Warehouse virtual simulation scheduling method and system based on digital twinning

Abstract

The invention relates to the technical field of digital twinning, in particular to a warehouse virtual simulation scheduling method and system based on digital twinning, and the method comprises the steps: collecting multi-source service data, such as the operation age limit, accumulated ultraviolet irradiation, historical thermal load circulation and current impact, of a decommissioned intelligent electric meter; predicting a shell friction coefficient through a material evolution model, and executing fatigue matching and thermal displacement correction to fit a centroid deviation coordinate; the fitting parameters are injected into the virtual entity model to execute acceleration gradient simulation, a dynamic distribution cloud picture for guiding speed limitation is generated, and a nonlinear prediction compensation value is calculated in combination with virtual and real pose deviation fusion; and finally, a cloud picture safety threshold value is extracted, a moving track is corrected, a dynamic execution torque is solved based on friction coefficient and centroid deviation constraint, and a sorting mechanism is driven to execute operation. According to the invention, through digital twinning attribute reconstruction, accurate self-adaptive sorting of the retired ammeters is realized.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to a warehouse virtual simulation scheduling method and system based on digital twins. Background Technology

[0002] With the deepening of smart grid construction, a large number of smart meters that have reached the end of their service life have entered the stage of recycling, testing, and reuse. In order to improve disposal efficiency, automated warehouse sorting systems are widely used for centralized processing of smart meters. In existing automated sorting operations, digital twin technology and virtual simulation methods are usually used. By establishing geometric and mechanical models of smart meters in virtual space, the movement paths and grasping actions during the sorting process are simulated in advance, thereby generating control commands to guide the physical sorting equipment to perform operations.

[0003] However, in actual operations, the electricity meters to be sorted are not brand new, but retired equipment that has been in service for several years. During long-term operation, the physical state of the electricity meters will undergo irreversible evolution due to the combined effects of various factors such as service environment and workload. Traditional virtual simulation scheduling methods are usually based on standardized physical parameters (such as rated friction coefficient, nominal centroid coordinates, etc.) under ideal conditions for modeling.

[0004] The lack of effective characterization of individual physical state deviations of electricity meters leads to significant twin distortions between the theoretical dynamic response generated by the virtual simulation system and the actual operating state on the physical sorting line. This distortion can cause problems such as grasping instability and trajectory deviation in sorting robots during high-acceleration operation or grasping due to the failure of dynamic constraint matching, which in turn affects the scheduling accuracy and operational reliability of the warehouse automated sorting system.

[0005] Therefore, how to solve the scheduling and control mismatch caused by the deviation of individual dynamic parameters due to the long-term service of the objects to be sorted in the process of high-precision automated sorting is a technical problem that urgently needs to be solved in the field of power logistics automation.

[0006] To address this, a warehouse virtual simulation scheduling method and system based on digital twins is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a warehouse virtual simulation scheduling method and system based on digital twins. Through digital twin attribute reconstruction, it achieves precise adaptive sorting of retired smart meters. This includes collecting multi-source service data such as the operating years, cumulative ultraviolet radiation, historical heat load cycles, and current surges of retired smart meters; predicting the outer shell friction coefficient using a material evolution model and performing fatigue matching and thermal displacement correction to fit the centroid deviation coordinates; injecting the fitted parameters into a virtual entity model to perform acceleration gradient simulation, generating a dynamic distribution cloud map to guide speed limits, and combining the virtual and real pose deviations to calculate nonlinear prediction compensation values; finally, extracting the cloud map safety threshold and correcting the running trajectory, and calculating the dynamic execution torque based on the friction coefficient and centroid deviation constraints to drive the sorting mechanism to perform operations.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A warehouse virtual simulation scheduling method based on digital twins includes: Acquire data on the operating years of smart meters, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current surge data; Based on the years of operation data and cumulative ultraviolet radiation dose, the predicted friction coefficient of the smart meter casing is calculated through a material evolution model. Combined with historical thermal load cycle data and cumulative load current impact data, fatigue matching and thermal displacement correction are performed to obtain the fitted centroid deviation coordinates. The predicted friction coefficient and the fitted centroid deviation coordinates are assigned to the virtual entity model. By performing multi-group sorting acceleration gradient simulation, a dynamic distribution cloud map is generated. The real-time pose data of the physical sorting line is obtained. The deviation between the real-time pose data and the theoretical pose data synchronously output by the virtual entity model is compared and fused to obtain the nonlinear prediction compensation value. Acceleration safety thresholds are extracted from dynamic distribution cloud maps, and nonlinear prediction compensation values ​​are combined to generate trajectory correction values. The dynamic execution torque of the sorting mechanism is calculated based on the predicted friction coefficient and the fitted centroid deviation coordinates. The trajectory correction values ​​and dynamic execution torque are summarized to generate scheduling control commands, which are then sent to the sorting driver to complete the sorting operation of smart meters.

