Intelligent factory dynamic optimization management system based on digital twinning and big data analysis

Abstract

The invention relates to the technical field of factory energy consumption management, in particular to a smart factory dynamic optimization management system based on digital twinning and big data analysis. Comprising a data acquisition and fusion module, a digital twinning construction module, a data analysis module, a dynamic optimization decision module and an anomaly diagnosis module. Constructing a digital twinborn model of a factory physical entity according to the collected data; constructing an energy consumption prediction model based on deep learning frameworks such as LTSM; when the energy consumption deviation exceeds the limit, abnormal root causes are positioned; and generating an energy consumption scheduling scheme based on a multi-objective optimization algorithm, and issuing an instruction to realize dynamic energy consumption adjustment. Through deep fusion of digital twinning and big data technologies, comprehensive and accurate simulation, multi-target collaborative optimization, rapid abnormality diagnosis and dynamic control of factory energy consumption are realized, the energy utilization efficiency is effectively improved, the cost is reduced, the intelligent level is improved, and the method has remarkable economic benefits and environmental benefits.

Description

Technical Field

[0001] The present invention relates to the technical field of factory energy consumption management, and specifically to a smart factory dynamic optimization management system based on digital twins and big data analysis. Background Art

[0002] With the rapid development of Industry 4.0 and intelligent manufacturing, factory production efficiency and capacity have significantly increased, but this has also led to increasingly prominent energy consumption issues. According to statistics, my country's manufacturing sector accounts for over 30% of the nation's total energy consumption. Energy waste is particularly severe in discrete factories due to their diverse equipment, complex processes, and fluctuating production schedules. Factory energy management has become a key concern for major companies and research institutions. Achieving energy conservation and consumption reduction while maintaining production efficiency is a key issue for the sustainable development of the manufacturing industry.

[0003] Traditional factory energy consumption management usually relies on manual statistics and empirical judgment, which is not only inefficient but also difficult to accurately grasp energy consumption data, making it impossible to achieve refined management and optimized control. Although existing factory energy consumption management systems have achieved the collection and analysis of energy consumption data to a certain extent, they still have the following shortcomings:

[0004] The lack of comprehensive digital modeling of the factory's physical entities makes it difficult to accurately simulate and predict energy consumption distribution, and is unable to reflect the coupling relationship between equipment and the impact of process parameter changes on energy consumption;

[0005] The data collection dimension is single, mainly focusing on energy consumption data such as electricity and water, and cannot fully consider the impact of multiple factors such as production process parameters, equipment operating status, and ambient temperature and humidity on energy consumption;

[0006] Most energy consumption optimization strategies are based on a single objective, making it difficult to achieve the coordinated optimization of multiple objectives such as energy consumption, cost, and equipment utilization, which can easily lead to poor results in actual applications.

[0007] The diagnosis and treatment of abnormal energy consumption lacks systematicity and intelligence, relying on manual investigation, making it difficult to quickly locate the root cause and take effective measures, resulting in delayed abnormal processing and energy waste.

[0008] In recent years, the rapid development of digital twin technology and big data analytics has provided new insights into factory energy management. Digital twin technology builds virtual models of physical entities, enabling real-time simulation and prediction of equipment, processes, and systems. Big data analytics, on the other hand, enables in-depth mining of multi-source, heterogeneous data to uncover hidden patterns in energy consumption and key influencing factors. However, integrating these two approaches to build a comprehensive, intelligent, and efficient factory energy optimization management system and methodology remains a pressing technical challenge. Summary of the Invention

[0009] In order to overcome the above-mentioned defects of the prior art, the present invention provides a smart factory dynamic optimization management system and method based on digital twins and big data analysis, which realizes precise management and optimized control of factory energy consumption through multi-source data fusion, digital twin modeling, intelligent analysis and prediction, and dynamic optimization decision-making.

