, the output is a scalar value estimate, the calculation formula is: ; Among them: MLP represents multi-layer perceptron, represents the graph pooling operation, is the node feature matrix, whose dimension is (N is the number of nodes, is the node feature dimension), each row represents the feature of a processing unit, which is usually composed of the components of the internal working condition vector of the unit; Only one heuristic value score HVS is above the preset threshold The candidate process topology adjacency matrix is sent to the next evaluation stage; the heuristic value score threshold The setting method is: First, calculate the mean of the heuristic value scores of all successful topologies in the historical run data and standard deviation , then set , where k is the screening strictness coefficient, which is used when more candidate solutions are needed. , when a few solutions need to be selected , when the system has extremely high quality requirements If there is a lack of historical data, The initial setting is 0.6, and it is dynamically adjusted according to the subsequent running results. The adjustment formula is: ; in: The heuristic value score of the best performing topology in the current period; This design enables the system to independently create new process flows, and through an efficient phased screening mechanism, it can quickly focus on a few high-potential candidate solutions from a huge space of possibilities, laying a solid foundation for subsequent accurate decision-making.

[0030] Example 4 In step 3, the 4F-LCA function is set to consist of the weighted sum of four sub-functions: material flow, energy flow, environmental flow, and economic flow; The weight factors corresponding to the four sub-functions are introduced to balance economic benefits, environmental impact, energy self-sufficiency rate and material conversion rate; The weighting factors are based on the market signal vector Make real-time adjustments to achieve the total desired value of the topology Dynamic calculation of In step 3, each candidate process topology adjacency matrix The instantiation steps include: In a multi-agent reinforcement learning system, the candidate process topology adjacency matrix Each functional node in is instantiated as an independent agent; The goal of the independent agents is set to find their respective optimal operating parameters through collaborative learning; The optimal operating parameters are used to maximize the adjacency matrix of the candidate process topology The total expected value of topology calculated based on the 4F-LCA function .

[0031] This example is a detailed explanation of the path value assessment process in Example 1. It also clarifies the relationship between the value proxy model in Example 3 and the 4F-LCA function in this example: the value proxy model is used for rapid pre-screening of candidate topologies. Its output, the heuristic value score (HVS), is an approximate estimate of the actual 4F-LCA function calculation result. By learning the mapping relationship between topological structures and final values ​​in historical data through a neural network, it provides efficient candidate solution screening before the full multi-agent optimization. The 4F-LCA function, on the other hand, is a precise value assessment calculated through a full multi-agent reinforcement learning process based on the candidate topologies screened by the value proxy model. The 4F-LCA function is set as the weighted sum of four sub-functions to calculate the total expected value of each topology ; Its specific function expression is: ; in: is the total expected value of the topology, are the branches, where each sub-function uses the min-max standardization method to unify the dimensions, and the standardization formula is , so that all normalized sub-functions have dimensionless values ​​in the range [0,1]; is the weight factor. The subscripts mat, eng, env, and eco represent material, energy, environmental, and economic flows, respectively. The corresponding weight factors are 、 、 、 , each weight factor is a dimensionless parameter and satisfies ; Weight factors corresponding to the four sub-functions Introduced to balance economic benefits, environmental impact, energy self-sufficiency and material conversion rate; these weight factors are based on the market signal vector Make real-time adjustments to achieve the total expected value of the topology Dynamic calculation of For each candidate process that passes the screening, the topological adjacency matrix , the multi-agent reinforcement learning system will have each of its functional nodes Instantiated as an independent agent; each agent adopts a deep Q-network architecture, the state space includes the operating parameters of its own node and the state information of adjacent nodes, and the action space is the discretized value of the adjustable operating parameters of the node; the communication mechanism based on neighborhood state sharing is adopted between agents, and each agent broadcasts an information vector containing its own state, reward and action to its adjacent nodes at each time step, where is the state of agent i, is the reward value, For an action, neighboring agent j takes it as part of its observation state. The convergence criterion for collaborative learning is that the average Q-value of all agents in 50 consecutive training episodes changes by less than 0.001, or the maximum number of training rounds is 5000. When the actions of multiple agents conflict, a priority scheduling mechanism is adopted. The priority is determined by the criticality of the node in the process flow, which is calculated as the weighted average of the node degree centrality and betweenness centrality. The goal of these independent agents is to find their own optimal operating parameters, such as temperature, pressure, etc., through collaborative learning; this optimal operating parameter combination is ultimately used to maximize the adjacency matrix of the candidate process topology. The total expected value of topology calculated based on the 4F-LCA function ; This evaluation method ensures that each candidate process is not only structurally sound, but also has its operating parameters optimized to the best state under current conditions, providing a refined and quantified value basis for the final decision.