[0009] Preferably, the process of obtaining the operating years data, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current impact data of the smart meter includes: scanning the barcode on the smart meter casing using an industrial vision sensor deployed at the front end of the sorting line to obtain a unique identifier; matching the unique identifier with the historical installation latitude and longitude coordinates and the daily solar radiation intensity in the historical meteorological database, and performing cumulative calculations in conjunction with a preset environmental shading coefficient to obtain the cumulative ultraviolet radiation dose; retrieving the historical current operating curve from the electricity information collection system and using a peak detection algorithm to extract the frequency of pulse events exceeding a preset rated current threshold to obtain the cumulative load current impact data; statistically analyzing the number of ambient temperature temperature difference cycles and the Joule thermal stress cycle characteristics caused by the load current during the smart meter's service life to obtain the historical heat load cycle data; and obtaining the operating years data by counting the cumulative number of days between the commissioning date and the dismantling date corresponding to the unique identifier recorded in the management system.

[0010] Preferably, the material evolution model includes: a baseline aging encoder, which receives the operating years data as input, processes it using a first multilayer perceptron, and generates a baseline aging feature vector; The stress intensity encoder receives the cumulative ultraviolet radiation dose and operating years data as input, calculates the derivative mapping of cumulative radiation dose with operating years, and generates a stress intensity feature vector that reflects the pulverization rate of the material surface. The response encoder receives the preset material attributes associated with the unique identifier as input, processes them using an embedded lookup table, transforms discrete material categories into continuous spatial constitutive vectors, and generates material constitutive feature vectors. The recalibration unit concatenates the stress intensity feature vector with the material constitutive feature vector and inputs it into the attention mechanism network to calculate the damage gain factor. The damage gain factor is then applied to the reference aging feature vector for dynamic recalibration, and the predicted friction coefficient of the smart meter casing is output.

[0011] Preferably, the process of obtaining the fitted centroid deviation coordinates includes: matching the number of cycles in the historical thermal load cycle data with a preset material fatigue damage index table to obtain the loosening displacement of the internal fasteners; performing linear interpolation calculation on the cumulative load current impact data to obtain the thermal deviation correction value of the internal heavy-duty components of the smart meter under thermal stress conditions; vector superimposing the loosening displacement and the thermal deviation correction value to obtain the actual offset vector of the internal heavy-duty components relative to the original installation position; compensating the actual offset vector to the original coordinates of the heavy-duty components retrieved from the standard three-dimensional model, and calculating the static moment disturbance deviation in combination with the mass of each component to obtain the fitted centroid deviation coordinates.

[0012] Preferably, the process of generating the dynamic distribution cloud map includes: initializing the digital twin attributes of a specific meter entity by assigning the predicted friction coefficient and the fitted centroid deviation coordinates to the virtual entity model; constructing a standard sorting path model consisting of straight segments, circular turning segments, and fork diversion segments in the virtual environment, and driving the virtual entity model to perform dynamic simulations under multiple speed gradients; determining the sorting performance boundaries of the meter under different sorting actions by extracting the critical acceleration values ​​of the virtual entity model when it experiences sideslip, overturning, and derailment in each path segment; and generating the dynamic distribution cloud map that guides actual scheduling speed limits by projecting the critical acceleration values ​​corresponding to each path segment onto the topology layout diagram of the sorting equipment and marking the safe speed limits of different areas.

[0013] Preferably, the process of obtaining the nonlinear prediction compensation value includes: acquiring the real-time spatial coordinates and offset angle of the smart meter through a visual sensor deployed on the sorting line, as the real-time pose data of the physical entity; calculating the spatiotemporal deviation vector of the real-time pose data relative to the theoretical pose of the virtual entity model, and extracting the variance fluctuation frequency of the sensor signal to evaluate the environmental noise intensity; comparing the environmental noise intensity with the parameter iteration convergence of the virtual entity model to calculate the pose confidence score reflecting the credibility ratio of virtual and real data; using the pose confidence score as a weighting factor to perform weighted fusion of the real-time measurement value and the theoretical prediction value, and calculating the nonlinear prediction compensation value in combination with the current running speed of the sorting line.