[0010] Smart factory dynamic optimization management system based on digital twin and big data analysis, including:

[0011] The data acquisition and fusion module is used to collect multi-source heterogeneous data from the factory's sensors, equipment, MES, and ERP systems in real time, and to clean, process, and fuse them;

[0012] The digital twin construction module builds a high-fidelity digital twin model of equipment, products, processes, and environments based on the geometric information, physical properties, process logic, and historical operation data of physical entities. It also synchronizes the status of the digital twin model with the physical entity in real time based on real-time collected data.

[0013] A data analysis module is used to train an energy consumption prediction model based on historical data. The model uses an attention mechanism to enhance the weight learning of key influencing factors.

[0014] Dynamic optimization decision-making module, which is used to generate energy consumption scheduling strategies based on the simulation and deduction capabilities of the digital twin model using a multi-objective intelligent optimization algorithm and verify the effectiveness of the strategy through the digital twin model;

[0015] The abnormality diagnosis module is used to trigger causal reasoning analysis and automatically adjust equipment operating parameters when the deviation between actual energy consumption and the twin model prediction value exceeds a threshold.

[0016] Preferably, the digital twin building module includes:

[0017] A device-level twin unit configured to simulate the energy consumption characteristics of a single device, including a power-load curve and a start-stop energy loss model;

[0018] A process-level twin unit configured to model the energy consumption coupling relationship of different process links, wherein the coupling relationship is represented by a coupling coefficient matrix;

[0019] A system-level twin unit configured to integrate a plant-wide energy flow network, the network including energy conversion nodes, transmission paths, and load terminals;

[0020] Preferably, the data analysis module includes:

[0021] an energy consumption feature extraction unit configured to calculate the equipment energy efficiency entropy value and the process energy consumption coupling degree;

[0022] a prediction model training unit configured to optimize an LSTM network using an attention mechanism that calculates the importance of input features through a weight matrix;

[0023] A model calibration unit is configured to correct the parameters of the digital twin model in real time based on a Kalman filter algorithm, wherein the correction period is 100ms;

[0024] Preferably, the dynamic optimization decision module is configured as follows:

[0025] Construct a multi-objective optimization function including total energy consumption, energy cost, and equipment utilization;

[0026] Set production cycle constraints, equipment start and stop time constraints, and energy supply upper limit constraints;

[0027] An improved NSGA-III algorithm is used to solve the Pareto optimal solution set, wherein the improvement includes introducing a reference point adaptive mechanism.

[0028] The dynamic optimization management method for smart factories based on digital twins and big data analysis includes:

[0029] Collect factory equipment energy consumption data and production process parameters through smart meters and vibration sensors;

[0030] Build a digital twin model of the factory's physical entities, including equipment energy consumption characteristic equations and process association logic;

[0031] An energy consumption prediction model is constructed using an LSTM network based on an attention mechanism that calculates the importance of input features through a weight matrix.

[0032] When the actual energy consumption deviates from the twin prediction value by more than 10%, the causal reasoning process is triggered to locate the root cause of the anomaly;

[0033] generating an energy consumption scheduling plan based on a multi-objective optimization algorithm, wherein the algorithm adopts a reference point adaptive mechanism;

[0034] Send optimization instructions to the PLC system to achieve dynamic adjustment of energy consumption.

[0035] Preferably, the construction of a digital twin model of the factory physical entity includes:

[0036] Constructing a physical model of device-level energy consumption that determines the functional relationship between device power, load, and speed through mechanism analysis;

[0037] Constructing a process-level energy transfer model that quantifies the energy coupling effects between different process links;

[0038] Constructing a system-level energy flow network model that simulates the energy distribution and conversion process throughout the plant;

[0039] Preferably, the LSTM network based on the attention mechanism constructs an energy consumption prediction model, including:

[0040] Decompose the input features into time series and extract trend terms, period terms and residual terms;

[0041] Calculate the importance weights of different time steps through the self-attention mechanism;

[0042] The model is trained using a quantile loss function that simultaneously optimizes prediction intervals at different confidence levels.