[0032] Example 5 In step 4, the path robustness score is calculated and restructuring switching costs The steps include: Construct a Monte Carlo simulation method to calculate the path robustness score by forward propagating the input uncertainty in multi-dimensional real-time working condition data ; Set up a comprehensive loss assessment model by evaluating the adjacency matrix of the switch from the current operating topology to the candidate process topology The energy, material and time losses generated during the reconstruction are calculated to obtain the reconstruction conversion cost. .

[0033] In step 4, the final decision score is calculated The steps include: Set a small regularization constant ; Restructuring switching costs Cost normalization factor After dimensionless processing, and the regularization small constant Add them together to get a denominator term; The total expected value of the topology and path robustness score Multiply them together to get a numerator term; Divide the numerator by the denominator to calculate the final decision score .

[0034] This embodiment inherits the total expected value of topology in Example 4. The calculation of the path robustness score required for optimal topology selection is further supplemented by and restructuring switching costs The calculation method and the final decision score are given The specific calculation formula of , thus completing the complete decision chain from candidate topology generation, value evaluation to optimal selection; A Monte Carlo simulation method is constructed to calculate the path robustness score by forward propagating the input uncertainty in multi-dimensional real-time operation data. , the specific calculation method is: Based on historical data analysis, the uncertainty distribution of input parameters is determined, and the raw material characteristic vector The components of are assumed to follow a truncated normal distribution , the market signal vector The components of are assumed to follow a lognormal distribution , internal working condition vector The components of are assumed to be uniformly distributed , where the distribution parameters are estimated based on the operating data of the past 6 months; N Monte Carlo samplings are performed (N ≥ 1000), and each sampling uses the Latin Hypercube Sampling (LHS) method to improve sampling efficiency and representativeness; the system output variance is calculated for each sampling result , the output indicators include material conversion rate, energy efficiency, environmental impact value and economic benefit, and the variance calculation formula is: ; in: is the dimensionless comprehensive output value of the i-th sampling, is the output mean; but ,in It is a dimensionless sensitivity coefficient. Its specific value is determined by the system's stability requirements. When the system requires high stability, , when medium stability is required , when lower stability requirements are required ; The convergence judgment standard is to increase the sampling rate by 100 times continuously. The change in is less than 1% of its current value.

[0035] A comprehensive loss assessment model is set up by evaluating the adjacency matrix of switching from the current operating topology to the candidate process topology The energy, material and time losses generated during the reconstruction are calculated to obtain the reconstruction conversion cost. ; Then, the final decision score The calculation steps are defined; a small regularization constant Set; restructure switching costs With a small regularization constant Add together to get the denominator; the total expected value of the topology and path robustness score Multiply them together to get the numerator; finally, divide the denominator by the numerator to calculate the final decision score. ; Its calculation formula is: ; in: is the final decision score; is the total expected value; It is the cost of restructuring switching; is the cost normalization factor, expressed in currency units (yuan), and its value is equal to the average annual operating cost of the system, so that is a dimensionless value, To prevent the denominator from being zero, the regularization constant is in the range of to , is a dimensionless value; is the path robustness score, a dimensionless value ranging from 0 to 1, which represents the system's resistance to disturbances; The final decision score The calculation not only takes into account the theoretical optimal value of the topology, but also weighs its stability risk and switching cost in actual operation, ensuring that the final selected process topology is optimal in theory, practice and economy.

