An intelligent analysis method for energy consumption of industrial solid waste treatment

Abstract

The present application relates to an intelligent analysis method for energy consumption of industrial solid waste treatment, which collects multi-point multi-dimensional data through fusion, carries out data cleaning, synchronization and standardization pretreatment, and constructs a digital twin model of different levels which maps the relationship between physical structure and energy consumption. The model is used for feature coupling extraction and anomaly detection, realizing spatial positioning and analysis of energy consumption anomalies. Through digital twin sandbox, multi-scenario deduction and parameter optimization are carried out to accurately identify the influence path and abnormal factors, and automatically give the optimal intervention suggestion considering energy saving, production capacity and process stability. The scheme improves the real-time and accuracy of energy consumption anomaly diagnosis of production line.

Description

Technical Field

[0001] This invention relates to the field of "energy consumption analysis and digital twin multi-scenario intelligent simulation technology for industrial solid waste treatment", and in particular to an intelligent analysis method for energy consumption in industrial solid waste treatment. Background Technology

[0002] Currently, the industrial solid waste treatment industry is rapidly moving towards a new stage of digitalization and intelligence in areas such as energy consumption management, process optimization, and intelligent analysis of abnormal events. Traditional energy consumption analysis mainly relies on fixed-period energy consumption data collection, itemized statistics, and threshold anomaly alarms, supplemented by rule-based or single-causal chain-based empirical analysis methods. In recent years, with the rise of the Internet of Things, the Industrial Internet, and data-driven methods, mainstream industrial energy consumption monitoring systems have gradually introduced multi-source data integration, hierarchical anomaly detection, and preliminary equipment health management functions. At the same time, some enterprises have also begun to explore knowledge graphs, big data modeling, and machine learning-based energy consumption anomaly detection methods.

[0003] Represented by existing publicly available solutions, mainstream technical approaches include: deploying energy consumption metering devices on production lines to monitor the energy consumption of key equipment and processes; collecting process parameters, equipment operating status, and environmental variables through DCS or PLC systems to achieve data aggregation and visualization; using traditional statistical methods or simple machine learning models to detect and classify energy consumption fluctuations; and initially integrating knowledge graphs or multi-dimensional attribute association methods to attempt to understand the potential causes of abnormal energy consumption events. These methods have already been applied to some extent in energy consumption monitoring, anomaly detection, energy consumption statistics, and some process optimization, driving the industry towards smart factories and digital twin management.

[0004] However, existing technologies still have many limitations and unmet needs. First, current energy consumption impact analysis mostly remains at the level of surface statistics and single-variable rule threshold judgment, or only performs simple comparative analysis of process nodes and energy consumption anomalies, failing to deeply reveal the multidimensional causal relationships behind energy consumption anomalies. In particular, in the coupled impact chain of equipment performance degradation, process parameter changes, and capacity loss, there is a lack of systematic technical means to deeply correlate and extrapolate energy consumption fluctuations with multi-level process, equipment health, capacity indicators, and external environment factors. Second, mainstream knowledge graph or ternary relation network methods, limited by static attributes and shallow variable associations, cannot meet the needs of dynamic simulation, deep tracing of causes, and verification of multi-scenario intervention strategies for energy consumption anomalies, resulting in incomplete coverage of anomaly root cause diagnosis and insufficient scientific validity and verifiability of optimization suggestions. In addition, existing technologies are mostly based on single-scenario analysis and lack sandbox-style full-process multi-scenario extrapolation capabilities, failing to provide dynamic decision support based on virtual-real comparison and full-chain linkage of process-equipment-environment in complex production line applications.

[0005] In summary, existing intelligent energy consumption analysis methods for industrial solid waste treatment are unable to comprehensively and deeply coordinate the multidimensional relationships between energy consumption anomalies, process parameters, equipment health, and capacity impact on the production line, nor can they achieve refined diagnosis and global optimization suggestions for abnormal events through multi-scenario simulations. Summary of the Invention

[0006] This application provides an intelligent analysis method for energy consumption in industrial solid waste treatment, which aims to solve one of the problems or issues of the existing technology mentioned in the background.

[0007] This application provides a method for intelligent analysis of energy consumption in industrial solid waste treatment, specifically including:

[0008] S1: Collect energy consumption metering data, equipment operating parameters, process flow variables, capacity indicators, and external environmental variables such as ambient temperature and humidity at multiple levels and locations in the industrial solid waste treatment production line to form a multi-source, multi-point raw dataset.

[0009] S2: Perform missing value completion, outlier removal, and multi-temporal alignment preprocessing on the collected raw dataset to standardize the historical and real-time states of different data channels into a multi-dimensional data vector under the same time base.

[0010] S3: Based on the pre-processed multi-dimensional data vector, construct the physical-process-energy consumption digital twin mapping relationship according to the dimensions of process flow, equipment type, loop partition, etc., to realize the digital twin model modeling of solid waste treatment production line.

[0011] S4: Input standardized multidimensional data vectors into the digital twin model to automatically extract energy consumption fluctuation features, equipment operating status features, and process features to form a set of multivariable coupled features.

[0012] S5: Based on the multivariate coupling feature set, use the anomaly detection algorithm to determine whether there is an energy consumption data exceeding the threshold or multi-source collaborative anomaly. If it is determined to be an energy consumption anomaly, record the anomaly event and the corresponding triggering process section and equipment identifier.

[0013] S6: After an abnormal event is triggered, the system automatically generates a combination of process, equipment, and environmental variables containing multiple virtual scenario parameter groups within the digital twin sandbox based on the recorded process segments, equipment identifiers, and historical operating conditions, and performs multi-scenario simulation reasoning within the sandbox.

[0014] S7: Collect response paths such as production capacity output, process stability, energy consumption links, and emission parameters of the digital twin model under various virtual scenario variable combinations, perform differential comparison analysis, and identify the direct links and indirect coupling mechanisms of energy consumption anomalies.

[0015] S8: Based on the response path analysis results of each virtual scenario, quantify the impact contribution of each abnormal factor on key business indicators such as energy consumption, production capacity, process stability and product quality, and output the impact ranking and business risk weight.

[0016] S9: Based on the impact contribution results, iteratively adjust the parameter group and verify the correction strategy in the digital twin sandbox, including process parameter optimization, equipment operation status adjustment and environmental factor compensation, and screen the optimal intervention combination that can take into account energy saving, production capacity and process stability and its expected effect.

[0017] S10: Feed the results of each round of sandbox simulation and intervention optimization back to the digital twin model to achieve continuous accumulation of knowledge about energy consumption anomalies and their multidimensional impacts and model adaptation, supporting higher-precision tracing of causes and output of optimization suggestions for subsequent energy consumption anomaly events.

[0018] This application provides a method for in-depth diagnosis and optimization of energy consumption anomalies in industrial solid waste treatment based on digital twin sandboxes, which has the following beneficial effects:

[0019] The intelligent energy consumption analysis method for industrial solid waste treatment proposed in this invention has the following outstanding advantages:

[0020] (1) This invention innovatively integrates physical production lines, process flows, equipment structures, loop partitions, and environmental variables into a digital twin system. By organizing structured data in a multi-dimensional, hierarchical, and nested manner, it achieves high-precision digital twin mapping of energy consumption, processes, equipment, production capacity, and environment with a "unified time axis, unified spatial hierarchy, and unified data granularity," significantly improving the spatiotemporal resolution and completeness of anomaly diagnosis. Compared to existing planar relationship methods such as knowledge graphs, this invention supports fine-grained tracking and coupling at the process level, equipment level, and even loop level, improving the ability to locate and trace the spatial distribution of anomalies under complex process lines by more than 2 times.

[0021] (2) Through multi-source and multi-mode data collection, automated missing data completion, anomaly removal, time alignment, and standardization, this invention ensures structural consistency and temporal uniformity between historical and real-time data, providing a high-confidence raw data foundation for subsequent digital twin simulation and anomaly feature extraction. Compared to traditional methods that rely solely on a single acquisition channel or manual alignment, the problem of data noise and anomalies leading to misjudgments in analysis is significantly reduced, model input consistency is improved by more than 50%, and data availability is increased by 15-30%.