[0043] When the prediction confidence is lower than 0.7, the model adaptive update is triggered.

[0044] Preferably, the causal reasoning process includes:

[0045] Construct a Bayesian network that includes equipment status, process parameters, and environmental variables;

[0046] Update network parameters based on real-time data and calculate the contribution of each factor to energy consumption anomalies using Shapley values.

[0047] When the contribution of a factor exceeds 30%, it is determined to be the root cause and adjustment suggestions are generated.

[0048] Preferably, the multi-objective optimization algorithm adopts an improved NSGA-III algorithm, and the improvement includes a reference point adaptive mechanism.

[0049] Compared with the prior art, the advantages of the present invention are:

[0050] By constructing a digital twin model of the factory's physical entity, this invention realizes comprehensive digital simulation and prediction of the factory's energy consumption, can accurately grasp the energy consumption distribution and changing trends, and provide a scientific basis for energy consumption optimization.

[0051] By using big data analysis technology, we conduct in-depth mining and analysis of multi-source heterogeneous data, extract key energy consumption characteristics, and improve the accuracy and reliability of energy consumption forecasts.

[0052] The introduction of attention mechanism and multi-objective optimization algorithm realizes the coordinated optimization of multiple objectives such as energy consumption, cost, and efficiency, improves energy utilization efficiency and reduces production costs.

[0053] Based on Bayesian networks and causal reasoning technology, rapid diagnosis and positioning of energy consumption anomalies are achieved, improving the reliability and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1This is the system architecture diagram of the smart factory dynamic optimization management system based on digital twin and big data analysis proposed in this invention. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0056] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0057] Example 1:

[0058] Reference Figure 1 The smart factory dynamic optimization management system based on digital twin and big data analysis provided in this embodiment includes a data acquisition and fusion module, a digital twin construction module, a data analysis module, a dynamic optimization decision module and an abnormality diagnosis module.

[0059] The data acquisition and fusion module includes smart meters, vibration sensors, and an edge computing gateway, configured to collect real-time data on factory equipment energy consumption, production process parameters, environmental parameters, and supply chain data. The smart meters collect equipment energy consumption data; the vibration sensors, with a range of 0-50g, monitor equipment operating status. The edge computing gateway, powered by an ARM Cortex-A53 processor, preprocesses and performs edge computing on the collected data, enabling data exchange with other devices via the Modbus protocol. Data preprocessing includes data cleaning, missing value interpolation, and outlier removal. Missing value interpolation uses cubic spline interpolation, while outlier removal uses the Tukey method based on the interquartile range.

[0060] The digital twin building block is configured to build a digital model of the factory's physical entities. This includes equipment-level twin units, process-level twin units, and system-level twin units. The equipment-level twin unit is configured to simulate the energy consumption characteristics of a single piece of equipment, including power-load curves and start-stop energy loss models.

[0061] In a preferred embodiment, the device-level twin unit is configured to simulate the energy consumption characteristics of motor-type devices, and its energy consumption characteristic equation is:

[0062]

[0063] Where, P rated is the rated power, η is the efficiency, T(t) is the real-time torque, T rated is the rated torque, n(t) is the real-time speed, n rated is the rated speed, a and b are exponential coefficients related to the motor type, E loss (t) is a function including iron loss, copper loss and mechanical loss.

[0064] The process-level twin unit is configured to model the energy consumption coupling relationship of different process links, which is represented by a coupling coefficient matrix.

[0065] In a preferred embodiment, the matrix elements of the coupling coefficient matrix are determined by the following formula:

[0066]

[0067] Where, X -ij Indicates the addition of E i and E j Other process parameters.

[0068] The system-level twin unit is configured to integrate the energy flow network of the entire plant. It is modeled using graph theory methods, abstracting energy conversion equipment as nodes and energy transmission paths as edges, and constructing a directed weighted graph G = (V, E, W), where the node v∈V represents the energy equipment, the edge e∈E represents the energy flow relationship, and the weight w∈W represents the energy conversion efficiency or transmission loss.

