A cross-domain collaborative energy cycle optimization method and system

Abstract

This invention discloses an energy cycle optimization method and system based on cross-domain collaboration, relating to the interdisciplinary fields of environmental protection and biomass energy. It includes: a collaborative data interaction unit for acquiring synchronous cross-domain data from water treatment and biomass natural gas sources; an AI cross-domain optimization unit for matching sludge supply and demand with wastewater reuse using a resource matching prediction model based on a standardized cross-domain dataset, optimizing global parameters of the matching results using a cross-domain process optimization model, and generating corresponding execution parameters based on the optimized global parameters using a microbial sludge adaptation model and a water quality / wastewater adaptation model, respectively, in conjunction with actual energy cycle scenarios; and an intelligent early warning unit for identifying cascading failure risks in cross-domain data based on a cross-domain fault correlation model. This invention can significantly improve cross-domain operational efficiency and resource utilization, achieving a closed-loop cycle optimization of the entire chain of cross-domain data, including sludge, energy, and water resources.

Description

Technical Field

[0001] This invention relates to the field of environmental protection and biomass energy intersection technology, and in particular to an energy cycle optimization method and system based on cross-domain collaboration. Background Technology

[0002] Based on the principle that residual sludge from water treatment can be used as raw material for anaerobic fermentation of biomass, and that high-concentration organic wastewater from biomass fermentation can be treated and reused as makeup water for water treatment systems, the harmless disposal of sludge in the water treatment industry and the raw material supply and wastewater disposal in the biomass natural gas industry can form a natural resource cycle complementary relationship.

[0003] However, in current technologies, the two industries have long operated independently and in a rudimentary manner, resulting in low accuracy in cross-domain resource matching and a lack of cross-domain collaborative optimization capabilities. Currently, there is no intelligent matching mechanism between the output, moisture content, and carbon-to-nitrogen ratio of wastewater treatment sludge and the demand for biomass fermentation raw materials. This easily leads to problems such as sludge oversupply causing accumulation and decay, secondary pollution, or insufficient supply leading to a shortage of fermentation raw materials and decreased gas production efficiency. Furthermore, the water quality of biomass fermentation wastewater does not match the water treatment makeup water standards, and direct reuse can easily impact the biochemical water treatment system. Pretreatment parameters are entirely implemented using existing single general-purpose algorithms, which cannot identify the cross-correlation between parameters from the two domains. This fails to address core issues such as cross-domain supply and demand matching, global multi-objective optimization, and cascading failure early warning, resulting in application effects far below the requirements for collaborative operation. In addition, during the collaborative optimization process, the water treatment process parameters and biomass fermentation parameters are isolated from each other and cannot achieve global optimization through coordinated adjustment. This often leads to contradictions such as excessive dehydration at the water treatment end, which increases energy consumption, while the biomass fermentation end needs to add water for dilution, and reduced chemical consumption at the water treatment end, which leads to a decrease in sludge organic matter and a significant decrease in fermentation gas production rate. Overall, energy consumption remains high. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide an energy cycle optimization method and system based on cross-domain collaboration, constructing an AI-driven, fully closed-loop cross-domain collaborative optimization system that can greatly improve cross-domain operational efficiency and resource utilization, and achieve full-chain closed-loop cycle optimization of various cross-domain data such as sludge, energy, and water resources.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0006] The first aspect of this invention provides an energy cycle optimization system based on cross-domain collaboration, comprising:

[0007] The collaborative data interaction unit is used to acquire synchronous cross-domain data from the water treatment end and the biomass natural gas end, and to preprocess the cross-domain data to obtain a standardized cross-domain dataset.

[0008] The AI ​​cross-domain optimization unit is used to match sludge supply and demand with wastewater reuse based on a standardized cross-domain dataset using a resource matching prediction model. The cross-domain process optimization model optimizes the global parameters of the matching results. The microbial sludge adaptation model and the water quality wastewater adaptation model are combined with the actual energy cycle scenario to generate corresponding execution parameters based on the optimized global parameters.

[0009] The intelligent early warning unit is used to identify the risk of cascading failures in cross-domain data based on a cross-domain fault association model, and to push early warning information and generate emergency dispatch plans according to the identification results at preset levels.

[0010] Furthermore, the collaborative data interaction unit includes:

[0011] The data acquisition module is used to collect synchronous cross-domain data between the water treatment end and the biomass natural gas end;

[0012] The data standardization module is used to preprocess synchronous cross-domain data. Specifically, the preprocessing includes outlier removal, missing value completion, and standardization operations.

[0013] The cross-domain data interface module is used to build a two-way data interaction interface based on the communication protocol.

[0014] Furthermore, in the AI ​​cross-domain optimization unit, the resource matching prediction model uses an LSTM algorithm with a cross-domain feature cross-attention mechanism to allocate weights between the water treatment end and the biomass natural gas end based on the causal relationship of the process; the cross-domain process optimization model uses a combination of reinforcement learning algorithm with dynamic weight multi-objective reward function and genetic algorithm to dynamically adjust the matching degree obtained by the resource matching prediction model; the microbial sludge adaptation model uses a fusion algorithm of CNN and LSTM with dual input branches to calculate the sludge pretreatment optimization parameters; and the water quality wastewater adaptation model uses a random forest algorithm with dynamic feature matching rules to calculate the fermentation wastewater pretreatment parameters.

[0015] Furthermore, the intelligent early warning unit includes:

[0016] The cross-domain fault association module is used to identify cascading risks using a cross-domain fault association model with a built-in cross-domain fault association graph.

[0017] The multi-level early warning module is used to generate a multi-level early warning mechanism based on preset fault thresholds and to generate early warning information based on fault identification results.

[0018] The emergency dispatch module is used to generate emergency dispatch plans based on early warning information and preset parameter rules.

[0019] Furthermore, the specific steps for identifying cascading risks using a cross-domain fault association model with a built-in cross-domain fault association graph are as follows:

[0020] The XGBoost algorithm was used to extract the core features of cross-domain faults;

[0021] By utilizing cross-domain fault propagation correlation graphs to analyze core features based on the propagation patterns of cross-domain faults, a fault graph structure model is constructed.

[0022] By using graph neural networks to learn graph features from fault graph structure models, we can identify the cross-domain propagation patterns, propagation speed, and chain reaction paths of faults.

[0023] Furthermore, it also includes a resource recycling execution unit, which is used to transport sludge and reuse wastewater according to execution parameters, and at the same time feeds the execution data back to the collaborative data interaction unit.

[0024] Furthermore, it also includes digital twin units, which are used to simulate and verify optimized parameters, and to perform remote parameter adjustments and emergency dispatch.

[0025] A second aspect of this invention provides an energy cycle optimization method based on cross-domain collaboration, comprising the following steps:

[0026] Acquire synchronous cross-domain data from the water treatment end and the biomass natural gas end, and preprocess the cross-domain data to obtain a standardized cross-domain dataset;

[0027] The resource matching prediction model is used to match sludge supply and demand with wastewater reuse based on a standardized cross-domain dataset. The cross-domain process optimization model optimizes the global parameters of the matching results. The microbial sludge adaptation model and the water quality wastewater adaptation model are used to generate corresponding execution parameters based on the optimized global parameters in combination with the actual energy cycle scenario.

[0028] Based on the cross-domain fault association model, the risk of cascading faults is identified in cross-domain data. According to the identification results, early warning information is pushed according to the preset level and an emergency dispatch plan is generated.

[0029] A third aspect of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing steps in the cross-domain collaborative energy cycle optimization method as described in the second aspect of the present invention.

[0030] A fourth aspect of the present invention provides a computer device comprising:

[0031] A processor, adapted to execute computer programs;

[0032] A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the energy cycle optimization method based on cross-domain collaboration as described in the second aspect of the present invention.

[0033] The above one or more technical solutions have the following beneficial effects:

[0034] This invention discloses an energy cycle optimization method and system based on cross-domain collaboration. The system includes a collaborative data interaction unit, an AI cross-domain optimization unit, and a resource cycle execution unit, all connected in sequence, as well as a digital twin unit and an intelligent early warning unit that interact bidirectionally with the aforementioned units, forming a fully closed-loop collaborative management and control system. Addressing the core shortcomings of existing technologies, such as limitations in single-domain optimization, lack of cross-domain algorithm adaptation, and weak module linkage, this invention structurally improves general AI algorithms for cross-domain scenarios, constructing a "prediction-optimization-adaptation-feedback" cross-domain collaborative algorithm system. It also clarifies the strongly bound data flow and causal relationships between modules. In some implementation examples, it can achieve precise matching of sludge and fermentation raw materials, and fermentation wastewater and water treatment makeup water, increasing the sludge fermentation conversion rate to over 85%, reducing overall cross-domain energy consumption by 20%-25%, and increasing wastewater reuse rate to over 85%. It can be widely applied to the collaborative operation of municipal or industrial water treatment sludge resource utilization and biomass natural gas projects.