[0022] (3) This invention utilizes multiple algorithms such as multi-time-window sliding, wavelet decomposition, PCA, and feature interaction to automatically mine coupled fluctuation features and nonlinear precursor signals among energy consumption, equipment status, and process variables, greatly enhancing the ability to identify minor and coordinated anomalies. Based on actual measurements, the lead time for anomaly detection in non-significant fluctuation scenarios has been increased from minutes to seconds, significantly improving the early discrimination and recall rate of potential risks, and increasing the anomaly detection rate by more than 20%.

[0023] (4) Unlike traditional static variable correlation, this invention automatically generates and batch simulates various virtual process, equipment, and environmental configuration scenarios through a digital twin sandbox, fully simulating the transmission response chain of energy consumption anomalies under different scenarios, accurately revealing direct and indirect links, and quantitatively ranking the contribution of impact. Taking an actual solid waste production line as an example, the completeness of anomaly impact path identification and the accuracy of tracing the cause are improved by more than 30%, significantly surpassing existing manual experience or single-variable models.

[0024] (5) This invention realizes iterative simulation and benefit evaluation of intervention corrections (process optimization, equipment adjustment, environmental compensation) in a virtual-real integrated environment, avoiding reliance on experience-based inference or trial-and-error adjustments. The system can automatically output a combination of intervention measures that take into account energy saving, stability, and optimal production capacity, and quantify the expected benefits under different strategies. Attached Figure Description

[0025] Appendix Figure 1 This is the main flowchart of the intelligent analysis method for energy consumption in industrial solid waste treatment proposed in this application. Detailed Implementation

[0026] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0027] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. In addition, examples of various specific processes and materials are provided in this invention, but those skilled in the art will recognize the application of other processes and the use of other materials.

[0028] As attached Figure 1As shown, this application provides an intelligent analysis method for energy consumption in industrial solid waste treatment, specifically including:

[0029] S1: Collect energy consumption metering data, equipment operating parameters, process flow variables, capacity indicators, and external environmental variables such as ambient temperature and humidity at multiple levels and locations in the industrial solid waste treatment production line to form a multi-source, multi-point raw dataset.

[0030] S2: Perform missing value completion, outlier removal, and multi-temporal alignment preprocessing on the collected raw dataset to standardize the historical and real-time states of different data channels into a multi-dimensional data vector under the same time base.

[0031] S3: Based on the pre-processed multi-dimensional data vector, construct the physical-process-energy consumption digital twin mapping relationship according to the dimensions of process flow, equipment type, loop partition, etc., to realize the digital twin model modeling of solid waste treatment production line.

[0032] S4: Input standardized multidimensional data vectors into the digital twin model to automatically extract energy consumption fluctuation features, equipment operating status features, and process features to form a set of multivariable coupled features.

[0033] S5: Based on the multivariate coupling feature set, use the anomaly detection algorithm to determine whether there is an energy consumption data exceeding the threshold or multi-source collaborative anomaly. If it is determined to be an energy consumption anomaly, record the anomaly event and the corresponding triggering process section and equipment identifier.

[0034] S6: After an abnormal event is triggered, the system automatically generates a combination of process, equipment, and environmental variables containing multiple virtual scenario parameter groups within the digital twin sandbox based on the recorded process segments, equipment identifiers, and historical operating conditions, and performs multi-scenario simulation reasoning within the sandbox.

[0035] S7: Collect response paths such as production capacity output, process stability, energy consumption links, and emission parameters of the digital twin model under various virtual scenario variable combinations, perform differential comparison analysis, and identify the direct links and indirect coupling mechanisms of energy consumption anomalies.

[0036] S8: Based on the response path analysis results of each virtual scenario, quantify the impact contribution of each abnormal factor on key business indicators such as energy consumption, production capacity, process stability and product quality, and output the impact ranking and business risk weight.

[0037] S9: Based on the impact contribution results, iteratively adjust the parameter group and verify the correction strategy in the digital twin sandbox, including process parameter optimization, equipment operation status adjustment and environmental factor compensation, and screen the optimal intervention combination that can take into account energy saving, production capacity and process stability and its expected effect.

[0038] S10: Feed the results of each round of sandbox simulation and intervention optimization back to the digital twin model to achieve continuous accumulation of knowledge about energy consumption anomalies and their multidimensional impacts and model adaptation, supporting higher-precision tracing of causes and output of optimization suggestions for subsequent energy consumption anomaly events.

[0039] Step S1 involves collecting energy consumption metering data, equipment operating parameters, process flow variables, capacity indicators, and external environmental variables such as ambient temperature and humidity from multiple levels and locations within the industrial solid waste treatment production line to form a multi-source, multi-point raw dataset. Specifically, this includes:

[0040] S1.1: Deploy energy consumption metering devices in each process section and equipment unit of the production line to obtain energy consumption metering data at different spatial levels, so as to achieve energy consumption data coverage at the process granularity and provide basic data for subsequent spatial distribution mapping of energy consumption fluctuations.

[0041] To address the multi-level energy consumption monitoring needs of industrial solid waste treatment production lines, the input objects are the spatial distribution information of each process section and equipment unit of the production line and the corresponding energy consumption signal channel configuration list.

[0042] A multi-level distributed energy consumption metering device deployment method is adopted (parameters include process segment division, equipment unit ID, spatial coordinates, and metering accuracy level). At key nodes of each process segment in the main production line and at auxiliary / supporting units, energy meters, smart circuit breakers, or circuit-level energy consumption acquisition modules are selected in layers to achieve multi-point energy consumption signal acquisition with vertical coverage (process-equipment-circuit) and horizontal coverage (main line-branch line-auxiliary system).

[0043] Through intelligent automatic addressing and device binding algorithms (parameters include device ID, PLC address, and communication protocol standard), the automatic mapping and attribution of metering devices to the production line DCS / PLC system is realized, achieving a unique correspondence between physical space structure and energy consumption data channels.

[0044] Furthermore, a time-series data sampling and remote automatic acquisition mechanism (parameters include sampling period T, data accuracy ΔP, and network transmission rate) is adopted to apply high-speed sampling tasks to the energy consumption devices of each process section, so as to realize the synchronous acquisition of multi-granularity energy consumption data such as ampere-hour power, real-time active / reactive power, and peak-valley power consumption. Key energy consumption indicators are temporarily buffered and initially aggregated through edge computing to reduce the risk of data packet loss.

[0045] Furthermore, by using data hierarchical integration and multi-channel mapping table generation methods, the raw energy consumption signals collected from each spatial level are automatically normalized to a unified energy consumption data format specification, and spatial level labels (such as process identifier, equipment level, and loop number) are added to generate a hierarchical energy consumption dataset with strong location attributes, thereby enabling spatial traceability and visualization mapping support for energy consumption signals.

[0046] Through the aforementioned multi-level distributed acquisition and intelligent mapping method, the spatial structure information of the aforementioned process sections and equipment units is transformed into a comprehensive and hierarchical energy consumption metering data channel, providing high-precision and full-granular basic data support for subsequent spatial distribution mapping of energy consumption fluctuations and source tracing of anomalies, thereby realizing a panoramic view of the energy consumption monitoring system for solid waste treatment production lines.

[0047] For example, in a large industrial solid waste treatment production line, the main line contains 7 key process sections. Each process section is equipped with one high-precision three-phase energy meter (0.2 class). Each process section contains 2-3 core equipment units, and each equipment unit is independently configured with a single-loop energy consumption acquisition module (0.5 class accuracy). All metering devices are connected to the production line's DCS system via Modbus / TCP communication protocol. Before installation, the clear correspondence between process section, equipment unit, and loop is obtained from the production line layout diagram, and a triplet mapping table of process-equipment-metering point is generated through automatic addressing. During data acquisition, the sampling period is set to 1 minute, and current and power signals at each point are collected remotely and continuously. The collected raw energy consumption data is automatically aggregated into a record structure of {process section ID, equipment ID, metering point number, sampling timestamp, current (A), phase voltage (V), and active power (P)}. After spatial tag normalization and hierarchical integration, the energy consumption level of each process section, equipment, and loop at any given time can be clearly tracked. During one actual production cycle, the system collected a total of 7 × 3 × 1440 = 30240 multi-point energy consumption data points, effectively supporting subsequent spatial location of anomalies in fine-grained energy consumption fluctuations and analysis of energy consumption anomalies in process segments. After applying this mechanism, the system achieved precise location capabilities for energy consumption anomalies and automated output of process-level energy consumption trend statistics.

