A sludge resource utilization path determination method, device and equipment

Abstract

The application provides a sludge resource utilization path determination method, device and equipment, and belongs to the technical field of computer information processing, and solves the problem that economic, environmental and social targets are difficult to dynamically and cooperatively optimize in the current sludge resource treatment process. The method comprises the following steps: obtaining original data, wherein the original data comprises at least one of sludge attribute data, process parameters and external dynamic data; preprocessing the original data to obtain sludge sample data; classifying and integrating the sludge sample data to obtain a multi-dimensional tensor; decomposing and dynamically updating the multi-dimensional tensor to obtain a decomposed factor matrix; inputting the decomposed factor matrix into a path optimization model for processing to obtain a resource optimization path set; and performing optimal verification processing on the resource optimization path set to obtain optimal utilization path data. The scheme realizes dynamic optimal balance of sludge treatment economic benefits, carbon emission reduction and social compliance.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer information processing, in particular to a sludge resource utilization path determination method, device and equipment. BACKGROUND

[0002] With the acceleration of urbanization, the production of sludge continues to rise, and its resource utilization has become a key issue in the field of environmental protection. Traditional optimization methods focus on single economic or environmental objectives, using static weight distribution or linear programming to solve, which is difficult to adapt to complex scenarios of dynamic external environment (such as energy price fluctuations) and multi-objective conflicts.

[0003] The prior art often loses key information due to dimension collapse or modal split when processing multi-source heterogeneous data, for example, the synergistic effect of process parameters and policy texts is not effectively modeled. In addition, most models rely on historical data offline training, lack of real-time feedback and dynamic updating mechanism, and are prone to decision failure due to data distribution deviation or environmental mutation. It is particularly worth noting that the existing method lacks quantitative evaluation of policy, and it is difficult to map abstract policy clauses to computable constraints, resulting in a risk of violating the resource utilization path. The above defects restrict the sustainability and actual landing efficiency of sludge resource utilization decision, and urgent technical breakthroughs are needed through cross-modal data fusion, dynamic game optimization and closed-loop feedback mechanism. SUMMARY

[0004] The present application provides a sludge resource utilization path determination method, device and equipment, which solves the problem that economic, environmental and social objectives are difficult to dynamically cooperate and optimize in the current sludge resource utilization process.

[0005] To solve the above technical problems, the technical solutions of the present application are as follows:

[0006] The present application provides a sludge resource utilization path determination method, device and equipment, which solves the problem that economic, environmental and social objectives are difficult to dynamically cooperate and optimize in the current sludge resource utilization process.

[0007] Obtaining original data, the original data including at least one of sludge attribute data, process parameters and external dynamic data;

[0008] Pretreating the original data to obtain sludge sample data;

[0009] Classifying and integrating the sludge sample data to obtain a multi-dimensional tensor;

[0010] Decomposing and dynamically updating the multi-dimensional tensor to obtain a decomposed factor matrix;

[0011] Inputting the decomposed factor matrix into a path optimization model for processing to obtain a resource optimization path set;

[0012] Optimally verifying the resource optimization path set to obtain optimal utilization path data.

[0013] Optionally, the obtaining of the original data comprises:

[0014] The sludge attribute data is obtained through a sensor array or laboratory detection, and the sludge attribute data comprises at least one of an organic matter content, a heavy metal concentration, and a water content;

[0015] The process parameters are obtained through real-time recording of a sludge treatment equipment monitoring system, and the process parameters comprise at least one of a reaction temperature, a pH value, a residence time, and a unit energy consumption;

[0016] The numerical data in the external dynamic data is obtained through a preset database or a preset platform;

[0017] The text data in the external dynamic data is obtained through preset text information;

[0018] The spatiotemporal data in the external dynamic data is obtained through association of a geographic information system and a time stamp marker.

[0019] Optionally, the original data is preprocessed to obtain sludge sample data, comprising:

[0020] The sludge attribute data, the process parameters, and the numerical data are normalized to obtain normalized data;

[0021] The text data is vectorized to obtain a semantic vector;

[0022] The spatiotemporal data is missing filled to obtain filled spatiotemporal data;

[0023] The normalized data, the semantic vector, and the filled spatiotemporal data are aligned according to a preset dimension to obtain sludge sample data.

[0024] Optionally, the sludge sample data is classified and integrated to obtain a multi-dimensional tensor, comprising:

[0025] The sludge sample data is classified according to a preset dimension to obtain a plurality of classification data, and the preset dimension comprises at least one of a sample dimension, a feature dimension, an external factor dimension, a time dimension, and a space dimension;

[0026] The plurality of classification data is integrated to obtain a multi-dimensional tensor, and the multi-dimensional tensor is represented as T(i, j, k, l, m);

[0027] Where i, j, k, l, m are natural numbers, and T(i, j, k, l, m) is the observed value of the j-th feature of the i-th sample under the influence of the k-th type of external factors, in the l-th time period and the m-th region.

[0028] Optionally, the multidimensional tensor is decomposed and dynamically updated to obtain the decomposed factor matrix, including:

[0029] The multidimensional tensor data is decomposed to obtain a set of low-rank factor matrices.

[0030] The time dimension factor matrix in the set of low-rank factor matrices is dynamically updated to obtain the decomposed factor matrix.

[0031] Optionally, the path optimization model includes at least one of an economic agent, an environmental agent, and a social agent;

[0032] The state space of the economic agent includes at least one of real-time energy price, sludge calorific value, and tensor decomposition eigenvector; the action space of the economic agent includes at least one of pyrolysis, composting, and incineration actions; and the reward function of the economic agent is: R E =E p -α·(E h +T c ), where α is the cost penalty coefficient, R E E represents the reward value for the economic agent. p For expected profits, E h For the expected processing cost, T c For transportation costs;

[0033] The state space of the environmental agent includes at least one of real-time carbon emission intensity, carbon tax unit price, and historical average carbon emission; the action space of the environmental agent includes a carbon emission constraint threshold; the reward function of the environmental agent is: R Env =-β·(C e ×C t ), where β is the environmental penalty coefficient, and R Env C represents the reward value for the environmental agent. e For carbon emissions, C t This refers to the unit price of carbon tax.

[0034] The state space of the social agent includes at least one of semantic vectors and historical violation counts; the action space of the social agent includes a compliance score; the reward function of the social agent is: R S =γ·C s γ is the compliance reward coefficient, R S C represents the reward value for the social intelligent agent. s For compliance scoring.

[0035] Optionally, the resource optimization path set is subjected to optimal verification processing to obtain optimal utilization path data, including:

[0036] The resource optimization path set is extracted to obtain a first candidate solution and a second candidate solution;

[0037] The first candidate solution and the second candidate solution are subjected to a domination relationship determination according to a preset domination condition to determine the dominated solution set and the non-dominated solution set.

[0038] The non-dominated solution set is sorted according to the preset target priority to obtain the optimal utilization path data.