[0035] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart of the energy cycle optimization method based on cross-domain collaboration in Embodiment 2 of the present invention. Detailed Implementation

[0038] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0039] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0040] 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 of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0041] Example 1:

[0042] Embodiment 1 of this invention provides an energy cycle optimization system based on cross-domain collaboration, including a collaborative data interaction unit, an AI cross-domain optimization unit, a resource cycle execution unit, a digital twin unit, and an intelligent early warning unit. The output of the collaborative data interaction unit is communicatively connected to the input of the AI ​​cross-domain optimization unit. The output of the AI ​​cross-domain optimization unit is communicatively connected to the inputs of the resource cycle execution unit, the digital twin unit, and the intelligent early warning unit, respectively. The feedback output of the resource cycle execution unit is bidirectionally connected to the input of the collaborative data interaction unit. Each unit achieves encrypted data communication with a 5G private network via industrial Ethernet, forming a fully closed-loop collaborative management and control system.

[0043] The cross-domain feature cross-attention mechanism in this embodiment employs a fusion operation of Hadamard product and feature concatenation, which quantifies process correlations while preserving single-domain information, addressing the pain point that general algorithms cannot identify cross-domain process couplings. A CNN-LSTM dual-branch architecture extracts sludge spatial features and microbial community time-series features respectively, achieving global optimization and refined adaptation. The XGBoost and GNN fusion architecture for fault models, combined with cross-domain fault maps, overcomes the limitations of single-algorithm recognition.

[0044] Specifically, it includes the following:

[0045] The collaborative data interaction unit, serving as the legitimate data source for the entire system, is a crucial prerequisite for the operation of all subsequent modules. It acquires synchronous cross-domain data from both the water treatment and biomass natural gas treatment ends, and preprocesses this cross-domain data to obtain a standardized cross-domain dataset. The collaborative data interaction unit comprises a data acquisition module, a data standardization module, and a cross-domain data interface module.

[0046] The collaborative data interaction unit includes:

[0047] The data acquisition module is used to collect synchronous cross-domain data from the water treatment end and the biomass natural gas end. Specifically, sensors for sludge production, moisture content, carbon-to-nitrogen ratio, influent COD, ammonia nitrogen, and dissolved oxygen are deployed at the water treatment end, while sensors for fermenter temperature, pH, C / N ratio, microbial activity, methane concentration, and fermentation wastewater quality are deployed at the biomass natural gas end. All sensors adopt a dual-redundancy design, with a uniform sampling frequency of 1 minute / time and timestamp accuracy at the millisecond level, ensuring that the data from the two systems are from the same source and synchronized, with a data accuracy of ≥99.5%.

[0048] The data standardization module is used to preprocess synchronous cross-domain data. Specifically, preprocessing includes outlier removal, missing value completion, and standardization. Specifically, a preprocessing workflow optimized for cross-domain parameters is adopted. Outliers are removed using a 3σ principle (±2σ threshold for sludge physicochemical parameters and ±1.5σ threshold for fermentation-sensitive parameters). Missing values ​​are completed using linear interpolation. Then, Min-Max standardization is used to convert data from different domains and scales into a standardized dataset in JSON format. This results in a standardized cross-domain dataset containing data such as sludge production, moisture content, carbon-to-nitrogen ratio, influent COD, and ammonia nitrogen from the water treatment end, as well as effective volume of the biomass-natural gas fermentation tank, microbial activity, daily raw material demand, and fermentation wastewater quality from the biomass-natural gas end.

[0049] The cross-domain data interface module is used to build a bidirectional data interaction interface based on a communication protocol. Specifically, based on the OPCUA industrial communication protocol, it builds an AES-256 encrypted bidirectional data interaction interface to support real-time data transmission between the water treatment end and the biomass natural gas end, with a transmission latency of ≤50ms, thus breaking down the data barriers between the two independent systems.

[0050] The AI ​​cross-domain optimization unit comprises a resource matching prediction model, a cross-domain process optimization model, a microbial sludge adaptation model, and a water quality / wastewater adaptation model, linked sequentially to form a closed-loop optimization logic. The AI ​​cross-domain optimization unit utilizes the resource matching prediction model to match sludge supply and demand with wastewater reuse based on a standardized cross-domain dataset. The cross-domain process optimization model performs global parameter optimization on the matching results. Finally, the microbial sludge adaptation model and the water quality / wastewater adaptation model, combined with the actual energy cycle scenario, generate corresponding execution parameters based on the optimized global parameters.

[0051] Specifically:

[0052] In the AI ​​cross-domain optimization unit, the resource matching prediction model utilizes an LSTM algorithm with a cross-domain feature cross-attention mechanism to allocate weights for the water treatment and biomass-natural gas ends based on the causal relationships of the processes. The input to the resource matching prediction model is a standardized cross-domain dataset from both the water treatment and biomass-natural gas ends. Abandoning the time-dimensional attention mechanism of general models, a cross-domain feature cross-attention weight allocation mechanism is designed. Through feature cross-operation, the correlation between parameters in the two domains is automatically identified, and the feature weights are dynamically adjusted to predict the matching degree between sludge supply and fermentation demand, and the adaptability between fermentation wastewater and water treatment makeup water for the next 24-72 hours. The prediction accuracy is ≥96%, and the model outputs preliminary optimization directions.

[0053] The cross-domain feature cross-attention weight allocation mechanism designed in this embodiment of the resource matching prediction model is a customized design that differs from general cross-domain attention mechanisms in other fields. Its core innovation lies in anchoring the causal relationship between the processes of water treatment and biomass natural gas, rather than simply the relationship between feature data. By constructing a cross-operation, correlation identification, and weight allocation system adapted to the characteristics of cross-domain processes, it solves the technical pain point that general algorithms cannot identify cross-domain process coupling relationships. The design, operation, identification, and calculation process of this mechanism all revolve around the core parameters of water treatment-biomass natural gas synergy. The specific steps are as follows:

[0054] In actual calculations, the dimensions and numerical ranges of sludge moisture content (percentage) and influent COD (mg / L) at the water treatment end differ significantly from those of the effective volume (m³) and microbial activity (percentage) of the biomass fermentation tank at the biomass end. Direct feature fusion would lead to the submergence of low-value features. Therefore, it is necessary to eliminate the influence of dimensions in attention allocation, which is the basis for achieving effective feature association. Furthermore, the parameters in these two domains are not simply numerically correlated, but rather exhibit clear causal process relationships. For example, the carbon-to-nitrogen ratio of sludge directly determines the metabolic activity of the fermentation microbial community, and the COD of the fermentation wastewater directly affects the biochemical load of the water treatment makeup system. This type of causal relationship is the core basis for attention weight allocation and a key difference from cross-domain attention mechanisms without process association in other domains. Secondly, the COD and ammonia nitrogen of the water treatment influent are affected by municipal or industrial wastewater and change dynamically in real time. The microbial activity at the biomass end is also in a state of continuous fluctuation due to the influence of fermentation temperature and pH. Therefore, the mechanism needs to have millisecond-level dynamic weight adjustment capabilities to match the parameter change rhythm of actual operation. In addition, different features have significantly different impacts on cross-domain resource matching results. For example, the influence weight of sludge moisture content and carbon-nitrogen ratio on fermentation raw material matching is much higher than that of the effective volume of the fermentation tank. The influence weight of fermentation wastewater COD and ammonia nitrogen on water treatment makeup adaptation is much higher than that of methane concentration. A mechanism is needed to automatically identify and strengthen the attention weight of highly coupled features.

[0055] The cross-domain feature cross-attention weight allocation mechanism in this embodiment uses a standardized cross-domain dataset after being standardized by the collaborative data interaction unit, including the water treatment end feature set. (Sludge production, moisture content, carbon-to-nitrogen ratio, influent COD, ammonia nitrogen, 5 dimensions) and biomass natural gas end-user characteristics set (Fermentation tank effective volume, microbial activity, daily raw material demand, fermentation wastewater COD, fermentation wastewater ammonia nitrogen, 5 dimensions) This type of data is the core feature data for cross-domain resource matching, which directly determines the matching effect of sludge fermentation raw materials and fermentation wastewater water treatment replenishment. It is the optimal feature selection after verification by a large number of process experiments and actual operation data.

[0056] This embodiment employs a fusion operation method combining Hadamard product and feature concatenation. This is a customized operation designed for cross-domain features of water treatment and biomass natural gas. Unlike the single Hadamard product or feature concatenation operation in other fields, its core advantage lies in realizing point-to-point process association learning of dual-domain features while retaining the original process information of single-domain features. This avoids information loss during the cross-domain feature fusion process and ensures that the model can simultaneously learn cross-domain coupling relationships and single-domain operating characteristics.

[0057] The specific process of cross operation is based on the 10-dimensional normalized feature vector after dimension alignment. (Water treatment end) and (Biomass end) Expand, the specific formula is:

[0058] .

[0059] in, The standardized feature vector for the water treatment end after dimension alignment is 1×10-dimensional and contains 5 core process features and 5 zero-value extended features. The dimensionally aligned, biomass-natural gas end-standardized feature vector is 1×10-dimensional and contains 5 core process features and 5 zero-value extended features. The core function of this operation is to quantify the point-to-point process correlation of dual-domain features, such as the element-wise multiplication of the sludge carbon-nitrogen ratio dimension and the microbial activity dimension, which directly reflects the process coupling relationship between the two. For feature concatenation, the 1×10 dimensional interaction vector obtained from the Hadamard product is combined with the original... , Horizontal concatenation is performed, resulting in a 1×30 dimensional cross-domain feature fusion vector. ; The final vector after cross-domain feature fusion is 1×30-dimensional. It contains both point-to-point interaction information of dual-domain features and retains the original single-domain feature information of water treatment end and biomass end, providing a complete feature foundation for subsequent correlation identification and weight allocation.