[0048] S1.2: Real-time acquisition of equipment operating parameters for each physical circuit and key equipment, including speed, current, voltage, vibration signals, etc., to capture the dynamic state of the equipment and provide a set of raw parameters for equipment performance status modeling.

[0049] For physical loops and key equipment units, the input objects include the unique identification information of each device (such as device ID, loop number, process section mark), deployment location, real-time signal channel configuration list, and corresponding sensor model.

[0050] A high-precision industrial-grade sensor acquisition system (parameter settings include sensor type, sampling channel, accuracy level, and installation point number) is adopted. Multiple types of dynamic status sensors are deployed on each piece of equipment and its connected circuits to achieve seamless acquisition of the equipment's core operating parameters, including rotational speed (RPM), operating current (A), operating voltage (V), and structural vibration signals (m / s²). 2 Key indicators such as (g) and temperature rise (°C).

[0051] Furthermore, the above-mentioned operating parameters are collected in real time through a multi-channel data synchronous acquisition algorithm (parameters: number of channels N, data integration protocol, unified timestamp mechanism). The sampling period T is set by the dynamic response characteristics of the equipment (such as motor rotational inertia, vibration frequency bandwidth) and the overall DCS / PLC data synchronization protocol of the production line. High-frequency signal buffering and FIFO buffering mechanisms are adopted to ensure that the parameters of each channel of the equipment can be concurrently entered into the database and maintain high data integrity.

[0052] Furthermore, by using data denoising filtering and transient change capture algorithms (such as adaptive Kalman filtering, window sliding mean filtering, and automatic outlier removal), the original acquired signal is denoised and preprocessed for abnormal changes to ensure the capture of minute dynamic changes in the equipment and achieve a high-confidence representation of the true operating conditions of each physical circuit and equipment unit.

[0053] Furthermore, a multi-index normalization mapping algorithm (parameters: reference baseline mean and standard deviation, process section rated parameters) is applied to standardize the collected operating parameters, eliminating the influence of different sensor dimensions and parameter distributions of different equipment models, and converting them into a structured equipment operating parameter vector:

[0054]

[0055] Furthermore, based on the mapping table between equipment type and process flow node, the above-mentioned standardized operating parameter vectors are assigned to the structured data records of the corresponding equipment-loop-process segment according to the acquisition timestamp, thereby realizing the automatic integration of multi-source spatiotemporal equipment status information.

[0056] Through multi-channel real-time dynamic acquisition, preprocessing, normalization, and structured assembly, the original signals of equipment operating parameters are transformed into a high-quality set of equipment performance modeling parameters, realizing a holographic expression of equipment status synchronized with time and space, and providing fine input for subsequent equipment health modeling and energy consumption anomaly tracing.

[0057] For example, a solid waste incineration line deploys three grate propellers and two fans as key energy-consuming devices. Each device is equipped with one high-precision Hall speed sensor (accuracy 0.01Hz), one set of three-phase current and voltage transformers (accuracy class 0.2), and one IEPE accelerometer (bandwidth 0.5Hz-5kHz, sensitivity ±2g), with a sampling period set to 10s. Real-time signals are synchronously acquired and stored in the memory of an edge industrial control computer via Ethernet Modbus / RTU. The raw current signals are smoothed and denoised using Kalman filtering, and abnormal abrupt changes, such as discontinuous current, are automatically removed by an outlier detection algorithm. Using the rated speed and rated current of the equipment as normalization benchmarks, each data point is standardized and converted into a unified format such as {equipment ID, loop number, timestamp, standardized speed, standardized current, standardized voltage, standardized vibration amplitude}. After data merging, a device performance status group is formed with time-equipment-loop index, realizing comprehensive dynamic status recording in both vertical (equipment layer) and horizontal (spatial layer), supporting subsequent energy consumption and equipment health analysis based on statistical modeling and anomaly detection. During the 30-day operation period, the solution collected 432,000 high-confidence equipment operating parameters, significantly improving the ability to capture abnormal equipment behavior and the granularity of energy consumption at the process stage.

[0058] S1.3: Based on the production line DCS system or PLC control point linkage process monitoring system, acquire key process variables (such as material ratio, flow rate, temperature, pH) and record them synchronously to ensure that the process variables and energy consumption metering data form a temporal correlation matrix.

[0059] S1.4: Design and set up capacity measurement points for each process of the production line, collect capacity indicators such as piecework output, cycle capacity, and production line efficiency, form a capacity data channel, and provide data support for the extraction of energy consumption and capacity coupling characteristics.

[0060] S1.5: Industrial environmental sensors are used to automatically collect external environmental variables such as temperature and humidity inside and outside the production line, and synchronize them with equipment operating parameters and energy consumption metering data as inputs of external variables affecting energy consumption and production capacity.

[0061] S1.6: Based on a unified timestamp and data synchronization protocol, energy consumption metering data, equipment operating parameters, process flow variables, production capacity indicators and environmental temperature and humidity data are collected in multiple dimensions to generate a structured multi-source and multi-point raw dataset, providing multi-channel spatiotemporal consistent input for subsequent preprocessing and digital twin modeling.

[0062] Step S2 involves preprocessing the collected raw dataset by imputing missing values, removing outliers, and performing multi-temporal alignment, standardizing the historical and real-time states of different data channels into a multi-dimensional data vector under the same time base. Specifically, this includes:

[0063] S2.1: Perform data integrity verification and missing value detection on the collected multi-source raw datasets (including energy consumption metering data, equipment operating parameters, process flow variables, production capacity indicators and environmental variables). Based on interpolation, model prediction or empirical mapping methods, fill in the missing values ​​of the detected missing data channels to obtain a complete set of data channels.

[0064] S2.2: Based on the data channel set after missing value completion, outlier detection and removal are performed on each data channel using methods such as coefficient of variation screening, box plot algorithm, and multidimensional collaborative filtering. By removing outlier data samples, a high-confidence data channel set with noise suppression and reasonable statistical distribution is generated.

[0065] S2.3: Based on the high-confidence data channel set after noise suppression, timestamp correction and sampling period synchronization are performed. Sliding window resampling and multi-temporal data alignment algorithms are adopted to uniformly map the historical asynchronous acquisition and real-time streaming acquisition data to the standard time base, and obtain a synchronous data matrix under the homogeneous time base of multiple channels.

[0066] S2.4: Utilize the synchronized data matrix to implement data normalization and standardization processing, including zero-mean normalization, range scaling and other multi-variable unified transformations, to ensure the uniformity of physical quantities between different data channels, eliminate scale differences, and obtain a set of standardized multi-dimensional vectors suitable for subsequent modeling.

[0067] S2.5: Based on the obtained standardized multidimensional vector set, perform data consistency verification and context validity screening, remove data slices with time-series aliasing or inconsistent with the process flow logic, and output high-quality unified time-based multidimensional data vectors for digital twin modeling, providing standard input for digital twin mapping and energy consumption anomaly detection.

[0068] Step S3: Based on the preprocessed multidimensional data vector, construct a physical-process-energy consumption digital twin mapping relationship according to dimensions such as process flow, equipment type, and loop partitioning, to realize the digital twin model modeling of the solid waste treatment production line. Specifically, this includes:

[0069] S3.1: Perform cluster analysis on the process flow variables in the multidimensional data vector, and use the process node identification algorithm to obtain the process flow node grouping that maps to the physical production line structure, so as to realize the unique correspondence between process flow nodes and energy consumption and equipment parameters.

[0070] S3.2: Based on process flow node grouping, extract the operating parameters of each equipment in the same physical production line, apply the equipment classification coding method, and bind the equipment parameters with the process flow node identifier to generate a joint index of equipment type and process flow.

[0071] The clustered process nodes and their corresponding standardized multidimensional data vectors have been obtained.

[0072] An equipment parameter extraction algorithm (parameters: complete set of equipment operating parameters and process flow node grouping results) is used to extract all equipment belonging to each node in the same physical production line by grouping them, thereby achieving comprehensive collection of equipment-level parameters.

[0073] Furthermore, by using the equipment classification coding method (parameters: equipment type, equipment physical code, process flow node ID), the extracted equipment operating parameters are categorized according to equipment type (such as motor, pump, fan, valve, etc.), and a unique identifier is assigned to each type of equipment to achieve standardized coding of equipment parameters.