[0014] Preferably, the process of sending the scheduling control command to the sorting driver includes: extracting the acceleration safety threshold of the current path segment based on the dynamic distribution cloud map, and correcting the theoretical pose by offset in combination with the nonlinear prediction compensation value to obtain the running trajectory correction amount; calculating the dynamic execution torque required for the sorting mechanism to perform actions based on the running trajectory correction amount and in combination with the physical constraints determined by the predicted friction coefficient and the fitted centroid deviation coordinates; summarizing the running trajectory correction amount and the dynamic execution torque to generate the final scheduling control command; and sending the scheduling control command to the sorting driver to complete the sorting operation of the smart meter by adjusting the output and position of the drive motor.

[0015] A warehouse virtual simulation scheduling system based on digital twins includes: Data acquisition module: acquires data on the operating years of smart meters, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current surge data; State fitting module: Based on the years of operation data and cumulative ultraviolet radiation dose, the predicted friction coefficient of the smart meter casing is calculated through the material evolution model. Combined with historical heat load cycle data and cumulative load current impact data, fatigue matching and thermal displacement correction are performed to obtain the fitted centroid deviation coordinates. Twin simulation module: Assigns the predicted friction coefficient and the fitted centroid deviation coordinates to the virtual entity model, generates a dynamic distribution cloud map by performing multi-group sorting acceleration gradient simulation; acquires the real-time pose data of the physical sorting line, compares the deviation between the real-time pose data and the theoretical pose data synchronously output by the virtual entity model and fuses them to obtain the nonlinear prediction compensation value. The scheduling and execution module extracts the acceleration safety threshold based on the dynamic distribution cloud map, generates the running trajectory correction amount by combining the nonlinear prediction compensation value, calculates the dynamic execution torque of the sorting mechanism based on the predicted friction coefficient and the fitted centroid deviation coordinates, summarizes the running trajectory correction amount and the dynamic execution torque to generate scheduling control instructions, and sends the scheduling control instructions to the sorting driver to complete the sorting operation of smart meters.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By integrating data such as years of operation, irradiation, and thermoelectric cycles, and using material evolution models and fatigue matching algorithms, a quantitative analysis of the deviation between the predicted friction coefficient and the fitted centroid of retired meters was achieved. This parameter reconstruction method reduces the difference in physical properties between the digital twin model and the physical entity, and reduces the initial calculation deviation caused by model parameter distortion.

[0017] 2. By mapping the fitted physical parameters to the simulation environment and performing multiple sets of acceleration gradient simulations, a dynamic distribution cloud map characterizing the operating limits of the equipment is generated. Based on the existing friction and center of gravity distribution of each meter, differentiated sorting acceleration critical values ​​can be determined. Under the premise of ensuring that the sorting action does not exceed the physical stability limit, the sorting efficiency of meters in good condition is improved, and the impact of the uncertainty of the physical state of retired equipment on the reliability of system operation is mitigated.

[0018] 3. By comparing the spatiotemporal deviations between real-time pose data and theoretical pose data, confidence-based weighted fusion and delay correction are performed, improving the accuracy of virtual-real synchronization during the sorting process. Compensation values ​​are calculated to reduce the trajectory lag of the actuator under dynamic operation. Even with fluctuations in sensor signals, a relatively stable pose correction value can still be output, enhancing the anti-interference capability of the closed-loop control system in high-speed operating environments.

[0019] 4. When generating scheduling instructions, friction coefficient constraints and center of mass offset characteristics are introduced into the dynamic solution process to calculate the dynamic execution torque. Based on the actual mass distribution characteristics of the meter, the torque output distribution of each degree of freedom can be optimized. By adjusting the output torque, the gravitational offset torque caused by the center of mass offset is offset, which reduces the overturning tendency of the meter during acceleration or gripping, makes the physical force state during sorting more balanced, and improves the mechanical operation stability under high-speed sorting conditions. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of a warehouse virtual simulation scheduling method based on digital twins according to the present invention. Figure 2 This is a schematic diagram of the material evolution model process of the present invention; Figure 3 This is a schematic diagram of the structure of a warehouse virtual simulation scheduling system based on digital twins according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figures 1 to 3 This invention provides a warehouse virtual simulation scheduling method and system based on digital twins, the technical solution of which is as follows: Example 1: A warehouse virtual simulation scheduling method based on digital twins, specifically as follows: Figure 1 As shown, it includes: Acquire data on the operating years of smart meters, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current surge data; Based on the years of operation data and cumulative ultraviolet radiation dose, the predicted friction coefficient of the smart meter casing is calculated through a material evolution model. Combined with historical thermal load cycle data and cumulative load current impact data, fatigue matching and thermal displacement correction are performed to obtain the fitted centroid deviation coordinates. The predicted friction coefficient and the fitted centroid deviation coordinates are assigned to the virtual entity model. By performing multi-group sorting acceleration gradient simulation, a dynamic distribution cloud map is generated. The real-time pose data of the physical sorting line is obtained. The deviation between the real-time pose data and the theoretical pose data synchronously output by the virtual entity model is compared and fused to obtain the nonlinear prediction compensation value. Acceleration safety thresholds are extracted from dynamic distribution cloud maps, and nonlinear prediction compensation values ​​are combined to generate trajectory correction values. The dynamic execution torque of the sorting mechanism is calculated based on the predicted friction coefficient and the fitted centroid deviation coordinates. The trajectory correction values ​​and dynamic execution torque are summarized to generate scheduling control commands, which are then sent to the sorting driver to complete the sorting operation of smart meters.