[0069] The data analysis module is configured to train an energy consumption prediction model based on historical data. The data analysis module also includes an energy consumption feature extraction unit, a prediction model training unit, and a model calibration unit. The energy consumption feature extraction unit is configured to calculate indicators such as the equipment energy efficiency entropy value, process energy consumption coupling degree, and carbon efficiency ratio. The calculation formula for the equipment energy efficiency entropy value (EES) is:

[0070]

[0071] Where p ij is the energy consumption ratio of equipment i in the jth working condition, which includes startup, standby, full load operation, etc. i When it is close to 1, it means that the energy consumption of the equipment is evenly distributed under various working conditions and the energy efficiency is stable. i When it is close to 0, it means that the equipment energy consumption is concentrated in a few working conditions and there is great potential for energy efficiency optimization.

[0072] The calculation formula of process energy consumption coupling (PEC) is:

[0073]

[0074] Where, Ea 、E b is the energy consumption sequence of process a and process b, Cov(·) is the covariance, and σ is the standard deviation. ab When ∣>0.7, it is determined to be strongly coupled. At this time, the adjustment of one process will significantly affect the energy consumption of the other process.

[0075] The prediction model training unit is configured to use the attention mechanism to optimize the LSTM network. Its structure includes an input layer, an attention layer, an LSTM layer, and an output layer. The attention layer calculates the importance weight of the input feature using the following formula:

[0076] α t =softmax(W a tanh(W h ·h t-1 +W x ·x t ));

[0077] Where x t is the input feature, h t-1 is the hidden state of the previous moment, W a 、W h 、W x is the trainable weight matrix, α t is the attention weight.

[0078] The model calibration unit is configured to correct the digital twin model parameters in real time based on the Kalman filter algorithm, with a correction period of 100ms.

[0079] The dynamic optimization decision module is configured to generate energy consumption scheduling strategies and verify their effectiveness through a digital twin model. Specifically, this involves constructing a multi-objective optimization function encompassing total energy consumption, energy cost, and equipment utilization, setting production cycle constraints, equipment start / stop time constraints, and energy supply upper limits, and employing an improved NSGA-III algorithm to find the Pareto optimal solution set.

[0080] Among them, the multi-objective optimization function is:

[0081]

[0082] Constraints include:

[0083]

[0084] Where R min 、R max are the minimum and maximum production cycles required by the process, T on,min 、T off,min Provided by the equipment manufacturer, P grid,max Determined by the grid agreement.

[0085] In a preferred embodiment, the improved NSGA-III algorithm introduces a reference point adaptation mechanism, which is implemented by the following steps:

[0086] Initialize the reference point set Z = {z 1 ,z 2 ,...z H ,}, where H is the number of reference points;

[0087] Calculate the distance d from each solution to each reference point i,j =‖x i -z j ‖;

[0088] Assign each solution to the nearest reference point and calculate the congestion degree δ of the reference point j ;

[0089] When the congestion of a reference point is too high, a new reference point is generated near it;

[0090] When a reference point is not assigned to a solution, it is deleted from the reference point set.

[0091] The anomaly diagnosis module is configured to trigger causal reasoning analysis and automatically adjust equipment operating parameters when the deviation between actual energy consumption and the twin model's prediction exceeds a threshold. Specifically, this involves building a Bayesian network that incorporates equipment status, process parameters, and environmental variables, updating network parameters based on real-time data, and calculating the contribution of each factor to energy anomalies. When the contribution exceeds a preset threshold, parameter adjustment instructions are generated and sent to the PLC system.