[0074] By using a parameter binding algorithm (parameters: unique code for equipment parameters and identifier for process flow nodes), the categorized and coded equipment operating parameters are associated one-to-one with the grouping results of process flow nodes. All equipment under each process flow node is assigned a two-level binding identifier, achieving full coverage of the mapping between equipment parameters and process flow node IDs.

[0075] A multidimensional index generation method (parameters: equipment type code, process flow node ID, physical equipment code) is adopted to concatenate and store the bound structured data into triples, and automatically generate a joint index of equipment type and process flow, so as to realize efficient retrieval and dynamic association of equipment parameters under the production line structure.

[0076] Through the above chain derivation, the equipment parameters at each level of the physical production line are precisely bound to the process flow nodes. The structured output of the equipment type-process flow joint index serves as the data foundation for subsequent loop energy consumption partitioning and digital twin modeling, thereby realizing the automation, standardization, and scalability of the process-equipment mapping process.

[0077] For example, for an industrial solid waste treatment line with five main process nodes (feeding, pretreatment, combustion, cooling, and discharge), a standardized multidimensional data vector covering the entire process has been acquired in step S2. Each node is equipped with 20 sets of equipment, belonging to three types: high-pressure blowers (equipment type A, equipment 1-5), centrifugal pumps (equipment type B, equipment 6-12), and circulating stirring motors (equipment type C, equipment 13-20). An equipment parameter extraction algorithm is applied to extract key operating parameters (speed, current, temperature, vibration, etc.) for all equipment from the database, generating an independent parameter set for each set of equipment. An equipment classification coding method is used, assigning equipment type A the code A01, B02, and C03, with individual equipment codes such as A01-1, B02-6, etc. A one-to-one mapping is established between each equipment parameter and its assigned process node (e.g., equipment A01-1 belongs to the feeding node Node-01). Subsequently, based on a multidimensional index generation method, a triplet index (equipment type code, process node ID, physical equipment code) was generated for each equipment record, such as (A01, Node-01, EID1001). All 20 sets of equipment obtained a unique composite index. In actual testing, using this composite index structure, dynamic binding and traceability queries of equipment parameters within any node can be completed within milliseconds. This meets the subsequent modeling requirements and link tracing requirements of the digital twin sandbox for multidimensional linkage of equipment-process-loop, and significantly improves the efficiency of accurate location and correlation impact analysis of abnormal energy consumption events.

[0078] S3.3: For the joint index, collect the energy consumption metering values ​​of the loop partition level, adopt the partition many-to-one mapping mechanism, and classify the energy consumption metering point data of each loop into the corresponding equipment type-process flow joint index to construct a physical-process-energy consumption three-dimensional mapping matrix.

[0079] For the generated equipment type-process flow joint index structure, collect energy consumption metering data at the loop partition level to obtain the energy consumption time series vector of each energy consumption metering point under the spatial distribution of the production line.

[0080] The loop partitioning identification method (parameters: process flow node grouping, equipment type coding, physical location of on-site energy consumption measurement points) is adopted to divide all energy consumption metering points in the production line into corresponding process-equipment loops, realizing the spatial mapping relationship between energy consumption collection points and physical loop structures.

[0081] Furthermore, through a partitioned many-to-one mapping algorithm (parameters: loop partition ID, equipment type code, process flow node ID), the energy consumption data groups of all energy consumption metering points within the same loop partition are centrally assigned to the equipment type-process flow joint index bound to them according to the preset mapping rules, thereby realizing the partitioned integration of energy consumption data.

[0082] A partitioned weighted aggregation method (parameters: energy consumption meter weights, sampling period) is used to generate a normalized energy consumption index that reflects the comprehensive energy consumption characteristics of the partition by applying weighted average, weighted accumulation, or maximum / minimum value merging algorithms to multiple energy consumption meters belonging to the same composite index. The specific calculation formula is as follows:

[0083]

[0084] Among them, E ij Let n be the total energy consumption of the partition under the joint index of the i-th process node and the j-th equipment type. ij To associate the number of energy consumption metering points, w ijk The weight coefficient for the k-th measurement point is... This represents the energy consumption value collected at time t at that metering point.

[0085] Through the above weighted aggregation, the spatial integration of energy consumption metering data under each partition is completed, and a normalized energy consumption data matrix based on loop partition is obtained.

[0086] Generate a three-dimensional mapping matrix structure (parameters: equipment type code, process flow node ID, loop partition ID), and record the high-dimensional nested relationship between physical entities, process structures and energy consumption metering points in a triplet manner to achieve integrated mapping of the entire space of physical-process-energy consumption.

[0087] By using a three-dimensional mapping matrix, the energy consumption metering data that has been partitioned and normalized is deeply integrated with the equipment type-process joint index and loop partition structure to form a structured energy consumption spatiotemporal distribution characteristic dataset, thereby realizing the accurate expression of the energy consumption characteristics of each layer of objects in the solid waste treatment production line.

[0088] For example, in a 150m long solid waste treatment main production line, eight loop partition energy consumption collection nodes are designed (such as the main blower power supply loop, feed pump loop, furnace electric heating loop, flue gas treatment, etc.), with 2-3 energy metering terminals deployed in each partition. Corresponding to the aforementioned three equipment types (high-pressure blower A01, centrifugal pump B02, circulating stirring motor C03) and five process flow nodes, 20 unique equipment-process binding units are defined through an equipment-process joint index. Using a loop partition identification method, each of the eight loop partitions is assigned a physical partition ID (RL01~RL08), and each energy consumption metering point is bound to its dedicated partition. Through a many-to-one partition mapping, the energy consumption data of all metering points (such as the two energy consumption terminals in the main blower loop) within each partition are aggregated into the joint index of the corresponding high-pressure blower-feeding node (A01, Node-01), etc. A weighting coefficient w is set. ijk The actual power capacity percentage for each metering point is automatically extracted by the system. For each energy consumption metering group under the joint index, a comprehensive energy consumption index E is generated for each time period according to the weighted aggregation formula described above.ij The final structured output is a three-dimensional mapping matrix data structure, with dimensions combined as [equipment type code × process flow node ID × loop partition ID], such as (A01, Node-01, RL01). This enables the collection of energy consumption data across the entire production line, multiple equipment, and multiple loop spatial structures, greatly improving the accuracy of subsequent digital twin modeling of equipment health, process flow, and abnormal energy consumption spatial distribution.

[0089] S3.4: Through a three-dimensional mapping matrix, the equipment health, loop energy consumption distribution and process flow status are integrated in a time sequence. A multi-level feature fusion algorithm is used to obtain a dynamic digital twin hierarchical structure covering the entire process and multiple granularities.

[0090] The input consists of the energy consumption metering dataset after partition normalization, which is indexed by the triplet index of equipment type code, process node ID and loop partition ID through a three-dimensional mapping matrix, as well as the corresponding standardized equipment health parameter sequence and process flow status variables.

[0091] A time-series synchronization fusion algorithm (parameters: equipment health sequence, loop energy consumption distribution sequence, process flow status sequence, unified timestamp reference) is adopted to achieve full-time domain alignment and synchronization integration of equipment health, loop energy consumption distribution and process flow status.

[0092] Furthermore, by using the Dynamic Time Warping (DTW) method (parameters: multiple sequence inputs, sampling period, and allowed time delay window), the micro-time difference between parameters from different sources is corrected, and the optimal alignment mapping relationship is established for the data points of each device and process section on the acquisition time axis, resulting in refined multidimensional synchronous sequence data.

[0093] Furthermore, a multi-level feature fusion algorithm (parameters: hierarchical aggregation rules, multi-dimensional feature input, hierarchical structure template) is adopted to perform hierarchical feature nesting. The equipment health status features, regional energy consumption distribution features, and process flow status features are aggregated step by step according to the physical-process-energy consumption spatial hierarchy to realize the multi-level nested expression of dynamic features in the whole scenario.

[0094] Furthermore, by using a feature importance hierarchical weight allocation method (parameters: feature category, location weight, equipment / process priority), the influence of fused features in their respective hierarchical structures is calculated, forming a weighted feature matrix that reflects spatial distribution, node hierarchy, and temporal changes.