[0023] Furthermore, the process of acquiring the operating years data, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current impact data of the smart meter includes: scanning the barcode on the smart meter casing using an industrial vision sensor deployed at the front end of the sorting line to obtain a unique identifier; matching the unique identifier with the historical installation latitude and longitude coordinates and the daily solar radiation intensity in the historical meteorological database, and performing cumulative calculations in conjunction with a preset environmental shading coefficient to obtain the cumulative ultraviolet radiation dose; retrieving the historical current operating curve from the electricity information collection system and using a peak detection algorithm to extract the frequency of pulse events exceeding a preset rated current threshold to obtain the cumulative load current impact data; statistically analyzing the number of ambient temperature temperature difference cycles and the Joule thermal stress cycle characteristics caused by the load current during the smart meter's service life to obtain the historical heat load cycle data; and obtaining the operating years data by counting the cumulative number of days between the commissioning date and the dismantling date corresponding to the unique identifier recorded in the management system.

[0024] Specifically, the sorting line is equipped with industrial-grade high-resolution vision sensors. When a smart meter to be sorted passes through the scanning station, the sensor captures the barcode or QR code on its casing. The image processing algorithm identifies and extracts the meter's unique identification code (i.e., unique identifier). This identifier serves as the index key for all subsequent data retrievals, triggering the association query between the asset management system and the production operation system. Based on the obtained unique identifier, the system retrieves the meter's historical service location (such as latitude and longitude coordinates) from the asset management platform. By associating these coordinates with the historical meteorological database, the system obtains the daily solar radiation intensity data from the date the meter was put into operation until the date it was removed. During the calculation process, the system introduces a preset environmental shading coefficient to characterize the solar radiation attenuation experienced by the meter under actual installation conditions (such as meter box shading or installation in the shade). The daily solar radiation intensity is multiplied by the shading coefficient to obtain the daily actual irradiance value. Finally, all the daily actual irradiance values ​​within the service period are accumulated to obtain the final cumulative ultraviolet radiation dose.

[0025] Accessing the electricity information collection system via a unique identifier, the historical load current curves during the meter's service life are retrieved. A preset current threshold (1.2 times the rated current in this embodiment) is set. Using a peak detection algorithm, all instantaneous pulse events exceeding this threshold are identified in the historical current sequence. By statistically analyzing the total frequency of these pulse events throughout the entire service life, cumulative load current impact data is generated. This indicator reflects the degree of electromagnetic force impact and thermal stress on the meter's internal electronic components and conductive contacts caused by instantaneous high current. Thermal cycle data is determined by integrating ambient temperature changes and internal heating conditions: the number of daily high and low temperature alternations in historical meteorological data is statistically analyzed to determine the frequency of ambient temperature difference cycles. Based on the historical current operating curves, the internal heating and cooling processes caused by large current fluctuations are identified. These two types of temperature change factors are superimposed to identify the total number of physical fatigue cycles formed by the alternating contraction and expansion of materials with different thermal expansion coefficients inside the meter, thereby obtaining historical thermal load cycle data. The commissioning date and dismantling date recorded in the management system are directly retrieved, and the cumulative number of natural days between the two dates is calculated as the basic time scale for meter material aging.

[0026] By utilizing industrial vision sensors to obtain unique identifiers and linking them with multi-source data such as historical geographical location, meteorological environment, and power load, a quantitative characterization of the service background of the meters to be sorted is achieved. Objective data such as irradiance, temperature difference cycle, and current pulse frequency are used to replace the single assessment of commissioning time, providing differentiated data input for calculating the degree of aging of the casing and the loosening of internal fasteners. By restoring the actual stress and aging history of the meters, random deviations in the parameter fitting process are reduced, laying the foundation for generating accurate scheduling and control commands.