[0060] The cross-domain feature cross-attention weight allocation mechanism in this embodiment is a three-layer neural network architecture that includes a dual-domain feature interaction layer, an attention layer, and a weight normalization layer. This architecture is a customized architecture designed for cross-domain process features. The number of neurons and activation functions in each layer are trained and optimized based on actual process data to achieve automatic identification of cross-domain feature correlation and dynamic allocation of attention weights.

[0061] The dual-domain feature interaction layer is used to extract process interaction information across domain features. Specifically, it extracts the 30-dimensional feature vector after cross-fusion. The input is a dual-domain feature interaction layer, a fully connected layer with 64 neurons and the ReLU activation function. Its core function is to extract cross-domain feature interaction information from the fused features, rather than simply numerical interaction information. The calculation formula is as follows:

[0062] .

[0063] in, The weight matrix (64×30 dimensions) for the dual-domain feature interaction layer was obtained through training with nearly one year of actual operational data. For bias vectors, The extracted cross-domain process interaction feature vector (1×64 dimensions) uses the ReLU activation function to avoid gradient vanishing and improve the model's ability to learn highly coupled features.

[0064] Attention layering is used to automatically identify cross-domain correlations and calculate attention scores. Specifically, it involves dividing the process interaction feature vectors... The input attention layer is a fully connected layer with 30 neurons (as opposed to...). (The dimensions are consistent), the activation function is the Sigmoid function, the core function is to automatically identify cross-domain feature correlation and calculate the attention score for each feature dimension, the calculation formula is:

[0065] .

[0066] in, To score attention, The weight matrix is ​​obtained based on the sharing of the dual-domain feature interaction layer, and higher initial weights are set for feature dimensions with high process coupling. This is the bias vector.

[0067] The Sigmoid function maps the result to the [0,1] interval, and the resulting αi is the attention score (1×30 dimensions) for each feature dimension. The value of the score directly represents the degree of cross-domain process correlation of that feature dimension, realizing automatic identification of correlation: a score ≥0.7 is a high correlation feature dimension (such as sludge carbon-nitrogen ratio-microbial activity, fermentation wastewater COD-water treatment influent COD), a score of 0.3-0.7 is a medium correlation feature dimension (such as sludge production-daily raw material demand, fermenter effective volume-sludge moisture content), and a score <0.3 is a low correlation feature dimension (such as methane concentration-sludge production, dissolved oxygen-fermentation wastewater ammonia nitrogen).

[0068] The recognition process at this layer is a dynamic self-learning process; as the system accumulates more and more data, the model will continue to optimize. and This improves the accuracy of correlation identification and ensures that the attention score is highly matched with the actual process coupling relationship.

[0069] The weight normalization layer is used to dynamically allocate attention weights. Specifically, the attention score αi is input into the weight normalization layer, and the attention score is normalized using the Softmax function. Its core function is to convert the attention score into the final attention weights, ensuring that the sum of the weights for all feature dimensions is 1, thus achieving dynamic allocation of attention weights. The calculation formula is as follows:

[0070] .

[0071] in, The final attention weights after normalization (1×30 dimensions). This represents the attention score for the k-th feature dimension. After normalization, the weight of highly relevant feature dimensions accounts for ≥80%, ensuring that the model focuses on feature parameters that play a core role in cross-domain resource matching during the prediction process, thereby improving prediction accuracy.

[0072] The calculation of cross-domain feature correlation is based on normalized attention weights. Cross-domain feature fusion vector The core objective is to quantify the process correlation between key process parameters at the water treatment and biomass ends, providing a quantitative basis for resource matching prediction. The correlation calculation results are presented in the form of a correlation matrix, facilitating model reading and subsequent invocation of cross-domain process optimization models. The specific steps are as follows:

[0073] 1. Calculation of correlation degree.

[0074] .

[0075] in, This is a cross-domain feature correlation matrix (30×30 dimensions). Cross-domain feature fusion vector The transpose of the matrix (30×1 dimension). Each element value in the array represents the cross-domain process correlation degree between the two feature dimensions. The element value ranges from [0,1]. The larger the element value, the tighter the process coupling relationship between the two.

[0076] 2. Extraction of core feature correlation.

[0077] because Including zero-value extended features, it needs to be from 30×30 dimensions. The correlation sub-matrix (5×5 dimension) of the core process features is extracted. This sub-matrix contains only the correlation between the five core features of the water treatment end and the five core features of the biomass end, and serves as the core basis for model prediction. It is denoted as... , The element values ​​are directly used to calculate the matching degree between sludge and fermentation raw materials, and between fermentation wastewater and water treatment makeup water.

[0078] 3. Application of correlation degree calculation results.

[0079] correlation calculation results ( and It has two main applications: First, it serves as the core quantitative basis for resource matching prediction models, calculating the matching degree of cross-domain resources based on the degree of correlation, with a prediction accuracy of ≥96%; second, it serves as the basis for weight adjustment in downstream cross-domain process optimization models, where the cross-domain process optimization model will adjust the weights according to the relevant factors. The correlation value is used to dynamically adjust the weight of the multi-objective reward function to ensure that the optimization direction is consistent with the actual process coupling relationship.

[0080] In the AI ​​cross-domain optimization unit, the cross-domain process optimization model uses a combination of reinforcement learning algorithm with dynamic weighted multi-objective reward function and genetic algorithm to dynamically adjust the matching degree obtained from the resource matching prediction model. The optimization objectives are to achieve the highest sludge fermentation conversion rate, the highest wastewater reuse rate, and the lowest overall energy consumption. The optimization variables are the sludge dewatering parameters at the water treatment end (flocculator dosage, dewatering machine speed) and the fermentation parameters at the biomass-natural gas end (temperature, pH, C / N ratio, stirring speed). Abandoning the fixed weight logic of general models, the weights of the reward function are dynamically adjusted based on the matching degree output by the resource matching prediction model, achieving global optimization for both systems, rather than local optimization.

[0081] Specifically, the dynamic weighted multi-objective reward function is as follows:

[0082] .

[0083] in , For the total reward, For sludge fermentation conversion rate, For wastewater reuse rate, The comprehensive energy consumption per unit of gas production, weighted , , The matching degree is dynamically adjusted based on the output of the resource matching prediction model.

[0084] In the AI ​​cross-domain optimization unit, the microbial sludge adaptation model uses a dual-input branch CNN and LSTM fusion algorithm to calculate sludge pretreatment optimization parameters. One branch extracts the physicochemical characteristics of the wastewater treatment sludge through CNN, while the other branch extracts the time-series features of fermentation microbial activity through LSTM. The outputs of the two branches are concatenated through a fusion layer to establish a real-time correlation between sludge parameters and fermentation microbial activity, dynamically adjusting the sludge pretreatment parameters to adapt the sludge raw material to the growth requirements of the microbial community, controlling the sludge carbon-nitrogen ratio at 25-30:1, the moisture content at 55%-60%, and the microbial activity stability rate at ≥95%.

[0085] The dual-input branch CNN-LSTM fusion algorithm used in the microbial sludge adaptation model in this embodiment is not an independently operating model, but rather a core component of the prediction-optimization-adaptation closed-loop logic in the AI ​​cross-domain optimization unit. It forms a strict input-output linkage with the upstream resource matching prediction model and the cross-domain process optimization model. Simultaneously, its output directly connects to the downstream resource loop execution unit and transmits operational feedback data back to the collaborative data interaction unit, providing data support for the parameter iteration of all upstream models. The linkage between this model and other models or units within the system is achieved based on industrial Ethernet + 5G private network, with a transmission latency ≤50ms, and the data format is standardized JSON. All inputs and outputs are fully automated and require no manual intervention.

[0086] Specifically, the upstream cross-domain process optimization model outputs the globally optimal range of process parameters. The microbial sludge adaptation model, through a dual-input branch CNN-LSTM fusion algorithm, refines this range of parameters into directly executable fine-grained operation parameters, achieving precise adaptation between the sludge physicochemical characteristics and the activity of fermentation microorganisms. This solves the technical pain point of the disconnect between global optimization parameters and actual process execution.

[0087] More specifically, the CNN branch includes 2 convolutional layers, 2 pooling layers, and 1 fully connected layer. The number of filters in the convolutional layers are 32 and 64, respectively, and the kernel size is 3×3. The pooling layer uses max pooling with a kernel size of 2×2 and the activation function is ReLU. Its core function is to extract spatial feature information of sludge physicochemical characteristics (such as the numerical distribution characteristics of water content and carbon-nitrogen ratio, and the spatial coupling characteristics between various parameters).

[0088] The LSTM branch consists of two hidden layers and one fully connected layer. Each hidden layer has 64 neurons and a time step of 60 (corresponding to 60 minutes of sampled data). The activation function is Tanh, and the forget gate activation function is Sigmoid. Its core function is to extract the time-series features of fermentation community activity (such as the fluctuation features of community activity over time and the time coupling features with fermentation parameters).