[0095] Through dynamic digital twin hierarchical structure generation and processing, the above-mentioned multi-level fusion features are merged and described as "physical structure - process level - energy consumption / health / process multi-feature fusion body" on the basis of three-dimensional mapping, realizing high-precision modeling and expression of the spatiotemporal correlation and dynamic response capabilities between objects in the entire process of solid waste treatment production line, multi-granularity, and multi-dimensional.

[0096] Through the aforementioned multi-layered and progressive data integration and feature fusion, the equipment health, loop energy consumption distribution, and process flow status are synchronized in time and nested in a structured manner within a three-dimensional mapping matrix. This results in a dynamic digital twin hierarchical structure covering the entire solid waste treatment production line and multiple granularities, enabling strong correlation and high-resolution digital mapping of production line objects. This lays the foundation for anomaly impact analysis, response chain tracing, and simulation reasoning.

[0097] For example, in a dynamic digital twin modeling scenario of a typical industrial solid waste treatment production line, a three-dimensional mapping matrix is ​​formed by 5 process flow nodes (Node-01 to Node-05), 3 equipment type codes (A01, B02, C03), and 8 loop partitions (RL01 to RL08). The data acquisition cycle is set to 1 minute, and the system collects health parameters (such as remaining lifespan and status score), corresponding loop energy consumption (normalized kWh value), and process flow status (stage identifier, batch flow direction, etc.) for 20 sets of equipment every minute. A time-series synchronous fusion algorithm is used to dynamically calibrate and merge the energy consumption data, equipment health parameters, and process status variables in each loop partition within a granularity of 0.1 seconds. The DTW algorithm is used to align and correct multi-source parameters with small sampling delays or time-series jumps, achieving optimal registration of all parameters within a maximum allowable time delay of 5 sampling cycles. In the multi-level feature fusion process, the health status and corresponding circuit energy consumption of all high-pressure blowers on Node-01 are first aggregated hierarchically, and then recursively aggregated to the process line level, expressing a hierarchical structure through feature nesting. During weight allocation, the blower health weight is configured as 0.4, energy consumption distribution as 0.4, and process flow status as 0.2, generating a hierarchical feature matrix. The final output dynamic digital twin hierarchical structure can reflect the multi-dimensional status of health, energy consumption, and flow of each node-equipment-partition unit in real time. Actual testing shows high sensitivity to health degradation and energy consumption anomalies within small time granularities, providing a high-throughput, full-space data foundation and structured input for subsequent energy consumption anomaly tracing and impact path simulation.

[0098] S3.5: Output the hierarchical structure of the dynamic digital twin as a unified digital twin model object, and use nested logic to express the high-precision digital correspondence between its physical entities, process nodes, equipment parameters and energy consumption metering points, providing a structured input basis for subsequent feature extraction and anomaly tracing.

[0099] Step S4: Input the standardized multidimensional data vector into the digital twin model to automatically extract energy consumption fluctuation features, equipment operating status features, and process features to form a multivariate coupled feature set. Specifically, this includes:

[0100] S4.1: Based on standardized multidimensional data vectors, the time-series window segmentation method is used to perform sliding window segmentation on continuous data of energy consumption metering points to obtain energy consumption fluctuation segments that reflect the dynamic changes of the solid waste treatment production line.

[0101] The input is a standardized multidimensional data vector output by step S3.5, which contains energy consumption metering data sequences that have been mapped and bound to process flow segments, equipment types, and loop partition levels.

[0102] The time-series window segmentation method is adopted (parameters: data input is standardized time-series data of energy consumption metering points, window length w, sliding step size s). For each energy consumption metering point of the target production line, the original energy consumption sequence is segmented by sliding window according to a unified time benchmark to realize the interval representation of data.

[0103] Furthermore, through an adaptive window parameter adjustment algorithm (parameters: adaptive window length criteria such as short-term energy consumption variance, mutation entropy, and data periodicity characteristics), the optimal window length w is set for different process sections and equipment partitions. i Dynamically match the process cycle and work rhythm to ensure that the signals of section anomalies and energy consumption fluctuations can be effectively captured within each window.

[0104] Furthermore, for each segmented sliding window energy consumption data fragment, a statistical feature extraction method within the window (parameters: mean, standard deviation, peak value, coefficient of variation, etc.) is used to calculate time-domain feature indicators reflecting the amplitude and stability of energy consumption fluctuations, and to provide high-confidence intervals for subsequent time-frequency analysis methods such as wavelet transform.

[0105] Furthermore, by employing a sliding window stacking algorithm (parameters: window overlap rate λ, step size s), the time-series window segmentation intervals are overlapped, forming a high-coverage, high-precision energy consumption fluctuation segment sequence, enhancing the ability to capture abnormal fluctuation intervals. The formula for calculating the sliding window overlap rate is as follows:

[0106]

[0107] Where s is the sliding step size, w is the window length, and λ is the overlap rate between windows.

[0108] Through the above chain derivation, a dataset of energy consumption fluctuation segments with high spatiotemporal resolution covering all energy consumption metering points of the complete production line is finally generated, laying the foundation for the next step of extracting energy consumption fluctuation feature parameters and diagnosing anomalies.

[0109] For example, on a solid waste treatment production line comprising 5 process nodes, 20 sets of equipment, and 8 loop partitions, standardized time-series energy consumption data with a sampling period of 1 minute is processed using a time-series window segmentation method. The window length is set to w=15, and the sliding step size to s=5. The window length is dynamically adjusted based on the sensitivity of energy consumption changes in each process segment (e.g., w=20 for the pretreatment node and w=12 for the combustion node), and the window overlap rate is set to λ=0.67. For 2000 minutes of historical data at each energy consumption metering point, approximately 400 segmented windows are generated. Descriptors such as mean, standard deviation, and peak value are calculated for each window. In practical applications, this method can effectively extract dynamic energy consumption fluctuation segments from the production line, distinguishing between stable operating conditions and short-term abnormal energy consumption abrupt changes. This provides high-timeliness and high-accuracy support for subsequent wavelet transform and root cause analysis of multi-frequency band energy consumption fluctuation characteristics. The final output is a dataset of energy consumption fluctuation segments with high spatiotemporal resolution, supporting subsequent micro-temporal domain analysis of energy consumption fluctuations.

[0110] S4.2: Adaptive wavelet transform algorithm is used for energy consumption fluctuation segments to extract energy consumption fluctuation characteristic parameters such as fundamental frequency and wavelet energy spectrum of different frequency bands in each process segment, and generate energy consumption fluctuation feature set that reflects the root cause of energy consumption anomalies.

[0111] The input is a dataset of energy consumption fluctuation segments obtained by the time-series window segmentation method, which includes standardized energy consumption sequence intervals under different process sections, equipment types and loop partitions of the solid waste treatment production line.

[0112] An adaptive wavelet transform algorithm (parameters: energy consumption fluctuation segment input, wavelet basis type, number of decomposition levels, adaptive frequency band selection rule) is adopted to realize multi-scale decomposition of energy consumption signal within the sliding window and extract time-frequency localized response features of energy consumption fluctuation in different frequency bands.

[0113] Furthermore, through a hierarchical wavelet decomposition process, multi-scale signal decomposition is performed on each energy consumption fluctuation segment to obtain sub-band signal coefficients at different levels (such as detail level and approximation level). The energy consumption signal x(t) can be expressed by wavelet decomposition as:

[0114]

[0115] Among them, a J,k Let d be the approximation coefficient of the Jth layer. j,k Let j be the detail coefficients of the j-th layer. Let ψ be the scaling function. j,k (t) is the wavelet function.

[0116] Furthermore, by automatically adjusting the wavelet decomposition level J and wavelet basis function parameters based on the adaptive frequency band analysis criterion of energy consumption fluctuations, the decomposition can fully adapt to the main oscillation range and abnormal energy consumption disturbance range of energy consumption signals in each process section, retain abnormal information to the maximum extent, and avoid information loss due to excessive filtering.

[0117] Furthermore, using the detail coefficients and approximation coefficients after wavelet decomposition at each level, the fundamental frequency characteristics and wavelet energy spectrum characteristics of instantaneous energy consumption are calculated respectively, forming a multi-band time-frequency parameter set that provides a criterion for the root cause of energy consumption anomalies. Fundamental frequency f b Extraction is based on the main energy peak frequency band, energy spectrum S j The calculation is as follows:

[0118]

[0119] Among them, S j Let be the energy consumption wavelet spectrum of the j-th decomposition layer.