[0092] Example 2:

[0093] The smart factory dynamic optimization management method based on digital twin and big data analysis provided in this embodiment includes the following steps:

[0094] S1. Collect factory equipment energy consumption data and production process parameters using smart meters and vibration sensors. Specifically, a DDSY1352 smart meter collects equipment energy consumption data, and a vibration sensor with a range of 0-50g monitors the equipment's operating status at a frequency of 100Hz. The collected data is transmitted via industrial Ethernet to an edge computing gateway, where data cleaning, feature extraction, and compression are performed. A sliding window filter algorithm is used for data cleaning, and a wavelet transform is used for feature extraction to convert time-domain signals into frequency-domain signals and extract the energy characteristics of different frequency components.

[0095] S2. Build a digital twin model of the factory's physical entity, which includes equipment energy consumption characteristic equations and process-related logic. The steps of building a digital twin model include:

[0096] Construct a physical model of device-level energy consumption and determine the functional relationship between device power, load, and speed through mechanism analysis.

[0097] Establish a process-level energy consumption transfer model to quantify the energy consumption coupling effect between different process links. Use the system dynamics method to establish the process-level energy consumption transfer equation:

[0098]

[0099] Where, E i (t) is the energy consumption of process i, α ij is the energy consumption transfer coefficient of process j to process i, β i is the energy consumption attenuation coefficient of process i, γ i To control the input gain, u i (t) is the control input.

[0100] Develop a system-level energy flow network model to simulate the energy distribution and conversion process of the entire plant. Use the Energy Hub modeling method to unify the conversion, storage, and distribution processes of different energy forms:

[0101]

[0102] Where, E e 、E h 、E c are the electrical, thermal and cold energy outputs, E e,in 、E h,in 、E c,in are the electrical, thermal and cold energy inputs respectively, x→y is the conversion efficiency from energy form x to energy form y.

[0103] S3. Build an energy consumption prediction model using an LSTM network based on an attention mechanism. The mechanism calculates the importance of input features through a weight matrix. The steps of building an energy consumption prediction model include:

[0104] Decompose the input features into time series to extract trend terms, cycle terms, and residual terms. Use the STL (Seasonal and Trend decomposition using Loss) decomposition method:

[0105] y t =T t +S t +R t ;

[0106] Where y t is the original time series, T t is the trend term, S tis a periodic term, R t is the residual term.

[0107] The importance weights of different time steps are calculated through the self-attention mechanism, which is implemented by the following formula:

[0108]

[0109] Where Q, K, and V are query matrix, key matrix, and value matrix, respectively. k is the dimension of the key vector.

[0110] The model is trained using a quantile loss function that simultaneously optimizes prediction intervals at different confidence levels:

[0111]

[0112] Where τ is the percentile value, y i is the actual value, is the predicted value.

[0113] When the prediction confidence (EPC) is lower than 0.7, the model adaptive update is triggered. The prediction confidence calculation formula is:

[0114]

[0115] S4: When the actual energy consumption deviates from the twin prediction value by more than 10%, the causal reasoning process is triggered to locate the root cause of the anomaly. The causal reasoning process includes:

[0116] A Bayesian network is constructed that includes equipment status, process parameters, and environmental variables. The network structure is learned using a scoring search method, and the Bayesian Information Criterion is used as the scoring function.

[0117] Update network parameters based on real-time data and calculate the contribution of each factor to energy consumption anomalies.

[0118] When the contribution of a factor exceeds 30%, it is determined to be the root cause and adjustment suggestions are generated.

[0119] S5. Generate an energy consumption scheduling plan based on a multi-objective optimization algorithm, wherein the algorithm adopts a reference point adaptive mechanism. The objective function of the multi-objective optimization algorithm is:

[0120]

[0121] Constraints include production cycle requirements, equipment start and stop time windows, and energy supply limits.

[0122] S6. Send optimization instructions to the PLC system to achieve dynamic energy consumption adjustment. The optimization instructions are transmitted to the PLC system through the OPCUA protocol to adjust the operating parameters of the equipment, such as motor speed, heating temperature, pump flow, etc., to achieve dynamic optimization control of energy consumption.