[0089] The fusion layer adopts a fusion method of feature concatenation and fully connected layer. It concatenates the 128-dimensional spatial feature vector output by the CNN branch with the 128-dimensional time series feature vector output by the LSTM branch to obtain a 256-dimensional fused feature vector. Then, it performs feature dimensionality reduction through a fully connected layer (128 neurons) to finally output a 64-dimensional adapted feature vector, which provides a basis for the generation of refined operation parameters.

[0090] The output layer is a fully connected layer with 8 neurons and a Linear activation function. It outputs 8 refined operation parameters for sludge pretreatment, which are directly connected to the resource recycling execution unit.

[0091] The basic process data inputs for the microbial community sludge adaptation model are the sludge physicochemical feature set and the fermentation microbial community activity feature set preprocessed by the collaborative data interaction unit, which are used as the inputs to the CNN branch and the LSTM branch, respectively.

[0092] The upstream model optimization parameters of the microbial community-sludge adaptation model are derived from the resource matching prediction model and the cross-domain process optimization model. These inputs serve as the basis for the model's adaptation direction, determining the model's refined adaptation target. They are all outputs of the upstream models and are transmitted in real-time to the fusion layer of the microbial community-sludge adaptation model to participate in the generation of adaptation feature vectors. Inputs from the resource matching prediction model may include sludge-fermentation feedstock matching degree (numerical range [0,1]), highly correlated feature pairs (such as sludge carbon-nitrogen ratio-microbial activity), and correlation matrix. This type of data determines the core adaptation direction of the model, namely, focusing on adapting to highly correlated feature pairs; the input from the cross-domain process optimization model can include the optimization range of sludge pretreatment parameters (such as moisture content 55%-60%, carbon-nitrogen ratio 25-30:1), the weight allocation results of the multi-objective reward function, and the globally optimal combination of process parameters. This type of data determines the parameter output range of the model, ensuring that the refined parameters output by the model are within the globally optimal range, and avoiding disconnection from the global optimization.

[0093] The two types of input data are deeply integrated within the model. The basic process data provides the actual process basis for parameter generation, while the upstream model optimization parameters provide the direction basis for parameter generation. Together, they ensure that the refined parameters output by the model not only fit the actual process operation status but also meet the global optimization goal.

[0094] The output of the microbial sludge adaptation model consists of eight refined operation parameters for sludge pretreatment. These parameters are directly executable quantitative parameters with no parameter range, and they are directly connected to the downstream resource recycling execution unit. At the same time, the output results are synchronously shared with the digital twin unit and the intelligent early warning unit to achieve parameter synchronization of multiple modules.

[0095] The model outputs eight parameters, which are customized operating parameters for the sludge co-transport module. All of them are quantitative values ​​that conform to the actual operating range of the equipment. Specifically, they include: sludge flocculant dosage (kg / ton of sludge), dewatering machine speed (r / min), sludge conditioning time (min), conditioning tank stirring speed (r / min), sludge transfer pump frequency (Hz), sludge preheating temperature (°C), straw addition (kg / ton of sludge), and composting agent addition (g / ton of sludge).

[0096] The refined operational parameters output by the model are directly sent to the sludge co-transport module of the resource recycling execution unit via an encrypted interface. All equipment in this module (flocculator dosing device, dewatering machine, transfer pump, etc.) operates fully automatically based on these parameters. The parameter sending frequency is once per minute, synchronized with the sampling frequency, ensuring that the equipment operating parameters can match process changes in real time. The model's output results are also synchronized to the digital twin unit and the intelligent early warning unit, realizing parameter synchronization across multiple modules for shared application. Specifically, as the core parameter basis for virtual-real linkage control in the digital twin unit, the digital twin platform updates the virtual model's operating status based on these parameters and performs simulation verification. If the simulation finds operational risks in the parameters, it will be fed back to the model in real time for parameter correction. As the process parameter basis for cross-domain fault early warning in the intelligent early warning unit, the intelligent early warning unit compares these parameters with the equipment's rated operating range. If the parameters exceed the range, it will trigger the corresponding level of warning to ensure equipment operation safety.

[0097] In the AI ​​cross-domain optimization unit, the water quality and wastewater adaptation model uses a random forest algorithm with dynamic feature matching rules to calculate the pretreatment parameters for fermentation wastewater. Using real-time fluctuation data of COD and ammonia nitrogen influent from the water treatment plant as core input features, the decision tree branching rules are optimized. The pretreatment parameters for fermentation wastewater are dynamically adjusted according to changes in the influent load, maintaining the neutralization pH at 6.5-7.5 and the filtration accuracy at 5-10μm. This ensures that the reused wastewater has a COD ≤300mg / L and an ammonia nitrogen ≤25mg / L, achieving a reuse compliance rate ≥99%.

[0098] All input data for the wastewater adaptation model originates from upstream modules within the system, with no independently collected data source. The input data has a two-tiered structure, including core process data input and related parameter reference input. The core process data serves as the core of the model's computation, while the related parameter reference input provides the basis for optimization direction. Specifically, the core process data input, derived from a standardized cross-domain dataset, forms the foundation for dynamic feature matching and parameter optimization. The related parameter reference input, derived from the resource matching prediction model and the cross-domain process optimization model, provides the optimization direction. These two types of input data are deeply integrated within the model. The core process data provides practical process guidance for parameter optimization, while the related parameter reference input provides global directional guidance. Together, they ensure that the preprocessed parameters output by the model accurately reflect actual water quality changes and align with the global goal of cross-domain collaboration.

[0099] The wastewater quality adaptation model is optimized based on dynamic feature matching rules, outputting refined operational parameters for fermentation wastewater pretreatment. These parameters are quantitative and executable, with no parameter range. They are primarily applied to the fermentation wastewater pretreatment module within the resource recycling execution unit. Simultaneously, the output results are shared with the digital twin unit and the intelligent early warning unit, enabling multi-module synchronous monitoring of the parameters. The model outputs seven customized operational parameters for the fermentation wastewater pretreatment process, all of which are quantitative values ​​consistent with the actual operating range of the equipment. These parameters include: neutralizing agent (sodium hydroxide / sulfuric acid) dosage (g / L), pH adjustment tank stirring speed (r / min), precision filter filtration accuracy (μm), filter backwashing frequency (times / hour), backwashing time (min), disinfectant (sodium hypochlorite) dosage (mg / L), and wastewater booster pump frequency (Hz).

[0100] Similar to the microbial sludge adaptation model, the wastewater adaptation model is simultaneously synchronized to both the digital twin unit and the intelligent early warning unit, achieving integrated linkage of execution, monitoring, and early warning. Specifically, as the basis for the virtual-real linkage control of the digital twin unit, the digital twin platform updates the virtual model operating status of the fermentation wastewater pretreatment module based on these parameters, and performs simulation verification of the pretreatment effect. If the simulation finds that the water quality of the reused wastewater fails to meet the standards, it will provide real-time feedback to the model, triggering parameter re-optimization. As the basis for the water quality parameters for cross-domain fault early warning of the intelligent early warning unit, the intelligent early warning unit compares the parameters output by the model with the equipment's rated operating range and water quality compliance standards. If the parameters exceed the equipment's operating range or the simulated water quality fails to meet the standards, it will immediately trigger the corresponding level of early warning to prevent unqualified wastewater from entering the water treatment makeup system.

[0101] In this embodiment, the wastewater adaptation model adopts a random forest algorithm with dynamic feature matching rules. The core innovation lies in the customized optimization of the decision tree branching rules, feature weights, and node partitioning of the traditional random forest algorithm. A dynamic feature matching rule that fits the fermentation wastewater reuse process is designed. This rule is not a simple numerical matching, but a triple dynamic matching based on changes in water treatment influent load, fluctuations in fermentation wastewater quality, and cross-domain process correlation. The optimization process is a fully automated closed-loop process, with a single round of optimization taking ≤10ms, ensuring the real-time performance of the model. The specific optimization steps are as follows:

[0102] Step 1: Feature selection and weight allocation.

[0103] The model first performs feature filtering on the input core process data, and then outputs a cross-domain feature correlation matrix based on the resource matching prediction model. The model selects fermentation wastewater features (such as COD, ammonia nitrogen, and pH) that are highly correlated with COD and ammonia nitrogen in the water treatment influent as core features of the random forest algorithm, accounting for more than 70% of the feature weight; the remaining features are auxiliary features, accounting for less than 30%. Simultaneously, the model dynamically adjusts the weight of the core features based on the real-time fluctuations in COD and ammonia nitrogen in the water treatment influent: the greater the fluctuation, the higher the weight of the core features, ensuring that the model focuses on features that play a crucial role in achieving water quality standards.

[0104] Step 2: Constructing dynamic decision tree branching rules.

[0105] The model is designed specifically for the characteristics of fermentation wastewater reuse processes, and features a customized dynamic decision tree branching rule system. Unlike the fixed branching rules of traditional random forest algorithms, the branching rules in this model are dynamically adjusted according to changes in the influent load and the quality of the fermentation wastewater. The branching rules are constructed based on the following:

[0106] Branch node division: The core branch nodes are COD (300 mg / L as threshold), ammonia nitrogen (25 mg / L as threshold), COD (800 mg / L as threshold), and ammonia nitrogen (60 mg / L as threshold) of fermentation wastewater. The node division threshold is dynamically adjusted according to the change of influent load.