[0027] Furthermore, the material evolution model includes: a baseline aging encoder, which receives the operating years data as input, processes it using a first multilayer perceptron, and generates a baseline aging feature vector; The stress intensity encoder receives the cumulative ultraviolet radiation dose and operating years data as input, calculates the derivative mapping of cumulative radiation dose with operating years, and generates a stress intensity feature vector that reflects the pulverization rate of the material surface. The response encoder receives the preset material attributes associated with the unique identifier as input, processes them using an embedded lookup table, transforms discrete material categories into continuous spatial constitutive vectors, and generates material constitutive feature vectors. The recalibration unit concatenates the stress intensity feature vector with the material constitutive feature vector, inputs the result to an attention mechanism network to calculate a damage gain factor, applies the damage gain factor to the baseline aging feature vector for dynamic recalibration, and outputs the predicted friction coefficient of the smart meter casing, as detailed below. Figure 2 As shown.

[0028] After acquiring the data, a pre-trained material evolution model is invoked to quantitatively predict the real-time friction characteristics of the smart meter casing during sorting operations. The specific implementation process is as follows: The acquired operating age data is input into a benchmark aging encoder. This encoder internally constructs a three-layer nonlinear mapping network. The input layer first performs Z-score standardization on the operating age data, that is, subtracting the mean of the training set from the original data and dividing by the standard deviation to normalize the feature distribution. The processed data is mapped to a first hidden layer with 64 neurons. Between the first and second hidden layers (containing 32 neurons), the ReLU linear rectified function is used as the activation function. The output layer uses the Sigmoid activation function to map the result to the interval between 0 and 1 for subsequent feature processing.

[0029] The stress intensity encoder quantifies the degree of damage caused by the external environment. The encoder simultaneously receives data on cumulative ultraviolet radiation dose and years of operation. Specifically, the encoder calculates the rate of change of cumulative radiation relative to the years of service by performing numerical difference calculation (or performing first-order difference calculation for the annual radiation sequence during the service period) by dividing the total cumulative radiation dose by the total years of operation. This reflects the average radiation intensity gradient experienced by the meter during its service period. After normalization, this gradient scalar is fed into a single-layer linear network containing 16 neurons for feature mapping. Through linear transformation performed by the learning weights and bias parameters within the network, it is converted into a 16-dimensional stress intensity feature vector that reflects the pulverization rate of the material surface caused by the influence of light. This process effectively captures the dynamic destructive effect of environmental stress on the physical properties of the material, thereby distinguishing the surface roughness differences caused by different environmental loads under the same years of service.

[0030] The encoder begins parsing the physical constitutive properties of the meter. Based on the meter's unique identifier, it retrieves the original factory parameters from a pre-set device attribute database and converts the material category (such as flame-retardant polycarbonate or ABS engineering plastic) into a corresponding discrete index code. Subsequently, using a pre-set embedding lookup table with N rows and 32 columns, the index code is mapped to a continuously distributed 32-dimensional constitutive vector. During the model training phase, this embedding lookup table is updated with parameters through a backpropagation algorithm, resulting in higher cosine similarity between materials with similar physicochemical properties in the vector space. This vector implicitly contains physical property constraints such as the tensile strength, heat distortion temperature, and photochemical stability of the specific material itself, thereby generating a material constitutive feature vector. This design ensures that the model can identify the differentiated mechanical performance evolution of materials with different constitutive properties when facing the same external stress.

[0031] The recalibration unit performs deep coupling of features and predicts the friction coefficient. This unit horizontally concatenates the 16-dimensional stress intensity feature vector and the 32-dimensional material constitutive feature vector in the feature dimension to form a 48-dimensional comprehensive stress response feature vector. This vector is then input into a single-head attention mechanism network. The single-head attention mechanism network calculates a vector with 16 weight centers that is consistent with the dimension of the baseline aging feature vector through an internal learnable linear transformation matrix. This vector serves as a damage gain factor that reflects the actual destructive effect of the current environmental stress on the specific material. Subsequently, this factor is applied to the initially generated 16-dimensional baseline aging feature vector and element-wise multiplication is performed to complete the weight correction and dynamic recalibration. The corrected feature vector is then dimensionally reduced through a fully connected layer and mapped to a scalar value. Finally, a nonlinear mapping is performed through the Sigmoid function, and a value that conforms to the physical dimensions and is within a preset friction range is output as the predicted friction coefficient of the smart meter casing. The physical principle of this prediction model is based on the aging evolution mechanism of the polymer material of the casing of retired electricity meters. Because ultraviolet radiation causes polymer molecular chains to break and gradually produce a powdery layer on the surface, it initially increases the surface micro-roughness and thus increases the friction coefficient. However, as aging intensifies, the powdery layer may peel off to expose a smooth inner layer, causing the friction coefficient to drop. This model uses the characteristics of the Sigmoid function, which is steep in the middle and saturated at both ends, to accurately simulate this nonlinear friction response relationship, which is violent in the early stage of aging and tends to saturate in the later stage. After benchmarking experiments on 300 electricity meters with different service years, the root mean square error between the predicted value and the measured value remained within 0.08, verifying the physical rationality of the prediction logic.