[0123] Example 3:

[0124] This embodiment provides a specific implementation plan of a smart factory dynamic optimization management system and method based on digital twin and big data analysis in an electronic equipment manufacturing plant, including:

[0125] Data collection and preprocessing:

[0126] First, multi-source data acquisition equipment was deployed. Smart meters that met industrial standards were installed on key equipment in each production workshop. These meters collect detailed data such as active power, reactive power, apparent power, and energy consumption in real time, supporting multiple communication protocols for easy data transmission. Furthermore, vibration sensors were installed at key locations on the equipment. MEMS vibration sensors, with a sensitivity of up to 100mV / g and a measurement range of ±5g, effectively capture vibration during operation, reflecting the equipment's operational status and potential faults. The edge computing gateway, an industrial-grade device equipped with a high-performance multi-core processor, communicates with various devices via the fieldbus. Data is collected in real time according to a set sampling period (every 5 seconds). The gateway then pre-processes the raw data using a wavelet threshold denoising algorithm to remove noise and improve data quality. The processed data is then transmitted to the data center via an encrypted industrial Ethernet channel.

[0127] Digital twin model construction:

[0128] Based on the collected data and the actual physical layout, process flow and other information of the factory, the twin units at each level in the digital twin modeling device are constructed.

[0129] In the equipment-level twin unit, for core equipment such as CNC machine tools, detailed energy consumption characteristic equations are constructed based on their mechanical transmission principles, electrical control system characteristics, and heat conduction mechanisms. For example, considering the impact of cutting force, spindle speed, and feed rate on energy consumption during tool cutting, the following energy consumption model is established:

[0130] E CNC (t) = k1·F c (t)·v(t)+k2·n(t) 3 +k3·ΔT(t)+E idle (t);

[0131] Where, ECNC (t) is the energy consumption of the CNC machine tool at time t, F c (t) is the cutting force, v(t) is the feed rate, n(t) is the spindle speed, ΔT(t) is the temperature change of the equipment caused by processing (data is obtained by installing temperature sensors at the key heat dissipation parts of the equipment), k1, k2, k3 are coefficients determined by experiments and theoretical analysis, E idle (t) is the basic energy consumption when the device is in standby mode.

[0132] For placement machines, an energy consumption model is constructed by considering factors such as the action frequency of the suction nozzle picking up components, the movement stroke, and the power of the vacuum pump, so as to accurately simulate the energy consumption characteristics of each device.

[0133] In the process-level twin unit, the energy consumption coupling relationship between screen printing, SMT, and reflow soldering processes in the circuit board assembly process is analyzed. For example, the ink thickness and printing speed of the screen printing process will affect the accuracy of component placement in the subsequent SMT process, which in turn affects the operating time and energy consumption of the SMT machine. The coupling coefficient between these processes is determined through regression analysis of large amounts of production data. Using the partial least squares (PLS) method, a correlation model is constructed between the energy consumption of each process to accurately characterize the degree of mutual influence of energy consumption in different process links.

[0134] In the system-level twin unit, the entire factory's power supply system, gas supply system, water supply system and other energy supply networks, as well as various production workshops and warehouses, are used as nodes, and energy transmission pipelines and lines are used as edges to construct a complex directed weighted energy flow network diagram. The energy conversion efficiency and transmission loss coefficient of each energy conversion equipment are marked in the diagram to comprehensively present the distribution, circulation and consumption of energy in the entire factory.

[0135] Data analysis and energy consumption forecasting:

[0136] The data analysis module conducts in-depth mining of historical data stored in the distributed data warehouse. The energy consumption feature extraction unit calculates the energy efficiency entropy (EES) of different devices. For example, it finds that the EES value of a batch of older test equipment is low, indicating that its energy consumption distribution is concentrated under a few specific operating conditions, indicating that there is significant room for energy efficiency optimization. When calculating the process energy consumption coupling (PEC), it is found that the PEC value between the wave soldering process and the subsequent cleaning process is high, indicating that adjusting the wave soldering process parameters will significantly affect the energy consumption of the cleaning process, and it is necessary to focus on the coordinated optimization of these two process links.