[0120] Furthermore, for each energy consumption fluctuation segment, the wavelet feature parameters of the above different frequency bands are aggregated to generate an energy consumption fluctuation feature set, including: fundamental frequency distribution, wavelet energy spectrum, hierarchical energy ratio, anomaly sensitivity coefficient, etc.

[0121] By using an energy consumption fluctuation feature parameter merging algorithm, various feature parameters extracted by multi-scale wavelets are structured and output according to process segmentation, equipment type, and spatial partitioning coding, forming an energy consumption fluctuation feature set that highly distinguishes the characteristics of energy consumption fluctuations. This provides a high signal-to-noise ratio and quantifiable data foundation for energy consumption anomaly attribution and subsequent equipment-process-environment multivariate correlation analysis, enabling accurate modeling of the root cause parameters of energy consumption anomalies.

[0122] For example, in a solid waste treatment production line, for the high-pressure blower section of the main combustion furnace process, the input is a standardized energy consumption fluctuation sequence with a sampling period of 1 minute and a window length of 20 minutes. The Daubechies-4 mother wavelet is used, with a wavelet decomposition level set to 4 levels. Adaptive decomposition automatically truncates the peak energy consumption frequency band within the window. For each window, the detail coefficients d for levels 1-4 are obtained. j,k and approximation coefficient a J,k .

[0123] For energy consumption fundamental frequency extraction, the fundamental frequency f is determined using the energy spectrum peak concentration layer. b For example, in the main error window, f b =0.025Hz. For the wavelet energy spectrum, the values ​​for each layer are calculated as follows: S1 = 5.8, S2 = 10.2, S3 = 18.6, S4 = 4.1 (unit normalized energy). The energy ratio of each layer is R. j,j+1 =S j / S j+1 , such as R 2,3=0.55. The final output feature set includes the fundamental frequency, energy spectrum and energy ratio of each layer, and the maximum anomaly sensitivity coefficient. It can be seen that before and after the anomaly occurred, the energy spectrum distribution of the combustion process section changed abruptly from S2 / S3 = 0.68 to 0.42, and the anomaly sensitivity index increased by 2.3 times.

[0124] In practical applications, this wavelet feature set can successfully distinguish three typical fluctuation scenarios: normal operating conditions, sudden changes in energy consumption, and periodic anomalies. The accuracy of subsequent anomaly root cause identification is improved to over 96%, achieving refined energy consumption anomaly diagnosis and data-driven support.

[0125] S4.3: Apply multidimensional statistical analysis methods (such as mean, variance, outlier, utilization rate, etc.) to the sequence of equipment operating parameters within the sliding window to extract equipment operating status characteristics, so as to achieve structured quantification of equipment status.

[0126] The input conditions are: a standardized multidimensional data vector output by step S3.5, which has embedded equipment type code, process flow node ID and partitioned normalized energy consumption metering data, and is accompanied by historical and real-time operating parameter sequences for each device, with a unified sampling period.

[0127] A multidimensional statistical analysis method (parameters: sliding window equipment operating parameter sequence, statistical type list) is adopted to perform window-by-window statistical analysis on key operating parameters (such as speed N(t), current I(t), voltage U(t), vibration a(t), temperature T(t), etc.) of each piece of equipment (such as high pressure blower, centrifugal pump, circulating stirring motor, etc.) within the sliding window, and extract typical statistical indicators of the original numerical domain.

[0128] Furthermore, by using the sliding statistic calculation method, the mean of each sequence of operating parameters X(t) is calculated. variance Maximum value x max Minimum value x min Coefficient of variation Cv X Utilization rate K run The specific formula is as follows:

[0129]

[0130] Where w is the number of sampling points in the sliding window interval, X i Let be the sampled value at time i.

[0131] Furthermore, a multi-parameter collaborative outlier detection algorithm (parameters: multi-dimensional equipment operating parameter vector) is used to perform joint outlier analysis among the operating parameters within a sliding window, and the Mahalanobis distance D is calculated. M Used to detect abnormal parameter linkage:

[0132]

[0133] in, This is the mean vector of the multidimensional parameters of this window. Let ∑ be the baseline mean vector of the historical parameters of the equipment, and let ∑ be the corresponding covariance matrix.

[0134] Furthermore, through a hierarchical equipment operation status index merging algorithm, the statistical characteristic parameters of each equipment within its process flow nodes and partitions are merged and output in a structured data manner, such as equipment-window-parameter index triples, which supports subsequent horizontal normalization comparison and vertical trend interpretation.

[0135] Through the above-mentioned multidimensional statistics and co-outlier degree processing, the segmented sliding window equipment operating parameter sequence is transformed into structured equipment operating status characteristics that reflect health status, abnormal operating conditions and operating efficiency. This injects high-confidence parameter expressions into the equipment-level digital twin model, realizing the quantification and standardization of production line equipment status.

[0136] For example, in the solid waste treatment production line Node-02 (pretreatment process section), a sliding window length of 15 (sampling period of 1 minute) is set, and statistical analysis is performed on the data series of rotational speed, line current, line voltage, and temperature of four circulating stirring motors (equipment type C03). Taking EID1013 (C03-13) as an example, within the 15-minute window, the mean of the rotational speed N(t) series is 690 rpm, the variance is 80, and the coefficient of variation is 0.13; the mean of the temperature is 64℃, the variance is 1.5, and the utilization rate K... run =1. Using the historical 90-day average rotational speed of 700 rpm and standard deviation of 30 as the baseline, calculate the Mahalanobis distance D. M =2.6. This value is higher than the set alarm threshold of 2.0, indicating that the current operating window of C03-13 deviates significantly from its historical healthy operating conditions, suggesting possible abnormal wear or stalling. Synchronous analysis was performed on equipment C03-14 / 15 / 16 in the same group, and the output statistical feature data was synchronously stored in the database, providing structured input for subsequent equipment anomaly link mapping and multivariate feature fusion in the digital twin sandbox. In actual operation, the above method can flatten and output more than 50 statistical indicators for each of the 20 equipment zones across the entire production line per minute, forming a high-resolution equipment operating status feature matrix. Sandbox simulation and subsequent anomaly tracing tests show that this feature set supports an early warning accuracy of equipment anomalies up to 98%, an improvement of approximately 15% compared to traditional single-parameter thresholds.

[0137] S4.4: Based on the process flow variables and loop partitioning relationships within the digital twin model, perform hierarchical principal component analysis (PCA) on the process variable sequence within the window to obtain the principal component features of the process and achieve multidimensional dimensionality reduction expression of the process state.

[0138] S4.5: Using energy consumption fluctuation feature set, equipment operating status feature and process principal component feature as input, a multi-source feature-level fusion method (such as feature splicing and interactive embedding algorithm) is adopted to generate a multivariable coupled feature set, which fully represents the nonlinear coupling relationship between objects at all levels of the production line.

[0139] S4.6: Using a meta-learning feature importance ranking algorithm, we perform contribution analysis on various features in the multivariate coupling feature set, and select key multivariate coupling features that can characterize the precursors of energy consumption anomalies and their linkage with equipment-process-environment, so as to output a high signal-to-noise ratio feature set for anomaly diagnosis and deep inference.

[0140] Step S5: Based on the multivariate coupling feature set, an anomaly detection algorithm is used to determine whether the energy consumption data exceeds the threshold or exhibits multi-source collaborative anomalies. If an energy consumption anomaly is determined, the anomaly event and the corresponding triggering process segment and equipment identifier are recorded. Specifically, this includes:

[0141] S5.1: Perform normalization processing on the extracted multivariate coupling feature set, compare it with the historical baseline based on the acquisition cycle, process flow node and equipment type to generate a normalized multivariate coupling feature vector, providing standardized input for subsequent anomaly detection algorithms.

[0142] S5.2: Based on the normalized multivariate coupled feature vector, clustering analysis algorithms (such as Gaussian Mixture Model, DBSCAN, etc.) are applied to divide the samples in the feature space into groups to automatically discover potential energy consumption behavior patterns and provide a statistical distribution basis for threshold setting and collaborative anomaly indicator identification.