[0039] Optionally, the method for determining the sludge resource utilization pathway further includes:

[0040] Obtain actual processing results data;

[0041] The actual processing effect data and the optimal utilization path data are input into a loss function for processing to obtain the deviation data.

[0042] Based on the deviation data, the decomposed factor matrix is ​​updated to obtain the updated factor matrix;

[0043] The path optimization model is updated based on the updated factor matrix to obtain the updated path optimization model.

[0044] This invention also provides a sludge resource utilization path determination device, comprising:

[0045] The acquisition module is used to acquire raw data, which includes at least one of sludge attribute data, process parameters, and external dynamic data.

[0046] The processing module is used to preprocess the raw data to obtain sludge sample data; classify and integrate the sludge sample data to obtain a multidimensional tensor; decompose and dynamically update the multidimensional tensor to obtain a decomposed factor matrix; and input the decomposed factor matrix into a path optimization model for processing to obtain a resource optimization path set.

[0047] The determination module is used to perform optimal verification processing on the set of resource optimization paths to obtain optimal utilization path data.

[0048] This invention also provides a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when run by the processor, executes the above-described method.

[0049] The technical solution of the present invention has at least the following effects:

[0050] The above-mentioned solution of the present invention achieves a dynamic optimal balance between the economic benefits of sludge treatment, carbon emission reduction, and social compliance by acquiring raw data, including at least one of sludge attribute data, process parameters, and external dynamic data; preprocessing the raw data to obtain sludge sample data; classifying and integrating the sludge sample data to obtain a multidimensional tensor; decomposing and dynamically updating the multidimensional tensor to obtain a decomposed factor matrix; inputting the decomposed factor matrix into a path optimization model for processing to obtain a resource optimization path set; and performing optimal verification on the resource optimization path set to obtain optimal utilization path data. Attached Figure Description

[0051] Figure 1 This is a flowchart of the sludge resource utilization path determination method provided in the embodiments of the present invention;

[0052] Figure 2 This is a structural diagram of the sludge resource utilization path determination device provided in an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of the structure of the computing device provided in an embodiment of the present invention. Detailed Implementation

[0054] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0055] like Figure 1 As shown, an embodiment of the present invention proposes a method for determining the resource utilization path of sludge, including:

[0056] Step 11: Obtain raw data, which includes at least one of sludge attribute data, process parameters, and external dynamic data;

[0057] Step 12: Preprocess the raw data to obtain sludge sample data;

[0058] Step 13: Classify and integrate the sludge sample data to obtain a multidimensional tensor;

[0059] Step 14: Decompose and dynamically update the multidimensional tensor to obtain the decomposed factor matrix;

[0060] Step 15: Input the decomposed factor matrix into the path optimization model for processing to obtain a set of resource optimization paths;

[0061] Step 16: Perform optimal verification processing on the resource optimization path set to obtain optimal utilization path data.

[0062] In this embodiment, the raw data encompasses at least one of the following: sludge attribute data, process parameters, and external dynamic data. Sludge attribute data includes the physical, chemical, and biological properties of the sludge. This data helps determine suitable treatment and resource utilization methods. For example, sludge with high water content may require dewatering, while sludge with excessively high heavy metal content requires special treatment processes to avoid environmental pollution. Process parameters involve operational parameters at each stage of the sludge treatment process, such as temperature, pressure, reaction time, stirring speed, and reagent dosage. Different process parameters significantly impact the treatment effect and resource utilization efficiency of the sludge. For example, in the pyrolysis process of sludge, the control of temperature and reaction time directly affects the quality and yield of pyrolysis products. External dynamic data includes market price information, policy and regulatory requirements, and environmental factors. These external factors dynamically adjust with time and environmental changes, significantly impacting the feasibility and economics of sludge resource utilization. For example, when the market price of sludge treatment products rises, it may prompt companies to adopt more efficient resource utilization processes to increase output. Raw data can be obtained through various means, including on-site sampling and testing, online monitoring equipment, laboratory analysis, market research, and policy document collection. This ensures the accuracy, completeness, and timeliness of the data.

[0063] After obtaining the raw data, it needs to be preprocessed to obtain sludge sample data. The preprocessing process includes: data cleaning to remove noise, outliers, and duplicate data from the raw data; data standardization, as different types of data have different dimensions and value ranges, requiring standardization to eliminate the influence of dimensions; handling missing values, as missing values ​​in the raw data can lead to biases in the data analysis results, thus requiring the selection of appropriate processing methods based on the actual situation of the data; and data transformation, performing appropriate transformations of the data according to the needs of subsequent analysis.

[0064] The process of classifying and integrating sludge sample data includes data classification and data integration. Data classification refers to categorizing sludge sample data according to its characteristics and uses. For example, it can be classified according to the source of the sludge (e.g., domestic sludge, industrial sludge), the treatment stage (e.g., pretreatment, advanced treatment), or the direction of resource utilization (e.g., fertilizer production, brick making, power generation). The purpose of classification is to better organize and manage the data, facilitating subsequent analysis and processing. Data integration refers to combining the classified data to construct a multidimensional data structure. A multidimensional tensor is an effective mathematical tool for representing multidimensional data, containing information from multiple dimensions simultaneously. For example, a three-dimensional tensor can be constructed, where the three dimensions represent the source of the sludge, the treatment process, and the resource utilization product. Each element corresponds to relevant data on the resource utilization product obtained under that specific source and treatment process, such as yield and quality indicators. By integrating sludge sample data into a multidimensional tensor, the various factors and their interrelationships in the sludge resource utilization process can be described more comprehensively, providing richer information for subsequent decomposition and dynamic updating.

[0065] After determining the multidimensional tensor, an appropriate tensor decomposition algorithm is used to decompose it. The purpose of tensor decomposition is to break down the complex multidimensional tensor into a product of multiple low-dimensional factor matrices, each factor matrix corresponding to one dimension of the tensor. These factor matrices contain the characteristic information of the original data in different dimensions, which helps to deeply understand the key factors and influencing factors in the sludge resource utilization process. Since the sludge resource utilization process is dynamic, the original data will change continuously over time. Therefore, the decomposed factor matrices need to be dynamically updated to reflect the latest changes in the data. The dynamic updating method can be selected according to the actual situation to ensure that the processing results are always consistent with the actual situation, thereby improving the accuracy and reliability of the sludge resource utilization path.

[0066] Based on the objectives and constraints of sludge resource utilization, a suitable path optimization model is constructed. Common optimization objectives include maximizing resource utilization efficiency, minimizing treatment costs, and reducing environmental pollution. Constraints include technological feasibility, equipment capacity limitations, and environmental regulations. The path optimization model can be constructed using mathematical programming methods. Specifically, the decomposed factor matrix is ​​used as input data and substituted into the path optimization model for solution. Through iterative calculations of the optimization algorithm, the optimal set of resource utilization paths that satisfy the optimization objectives and constraints is found. For example, considering the objectives of maximizing resource utilization efficiency and minimizing treatment costs, the optimization model may recommend multiple different combinations of treatment processes and resource utilization schemes, forming a set of resource optimization paths. These path sets contain optimal or suboptimal solutions under different conditions, providing a scientific basis for sludge resource utilization decisions.