[0107] Branch path optimization: Based on the classification of inflow load (low, medium, high), optimize the branch paths of the decision tree. When the load is high, reduce the branch paths of non-core features to improve the computational efficiency of the model; when the load is low, increase the branch paths to improve the accuracy of parameter optimization.

[0108] Step 3: Training and prediction of the random forest model.

[0109] The model is trained and predicted using a random forest algorithm based on the selected features and dynamically constructed decision tree branching rules: the number of decision trees is set to 100, the decision tree depth to 10, the minimum number of sample splits to 2, and the maximum number of features to 6 (consistent with the core process feature dimensions); the model is trained using actual operational data from the past year, with the training set accounting for 80% and the test set accounting for 20%, and the model's prediction accuracy is ≥98%; based on the trained model, preliminary predictions are made for the optimized values ​​of the fermentation wastewater pretreatment parameters, and the predicted parameter values ​​are obtained.

[0110] Step 4: Dynamic feature matching verification.

[0111] The model is based on a triple dynamic matching rule to verify the initially predicted parameter values. If the verification is successful, it proceeds to the next step; if the verification fails, it returns to step 2 to readjust the decision tree branching rules. The triple dynamic matching rule is the core innovation of the model, as detailed below:

[0112] Water quality compliance matching: The pretreated wastewater quality corresponding to the parameter prediction values ​​must meet the water treatment makeup water standards (COD≤300mg / L, ammonia nitrogen≤25mg / L), which is the basic matching rule;

[0113] Influent load matching: The predicted parameter values ​​need to match the real-time influent load at the water treatment end. That is, when the influent load is high, the pretreatment parameters need to improve the wastewater treatment accuracy to avoid impacting the water treatment biological system; when the influent load is low, the pretreatment parameters can be appropriately reduced to save energy.

[0114] Cross-domain process correlation matching: The predicted parameter values ​​must match the global optimization objectives output by the cross-domain process optimization model, that is, ensure that the wastewater reuse rate is ≥85% while controlling the pretreatment energy consumption, in line with the goal of reducing the overall cross-domain energy consumption by 20%-25%.

[0115] The resource recycling execution unit serves as the implementation platform for the entire system and is a crucial component in achieving the desired technical effects. It is used for sludge transport and wastewater reuse based on execution parameters, while simultaneously feeding back the execution data to the collaborative data interaction unit. The resource recycling execution unit comprises a sludge co-transport module, a fermentation wastewater pretreatment module, and a resource recycling feedback module. The control terminals of the sludge co-transport module and the fermentation wastewater pretreatment module are communicatively connected to the output terminal of the AI ​​cross-domain optimization unit, while the output terminal of the resource recycling feedback module is connected to the input terminal of the collaborative data interaction unit.

[0116] The sludge co-transport module includes a sludge moisture content adjustment device, an intelligent transport pump set, and a flow metering device. The control parameters come from the output of the AI ​​cross-domain optimization unit, which accurately controls the sludge moisture content at 55%-60% and transports it to the biomass fermentation tank in a timely and quantitative manner according to the fermentation requirements.

[0117] The fermentation wastewater pretreatment module consists of an automatic pH adjustment tank, a precision filter, and a disinfection device. The control parameters are derived from the output of the AI ​​cross-domain optimization unit, which automatically adjusts the dosage of neutralizing agent and the backwashing frequency of the filter to ensure that the pretreated wastewater meets the standards.

[0118] The resource recycling feedback module deploys flow, water quality, and moisture content sensors to collect real-time data on sludge delivery volume, wastewater reuse volume, and treated water quality. This data is then transmitted back to the collaborative data interaction unit, providing data support for model iteration in the AI ​​cross-domain optimization unit and forming a closed-loop optimization.

[0119] The digital twin unit is used to simulate and verify optimized parameters, and to remotely adjust parameters and perform emergency dispatch. The digital twin unit includes a cross-domain virtual modeling module, a collaborative simulation optimization module, and a virtual-real linkage control module. Its data sources come from the collaborative data interaction unit and the AI ​​cross-domain optimization unit.

[0120] The cross-domain virtual modeling module adopts CFD+AI fusion modeling technology to restore the physical scene of the entire water treatment process and the entire biomass natural gas chain in a 1:1 ratio. The data source comes from the collaborative data interaction unit, realizing real-time synchronization between the physical scene and the virtual scene with a data delay of ≤1 minute.

[0121] The collaborative simulation optimization module simulates the cross-domain operation effect under different sludge supply and wastewater quality based on the parameter combination output by the AI ​​cross-domain optimization engine. It compares the resource utilization rate and energy consumption of different parameter combinations, outputs the optimal collaborative solution, and avoids the risk of trial and error on site.

[0122] The virtual-real linkage control module supports remote adjustment of cross-domain process parameters on a virtual platform and sends instructions to the resource cycle execution unit to achieve integrated management and control of virtual monitoring, remote decision-making, and on-site execution.

[0123] The intelligent early warning unit is a crucial safety component of the entire system, tightly integrated with and inseparable from the front-end modules. It includes a cross-domain fault correlation model, a multi-level early warning module, and an emergency dispatch module. The intelligent early warning unit identifies cascading fault risks in cross-domain data based on the cross-domain fault correlation model, pushes early warning information according to preset levels based on the identification results, and generates emergency dispatch plans. The intelligent early warning unit's cross-domain fault correlation graph incorporates 12 types of cross-domain fault propagation paths, enabling it to identify cascading risks 72 hours in advance and reducing fault handling response time by 60%. The multi-level early warning module employs a three-tiered early warning mechanism (yellow, orange, and red), and the emergency dispatch module's dispatch plan is generated based on the parameter rules of the AI ​​cross-domain optimization unit.

[0124] Specifically, the cross-domain fault association module is used to identify cascading risks using a cross-domain fault association model with a built-in cross-domain fault association graph. The cross-domain fault association model extracts the core features of cross-domain faults using the XGBoost algorithm, understands the cross-domain propagation patterns of faults based on the cross-domain fault propagation association graph, and achieves accurate identification and early warning of cross-domain cascading faults through a fusion algorithm of XGBoost and GNN. The XGBoost and GNN fusion algorithm, based on the characteristics and propagation patterns of cross-domain faults, designs a five-layer fusion architecture: feature extraction layer (XGBoost), graph structure construction layer, graph feature learning layer (GNN), fault identification layer, and early warning output layer. The function and algorithm parameters of each layer are trained and optimized based on actual cross-domain fault data, with a single round of fault identification taking ≤20ms, ensuring real-time early warning.

[0125] The specific steps are as follows:

[0126] 1. The XGBoost algorithm is used to extract the core features of cross-domain faults.

[0127] The feature extraction layer is the foundational layer of the fusion algorithm. It adopts the XGBoost extreme gradient boosting algorithm. Its core function is to extract the core features of cross-domain faults from massive cross-domain operation data, eliminate invalid features, and provide high-quality feature basis for subsequent graph structure construction and graph feature learning.

[0128] Specifically, the input is a standardized cross-domain dataset. The number of decision trees is set to 200, the learning rate to 0.1, the maximum tree depth to 12, the minimum number of sample splits to 3, and the loss function to be Logistic Loss. Using the XGBoost algorithm to rank the features by importance, the 20 core fault features with the greatest impact on cross-domain faults are extracted (10 from the water treatment end and 10 from the biomass end), such as sludge dewatering machine failure, fermenter agitation failure, sudden increase in influent COD, and sudden decrease in bacterial activity. The importance weights of each core fault feature are output, providing a weighting basis for graph structure construction.

[0129] 2. Utilize cross-domain fault propagation correlation graphs to analyze core features based on the propagation patterns of cross-domain faults and construct a fault graph structure model.

[0130] The graph structure construction layer is the core connecting layer of the fusion algorithm. Its core function is to construct a cross-domain fault graph structure model G=(V,E,W) based on the core fault features extracted by XGBoost and the cross-domain fault propagation association graph. This transforms the propagation law of cross-domain faults into a computer-recognizable graph structure. In this model, node V corresponds to 20 core fault features, and each node represents a cross-domain fault event (e.g., node V1 is a fermenter stirring fault, and node V2 is a sludge consumption stagnation fault). Edge E corresponds to the fault propagation relationship between nodes. If fault event A can trigger fault event B, then a directed edge E (A→B) is constructed between node A and node B, representing the fault propagating from A to B. The weight W corresponds to the fault propagation probability of the edge, which is obtained by training based on the cross-domain fault propagation association graph and actual fault data. The weight value ranges from [0,1], and the larger the weight value, the higher the probability of the propagation path occurring.

[0131] The cross-domain fault propagation correlation graph consists of a three-layer structure: a core layer, a middle layer, and an outer layer. Each layer corresponds to a different type of fault node, and the layers are connected by directed edges, clearly showing the source, process, and impact of fault propagation. The specific structure is as follows:

[0132] Core layer (fault source nodes): There are 12 nodes in total. These are the core fault events that are prone to triggering cross-domain cascading failures, such as fermenter stirring failure, sludge dewatering machine failure, sudden increase in influent COD, and sudden decrease in bacterial activity. These are the core analysis objects of the graph.