[0036] Example 6 In step 4, the adaptive execution strategy is a hierarchical dynamic reconstruction strategy. The execution of this strategy includes: Set a low-cost reconstruction threshold ; Optimal process topology Corresponding reconstruction conversion cost With low-cost reconstruction threshold Make comparisons; When refactoring switching costs Below the low-cost reconstruction threshold When , a first-level reconstruction is triggered to perform smooth switching; When refactoring switching costs Above the low-cost reconstruction threshold , triggers the secondary reconstruction to start the planned reconstruction and generate an optimal transition operation sequence.

[0037] This embodiment is a specific description of the adaptive execution strategy in embodiment 1; the strategy is designed as a hierarchical dynamic reconstruction strategy; a low-cost reconstruction threshold is set; optimal process topology Corresponding reconstruction conversion cost With this threshold To compare; The superscript * indicates the optimal topology selected after optimization. The subscript C1 represents the first-level cost threshold, which is set at 5%-15% of the system's average daily operating cost; when the reconstruction conversion cost Below the low-cost reconstruction threshold When the cost of the reconstruction is high, the first level reconstruction is triggered to perform a smooth switch, which usually involves adjusting valves online, activating or dormant a few processing units, and not interrupting the main process. Above the low-cost reconstruction threshold When the secondary reconstruction is triggered, it starts the planned reconstruction; this process may require a short downtime maintenance to achieve a more fundamental process change. The system will generate an optimal transition operation sequence to minimize the downtime loss. The operation sequence is calculated by the dynamic programming algorithm, whose goal is to find a set of optimal operation sequences. , the objective function is: ; in: 、 、 They represent the production loss, energy loss, and material loss at time t, respectively, and T is the total reconstruction time. To ensure dimensional consistency, all loss items are converted into equivalent monetary units during calculation. The state of dynamic programming is defined as: ; in: It represents the equipment configuration state, material inventory state and energy reserve state at time t; the state transfer equation is ,in is the operation action at time t, is an external disturbance; the boundary conditions are set to the initial state is the current running state, the target state Optimal process topology The corresponding steady-state operating state; The recursive relation of dynamic programming is: ; in: is the minimum total loss starting from state S at time t, is the instantaneous loss function; the time complexity of the algorithm is ,in and are the sizes of state space and action space respectively, and the space complexity is ; The construction of this hierarchical strategy safely and efficiently transforms the "optimal topology" at the computing level into the actual production process in the physical world, and uses continuous feedback for self-optimization, ultimately achieving dynamic process reconstruction at the entire system level, reflecting a comprehensive consideration of cost-effectiveness and operational feasibility.

[0038] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A four-stream coupled full life cycle assessment method for agricultural waste resource treatment based on graph neural network, characterized by: include: Step 1: Acquire multi-dimensional real-time operating data representing the system status, including raw material characteristic vectors, market signal vectors, and internal operating condition vectors, and construct a heterogeneous process map with preset processing units as nodes and potential process paths as edges. ; The quantitative values ​​of the material flow, energy flow, environmental flow and economic flow of the entire life cycle calculated based on the multi-dimensional real-time working condition data are set as the heterogeneous process map The 4D weight tensor of each edge in ; Step 2: Construct a topology generation network to map the heterogeneous process The multi-dimensional real-time working condition data is input into the topology generation network to generate a set of candidate process topology adjacency matrices ; Step 3: Build a multi-agent reinforcement learning system and map each candidate process topology adjacency matrix Instantiate to collaboratively learn the optimal operating parameter combination, and use the optimal operating parameter combination and the preset 4F-LCA function to calculate the adjacency matrix of each candidate process topology The corresponding topological total expected value ; Step 4: Calculate the topological adjacency matrix of each candidate process Path robustness score and restructuring switching costs ; Combined with the total expected value of the topology , path robustness score and restructuring switching costs , calculate the final decision score , and select the final decision score The highest candidate process topology adjacency matrix As the optimal process topology , to output the optimal process topology Corresponding adaptive execution strategy.

2. The method according to claim 1, characterized in that In the step 1, the step of obtaining the multi-dimensional real-time operating condition data includes: Obtaining the raw material characteristic vector by online sensor measurement, wherein the raw material characteristic vector includes carbon-nitrogen ratio, moisture content and lignin content; obtaining the market signal vector through an external interface, wherein the market signal vector includes a methanol market price, a carbon credit price, and a grid electricity price; The internal operating condition vector is acquired through sensor collection of each processing unit, and the internal operating condition vector includes temperature, pressure and flow.