[0137] The prediction model training unit uses an LSTM network integrated with an attention mechanism to build an energy consumption prediction model. Within the attention mechanism, a multi-head attention mechanism is implemented to enhance focus on different feature dimensions. The model is trained using extensive historical data, with root mean square error (RMSE) and mean absolute percentage error (MAPE) used as evaluation metrics. Model parameters are continuously optimized to ensure that the model can accurately predict energy consumption across equipment and process steps even in the face of complex production conditions.

[0138] Abnormal diagnosis and dynamic optimization:

[0139] When the deviation between actual energy consumption and the energy consumption predicted by the digital twin model exceeds a set threshold of 8%, the anomaly diagnosis process is initiated. For example, during one production run, the actual energy consumption of a particular production line was found to be significantly higher than predicted. The Bayesian network inference module in the anomaly diagnosis module quickly intervened. This Bayesian network is constructed based on the factory's long-term accumulated historical data, including equipment failure records, process parameter anomalies, and changes in environmental factors. It incorporates numerous node variables, such as the degree of wear on key equipment components, the fluctuation range of process parameters, and the workshop humidity and temperature. By updating the network parameters in real time and using variational Bayesian inference methods to calculate the contribution of each factor to the energy consumption anomaly, the team determined that the recent increase in workshop humidity had caused the insulation performance of some electrical equipment to deteriorate, resulting in leakage, which in turn led to increased energy consumption. This factor contributed approximately 40% to the anomaly, making it the root cause.

[0140] Subsequently, the dynamic optimization decision module constructs a multi-objective optimization function that includes total energy consumption, energy costs, and comprehensive equipment performance, based on current energy consumption anomalies, production plans, equipment status, and other factors. It sets constraints such as the production cycle requirements for each production process, the minimum continuous operating time of the equipment, and the minimum downtime of the equipment. It then uses an improved particle swarm optimization algorithm (introducing adaptive inertia weights and a dynamic shrinkage factor mechanism) to solve the Pareto optimal solution set and generate an energy consumption scheduling strategy tailored to the current situation. For example, it adjusts the operating order of some equipment, appropriately reduces the operating power of some non-critical process equipment, and increases the operating time of equipment that controls ambient humidity.

[0141] Finally, optimization instructions are accurately sent to the corresponding PLC control system through the OPC UA secure communication protocol. The PLC system accurately adjusts the operating parameters of each device based on the instructions, such as the spindle speed of the CNC machine tool, the placement speed of the placement machine, and the dehumidification power of the air conditioning system, thereby achieving dynamic optimization and adjustment of the energy consumption of the entire factory.

[0142] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0143] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A smart factory dynamic optimization management system based on digital twins and big data analysis, characterized by: include: The data acquisition and fusion module is used to collect multi-source heterogeneous data from the factory's sensors, equipment, MES, and ERP systems in real time, and to clean, process, and fuse them; The digital twin construction module is used to build a high-fidelity digital twin model of equipment, products, processes, and environments based on the geometric information, physical properties, process logic, and historical operation data of physical entities. It also synchronizes the status of the digital twin model with the physical entity in real time based on real-time collected data. A data analysis module is used to train an energy consumption prediction model based on historical data. The model uses an attention mechanism to enhance the weight learning of key influencing factors. Dynamic optimization decision-making module, which is used to generate energy consumption scheduling strategies based on the simulation and deduction capabilities of the digital twin model using a multi-objective intelligent optimization algorithm and verify the effectiveness of the strategy through the digital twin model; The abnormality diagnosis module is used to trigger causal reasoning analysis and automatically adjust equipment operating parameters when the deviation between actual energy consumption and the twin model prediction value exceeds a threshold.