[0133] Intermediate layer (propagation nodes): There are 20 nodes in total. These are intermediate events in the fault propagation process, such as sludge consumption stagnation, sudden drop in fermentation gas production rate, wastewater reuse water quality exceeding standards, and sudden increase in water treatment chemical consumption. They serve as a bridge connecting the core layer and the outer layer.

[0134] Outer layer (affected nodes): There are 14 nodes in total. These are the final impact events of fault propagation, such as the accumulation of sludge in water treatment, interruption of biomass natural gas production, collapse of water treatment biochemical system, and sudden increase in cross-domain overall energy consumption. They are the key targets for fault early warning and handling.

[0135] It is important to note that the 12 types of cross-domain fault propagation paths identified in this embodiment all originate from the core layer of the cross-domain fault propagation correlation map. These are high-risk, high-probability cross-domain fault propagation paths verified through process analysis and data validation. They cover the entire process chain of water treatment-biomass natural gas synergy, including sludge transportation, wastewater reuse, microbial culture, and process control. The paths are categorized into three types: propagation from biomass to water treatment (7 paths), propagation from water treatment to biomass (4 paths), and bidirectional cross-propagation (1 path). Each path clearly defines the fault propagation direction, core nodes, and propagation probability, with a propagation probability ≥80%. This provides clear quantitative evidence for cross-domain fault early warning and handling. The 12 specific paths are as follows:

[0136] Seven pathways for fault propagation from the biomass end to the water treatment end. This type of pathway, where a failure at the biomass end triggers a failure at the water treatment end, is the main type of cross-domain fault propagation, accounting for nearly 60%. Specific pathways include:

[0137] Fermentation tank stirring failure → sudden drop in bacterial activity → sludge consumption stagnation → sludge accumulation failure in water treatment (95% probability of propagation).

[0138] Fermentation tank temperature control failure → sudden drop in fermentation efficiency → fermentation wastewater quality exceeding standards → water treatment makeup system failure (propagation probability 92%).

[0139] Methane purification unit malfunction → gas production interruption → sludge transport suspension → water treatment sludge disposal system paralysis (probability of propagation 88%).

[0140] Fermentation feed ratio error → carbon-nitrogen ratio imbalance → mass bacterial death → zero sludge consumption → accumulation of water treatment sludge (probability of transmission 87%).

[0141] Fermentation wastewater pretreatment device malfunction → excessive suspended solids in wastewater → clogged water treatment filter → filtration system malfunction (85% probability of propagation).

[0142] Biomass-side chemical dosing device malfunction → fermentation pH imbalance → sudden drop in gas production rate → slowed sludge transport → overflow of water treatment sludge temporary storage tank (probability of transmission 83%).

[0143] Fermentation tank sealing failure → biogas leakage → system shutdown → sludge transport interruption → water treatment sludge disposal system failure (propagation probability 80%).

[0144] Four pathways for fault propagation from water treatment to biomass. This type of pathway, where a fault at the water treatment end leads to a fault at the biomass end, is a secondary type of cross-domain fault propagation. Specific pathways include:

[0145] A sudden increase in influent COD / ammonia nitrogen leads to a sudden change in the physicochemical characteristics of sludge, which in turn inhibits the activity of the microbial community and causes a sharp drop in the biomass fermentation gas production rate (93% probability of transmission).

[0146] Sludge dewatering machine malfunction → excessive sludge moisture content → poor feeding into fermentation tank → decreased fermentation efficiency (90% probability of transmission).

[0147] Water treatment dosing device malfunction → sludge organic matter content drops sharply → fermentation raw material quality declines → gas production rate drops sharply (probability of propagation 86%).

[0148] Sludge transfer pump failure → sludge supply interruption → fermentation raw material shortage → partial shutdown of biomass production (probability of transmission 82%).

[0149] Type 1 path with bidirectional cross-propagation. This type of path involves faults at both ends mutually triggering and propagating across each other. It is the most severe type of cross-domain fault and is prone to causing systemic operational risks. Specific path:

[0150] Water treatment biological system collapse → effluent quality fails to meet standards → reclaimed water quality exceeds standards → fermentation bacteria activity drops sharply → gas production rate is zero → sludge transport is interrupted → water treatment sludge accumulates → biological system collapses further (probability of propagation 98%).

[0151] 3. Utilize graph neural networks to learn graph features from fault graph structure models to identify the cross-domain propagation patterns, propagation speed, and chain reaction paths of faults.

[0152] The graph feature learning layer is the core layer of the fusion algorithm, employing a Graph Convolutional Network (GCN) from the Graph Neural Network (GNN) family. Its core function is to learn graph features from the constructed cross-domain fault graph structure model, identifying the cross-domain propagation patterns, propagation speed, and chain reaction paths of faults. The GCN architecture consists of two graph convolutional layers and one fully connected layer. Each graph convolutional layer has 64 neurons with ReLU activation, and the fully connected layer has 32 neurons. It learns the fault feature embedding vector for each node, quantifies the fault propagation correlation between nodes, and identifies the critical paths and core nodes (i.e., key fault events that easily trigger cross-domain chain failures) of fault propagation. It outputs a 64-dimensional fault feature embedding vector for each node and a fault propagation correlation matrix between nodes, providing the core basis for the fault identification layer.

[0153] 4. Accurate identification of cross-domain cascading faults based on the cross-domain propagation patterns, propagation speed, and chain reaction paths of faults.

[0154] The fault identification layer, based on the output of the graph feature learning layer and combined with the cross-domain fault propagation association graph, achieves accurate identification of cross-domain cascading faults, including four-dimensional identification of fault type, fault propagation path, fault impact range, and fault development trend.

[0155] Specifically, based on fault feature embedding vectors, a logistic regression classifier is used to identify the type of the initial fault, determine the initial domain (water treatment end / biomass end) and the specific fault event. Based on the fault propagation correlation matrix, preset propagation paths in the cross-domain fault propagation correlation graph are matched to identify potential cascading fault paths that the initial fault may trigger, and the propagation probability of each path is marked. A propagation probability threshold of ≥80% is set as a high-risk path, and high-risk paths are analyzed in detail. The impact range of the fault is determined in conjunction with process coupling relationships, categorized into single-domain impact (only the initial domain) and cross-domain impact (affecting another domain).

[0156] A quantitative model based on fault propagation speed (trained from actual fault data) predicts the development trend of faults, dividing them into three stages: short-term (0-24h), medium-term (24-48h), and long-term (48-72h), and identifying the fault events that may occur in each stage.

[0157] 5. Output multi-dimensional early warning information.

[0158] The early warning output layer is the final output layer of the fusion algorithm. Its core function is to transform the results of the fault identification layer into standardized, multi-dimensional early warning information, which is then transmitted to the multi-level early warning module and emergency dispatch module of the intelligent early warning unit. The early warning information is output in a structured data format, containing seven core dimensions: initial fault type, initial occurrence area, potential chain fault paths, propagation probability of each path, fault impact range, fault development trend, and fault occurrence time prediction. This ensures that downstream modules can quickly obtain core early warning information and carry out subsequent handling.

[0159] The multi-level early warning module is used to generate a multi-level early warning mechanism based on preset fault thresholds and to generate early warning information based on fault identification results. The multi-level early warning module sets up a three-level early warning mechanism: yellow warning (parameter deviation from the optimal range of 5%), orange warning (deviation of 10%), and red warning (deviation of 15% or equipment failure), and pushes early warning information through SMS, APP, and central control screen.

[0160] The cross-domain fault association model classifies identified fault risks into three levels: low, medium, and high, corresponding to the yellow, orange, and red three-level early warning mechanisms of the intelligent early warning unit. The early warning information output by the model serves as the triggering basis for early warnings provided by the multi-level early warning modules, with specific linkages as follows:

[0161] Low risk (yellow alert): Only initial signs of failure are detected, with no obvious cross-domain propagation trend and a propagation probability of <60%. The model triggers a yellow alert, and the alert information is pushed through the central control screen to remind maintenance personnel to pay attention.

[0162] Medium risk (orange alert): A clear cross-domain propagation path has been identified, with a propagation probability of 60%-80%. The fault may affect another area within 24-48 hours. The model triggers an orange alert and pushes the alert information through the central control screen or APP, requiring maintenance personnel to take preventive measures.

[0163] High risk (red alert): A high-risk cross-domain propagation path has been identified, with a propagation probability of ≥80%. The fault will trigger a cross-domain cascading failure within 48-72 hours. The model triggers a red alert and pushes the alert information to multiple terminals such as the central control screen, APP, or SMS to operation and maintenance personnel and managers, requiring them to immediately initiate emergency response.

[0164] Meanwhile, the model transmits fault development trend data to the multi-level early warning module in real time, enabling dynamic adjustment of the early warning level. For example, if a low-risk fault further develops into a medium-risk fault, it will automatically upgrade from a yellow warning to an orange warning.

[0165] The emergency dispatch module is used to generate emergency dispatch plans based on early warning information and preset parameter rules to ensure stable system operation.

[0166] The emergency dispatch module generates customized emergency dispatch solutions based on preset process control rules and parameter rules of the AI ​​cross-domain optimization unit, combined with the specific type and propagation path of the fault. The solutions include core components such as equipment control parameters, process adjustment measures, material transportation strategies, and personnel handling requirements.