[0067] To determine the optimal utilization path from the set of resource optimization paths, appropriate validation indicators need to be selected. Validation indicators should be relevant to the optimization objectives and comprehensively reflect the advantages and disadvantages of resource utilization paths. For example, the yield, quality, economic benefits, and environmental benefits of resource-based products can be selected as validation indicators. Validation methods can include experimental validation, simulation validation, and case studies to validate each path in the set of resource optimization paths. Experimental validation can assess the feasibility and effectiveness of a path through actual small-scale, pilot-scale, or production trials; simulation validation can use computer simulation software to simulate the operation of the path and analyze its performance indicators; case studies can refer to existing similar project cases to compare the actual effects of different paths. By comprehensively using multiple validation methods, the validation indicators of each path are evaluated and compared. Based on the validation results, the path that performs best in terms of validation indicators is selected as the optimal utilization path. Relevant data on the optimal utilization path, such as treatment process parameters, resource-based product information, economic benefits, and environmental benefits, are organized and recorded to form optimal utilization path data. This data can provide detailed guidance for the actual implementation of sludge resource utilization projects, ensuring that the projects achieve their expected goals and effects.

[0068] The embodiments of the present invention construct a spatiotemporal correlation tensor by classifying and preprocessing multi-source data and fusing features; extract latent features based on sparse nonnegative decomposition and dynamically update the weights of external factors; design economic, environmental, and social agents to drive path selection, carbon emission constraints, and compliance scoring respectively, and generate a frontier solution set through an equilibrium strategy; finally, construct a multi-objective feedback loss function using actual processing effect data, and synchronously update the tensor model parameters and decision-making strategies to form a closed-loop adaptive optimization system of "data-model-decision". Combined with a multi-agent reinforcement learning collaborative game mechanism, it achieves multi-objective dynamic optimization of economic profit, carbon emissions, and policy compliance, and improves the robustness and sustainability of resource utilization paths in complex dynamic environments.

[0069] In an optional embodiment of the present invention, step 11 may include:

[0070] Step 111: Obtain the sludge property data through a sensor array or laboratory testing. The sludge property data includes at least one of organic matter content, heavy metal concentration, and moisture content.

[0071] Step 112: Obtain the process parameters through real-time recording of the sludge treatment equipment monitoring system. The process parameters include at least one of reaction temperature, pH value, residence time, and unit energy consumption.

[0072] Step 113: Obtain numerical data from the external dynamic data through a preset database or preset platform;

[0073] Step 114: Obtain text data from the external dynamic data using preset text information;

[0074] Step 115: Obtain spatiotemporal data from the external dynamic data by associating it with the geographic information system and timestamp markers.

[0075] In this embodiment, step 111 involves deploying multiple sensors at the sludge treatment site to monitor key physicochemical indicators of the sludge in real time, such as organic matter content, heavy metal concentration, and moisture content. Sensor data is transmitted to a data processing center wirelessly or via wired connection. For indicators that cannot be monitored in real time by sensors, or for which higher precision measurement results are required, sludge samples can be collected periodically and sent to a laboratory for testing. The laboratory test results serve as a supplement to or verification of the sensor data.

[0076] In step 112, the monitoring system equipped with the sludge treatment equipment records key process parameters in real time, such as reaction temperature, pH value, residence time, and unit energy consumption. The monitoring system transmits the recorded process parameters to the data processing center via sensors and data acquisition modules. The data acquisition frequency can be adjusted according to actual needs to ensure the real-time nature and accuracy of the data. The acquired process parameters are verified to ensure data integrity and consistency. Abnormal data or missing values ​​must be marked and processed to avoid affecting subsequent data analysis and optimization decisions.

[0077] In step 113, the first step is to establish or access a database containing economic and environmental parameters such as energy market prices and carbon tax unit prices. The database needs to be updated regularly to ensure the timeliness and accuracy of the data. Relevant numerical data can be obtained using pre-set platforms such as government public databases and industry trading platforms. These platforms typically provide data interfaces or APIs to facilitate automated data collection and integration. The numerical data obtained from the pre-set database or platform will be integrated with sludge attribute data and process parameters to form a complete dataset.

[0078] In step 114, relevant environmental protection policy and regulatory texts are first collected and organized to form a pre-defined text information database. This text information may come from government official websites, industry association publications, and other channels. The pre-defined text information is then parsed to extract key information, such as policy clauses, implementation dates, and scope of impact. Natural language processing technology can be used during the parsing process to improve the accuracy and efficiency of information extraction. The parsed textual data is then correlated with sludge attribute data, process parameters, and numerical data to form a multi-dimensional dataset. The logical and temporal relationships between the data must be considered during the correlation process.

[0079] In step 115, GIS technology is first used to geocode the sludge source area, recording the geographical location information of the sludge. The GIS system provides functions such as map display and spatial analysis of the sludge source area. Key events and data in the sludge treatment process are timestamped to record the time series information of the treatment stages. The timestamps can be used to analyze the temporal changes and patterns in the sludge treatment process. Then, the geocoding information provided by GIS is fused with the time series information marked by timestamps to form spatiotemporal data. The spatiotemporal data can be used to analyze the geographical distribution characteristics of the sludge source area and the temporal change patterns of the treatment process, providing strong support for optimizing the sludge resource utilization path.

[0080] In an optional embodiment of the present invention, step 12 may include:

[0081] Step 121: Normalize the sludge attribute data, the process parameters, and the numerical data to obtain normalized data;

[0082] Step 122: The text data is vectorized to obtain a semantic vector;

[0083] Step 123: Perform missing data filling on the spatiotemporal data to obtain the filled spatiotemporal data;

[0084] Step 124: Align the normalized data, the semantic vector, and the padded spatiotemporal data according to a preset dimension to obtain sludge sample data.

[0085] In this embodiment, step 121 is used to normalize sludge attribute data (such as organic matter content, heavy metal concentration, moisture content, etc.), process parameters (such as reaction temperature, pH value, residence time, unit energy consumption, etc.), and numerical data (such as energy market price, carbon tax unit price, etc.) from external dynamic data to eliminate dimensional differences and enable data to be compared and analyzed on the same scale. The normalization method uses the min-max normalization method to linearly map the original data to the [0,1] interval. For numerical data with missing values, linear interpolation is performed based on data from adjacent time periods of the same sludge sample. Specifically, if data for a certain time period is missing, linear interpolation is performed using valid data from adjacent time periods to estimate the missing value. The normalized data will have a uniform scale, facilitating subsequent feature fusion and model training.