[0143] S5.3: Based on the clustering analysis results, use multi-channel anomaly detection algorithms (such as isolated forest and multivariate statistical process control MSPC) to perform threshold judgment and collaborative anomaly analysis on the energy consumption characteristics of multiple sources at the same time, thereby identifying whether there are any out-of-limit anomalies and multivariate collaborative anomalies in the energy consumption data.

[0144] S5.4: For feature vectors identified as energy consumption anomalies, immediately index the corresponding process flow segment and equipment entity identifier in the digital twin model. Based on the feature mapping relationship, assign the anomaly to the specific process segment and equipment identifier, and associate it with the timestamp and anomaly feature value to achieve event-level precise localization.

[0145] S5.5: Generate structured abnormal event records from attributed energy consumption anomalies, process segment identifiers, and equipment identifiers, and store them in the event management database. This provides an abnormal event input interface for subsequent sandbox multi-scenario reasoning and impact chain analysis, enabling closed-loop management of intelligent diagnosis and simulation.

[0146] Step S6: After an abnormal event is triggered, the system automatically generates a combination of process, equipment, and environmental variables containing multiple virtual scenario parameter groups within the digital twin sandbox based on the recorded process segments, equipment identifiers, and historical operating conditions, and performs multi-scenario simulation reasoning within the sandbox. Specifically, this includes:

[0147] S6.1: Based on the process segment identifier, equipment identifier, and historical operating condition parameters of energy consumption anomaly events, automatically retrieve and obtain the process parameter set, equipment status parameter set, and environmental variable set in the digital twin sandbox to form the input basis for multi-scenario parameter generation.

[0148] S6.2: Utilize multi-scenario generation rules to reorganize the process parameter set, equipment status parameter set, and environmental variable set in multiple dimensions. Employ system variable interference injection and typical extreme value perturbation algorithms to generate batch process parameter groups, equipment operating status groups, and environmental variable groups that include normal operating conditions, boundary operating conditions, and extreme operating conditions, thereby achieving system coverage of multi-scenario parameter combinations.

[0149] S6.3: For each group of process parameters, equipment operating status, and environmental variables, a virtual instance is established based on the digital twin sandbox. The historical operating conditions and current abnormal features are injected into the virtual production process through the parameter mapping engine to ensure high-fitness scene reproduction of the abnormal context.

[0150] S6.4: Within the sandbox platform, based on the aforementioned virtual instance, a multi-objective dynamic inference engine is used to conduct sandbox scenario simulation and deduction for the parameter group. The energy consumption response chain, production capacity output chain, process stability index chain, and equipment health causal chain under different parameter groups are recorded in real time to obtain the full-process multi-scenario inference results.

[0151] S6.5: Compare and analyze the simulation results of each scenario one by one, extract the key indicator change chain related to energy consumption anomaly events, realize the causal correspondence between process parameter group, equipment operating status group, environmental variable group and energy consumption anomaly chain, and provide multi-dimensional input basis for subsequent impact measurement and correction verification.

[0152] Step S7: Collect the response paths of the digital twin model under various virtual scenario variable combinations, including production capacity output, process stability, energy consumption links, and emission parameters; perform differential comparison analysis to identify the direct links and indirect coupling mechanisms of energy consumption anomalies. Specifically, this includes:

[0153] S7.1: Based on the virtual scenario parameter group generated in the digital twin sandbox, input the combination of virtual scenario variables into the digital twin model for dynamic simulation, collect multiple response paths such as production capacity output, process stability, energy consumption link and emission parameters under the corresponding virtual scenario, and obtain the virtual scenario response path dataset as the input basis for subsequent analysis.

[0154] S7.2: Apply a multidimensional path difference analysis algorithm to the collected virtual scenario response path dataset. Based on technical terms such as production capacity output, process stability, energy consumption link, and emission parameters, calculate the quantitative difference index of each response path under different combinations of virtual scenario variables in order to extract the significant characteristic changes under the influence of energy consumption anomalies.

[0155] S7.3: Using the differences in capacity output, process stability, energy consumption links and emission parameters, combined with causal inference modeling, causal chain reasoning is performed on the path of changes in technical terms between adjacent variable combinations to initially screen out the set of response paths of direct links (such as single energy consumption link breakpoints or equipment-side impacts) that affect energy consumption anomalies.

[0156] S7.4: Further incorporate process change history and equipment operating status information into the initially screened set of direct link response paths. Employ a multivariate coupled tracing algorithm to conduct in-depth correlation analysis on the complex coupling effects between energy consumption links, production capacity output, and process stability. Identify the secondary propagation mechanism of energy consumption anomalies through indirect paths (such as process linkage or environmental factor coupling) to the response path.

[0157] S7.5: Summarize and organize all direct links and indirect coupling mechanisms caused by changes in the combination of virtual scenario variables to form a full-process abnormal impact path map of "energy consumption link - production capacity output - process stability - emission parameters". This provides a detailed multi-scenario response link basis for subsequent quantitative assessment of impact contribution and is passed as data input to the next technical processing step.

[0158] Step S8: Based on the response path analysis results of each virtual scenario, quantify the contribution of each abnormal factor to key business indicators such as energy consumption, production capacity, process stability, and product quality, and output the impact ranking and business risk weights. Specifically, this includes:

[0159] S8.1: Based on the response path data collected after multi-scenario simulation and reasoning in the sandbox, and according to the capacity output, process stability parameters, energy consumption links and product quality monitoring results contained in the response path, extract the multi-dimensional response feature set under the influence of each abnormal factor to form the input feature vector required for quantitative analysis.

[0160] S8.2: Based on the extracted multidimensional response feature set, the principal component analysis (PCA) algorithm is used to reduce the dimensionality of data such as capacity output, process stability, product quality and energy consumption fluctuations, and obtain the response principal component index that reflects the weight of the joint influence of each abnormal factor, providing a feature basis with a unified dimension for further contribution calculation.

[0161] S8.3: For the principal component indicators of the response, apply regression analysis or influence contribution decomposition algorithm to calculate the independent influence weight and cumulative contribution of each abnormal factor on the key business indicators (energy consumption, production capacity, process stability, product quality), and form the indicator influence contribution matrix of abnormal factors.

[0162] S8.4: Using the impact contribution matrix, each abnormal factor is weighted and sorted according to the comprehensive risk weights on energy consumption and business objectives, generating a list of the risk impacts of the abnormal factors after sorting, and verifying the rationality of the weight distribution in conjunction with previous historical simulation results.

[0163] S8.5: Based on the sorted list of abnormal factors' risk impacts and contribution matrix, automatically output the impact ranking of each abnormal factor, business risk weights, and quantitative assessment reports, providing the optimal priority for handling abnormal energy consumption events and the basis for risk warnings for subsequent intervention optimization and decision support systems.

[0164] Step S9: Based on the impact contribution results, iteratively adjust the parameter group and verify the correction strategy sequentially within the digital twin sandbox, including process parameter optimization, equipment operating status adjustment, and environmental factor compensation, to screen the optimal intervention combination that can balance energy saving, production capacity, and process stability, and its expected effects. Specifically, this includes:

[0165] S9.1: Based on the ranking results of influence contribution, automatically obtain the main abnormal factors identified in the sandbox simulation and their corresponding digital twin variable sets, and use the abnormal factor parameter group as the input object for iterative adjustment to construct the parameter space matrix to be optimized.

[0166] S9.2: For the matrix to be optimized in the parameter space, use process parameter optimization algorithms (such as multi-objective genetic algorithm or particle swarm optimization) to generate candidate process parameter groups in batches, and input them into the digital twin sandbox for virtual process simulation to obtain the energy consumption link simulation results after each set of process parameters is adjusted.

[0167] S9.3: Based on the optimization results of process parameters, further combine with the adjustment of equipment operating status. Through the equipment status simulator, perform sensitivity analysis and dynamic scheduling on key operating parameters of the equipment to obtain the combined impact response data of equipment operating status adjustment on energy consumption links and production capacity output, and expand the response dimension of digital twin variables.

[0168] S9.4: Based on the comprehensive process parameter set and equipment operation status adjustment, an environmental factor compensation mechanism is superimposed. Environmental variable disturbance simulation (such as temperature and humidity control algorithm) is adopted to evaluate the synergistic optimization effect of different environmental parameter values ​​on energy consumption link, process stability and production capacity output, and generate a multi-dimensional variable compensation response matrix.