[0167] The emergency dispatch plan is sent to the resource cycle execution unit through the virtual-physical linkage control module to realize the automated emergency adjustment of equipment and processes, and at the same time, it is pushed to the operation terminal of the operation and maintenance personnel to guide manual handling;

[0168] The model monitors the execution effect of the emergency dispatch plan in real time. If the trend of the fault development is not contained, the identification results will be updated in real time. The emergency dispatch module will regenerate the dispatch plan based on the new results to achieve dynamic optimization of the dispatch plan.

[0169] This invention constructs a fully closed-loop system encompassing "collaborative data interaction - AI cross-domain optimization - resource cyclic execution - digital twin - intelligent early warning," with each unit strongly bound and linked, and data flow and causal relationships clearly defined. Millisecond-level data transmission is achieved through industrial Ethernet and a 5G private network, a sampling frequency of 1 minute / time ensures real-time performance, and a dynamic weight adjustment mechanism achieves global optimization rather than local optimization, overcoming the shortcomings of existing technologies with fragmented modules.

[0170] The cross-domain fault association model of this invention can achieve 72-hour advance warning in some implementation examples, which is 3-6 times faster than the warning time of existing industry models that range from a few hours to one day. Based on a dynamic graph constructed from 2,000+ fault cases over 3 years, it accurately quantifies 12 types of cross-domain fault propagation paths, reducing fault handling response time by 60% and lowering the system failure rate from 15% to below 3%.

[0171] Verification through multiple embodiments shows that the sludge fermentation conversion rate of the present invention is ≥85%, the wastewater reuse rate is ≥85%, the cross-domain energy consumption is reduced by 20%-25%, and the wastewater reuse compliance rate is ≥99%, with all indicators far exceeding the existing technology level.

[0172] Example 2:

[0173] Embodiment 2 of the present invention provides an energy cycle optimization method based on cross-domain collaboration, such as... Figure 1 As shown, it includes the following steps:

[0174] Acquire synchronous cross-domain data from the water treatment end and the biomass natural gas end, and preprocess the cross-domain data to obtain a standardized cross-domain dataset;

[0175] The resource matching prediction model is used to match sludge supply and demand with wastewater reuse based on a standardized cross-domain dataset. The cross-domain process optimization model optimizes the global parameters of the matching results. The microbial sludge adaptation model and the water quality wastewater adaptation model are used to generate corresponding execution parameters based on the optimized global parameters in combination with the actual energy cycle scenario.

[0176] Based on the cross-domain fault association model, the risk of cascading faults is identified in cross-domain data. According to the identification results, early warning information is pushed according to the preset level and an emergency dispatch plan is generated.

[0177] To better illustrate the superiority of the above method, the following experiment was conducted in this embodiment:

[0178] Experiment 1:

[0179] 1. System parameters and raw material conditions:

[0180] Water treatment capacity: 20,000 tons / day, sludge production: 50 tons / day, initial moisture content: 85%, carbon-nitrogen ratio: 22:1;

[0181] Biomass natural gas scale: 150,000 m³ / year, effective volume of fermentation tank 1000 m³, microbial community type is methanogens (initial activity 85%), fermentation raw material requirements: daily consumption of 45-50 tons of sludge, moisture content 55%-60%, carbon-nitrogen ratio 25-30:1;

[0182] Fermentation wastewater quality: COD=800mg / L, ammonia nitrogen=60mg / L; water treatment makeup water standard: COD≤300mg / L, ammonia nitrogen≤25mg / L.

[0183] 2. System closed-loop operation process:

[0184] Step 1: The collaborative data interaction unit synchronously collects 12 core parameters, including sludge production, moisture content, carbon-nitrogen ratio, and influent water quality at the water treatment end, and temperature, pH, bacterial activity, and fermentation wastewater quality at the biomass end, through dual redundant sensors. The sampling frequency is 1 minute / time. After outliers are removed by the graded σ principle and the data is standardized, it is transmitted in real time to the AI ​​cross-domain optimization unit through the OPC UA interface with a transmission delay of ≤40ms.

[0185] Step 2: The resource matching prediction model, based on a cross-domain dataset, predicts a 97% match between sludge supply and fermentation demand over the next 48 hours through a cross-domain feature cross-attention mechanism. It also indicates that the adaptability of fermentation wastewater reuse needs optimization, outputting preliminary directions for a neutral pH of 7.0 and a filtration accuracy of 8μm. The cross-domain process optimization model, based on the matching data, dynamically adjusts the reward function weights and outputs the globally optimal parameter combination: 0.8 kg / ton of sludge flocculant for sludge dewatering, 1000 r / min dewatering machine speed; 37℃ biomass fermentation temperature, 50 r / min stirring speed; and straw supplementation to adjust the carbon-nitrogen ratio to 28:1. The execution parameters are then refined using the microbial sludge adaptation model and the water quality wastewater adaptation model.

[0186] Step 3: Based on the AI ​​output parameters, the resource recycling execution unit and the sludge co-transport module transport 48 tons of dewatered sludge with a moisture content of 58% to the fermentation tank. The fermentation wastewater pretreatment module treats the wastewater at pH=7.0 and a filtration accuracy of 8μm, and reuses it in the water treatment makeup system (reuse volume 120m³ / day). The resource recycling feedback module collects execution data in real time and transmits it back to the collaborative data interaction unit to complete the model iteration.

[0187] Step 4: The digital twin unit synchronizes real-time data, restores the physical scene 1:1, performs simulation verification on the parameter combination, confirms that there is no operational risk, and the operation and maintenance personnel monitor the entire process operation status in real time through the platform.

[0188] Step 5: The intelligent early warning unit analyzes cross-domain equipment and process parameters in real time, identifies risks through cross-domain fault correlation graphs, and has no abnormalities throughout the process, so no early warning is triggered.

[0189] The results of the above steps are compared with those of conventional techniques, as shown in Table 1.

[0190] Table 1. Comparison of Operation Results with Existing Conventional Technologies

[0191]

[0192] Experiment 2:

[0193] 1. System parameters and raw material conditions:

[0194] Water treatment capacity: 15,000 tons / day, sludge production: 35 tons / day, initial moisture content: 82%, carbon-nitrogen ratio: 20:1;

[0195] Biomass natural gas scale: 100,000 m³ / year, effective volume of fermentation tank: 800 m³, initial activity of microbial community: 83%, fermentation raw material requirements: daily consumption of 30-35 tons of sludge, moisture content: 55%-60%, carbon-nitrogen ratio: 25-30:1;

[0196] Fermentation wastewater quality: COD=750mg / L, ammonia nitrogen=55mg / L; water treatment makeup water standard: COD≤300mg / L, ammonia nitrogen≤25mg / L.

[0197] 2. Results:

[0198] The sludge fermentation conversion rate was 85.1%, the biomass gas production rate was 0.47 m³ / kgVS (an increase of 19.8%), the water treatment chemical consumption was reduced by 19.5%, the fermentation stirring energy consumption was reduced by 15.3%, the wastewater reuse rate was 86.3%, the cross-domain fault early warning rate was 73 hours, and the system failure rate was 3.0%.

[0199] Experiment 3:

[0200] 1. System parameters and raw material conditions:

[0201] Water treatment capacity: 30,000 tons / day, sludge production: 65 tons / day, initial moisture content: 88%, carbon-nitrogen ratio: 24:1;

[0202] Biomass natural gas scale: 200,000 m³ / year, effective volume of fermentation tank 1200 m³, initial activity of microbial community 87%, fermentation raw material requirements: daily consumption of 60-65 tons of sludge, moisture content 55%-60%, carbon-nitrogen ratio 25-30:1;

[0203] Fermentation wastewater quality: COD=850mg / L, ammonia nitrogen=65mg / L; water treatment makeup water standard: COD≤300mg / L, ammonia nitrogen≤25mg / L.

[0204] 2. Results:

[0205] The sludge fermentation conversion rate was 87.5%, the biomass gas production rate was 0.49 m³ / kgVS (an increase of 23.1%), the water treatment chemical consumption was reduced by 22.7%, the fermentation stirring energy consumption was reduced by 17.8%, the wastewater reuse rate was 90.2%, the cross-domain fault early warning rate was 76 hours, and the system failure rate was 2.5%.

[0206] The above experiments demonstrate that the method of the present invention can be adapted to water treatment and biomass natural gas projects of different scales, stably achieving unexpected technical effects, and possesses strong practicality and scalability.

[0207] Example 3:

[0208] Embodiment 3 of the present invention provides a computer-readable storage medium storing a computer program adapted for loading by a processor and executing the steps in the cross-domain collaborative energy cycle optimization method as described in Embodiment 2 of the present invention.

[0209] Example 4:

[0210] Embodiment 4 of the present invention provides a computer device, the device comprising:

[0211] A processor, adapted to execute computer programs;

[0212] A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps in the energy cycle optimization method based on cross-domain collaboration as described in Embodiment 2 of the present invention.

[0213] The steps and methods involved in Examples 2, 3 and 4 above correspond to those in Example 1. For specific implementation details, please refer to the relevant description section of Example 1.