[0086] In step 122, textual data (such as environmental policy and regulatory texts) from external dynamic data is vectorized, converting the textual data into numerical data that can be processed by a computer, i.e., semantic vectors. Specifically, the process includes: inputting the policy text into a pre-trained language model (such as BERT), and extracting the hidden layer output labeled [CLS] as the global semantic vector. Through the attention mechanism of the BERT model, keywords in the policy text (such as "prohibition of sludge landfill" and "preferential use of anaerobic digestion technology") are reinforced and encoded, thereby capturing the constraints of policy guidance on path selection. The resulting semantic vector reflects the semantic information of the policy text, providing policy constraints for subsequent multi-agent reinforcement learning.

[0087] In step 123, spatiotemporal data in the external dynamic data, such as the geocoding of the sludge source area and the time series information of the treatment stage, undergoes missing data imputation to address the data gap problem and ensure data integrity and continuity. The specific process includes: dynamic imputation using a sliding time window mechanism. The window size is defined (e.g., w=5, i.e., 5 time periods before and after), and the mean of the valid data within the window is calculated as the imputation value. For example, if the pH value at time stamp t is missing, it is filled based on the arithmetic mean of the non-missing values ​​from time period t-5 to t+5. During the imputation process, the spatiotemporal correlation of the data must be preserved to ensure that the imputed data reflects the spatiotemporal variation patterns of the sludge treatment process. The imputed spatiotemporal data will have complete spatiotemporal information, providing a foundation for subsequent multidimensional data tensor construction.

[0088] In step 124, the normalized sludge attribute data, process parameters, numerical external data, semantic vectors, and filled spatiotemporal data are aligned according to the preset dimensional structure. For example, the data for each sludge sample is organized into structured data containing sample number, feature vector, timestamp, spatial code, and external factor vector. The aligned data is then integrated to form a complete sludge sample dataset. The dataset should contain all necessary information for subsequent multidimensional data tensor construction and model training. The resulting sludge sample data will have a unified format and structure, facilitating subsequent data processing and analysis.

[0089] In an optional embodiment of the present invention, step 13 may include:

[0090] Step 131: Classify the sludge sample data according to a preset dimension to obtain multiple classification data; wherein, the preset dimension includes at least one of the following: sample dimension, feature dimension, external factor dimension, time dimension, and spatial dimension;

[0091] Step 132: Integrate the multiple classification data to obtain a multidimensional tensor, which is represented as: T(i, j, k, l, m);

[0092] Where i, j, k, l, m are natural numbers, and T(i, j, k, l, m) is the observed value of the j-th feature of the i-th sample under the influence of the k-th type of external factors, in the l-th time period and the m-th region.

[0093] In step 131 of this embodiment, the sample dimension refers to each sludge sample corresponding to a processing batch or record, and the sample dimension data identifies the source and processing batch of the sludge. For example, a batch of sludge produced by a wastewater treatment plant in a single day. The feature dimension includes sludge attribute data (such as organic matter content, heavy metal concentration, and moisture content) and process parameters (such as reaction temperature, pH value, residence time, and unit energy consumption). These features reflect the physicochemical properties of the sludge and key parameters in the treatment process. The external factor dimension refers to integrating dynamic external variables, including numerical data (such as energy market prices and carbon tax unit prices) and textual data (such as semantic vectors of environmental policy and regulatory texts). These external factors have a significant impact on the selection of sludge resource utilization pathways. The time dimension refers to dividing the process into multiple time slots according to the treatment stage, recording the continuous changes of various parameters during the sludge treatment process. Time dimension data helps to analyze the temporal evolution of the sludge treatment process. The spatial dimension refers to dividing the process according to the geographic coding of the sludge source area, recording the geographic location information of the sludge. Spatial dimension data helps to analyze the geographic distribution characteristics of the sludge source area. Based on the above five dimensions of classification, the sludge sample data is divided into multiple classification datasets, with each dataset corresponding to specific data under one dimension.

[0094] In step 132, a five-dimensional tensor is constructed based on the classified dataset. Each dimension of the tensor corresponds to one of the five dimensions: sample, feature, external factor, time, and space. Each element T(i, j, k, l, m) in the tensor represents an observation under specific sample (i), feature (j), external factor (k), time (l), and space (m) conditions. These observations may be sludge attribute values, process parameter values, external factor values, or combinations thereof. The classified dataset is then populated according to the tensor structure to form a complete multidimensional tensor. During the integration process, the accuracy and consistency of the data must be ensured to avoid data misalignment or omission.

[0095] Multidimensional tensors can simultaneously model the multidimensional interaction relationships of data in samples, time, space and external environment, revealing the intrinsic correlation between data; in addition, multidimensional tensors also provide high-fidelity feature inputs for subsequent multi-agent reinforcement learning game optimization, which helps to generate more scientific and sustainable sludge resource utilization paths.

[0096] In an optional embodiment of the present invention, step 14 may include:

[0097] Step 141: Decompose the multidimensional tensor data to obtain a set of low-rank factor matrices;

[0098] Step 142: Dynamically update the time dimension factor matrix in the set of low-rank factor matrices to obtain the decomposed factor matrix.

[0099] In step 141 of this embodiment, a tensor decomposition model with nonnegativity constraints and sparse regularization is used to decompose multidimensional tensors. Decomposed into five low-rank factor matrices A (1) A (2) A (3) t (4) and A (5) These correspond to the sample, feature, external factors, and time and space dimensions, respectively. Specifically, the decomposition objective function is defined as:

[0100]

[0101] in, The tensor reconstruction operator, ||·|| F Let be the Frobenius norm, and λ be the sparse regularization coefficient. It should be noted that the nonnegativity constraint (A...) (n) ≥0) Ensures the physical interpretability of the factor matrix, such as the time-dimensional factor matrix A. (4) Each column can be viewed as a weight representing the contribution of different time periods to the latent features; sparse regularization term ||A (n) ||1 is used to remove redundant features and improve the model's generalization ability.

[0102] In step 142, to adapt to real-time changes in the external environment (such as policy changes and energy price fluctuations), an incremental learning strategy is adopted to adjust the time dimension factor matrix A. (4) Perform dynamic updates. For example, at fixed time intervals Δt (e.g., 1 hour), perform the following operations:

[0103] (1) Data window sliding: Remove the data slice from the oldest time window (e.g., the lth time period). And add data slices for new time periods.

[0104] (2) Local gradient descent update: fix other factor matrices A (1) A (2) A (3) A (5) Only for A (4) After iterative optimization, the updated formula is as follows:

[0105]

[0106] Where η is the learning rate. This is the updated time window data. Understandably, this method significantly reduces the computational cost of full retraining while maintaining model stability.

[0107] It should be noted that the factor matrix obtained from the decomposition contains potential characteristics of the sludge resource utilization process. Specifically:

[0108] Sample dimension factor matrix A (1) Each row represents the distribution of a single sample across various latent features, which can be used for cluster analysis to process batches similarly.

[0109] Feature dimension factor matrix A (2) This study aims to reveal the implicit correlation between sludge properties and process parameters, such as the synergistic effect between heavy metal concentration and reaction temperature.