[0032] During model training, the Xavier initialization method was used to assign initial values ​​to the network weights, ensuring consistent variance in the data distribution across layers. The Adam optimizer was applied during training, with an initial learning rate of 0.001 and relevant momentum parameters set to 0.9 and 0.999, respectively. The training batch size was fixed at 32. To prevent overfitting, Dropout regularization was applied to the first and second hidden layers, with a dropout rate of 0.2 for both. An early stopping mechanism was also introduced: if the loss function on the validation set did not decrease for 10 consecutive iterations, training was automatically stopped, thus ensuring good generalization ability of the model.

[0033] By employing multi-dimensional encoding and recalibration in the material evolution model, a nonlinear quantitative characterization of the frictional properties of the meter casing was achieved. This process utilizes a stress intensity encoder to extract the derivative mapping of irradiance with age, quantifying the surface pulverization rate of the material under environmental stress, and addresses the differences in aging evolution paths among different casing materials by incorporating material constitutive characteristics. An attention mechanism is introduced to calculate the damage gain factor and perform dynamic recalibration, ensuring that the output predicted friction coefficient accurately reflects the individualized degree of service damage. This effectively alleviates the discrepancy between nominal parameters and the actual state of decommissioned meters in traditional modeling, providing physical boundary conditions consistent with the actual condition of the meters for subsequent anti-slip stability analysis in sorting path simulations.

[0034] Further, the process of obtaining the fitted centroid deviation coordinates includes: matching the number of cycles in the historical thermal load cycle data with a preset material fatigue damage index table to obtain the loosening displacement of the internal fasteners; performing linear interpolation calculation on the cumulative load current impact data to obtain the thermal deviation correction value of the internal heavy-duty components of the smart meter under thermal stress conditions; vector superimposing the loosening displacement and the thermal deviation correction value to obtain the actual offset vector of the internal heavy-duty components relative to the original installation position; compensating the actual offset vector to the original coordinates of the heavy-duty components retrieved from the standard three-dimensional model, and combining the static moment disturbance deviation calculation with the mass of each component to obtain the fitted centroid deviation coordinates.

[0035] Specifically, firstly, the historical thermal load cycle counts statistically analyzed in the aforementioned steps are extracted and used as index values ​​for matching in a pre-defined material fatigue damage index table. This index table is based on an empirical database obtained from accelerated aging tests of meters made of the same material, recording the stress relaxation degree at the connection between the internal plastic bracket and the metal screws of the meter under different temperature change cycles. Through matching, the corresponding loosening displacement is obtained. This displacement represents the probability amplitude of the component's minute displacement in three-dimensional space caused by the decrease in the fastening force of internal structural components due to long-term thermal expansion and contraction. The method for establishing this material fatigue damage index table is as follows: through typical intelligent... The materials used for the meter casing, including flame-retardant polycarbonate, ABS engineering plastic, and glass fiber reinforced polyester, are obtained through accelerated aging tests. In these tests, material samples are placed in a constant temperature and humidity environment with a temperature range of -20°C to +60°C and a humidity range of 65% to 75% for multiple thermal cycles. The average loosening displacement value of M3 or M4 screws (with a preload torque in the range of 0.3 to 0.7 Nm) is recorded for each material at different cycles. The error range is typically maintained between ±0.03 and 0.08 mm, thus ensuring the applicability and effectiveness of the meter in the case of meter casings with a thickness of 2 to 4 mm.

[0036] Secondly, the cumulative load current impact data is obtained, which is the frequency of pulses exceeding the rated current during the service life. Because the heavy-load components inside the smart meter (such as manganese copper shunts, high-power relays, or current transformers) generate severe Joule heating when subjected to instantaneous high current, resulting in local thermal stress deformation, linear interpolation calculations are performed for the current impact frequency using a preset thermal stress-deformation response curve. Through interpolation, the irreversible thermal bias value generated by the accumulation of multiple thermal shocks of the heavy-load components can be determined, thereby quantifying the physical offset tendency of these high-density components relative to the design position. In the actual interpolation, the material type of the current meter casing is first determined, and then the closest data point is retrieved from the index table based on the historical thermal load cycle number. If the calculated cycle number is between two discrete data points in the table, the intermediate value is calculated using a linear interpolation algorithm. For cases where there are multiple fastening screws inside the meter, table lookup calculations are performed separately to obtain the loosening displacement of each individual screw, which is then converted into a three-dimensional vector for synthesis to improve the accuracy of centroid fitting.