2. The smart factory dynamic optimization management system based on digital twin and big data analysis according to claim 1 is characterized in that: The digital twin building blocks include: A device-level twin unit configured to simulate the energy consumption characteristics of a single device, including a power-load curve and a start-stop energy loss model; A process-level twin unit configured to model the energy consumption coupling relationship of different process links, wherein the coupling relationship is represented by a coupling coefficient matrix; A system-level twin unit is configured to integrate a plant-wide energy flow network, including energy conversion nodes, transmission paths, and load terminals.

3. The smart factory dynamic optimization management system based on digital twin and big data analysis according to claim 1 is characterized in that: The data analysis module includes: an energy consumption feature extraction unit configured to calculate the equipment energy efficiency entropy value and the process energy consumption coupling degree; a prediction model training unit configured to optimize an LSTM network using an attention mechanism that calculates the importance of input features through a weight matrix; The model calibration unit is configured to correct the digital twin model parameters in real time based on the Kalman filter algorithm, and the correction period is 100ms.

4. The smart factory dynamic optimization management system based on digital twin and big data analysis according to claim 1 is characterized in that: The dynamic optimization decision module is configured as follows: Construct a multi-objective optimization function including total energy consumption, energy cost, and equipment utilization; Set production cycle constraints, equipment start and stop time constraints, and energy supply upper limit constraints; An improved NSGA-III algorithm is used to solve the Pareto optimal solution set, wherein the improvement includes introducing a reference point adaptive mechanism.

5. A smart factory dynamic optimization management method based on digital twins and big data analysis is characterized by: include: Collect factory equipment energy consumption data and production process parameters through smart meters and vibration sensors; Build a digital twin model of the factory's physical entities, including equipment energy consumption characteristic equations and process association logic; An energy consumption prediction model is constructed using an LSTM network based on an attention mechanism that calculates the importance of input features through a weight matrix. When the actual energy consumption deviates from the twin prediction value by more than 10%, the causal reasoning process is triggered to locate the root cause of the anomaly; generating an energy consumption scheduling plan based on a multi-objective optimization algorithm, wherein the algorithm adopts a reference point adaptive mechanism; Send optimization instructions to the PLC system to achieve dynamic adjustment of energy consumption.

6. The method for dynamic optimization management of smart factories based on digital twins and big data analysis according to claim 5 is characterized in that: The construction of a digital twin model of the factory physical entity includes: Constructing a physical model of device-level energy consumption that determines the functional relationship between device power, load, and speed through mechanism analysis; Constructing a process-level energy transfer model that quantifies the energy coupling effects between different process links; A system-level energy flow network model is constructed, which simulates the energy distribution and conversion process of the entire plant.

7. The method for dynamic optimization management of smart factories based on digital twins and big data analysis according to claim 5 is characterized in that: The energy consumption prediction model constructed by the LSTM network based on the attention mechanism includes: Decompose the input features into time series and extract trend terms, period terms and residual terms; Calculate the importance weights of different time steps through the self-attention mechanism; The model is trained using a quantile loss function that simultaneously optimizes prediction intervals at different confidence levels. When the prediction confidence is lower than 0.7, the model adaptive update is triggered.

8. The method for dynamic optimization management of smart factories based on digital twins and big data analysis according to claim 5 is characterized in that: The causal reasoning process includes: Construct a Bayesian network that includes equipment status, process parameters, and environmental variables; Update network parameters based on real-time data and calculate the contribution of each factor to energy consumption anomalies using Shapley values. When the contribution of a factor exceeds 30%, it is determined to be the root cause and adjustment suggestions are generated.

9. The method for dynamic optimization management of smart factories based on digital twins and big data analysis according to claim 5 is characterized in that: The multi-objective optimization algorithm adopts an improved NSGA-III algorithm, and the improvement includes a reference point adaptive mechanism.

Images