[0110] External Factor Dimension Factor Matrix A (3) : Quantify the impact weight of dynamic variables such as policies and energy prices on the processing path. For example, the feature activation value corresponding to the keyword "carbon emission reduction" in the policy semantic vector is relatively high.

[0111] Time dimension factor matrix A (4) This reflects the temporal evolution of potential characteristics during the processing, such as the increased preference for pyrolysis processes during periods of low electricity prices at night.

[0112] Spatial Dimension Factor Matrix A (5) This characterizes the impact of regional differences on treatment pathway selection; for example, incineration processes have lower weights in areas with high emission limits.

[0113] Through the tensor decomposition and dynamic update mechanism described above, this embodiment achieves efficient dimensionality reduction and feature extraction of multi-source heterogeneous data, providing high-fidelity feature input for subsequent multi-objective optimization.

[0114] In an optional embodiment of the present invention, step 15 includes:

[0115] Step 151: Input the decomposed factor matrix into the path optimization model for processing to obtain intermediate results;

[0116] Step 152: Filter the intermediate results to obtain a set of resource optimization paths;

[0117] In step 151 of this embodiment, the path optimization model includes an economic agent, an environmental agent, and a social agent, corresponding to the objectives of profit maximization, carbon emission minimization, and compliance maximization, respectively. It should be noted that each agent shares the same environmental state but makes independent decisions, achieving a globally optimal strategy combination through game theory interaction. In one possible implementation, the state space s of the economic agent... E include:

[0118] Real-time energy prices: Normalized electricity or natural gas prices obtained from external dynamic data;

[0119] Sludge calorific value: Calorific value per unit mass calculated based on sludge property data;

[0120] Tensor decomposition eigenvectors f NTF Step 14 decomposes the feature matrix to obtain potential features related to economic benefits, such as process correlation weights.

[0121] Specifically, its action space Defined as resource recovery path selection, it includes three discrete actions: pyrolysis, composting, and incineration. For example, the reward function R... E Designed as follows:

[0122] R E =E p -α·(E h +T c )

[0123] Where α is the cost penalty coefficient, determined by fitting historical data, and R E E represents the reward value for the economic agent. p For expected profits, E h For the expected processing cost, T c This represents transportation costs. Understandably, this function drives the agent to choose the high-yield path by positively incentivizing profit growth and negatively suppressing cost expenditures.

[0124] As an option, the state space s of the environmental agent Env include:

[0125] Real-time carbon emission intensity: Carbon emissions per unit of treatment volume calculated based on the type of treatment process and energy consumption;

[0126] Carbon tax unit price: Normalized carbon tax price obtained from external dynamic data;

[0127] Historical average carbon emissions: The average carbon emissions within a sliding time window (e.g., the last 24 hours).

[0128] Its action space Defined as the carbon emission constraint threshold a Env ∈[0,1], mapped to a specific carbon emission cap via the Sigmoid function. For example, the reward function R... Env Designed as follows:

[0129] R Env =-β·(C e ×C t );

[0130] Where β is the environmental penalty coefficient, R Env C represents the reward value for the environmental agent. e For carbon emissions, C t This is the unit price for carbon tax. It should be noted that the negative incentive mechanism forces agents to prioritize low-carbon emission processes, such as composting instead of incineration.

[0131] In one possible implementation, the state space s of the social intelligent agent S include:

[0132] Policy semantic vector v policy The policy text encoding generated by the BERT model in step 12;

[0133] Historical violation count: Statistics on the number of times the processing path has been penalized for violations recently.

[0134] Its action space Defined as compliance score a S ∈[0,1], and are transformed into compliance probabilities for different paths using the Softmax function. For example, the reward function R... S Designed as follows:

[0135] R S =γ·C s

[0136] Where γ is the compliance reward coefficient, R S C represents the reward value for the social intelligent agent. s This function scores compliance. Understandably, it ensures decisions comply with policy constraints by amplifying the reward signal for high-scoring paths.

[0137] In step 152, each agent updates its action value function based on the Q-learning algorithm and converges to a stable solution through an equilibrium strategy:

[0138] (1) Q-value function update:

[0139] Each agent selects action a based on its current state s. i Get reward R after interaction i And transition to the next state s ′ The updated formula is:

[0140]

[0141] Among them, Q i (s,a i ) indicates that agent i chooses action a in state s. i The value of R; i η is the immediate reward for agent i; δ is the learning rate; and s is the discount factor. ′ To perform action a i The next state after transition; a i ′ This indicates that agent i is in state s. ′ The actions that can be taken. It should be noted that... Indicates the agent's state s in the next state.′ The maximum expected cumulative reward.

[0142] (2) Equilibrium strategy convergence:

[0143] Through iterative game theory, the strategies of each agent are combined. The equilibrium condition is met:

[0144]

[0145] in, Represent the action spaces of the economic intelligent agent, the environmental intelligent agent, and the social intelligent agent, respectively; a E a Env a S These represent any candidate actions of the economic agent, environmental agent, and social agent in their action space, respectively. These represent the optimal actions of the economic agent, environmental agent, and social agent under the equilibrium strategy, respectively.

[0146] For example, when an economic agent chooses the incineration path, an environmental agent may raise the carbon emission constraint threshold to offset its environmental costs, while a social agent dynamically adjusts the compliance score weights based on policy vectors.

[0147] Through the aforementioned multi-agent game mechanism, this embodiment achieves dynamic trade-offs and conflict resolution among multiple objectives, resulting in a set of resource optimization paths, denoted as... The process includes:

[0148] Input a graph structure (nodes represent positions, edges represent connections) and a start point and an end point;

[0149] Generate an initial path from the starting point, such as a single-node path, and store it in the open list;

[0150] Iteratively select paths from the open list (e.g., prioritizing the lowest cost), expand the adjacent nodes of their terminal nodes, and generate new path branches;

[0151] The cost of the new path is evaluated using a formula:

[0152] f(p) = g(p) + h(p)

[0153] Where g(p) is the actual cost from the starting point to the current node, such as distance; h(p) is a heuristic function, such as the estimated value of the Euclidean distance to the endpoint.

[0154] If the expanded path does not reach the destination, it is added back to the open list as a candidate path; if it has reached the destination, it is output as a feasible path.

[0155] Filtering and Termination: Invalid paths are removed through collision detection, such as grid map obstacle verification; the algorithm continues to iterate until the termination condition is met, such as finding N paths, the open list being empty, or resources being exhausted.

[0156] Furthermore, the present invention also provides steps for constructing the above-mentioned framework, including:

[0157] (1) Framework initialization and parameter configuration

[0158] Alternatively, the construction of a multi-agent framework begins with the following initialization operations:

[0159] Economic intelligent agent: Initialize the Q-value table with dimension |s E |×3(state space size×number of actions), where the state space elements include real-time energy price, sludge calorific value and tensor decomposition feature vector as mentioned above, and the actions are three types of path selection: pyrolysis, composting and incineration.