[0037] Subsequently, the obtained fastener loosening displacement and the thermal offset correction value of the heavy-duty component are vector-superimposed. During the calculation, a three-dimensional Cartesian coordinate system is established with the geometric center of the meter as the origin. Different directional weights are assigned to the displacement based on the installation orientation of the components. Through spatial vector summation, the actual offset vector of each heavy-duty component relative to its original installation position (i.e., the theoretical design position) is obtained. This transforms the damage caused by multi-source stress into coordinate updates of discrete components in virtual space. The directional weights are determined based on the geometric installation orientation of the heavy-duty component inside the meter. For example, for a horizontally installed relay, its thermal offset correction value is mainly... The weights are assigned to the XY plane in a three-dimensional Cartesian coordinate system. Finally, the original coordinates and mass attributes of each heavy-duty component are retrieved from the standard three-dimensional model (CAD model). The actual offset vectors are then compensated to the corresponding original coordinates to update the real-time spatial position of the heavy-duty components. Subsequently, using the principle of static moment balance in mechanics, the algebraic sum of the product of the mass of all components and their updated coordinates is calculated and divided by the total mass of the meter. This yields the displacement deviation of the meter's centroid relative to the nominal centroid. The final fitted centroid deviation coordinates are then updated in real time to the digital twin entity model for subsequent dynamic balance analysis of sorting operations.

[0038] By converting historical thermal load cycle and current impact data into the physical displacement of internal components, a quantitative fitting of the smart meter's center of gravity deviation state was achieved. Using the vector superposition of fatigue damage matching and thermal stress correction, the failure of internal fasteners and the pose drift of heavy-load components caused by long-term service were quantified. By performing static torque disturbance deviation calculations, the microscopic displacements of local components were mapped to the macroscopic deviation coordinates of the entire meter's center of gravity. This process solves the problem of dynamic modeling distortion caused by neglecting the evolution of internal structures in traditional simulations, effectively reduces the risk of torque imbalance caused by inaccurate center of gravity prediction during the high-speed start-up and shutdown phases of the sorting mechanism, and enhances the adaptability of virtual scheduling commands to the physical characteristics of individual physical objects.

[0039] Furthermore, the process of generating the dynamic distribution cloud map includes: initializing the digital twin attributes of a specific meter entity by assigning the predicted friction coefficient and the fitted centroid deviation coordinates to the virtual entity model; constructing a standard sorting path model consisting of straight segments, circular turning segments, and fork diversion segments in the virtual environment, and driving the virtual entity model to perform dynamic simulations under multiple speed gradients; determining the sorting performance boundaries of the meter under different sorting actions by extracting the critical acceleration values ​​of the virtual entity model when it experiences sideslip, overturning, and derailment in each path segment; and generating the dynamic distribution cloud map that guides actual scheduling speed limits by projecting the critical acceleration values ​​corresponding to each path segment onto the topology layout diagram of the sorting equipment and marking the safe speed limits of different areas.

[0040] Specifically, the calculated predicted friction coefficient and the fitted centroid deviation coordinates are used as physical variables and assigned to the standard 3D model of the meter in the virtual environment. In practice, the surface physical material properties (such as dynamic friction coefficient and static friction coefficient) of the virtual entity model are modified to achieve a quantitative mapping of surface aging. At the same time, based on the centroid deviation coordinates, a displacement offset relative to the geometric center is set in the gravity center calculation component of the model. In this way, the originally standardized virtual model is transformed into a digital twin of a retired meter with individual service characteristics, enabling it to generate force feedback consistent with the real object in dynamic simulation. To achieve accurate spatial speed limit indexing, a three-dimensional Cartesian coordinate system is established based on the mechanical reference coordinate system of the sorting equipment. The operating space of the entire sorting line is divided into uniform cubic grid units, with a grid resolution set to 0.1 m x 0.1 m x 0.1 m. The logic for setting this resolution is as follows: the sorting speed is usually between 0.5 m / s and 2.0 m / s, the system control cycle is 50 milliseconds, and the distance the meter moves within one control cycle is between 25 mm and 100 mm. A grid resolution of 0.1 m can provide sufficient positional accuracy while ensuring that the amount of computation is within an acceptable range for real-time processing.