3. The method according to claim 1, characterized in that In the step 1, the four-dimensional weight tensor is set The steps include: The transmission coefficient that characterizes the conversion efficiency of matter between two nodes is set as the four-dimensional weight tensor Material flow weight in ; The value representing the energy transfer efficiency or loss on the path is set as the four-dimensional weight tensor The energy flow weight in ; The equivalent value representing the environmental load generated by the path is set as the four-dimensional weight tensor Environmental flow weights in; The numerical value representing the operational benefit associated with the path is set as the four-dimensional weight tensor The economic flow weight in .

4. The method according to claim 1, wherein In the step 2, the topology generation network includes a graph variational autoencoder and a generative decoder; The graph variational autoencoder is used to transform the heterogeneous process graph The multi-dimensional real-time working condition data is encoded into a topological distribution latent space middle; The generative decoder is used to distribute the latent space from the topology and generate a set of candidate process topology adjacency matrices based on the multi-dimensional real-time working condition data. .

5. The method according to claim 4, characterized in that The second step is to generate the set of candidate process topology adjacency matrix After that, it also includes: For each candidate process topology adjacency matrix Perform structural and physical constraint checks and discard candidate process topology adjacency matrices that fail the checks; Construct a value proxy model and use the value proxy model to analyze the topological adjacency matrix of the candidate processes that have passed the verification. Calculate its heuristic value score HVS; Set a heuristic value score threshold , and select the heuristic value score HVS higher than the heuristic value score threshold The candidate process topology adjacency matrix , used for processing in subsequent steps.

6. The method according to claim 1, characterized in that In step 3, the 4F-LCA function is set to be composed of the weighted sum of four sub-functions: material flow, energy flow, environmental flow, and economic flow; The weight factors corresponding to the four sub-functions are introduced to balance economic benefits, environmental impact, energy self-sufficiency rate and material conversion rate; The weight factor is based on the market signal vector Make real-time adjustments to achieve the total desired value of the topology Dynamic calculation.

7. The method according to claim 6, characterized in that In step 3, each candidate process topology adjacency matrix The instantiation steps include: In the multi-agent reinforcement learning system, the candidate process topology adjacency matrix Each functional node in is instantiated as an independent agent; The goal of the independent agents is to find their respective optimal operating parameters through collaborative learning; The optimal operating parameters are used to maximize the adjacency matrix of the candidate process topology The total expected value of the topology calculated based on the 4F-LCA function .

8. The method according to claim 1, characterized in that In step 4, the path robustness score is calculated The cost of the reconstruction conversion The steps include: Construct a Monte Carlo simulation method to calculate the path robustness score by forward propagating the input uncertainty in the multi-dimensional real-time working condition data ; Set up a comprehensive loss assessment model, by evaluating the adjacency matrix of switching from the current operating topology to the candidate process topology The energy, material and time losses generated during the reconstruction are calculated to obtain the reconstruction conversion cost. .

9. The method according to claim 8, characterized in that In step 4, the final decision score is calculated The steps include: Set a small regularization constant ; Restructuring switching costs Cost normalization factor After dimensionless processing, the regularization constant Add them together to get a denominator term; The total expected value of the topology and the path robustness score Multiply them together to get a numerator term; Divide the numerator by the denominator to calculate the final decision score .

10. The method according to claim 1, characterized in that In step 4, the adaptive execution strategy is a hierarchical dynamic reconstruction strategy, and the execution of the strategy includes: Set a low-cost reconstruction threshold ; The optimal process topology Corresponding reconstruction conversion cost With the low-cost reconstruction threshold Make comparisons; When the reconstruction conversion cost Below the low-cost reconstruction threshold When , a first-level reconstruction is triggered to perform smooth switching; When the reconstruction conversion cost Above the low-cost reconstruction threshold , triggers the secondary reconstruction to start the planned reconstruction and generate an optimal transition operation sequence.

Images