[0160] Environmental intelligent agent: It adopts a neural network to approximate the continuous action space (carbon emission constraint threshold). The dimension of the network input layer matches the state space (carbon emission intensity, carbon tax unit price, historical carbon emission average), and the output layer is a single neuron representing the threshold.

[0161] Social intelligent agent: Initializes the compliance score generator, which maps policy semantic vectors and historical violation records to 0-1 score values ​​through a fully connected layer.

[0162] It should be noted that the learning rate η and the discount factor δ are set uniformly before training.

[0163] (2) Interaction mechanism and action execution

[0164] In one possible implementation, the action execution flow of each agent is as follows:

[0165] Action selection: The economic agent selects a path based on an ∈-greedy strategy (e.g., ∈ = 0.1 exploration probability); the environmental agent outputs a carbon emission threshold through a neural network; and the social agent generates a compliance score.

[0166] Environment simulator feedback: Combining actions (a E ,a Env ,a S Input the simulator and calculate multi-objective feedback:

[0167] Economic feedback: If path a E Carbon emissions exceeded threshold a env If so, the profit will be deducted according to the excess ratio;

[0168] Compliance verification: If the score is a S If the value is below a preset threshold, a violation will be marked and a penalty will be triggered.

[0169] (3) Model update and equilibrium convergence

[0170] Specifically, the update logic for each agent is designed differently:

[0171] Economic agent: The discrete action value is updated using table Q learning. The update formula is as shown above, but it is limited to path selection actions.

[0172] Environmental and social intelligent agents: The parameters of the neural network are updated using gradient descent, and the loss function is the mean square error between the predicted action value and the target value.

[0173] It should be noted that the equilibrium convergence condition is verified through periodic policy evaluation. When the policy change rate of each agent is less than 1% in 10 consecutive iterations, it is determined to be converged.

[0174] In an optional embodiment of the present invention, step 16 may include:

[0175] Step 161: Extract the resource optimization path set to obtain the first candidate solution and the second candidate solution;

[0176] Step 162: Determine the dominance relationship between the first candidate solution and the second candidate solution according to the preset dominance conditions, and determine the dominated solution set and the non-dominated solution set;

[0177] Step 163: Sort the non-dominated solution set according to the preset target priority to obtain the optimal utilization path data.

[0178] In step 161 of this embodiment, the candidate path set is denoted as... Each solution It contains three objective function values, among which, To maximize the profit of the k-th path, its calculation is based on the difference between processing revenue and cost.

[0179] The carbon emissions of the k-th path need to be minimized, which is obtained by multiplying the process energy consumption and the carbon emission coefficient.

[0180] The compliance score for the k-th path needs to be maximized and is dynamically generated by the social agent based on the policy semantic vector.

[0181] For example, if the profit of the pyrolysis pathway is 100,000 yuan, the carbon emission is 500 kg, and the compliance score is 0.9, then its solution vector is p. k = (10, 500, 0.9).

[0182] Extracting from the set of resource optimization paths yields any two solutions p. iWith p j That is, the first candidate solution and the second candidate solution;

[0183] In step 162, for any two solutions p i With p j If the following preset dominance conditions are met:

[0184] (1) Economic objectives are no worse than:

[0185] (2) Environmental targets are no worse than:

[0186] (3) Social goals are no worse than:

[0187] (4) At least one objective is strictly better than: at least one of the above three inequalities is a strict inequality (> or <); then p is called i Dominate p j (recorded as) ), and remove the dominated solution p j .

[0188] For example, if the solutions p1 = (12, 600, 0.8) and p2 = (10, 500, 0.9):

[0189] Economic objective: 12 ≥ 10 (p1 is better);

[0190] Environmental target: 600 ≥ 500 (p2 is better);

[0191] Social objective: 0.8 ≤ 0.9 (p2 is better).

[0192] Since p1 and p2 do not dominate each other, both are retained.

[0193] As an alternative, frontier solution set Defined as the set of all candidate solutions that are not dominated by any other solution, mathematically expressed as:

[0194]

[0195] in, This indicates "does not exist", meaning for solution p k If there is no other solution p m Dominate p k Then p k This belongs to the Pareto front. Understandably, solutions in this set cannot further optimize a particular objective without harming others, representing the optimal trade-off surface in multi-objective optimization. Through the above process, the dominated solution set p is determined. k Non-dominated solution set p k ;

[0196] In step 163, the frontier solution set is sorted according to a preset objective priority and then output. For example, if the priority is economic > environmental > social, then the preset objective priority is:

[0197] (1) First priority: Press f econ Sort in descending order;

[0198] (2) Second priority: for f econ For the same solution, press f env Sort in ascending order;

[0199] (3) Third priority: for f econ with f env All solutions are identical, according to f soc Sort in descending order.

[0200] It should be noted that this ranking mechanism allows decision-makers to quickly identify the optimal solution based on actual needs (such as policy inclinations or short-term benefit objectives) without having to traverse the entire solution set.

[0201] Through the above process, this embodiment achieves the scientific screening and visualization output of multi-objective optimization solution sets, and obtains the optimal utilization path data.

[0202] In an optional embodiment of the present invention, the method for determining the sludge resource utilization path further includes:

[0203] Step 171: Obtain actual processing effect data;

[0204] Step 172: Input the actual processing effect data and the optimal utilization path data into the loss function for processing to obtain the deviation data;

[0205] Step 173: Based on the deviation data, update the decomposed factor matrix to obtain the updated factor matrix;

[0206] Step 174: Update the path optimization model according to the updated factor matrix to obtain the updated path optimization model.

[0207] In this embodiment, the real-time feedback data update step is used to achieve dynamic closed-loop optimization of the model and policy. Its core lies in using actual processing effect data to correct prediction biases, ensuring the system's continuous adaptability to dynamic environments. Specifically, this step forms an iterative enhancement loop of "data-model-decision" through multi-objective loss function construction, gradient-driven parameter updates, and policy synchronization mechanisms.

[0208] In step 171, the actual processing effect data includes the actual profit R. real Actual carbon emissions R realand actual compliance score C real These correspond to the results of achieving economic, environmental, and social goals, respectively. It should be noted that the data sources include:

[0209] Economic data; obtain energy recovery revenue, processing costs, and transportation costs for the processing path from the financial system, and calculate net profit;

[0210] Environmental data; real-time monitoring of process energy consumption through IoT sensors, combined with carbon emission coefficients to calculate total carbon emissions;

[0211] Compliance data; compliance audit results from regulatory authorities are converted into a standardized score of 0-1.

[0212] In step 172, a weighted mean squared error loss function is constructed to quantify the model's predicted value (R0). pred E pred C pred Deviation from actual value:

[0213]

[0214] Among them, R pred E pred C pred These represent the model's predicted profit, carbon emissions, and compliance score, respectively; α1, α2, and α3 are the feedback weights for each objective. It should be noted that the weighting coefficients are dynamically adjusted based on business priorities; for example, α2 is increased to strengthen carbon emission control when environmental policies tighten.