[0214] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0215] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0216] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An energy cycle optimization system based on cross-domain collaboration, characterized in that, include: The collaborative data interaction unit is used to acquire synchronous cross-domain data from the water treatment end and the biomass natural gas end, and to preprocess the cross-domain data to obtain a standardized cross-domain dataset. The cross-domain data includes sludge production, moisture content, carbon-nitrogen ratio, influent COD, and ammonia nitrogen from the water treatment end, and fermentation tank temperature, pH, C / N ratio, microbial activity, methane concentration, and fermentation wastewater quality from the biomass natural gas end. The AI ​​cross-domain optimization unit is used to match sludge supply and demand with wastewater reuse based on a standardized cross-domain dataset using a resource matching prediction model. The cross-domain process optimization model optimizes the global parameters of the matching results. The global parameters include sludge dewatering parameters at the water treatment end and fermentation parameters at the biomass-natural gas end. The microbial sludge adaptation model is used in conjunction with the actual energy cycle scenario to generate sludge pretreatment operation parameters based on the optimized biomass-natural gas end fermentation parameters. The water quality-wastewater adaptation model is used in conjunction with the actual energy cycle scenario to generate fermentation wastewater pretreatment operation parameters based on the optimized water treatment end sludge dewatering parameters. The sludge pretreatment operation parameters include sludge flocculant dosage, dewatering machine speed, sludge conditioning time, conditioning tank stirring speed, sludge conveying pump frequency, sludge preheating temperature, straw addition amount, and composting agent addition amount. The fermentation wastewater pretreatment operation parameters include neutralizing agent dosage, pH adjustment tank stirring speed, precision filter filtration accuracy, filter backwashing frequency, backwashing time, disinfectant dosage, and wastewater lift pump frequency. The intelligent early warning unit is used to identify the risk of cascading failures in cross-domain data based on the cross-domain fault association model, and to push early warning information according to the preset level and generate emergency dispatch plan based on the identification results. Among them, the resource matching prediction model is used to use the LSTM algorithm with cross-domain feature cross-attention mechanism to perform Hadamard product operation and feature splicing fusion on the water treatment end feature vector and biomass natural gas end feature vector in the standardized cross-domain dataset. The cross-domain process correlation degree is automatically identified and the attention weight is dynamically allocated through the dual-domain feature interaction layer, attention splitting layer and Softmax weight normalization layer. The sludge supply and demand and wastewater reuse are matched according to the identified cross-domain process correlation degree. The cross-domain process optimization model utilizes reinforcement learning and genetic algorithms with dynamic weighted multi-objective reward functions to optimize sludge fermentation conversion rate, wastewater reuse rate, and overall energy consumption. The reward function weights are dynamically adjusted based on the matching degree output by the resource matching prediction model to globally optimize sludge dewatering parameters at the water treatment end and fermentation parameters at the biomass natural gas end. The microbial community sludge adaptation model is used to utilize a dual-input branch CNN and LSTM fusion algorithm. One branch extracts spatial feature information of sludge physicochemical characteristics through CNN, and the other branch extracts time series feature information of fermentation microbial community activity through LSTM. The outputs of the two branches are concatenated and dimensionality reduced through a fusion layer to establish a real-time correlation between sludge parameters and fermentation microbial community activity. Sludge pretreatment operation parameters are generated based on the globally optimized parameters. The water quality and wastewater adaptation model is used to utilize a random forest algorithm with dynamic feature matching rules. It takes real-time fluctuation data of COD and ammonia nitrogen in the water treatment terminal as the core input features, selects core features based on the cross-domain feature correlation matrix and dynamically adjusts the feature weights, dynamically adjusts the decision tree branching rules based on the changes in influent load, and generates fermentation wastewater pretreatment operation parameters based on the globally optimized parameters. The cross-domain fault association model uses the XGBoost algorithm to extract the core features of cross-domain faults, constructs a fault graph structure model based on the cross-domain fault propagation association graph, and uses a graph neural network to learn the graph features of the fault graph structure model to identify the cross-domain propagation rules, propagation speed and chain reaction path of the fault.

2. The energy cycle optimization system based on cross-domain collaboration as described in claim 1, characterized in that, The collaborative data interaction unit includes: The data acquisition module is used to collect synchronous cross-domain data between the water treatment end and the biomass natural gas end; The data standardization module is used to preprocess synchronous cross-domain data. Specifically, the preprocessing includes outlier removal, missing value completion, and standardization operations. The cross-domain data interface module is used to build a two-way data interaction interface based on the communication protocol.

3. The energy cycle optimization system based on cross-domain collaboration as described in claim 1, characterized in that, The intelligent early warning unit includes: The cross-domain fault association module is used to identify cascading risks using a cross-domain fault association model with a built-in cross-domain fault association graph. The multi-level early warning module is used to generate a multi-level early warning mechanism based on preset fault thresholds and to generate early warning information based on fault identification results. The emergency dispatch module is used to generate emergency dispatch plans based on early warning information and preset parameter rules.

4. The energy cycle optimization system based on cross-domain collaboration as described in claim 1, characterized in that, It also includes a resource recycling execution unit, which is used to transport sludge and reuse wastewater according to execution parameters, and at the same time feeds the execution data back to the collaborative data interaction unit.

5. The energy cycle optimization system based on cross-domain collaboration as described in claim 1, characterized in that, It also includes a digital twin unit, which is used to simulate and verify the optimized parameters, and to perform remote parameter adjustment and emergency dispatch.

6. An energy cycle optimization method based on cross-domain collaboration, characterized in that, Includes the following steps: Synchronous cross-domain data from the water treatment end and the biomass natural gas end are acquired, and the cross-domain data is preprocessed to obtain a standardized cross-domain dataset. The cross-domain data includes sludge production, moisture content, carbon-nitrogen ratio, influent COD, and ammonia nitrogen from the water treatment end, and fermentation tank temperature, pH, C / N ratio, microbial activity, methane concentration, and fermentation wastewater quality from the biomass natural gas end. A resource matching prediction model is used to match sludge supply and demand with wastewater reuse based on a standardized cross-domain dataset. A cross-domain process optimization model optimizes the global parameters of the matching results. The global parameters include sludge dewatering parameters at the water treatment end and fermentation parameters at the biomass-natural gas end. A microbial sludge adaptation model is used in conjunction with an actual energy cycle scenario to generate sludge pretreatment operation parameters based on the optimized biomass-natural gas end fermentation parameters. A water quality-wastewater adaptation model is used in conjunction with an actual energy cycle scenario to generate fermentation wastewater pretreatment operation parameters based on the optimized water treatment end sludge dewatering parameters. The sludge pretreatment operation parameters include sludge flocculant dosage, dewatering machine speed, sludge conditioning time, conditioning tank stirring speed, sludge conveying pump frequency, sludge preheating temperature, straw addition amount, and composting agent addition amount. The fermentation wastewater pretreatment operation parameters include neutralizing agent dosage, pH adjustment tank stirring speed, precision filter filtration accuracy, filter backwashing frequency, backwashing time, disinfectant dosage, and wastewater lift pump frequency. Based on the cross-domain fault association model, the risk of cascading faults in cross-domain data is identified, and early warning information is pushed according to the preset level and an emergency dispatch plan is generated based on the identification results. Among them, the resource matching prediction model is used to use the LSTM algorithm with cross-domain feature cross-attention mechanism to perform Hadamard product operation and feature splicing fusion on the water treatment end feature vector and biomass natural gas end feature vector in the standardized cross-domain dataset. The cross-domain process correlation degree is automatically identified and the attention weight is dynamically allocated through the dual-domain feature interaction layer, attention splitting layer and Softmax weight normalization layer. The sludge supply and demand and wastewater reuse are matched according to the identified cross-domain process correlation degree. The cross-domain process optimization model utilizes reinforcement learning and genetic algorithms with dynamic weighted multi-objective reward functions to optimize sludge fermentation conversion rate, wastewater reuse rate, and overall energy consumption. The reward function weights are dynamically adjusted based on the matching degree output by the resource matching prediction model to globally optimize sludge dewatering parameters at the water treatment end and fermentation parameters at the biomass natural gas end. The microbial community sludge adaptation model is used to utilize a dual-input branch CNN and LSTM fusion algorithm. One branch extracts spatial feature information of sludge physicochemical characteristics through CNN, and the other branch extracts time series feature information of fermentation microbial community activity through LSTM. The outputs of the two branches are concatenated and dimensionality reduced through a fusion layer to establish a real-time correlation between sludge parameters and fermentation microbial community activity. Sludge pretreatment operation parameters are generated based on the globally optimized parameters. The water quality and wastewater adaptation model is used to utilize a random forest algorithm with dynamic feature matching rules. It takes real-time fluctuation data of COD and ammonia nitrogen in the water treatment terminal as the core input features, selects core features based on the cross-domain feature correlation matrix and dynamically adjusts the feature weights, dynamically adjusts the decision tree branching rules based on the changes in influent load, and generates fermentation wastewater pretreatment operation parameters based on the globally optimized parameters. The cross-domain fault association model uses the XGBoost algorithm to extract the core features of cross-domain faults, constructs a fault graph structure model based on the cross-domain fault propagation association graph, and uses a graph neural network to learn the graph features of the fault graph structure model to identify the cross-domain propagation rules, propagation speed and chain reaction path of the fault.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in claim 6, which is based on cross-domain collaborative energy cycle optimization.

8. A computer device, characterized in that, include: A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the energy cycle optimization method based on cross-domain collaboration as described in claim 6.

Images