[0169] S9.5: For each set of digital twin simulation scenarios after parameter iteration adjustment, perform multi-objective comprehensive evaluation (such as weighted scoring mechanism), calculate the comprehensive performance score under multiple indicators such as energy saving effect, production capacity impact and process stability, and form a global intervention scheme performance ranking list.

[0170] S9.6: From the global intervention scheme performance ranking list, use the decision optimization algorithm to automatically select the top-ranked optimal intervention combination that can balance energy consumption reduction, production capacity guarantee and process stability, output the corresponding process operation suggestions, equipment adjustment measures and environmental compensation parameter groups, and clarify its expected virtual simulation effect.

[0171] Step S10: Feed the results of each round of sandbox simulation and intervention optimization back to the digital twin model to achieve continuous accumulation of knowledge about energy consumption anomalies and their multidimensional impacts, and model self-adaptation, supporting higher-precision attribution and optimization suggestions for subsequent energy consumption anomaly events. Specifically, this includes:

[0172] S10.1: Collect and process the response characteristics of the process parameter optimization results, equipment adjustment measures and environmental compensation measures obtained in each round of virtual simulation in the digital twin sandbox to generate influence factor-result mapping knowledge tuples, and establish a benchmark for subsequent model knowledge base expansion.

[0173] S10.2: Based on the knowledge tuples of the influence factor-result mapping obtained by aggregation, an incremental knowledge update algorithm is used to dynamically iterate the knowledge base of the digital twin model, and automatically generate new sub-model parameters with adaptive characteristics to support the parallel evolution of model structure and parameters.

[0174] S10.3: Perform global consistency verification on the correlation between process and energy consumption for all newly added sub-model parameters in the knowledge base, and use a feature constraint detection mechanism to remove redundant or abnormal knowledge tuples to ensure the integrity and robustness of the digital twin model knowledge base.

[0175] S10.4: Utilize the updated and validated digital twin model knowledge base to reconstruct the causal chain of energy consumption anomalies, achieve adaptive optimization of the multidimensional impact tracing path of energy consumption anomaly events supported by new knowledge, and output the latest anomaly diagnosis mode.

[0176] S10.5: For new energy consumption anomalies discovered in the future, the evolved digital twin model knowledge base is used to automatically match the influencing factor patterns in historical projections, and the optimal intervention combination generation module is used to intelligently recommend correction strategies and expected business benefits, thus completing a high-precision closed-loop evolution between the model and reality.

[0177] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0178] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0179] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered 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. A method for intelligent analysis of energy consumption in industrial solid waste treatment, specifically including: S1: Collect raw datasets from multiple levels and locations within the industrial solid waste treatment production line; S2: Preprocess the collected raw dataset to standardize the historical and real-time states of different data channels into a standardized multidimensional data vector under the same time base; S3: Based on the multidimensional data vector, construct a physical-process-energy consumption digital twin mapping relationship according to the dimensions of process flow, equipment type, and loop partitioning, so as to realize the modeling of the digital twin model; S4: Input the standardized multidimensional data vector into the digital twin model to automatically extract energy consumption fluctuation features, equipment operating status features and process features to form a multivariable coupled feature set; Based on standardized multidimensional data vectors, a sliding window segmentation process is performed on continuous data from energy consumption metering points to obtain energy consumption fluctuation segments that reflect the dynamic changes in the solid waste treatment production line. For energy consumption fluctuation segments, extract the fundamental frequency and wavelet energy spectrum energy consumption fluctuation characteristic parameters of different frequency bands in each process section to generate an energy consumption fluctuation feature set that reflects the root cause of energy consumption anomalies; extract equipment operating status characteristics from the equipment operating parameter sequence within the sliding window. Based on the process flow variables and loop partitioning relationships within the digital twin model, hierarchical principal component analysis is performed on the process process variable sequence within the window to obtain the principal component features of the process. Using energy consumption fluctuation feature set, equipment operating status feature and process principal component feature as input, a multi-source feature-level fusion method is adopted to generate a multivariate coupled feature set; the contribution of various features in the multivariate coupled feature set is analyzed, and key multivariate coupled features that can characterize the precursors of energy consumption anomalies and their linkage with equipment-process-environment are selected, so as to output a high signal-to-noise ratio feature set for anomaly diagnosis and deep inference. S5: Based on the multivariate coupling feature set, determine whether the energy consumption data exceeds the threshold or has multi-source collaborative anomalies. If it is determined to be an energy consumption anomaly, record the abnormal event and the corresponding triggering process segment and equipment identifier. S6: After an abnormal event is triggered, the process, equipment, and environmental variable combinations containing multiple virtual scenario parameter groups are generated in the digital twin sandbox based on the recorded process sections, equipment identifiers, and historical operating conditions, and multi-scenario simulation reasoning is performed in the sandbox. S7: Collect the production capacity output, process stability, energy consumption chain, and emission parameter response path of the digital twin model under various combinations of virtual scenario variables, and perform a comparative analysis of differences; S8: Based on the response path analysis results of each virtual scenario, quantify the impact contribution of each abnormal factor on key business indicators such as energy consumption, production capacity, process stability and product quality, and output the impact ranking and business risk weight. S9: Based on the impact contribution results, iteratively adjust the parameter group and verify the correction strategy in the digital twin sandbox, and screen the optimal intervention combination that can take into account energy saving, production capacity and process stability and its expected effect.

2. The intelligent energy consumption analysis method for industrial solid waste treatment according to claim 1, characterized in that: Step S9 is followed by: S10: Feed the results of each round of sandbox simulation and intervention optimization back to the digital twin model to achieve continuous accumulation of knowledge about energy consumption anomalies and their multidimensional impacts and model adaptation, supporting higher-precision tracing of causes and output of optimization suggestions for subsequent energy consumption anomaly events.

3. The intelligent energy consumption analysis method for industrial solid waste treatment according to claim 1, characterized in that: The original dataset includes: energy consumption metering data, equipment operating parameters, process flow variables, production capacity indicators, and external environmental variables such as ambient temperature and humidity.

4. The intelligent energy consumption analysis method for industrial solid waste treatment according to claim 1, characterized in that: Step S9 involves iteratively adjusting the parameter group and sequentially verifying the correction strategy within the digital twin sandbox, including: process parameter optimization, equipment operating status adjustment, and environmental factor compensation.

5. The intelligent energy consumption analysis method for industrial solid waste treatment according to claim 1, characterized in that: Step S2 involves preprocessing the collected raw dataset, including filling in missing values, removing outliers, and performing multi-temporal alignment preprocessing.

6. The intelligent energy consumption analysis method for industrial solid waste treatment according to claim 1, characterized in that: Step S3 specifically includes: Cluster analysis is performed on the process flow variables in the multidimensional data vector to obtain process flow node groups that map to the physical production line structure; based on the process flow node groups, the operating parameters of each piece of equipment in the same physical production line are extracted, and the equipment classification coding method is applied to bind the equipment parameters with the process flow node identifier to generate a joint index of equipment type-process flow. For the joint index, the energy consumption metering values ​​of the collection loop partition level are collected. The partition many-to-one mapping mechanism is adopted to classify the energy consumption metering point data of each loop into the corresponding equipment type-process flow joint index and construct a physical-process-energy consumption three-dimensional mapping matrix. The three-dimensional mapping matrix is ​​used to integrate the equipment health, loop energy consumption distribution and process flow status in a time sequence, thereby obtaining a dynamic digital twin hierarchical structure that covers the entire process and multiple granularities. The dynamic digital twin hierarchical structure is output as a unified digital twin model object, and the high-precision digital correspondence between its physical entities, process nodes, equipment parameters and energy consumption metering points is expressed by nested logic.

7. The intelligent energy consumption analysis method for industrial solid waste treatment according to claim 1, characterized in that: Step S6 generates a combination of process, equipment, and environmental variables containing multiple virtual scenario parameter sets, including: multidimensionally recombining the process parameter set, equipment status parameter set, and environmental variable set to generate batch process parameter sets, equipment operating status sets, and environmental variable sets including normal operating conditions, boundary operating conditions, and extreme operating conditions.

Images