[0215] In step 173, the five factor matrices A of the dynamic tensor decomposition model are updated using the gradient descent method. (1) ,…,A (5) The updated formula is:

[0216]

[0217] Where, η fb To provide feedback on the learning rate, its range is limited to 0.001-0.1 to prevent gradient explosion. It can be understood that the update direction of the factor matrix is ​​determined by the partial derivatives of the loss function with respect to each matrix, thereby minimizing prediction bias. For example, the time dimension factor matrix A... (4) The update will improve the model's ability to predict the time-series evolution of process parameters.

[0218] In step 174, the updated factor matrix is ​​input into the path optimization model, triggering policy retraining:

[0219] (1) Eigenvector reconstruction: based on the new factor matrix A (1) ,…,A (5) Recalculate the tensor eigenvector fNTF Replace the original state input of the economic intelligent agent;

[0220] (2) Q-value table reset: 80% of the historical Q-values ​​are retained as prior knowledge, and the remaining 20% ​​are re-initialized according to the new feature distribution to balance experience inheritance and exploration ability;

[0221] (3) Strategy weight adjustment: The parameters of the carbon emission constraint threshold generation network of the environmental agent and the compliance scoring network of the social agent are updated synchronously to adapt to the latest tensor features.

[0222] It should be noted that this synchronization mechanism avoids policy oscillations caused by directly resetting the agent, thus ensuring the consistency of decision-making.

[0223] The above process enables the path optimization model to be updated.

[0224] like Figure 2 As shown, this embodiment of the invention also provides a sludge resource utilization path determination device 20, comprising:

[0225] Acquisition module 21 is used to acquire raw data, which includes at least one of sludge attribute data, process parameters, and external dynamic data;

[0226] Processing module 22 is used to preprocess the raw data to obtain sludge sample data; classify and integrate the sludge sample data to obtain a multidimensional tensor; decompose and dynamically update the multidimensional tensor to obtain a decomposed factor matrix; and input the decomposed factor matrix into a path optimization model for processing to obtain a resource optimization path set.

[0227] The determination module 23 is used to perform optimal verification processing on the resource optimization path set to obtain optimal utilization path data.

[0228] Optionally, module 21 is specifically used for:

[0229] The sludge property data is obtained through sensor arrays or laboratory testing, and the sludge property data includes at least one of organic matter content, heavy metal concentration, and moisture content.

[0230] The process parameters are obtained through real-time recording by the sludge treatment equipment monitoring system. The process parameters include at least one of reaction temperature, pH value, residence time, and unit energy consumption.

[0231] Numerical data from the external dynamic data is obtained through a preset database or preset platform;

[0232] By using preset text information, text-type data can be obtained from the external dynamic data;

[0233] Spatiotemporal data from the external dynamic data is obtained by associating it with a geographic information system and a timestamp marker.

[0234] Optionally, processing module 22 is specifically used for:

[0235] The sludge attribute data, the process parameters, and the numerical data are normalized to obtain normalized data.

[0236] The text data is vectorized to obtain semantic vectors;

[0237] The spatiotemporal data is filled with missing data to obtain the filled spatiotemporal data.

[0238] The normalized data, the semantic vector, and the padded spatiotemporal data are aligned according to a preset dimension to obtain sludge sample data.

[0239] Optionally, the processing module 22 is also specifically used for:

[0240] The sludge sample data is classified according to preset dimensions to obtain multiple classification data; wherein, the preset dimensions include at least one of sample dimension, feature dimension, external factor dimension, time dimension and spatial dimension;

[0241] The multiple classification data are integrated to obtain a multidimensional tensor, which is represented as T(i, j, k, l, m).

[0242] Where i, j, k, l, m are natural numbers, and T(i, j, k, l, m) is the observed value of the j-th feature of the i-th sample under the influence of the k-th type of external factors, in the l-th time period and the m-th region.

[0243] Optionally, the processing module 22 is also specifically used for:

[0244] The multidimensional tensor data is decomposed to obtain a set of low-rank factor matrices.

[0245] The time dimension factor matrix in the set of low-rank factor matrices is dynamically updated to obtain the decomposed factor matrix.

[0246] Optionally, the path optimization model includes at least one of an economic agent, an environmental agent, and a social agent;

[0247] The state space of the economic agent includes at least one of real-time energy price, sludge calorific value, and tensor decomposition eigenvector; the action space of the economic agent includes at least one of pyrolysis, composting, and incineration actions; and the reward function of the economic agent is: R E=E p -α·(E h +T c ), where α is the cost penalty coefficient, R E E represents the reward value for the economic agent. p For expected profits, E h For the expected processing cost, T c For transportation costs;

[0248] The state space of the environmental agent includes at least one of real-time carbon emission intensity, carbon tax unit price, and historical average carbon emission; the action space of the environmental agent includes a carbon emission constraint threshold; the reward function of the environmental agent is: R Env =-β·(C e ×C t ), where β is the environmental penalty coefficient, and R Env C represents the reward value for the environmental agent. e For carbon emissions, C t This refers to the unit price of carbon tax.

[0249] The state space of the social agent includes at least one of semantic vectors and historical violation counts; the action space of the social agent includes a compliance score; the reward function of the social agent is: R S = γ·Compliance Score, where γ is the compliance reward coefficient, R S C represents the reward value for the social intelligent agent. s For compliance scoring.

[0250] Optionally, module 23 is specifically used for:

[0251] The resource optimization path set is extracted to obtain a first candidate solution and a second candidate solution;

[0252] The first candidate solution and the second candidate solution are subjected to a domination relationship determination according to a preset domination condition to determine the dominated solution set and the non-dominated solution set.

[0253] The non-dominated solution set is sorted according to the preset target priority to obtain the optimal utilization path data.

[0254] Optionally, the sludge resource utilization path determination device 20 further includes:

[0255] The update module 24 is used to acquire actual processing effect data; input the actual processing effect data and the optimal utilization path data into a loss function for processing to obtain deviation data; update the decomposed factor matrix according to the deviation data to obtain an updated factor matrix; and update the path optimization model according to the updated factor matrix to obtain an updated path optimization model.

[0256] It should be noted that this device is a device corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0257] like Figure 3 As shown, this embodiment of the invention also provides a computing device 30, including a processor 31, a memory 32, and a program or instructions stored in the memory 32 and executable on the processor 31. When the program or instructions are executed by the processor 31, they implement the various processes of the above-described sludge resource utilization path determination method embodiment and achieve the same technical effects. To avoid repetition, they will not be described again here. It should be noted that the computing device in this embodiment of the invention includes the above-described mobile electronic devices and non-mobile electronic devices.

[0258] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein 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 implementations should not be considered beyond the scope of this invention.

[0259] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0260] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0261] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0262] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0263] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0264] Furthermore, it should be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above-described series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0265] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing device. The computing device can be a known general-purpose device. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code for implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps for performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0266] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for determining the resource utilization pathway of sludge, characterized in that, include: Acquire raw data, which includes at least one of sludge property data, process parameters, and external dynamic data; The raw data is preprocessed to obtain sludge sample data; The sludge sample data is classified and integrated to obtain a multidimensional tensor; The multidimensional tensor is decomposed and dynamically updated to obtain the decomposed factor matrix; The decomposed factor matrix is ​​input into the path optimization model for processing to obtain a set of resource optimization paths; The optimal resource optimization path set is subjected to optimal verification processing to obtain optimal utilization path data; The acquisition of raw data includes: The sludge property data is obtained through sensor arrays or laboratory testing, and the sludge property data includes at least one of organic matter content, heavy metal concentration, and moisture content. The process parameters are obtained through real-time recording by the sludge treatment equipment monitoring system. The process parameters include at least one of reaction temperature, pH value, residence time, and unit energy consumption. Numerical data from the external dynamic data is obtained through a preset database or preset platform; By using preset text information, text-type data can be obtained from the external dynamic data; And by associating the data with timestamps through a geographic information system, spatiotemporal data can be obtained from the external dynamic data; The path optimization model includes an economic agent, an environmental agent, and a social agent; wherein, the state space of the economic agent includes at least one of real-time energy price, sludge calorific value, and tensor decomposition eigenvectors; the action space of the economic agent includes at least one of pyrolysis action, composting action, and incineration action; and the reward function of the economic agent is: R E = E p - α ·( E h + T c ), α This is the cost penalty coefficient. R E The reward value for the economic intelligent agent. E p For expected profits, E h For the expected processing cost, T c For transportation costs; The state space of the environmental agent includes at least one of real-time carbon emission intensity, carbon tax unit price, and historical average carbon emission; the action space of the environmental agent includes a carbon emission constraint threshold; the reward function of the environmental agent is: R Env =- β ·( C e × C t ), β This is the environmental penalty coefficient. R Env The reward value for the environmental intelligent agent. C e For carbon emissions, C t This refers to the unit price of carbon tax. The state space of the social agent includes at least one of semantic vectors and historical violation counts; the action space of the social agent includes a compliance score; the reward function of the social agent is: R S = γ · C s , γ For compliance reward coefficient, R S The reward value for the social intelligent agent. C s For compliance scoring.

2. The method for determining the resource utilization path of sludge according to claim 1, characterized in that, The raw data is preprocessed to obtain sludge sample data, including: The sludge attribute data, the process parameters, and the numerical data are normalized to obtain normalized data. The text data is vectorized to obtain semantic vectors; The spatiotemporal data is filled with missing data to obtain the filled spatiotemporal data. The normalized data, the semantic vector, and the padded spatiotemporal data are aligned according to a preset dimension to obtain sludge sample data.

3. The method for determining the resource utilization path of sludge according to claim 1, characterized in that, The sludge sample data is classified and integrated to obtain a multidimensional tensor, including: The sludge sample data is classified according to preset dimensions to obtain multiple classification data; wherein, the preset dimensions include at least one of sample dimension, feature dimension, external factor dimension, time dimension and spatial dimension; The multiple categorical data are integrated to obtain a multidimensional tensor, which is represented as follows: T ( i , j , k , l , m ); in, i , j , k , l , m For natural numbers, T ( i , j , k , l , m ) is the first i The first sample j Item feature in the first k Under the influence of external factors, the first l The time period and the first m Observations for each region.

4. The method for determining the resource utilization path of sludge according to claim 1, characterized in that, The multidimensional tensor is decomposed and dynamically updated to obtain the decomposed factor matrix, including: The multidimensional tensor data is decomposed to obtain a set of low-rank factor matrices. The time dimension factor matrix in the set of low-rank factor matrices is dynamically updated to obtain the decomposed factor matrix.

5. The method for determining the resource utilization path of sludge according to claim 1, characterized in that, The optimal resource optimization path set is subjected to optimal verification processing to obtain optimal utilization path data, including: The resource optimization path set is extracted to obtain a first candidate solution and a second candidate solution; The first candidate solution and the second candidate solution are subjected to a domination relationship determination according to a preset domination condition to determine the dominated solution set and the non-dominated solution set. The non-dominated solution set is sorted according to the preset target priority to obtain the optimal utilization path data.

6. The method for determining the resource utilization path of sludge according to claim 1, characterized in that, Also includes: Obtain actual processing results data; The actual processing effect data and the optimal utilization path data are input into a loss function for processing to obtain the deviation data. Based on the deviation data, the decomposed factor matrix is ​​updated to obtain the updated factor matrix; The path optimization model is updated based on the updated factor matrix to obtain the updated path optimization model.

7. A device for determining the path of sludge resource utilization, characterized in that, include: The acquisition module is used to acquire raw data, which includes at least one of sludge attribute data, process parameters, and external dynamic data. The processing module is used to preprocess the raw data to obtain sludge sample data; The sludge sample data is classified and integrated to obtain a multidimensional tensor; The multidimensional tensor is decomposed and dynamically updated to obtain the decomposed factor matrix; the decomposed factor matrix is ​​then input into the path optimization model for processing to obtain a set of resource optimization paths. The determination module is used to perform optimal verification processing on the resource optimization path set to obtain optimal utilization path data; The acquisition of raw data includes: The sludge property data is obtained through sensor arrays or laboratory testing, and the sludge property data includes at least one of organic matter content, heavy metal concentration, and moisture content. The process parameters are obtained through real-time recording by the sludge treatment equipment monitoring system. The process parameters include at least one of reaction temperature, pH value, residence time, and unit energy consumption. Numerical data from the external dynamic data is obtained through a preset database or preset platform; By using preset text information, text-type data can be obtained from the external dynamic data; And by associating the data with timestamps through a geographic information system, spatiotemporal data can be obtained from the external dynamic data; The path optimization model includes an economic agent, an environmental agent, and a social agent; wherein, the state space of the economic agent includes at least one of real-time energy price, sludge calorific value, and tensor decomposition eigenvectors; the action space of the economic agent includes at least one of pyrolysis action, composting action, and incineration action; and the reward function of the economic agent is: R E = E p - α ·( E h + T c ), α This is the cost penalty coefficient. R E The reward value for the economic intelligent agent. E p For expected profits, E h For the expected processing cost, T c For transportation costs; The state space of the environmental agent includes at least one of real-time carbon emission intensity, carbon tax unit price, and historical average carbon emission; the action space of the environmental agent includes a carbon emission constraint threshold; the reward function of the environmental agent is: R Env =- β ·( C e × C t ), β This is the environmental penalty coefficient. R Env The reward value for the environmental intelligent agent. C e For carbon emissions, C t This refers to the unit price of carbon tax. The state space of the social agent includes at least one of semantic vectors and historical violation counts; the action space of the social agent includes a compliance score; the reward function of the social agent is: R S = γ · C s , γ For compliance reward coefficient, R S The reward value for the social intelligent agent. C s For compliance scoring.

8. A computing device, characterized in that, include: A processor, a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method as described in any one of claims 1 to 6.

Images