[0041] Secondly, in a virtual simulation platform (such as a physics engine-based robot dynamics simulation system), a standard sorting path 3D model is constructed, consisting of a straight acceleration section, a circular turning section, and a fork diversion section. This digital twin is then driven to run on the path model, and multiple sets of simulation experiments are performed using a velocity-acceleration gradient stepping method. Specifically, within each path segment, the acceleration value of the virtual drive mechanism is gradually increased in preset increments (e.g., from 1.0 m / s²). 2 Starting at 0.1 m / s 2 (Incrementing step size) to simulate the dynamic response of the meter under different workloads.

[0042] Subsequently, during the simulation, the pose and contact force data of the digital twin are monitored in real time. When the acceleration reaches a certain threshold, causing the model to enter any of the following states, the instantaneous acceleration at this time is recorded as the critical value: Side slip: The relative displacement between the model and the sorting mechanism exceeds a preset threshold (e.g., the friction is insufficient to counteract the inertial force). Overturning: The bottom of the model detaches from the supporting plane and the tilt angle exceeds the critical instability angle (e.g., the center of mass offset causes torque imbalance). Derailment: The model completely deviates from the preset running trajectory; by repeating the above experiment for different path segments (straight line, curve, fork), a set of differentiated sorting performance boundary data was determined for this specific meter.

[0043] The overturning stability of the meter is determined by the balance between the torque of the moment formed by gravity and lateral inertial force and the frictional torque experienced by the meter rotating around the contact edge. This scheme emphasizes the necessity of fitting the centroid deviation because the offset of the centroid in the horizontal plane will significantly reduce the stabilizing arm, thereby greatly reducing the critical acceleration required for overturning. Experimental data shows that when the centroid deviation vector length exceeds 1.5 mm, the instability rate of the meter in high-speed sorting will begin to increase significantly. Therefore, 1.5 mm is set as the core threshold for triggering the motion curve flexible correction program.

[0044] Subsequently, a dynamic projection algorithm for the acceleration critical value is executed to map the simulated acceleration critical value to the above grid space: for each grid cell located on the sorting path, the corresponding critical value is retrieved according to the path type it is in; if a grid cell crosses two different path segments at the same time, the system adopts a conservative strategy and takes the smaller of the two critical values ​​as the safety benchmark for that grid.

[0045] Finally, the critical acceleration values ​​extracted from each path segment (straight line, arc, and fork) are aligned with the global three-dimensional topological coordinate system of the sorting equipment. By establishing the coordinate mapping relationship between the physical path and the simulation environment, the discrete dynamic simulation results are converted into a constraint dataset with spatial attributes. Subsequently, the physical path is divided into several functional grid units, and the critical acceleration value corresponding to each unit is assigned to the spatial coordinate point, providing underlying data support for subsequent speed-limited scheduling.

[0046] Considering the uncertainties such as vibration and mechanical wear in the actual working environment, a preset safety margin coefficient (0.85 in this embodiment) is introduced. For each functional grid unit, the maximum allowable speed limit is derived in reverse, based on the corresponding acceleration critical value and path geometric features (such as the radius of curvature of the arc segment) of that area. In this way, the complex dynamic constraints are transformed into speed control parameters that can directly guide the operation of the drive motor.

[0047] Finally, using heatmap rendering technology in conjunction with data matrix annotation technology, the cloud map is generated in real time in the topology layout diagram of the sorting equipment. The heatmap rendering intuitively represents the stability distribution of the entire sorting line under the current meter status through changes in color intensity (for example, red represents the low-speed limit zone and green represents the high-speed passage zone). At the same time, the data matrix annotation technology accurately marks the specific speed limit values ​​at key control nodes (such as the fork entry and the center of the curve). This cloud map, which combines macro trend indications with micro precise instructions, can guide the actual physical sorting line to dynamically adjust the operating rate of each execution node according to the individual physical characteristics of the current passing meter.

[0048] During the generation process, the system introduces a safety margin coefficient of 0.85 to correct the upper limit of acceleration. This value is determined based on actual comparative tests of 500 meters with different service years, which can ensure that the actual instability rate is reduced to zero under the statistical distribution of unstable acceleration. The corrected upper limit of speed is dynamically rendered by color depth, and the color scheme adopts a red-yellow-green mode: the area with the upper limit of speed below 0.3 m / s is rendered as dark red (high risk); the area between 0.6 and 0.9 m / s is yellow; and the area above 1.2 m / s is dark green (low risk area). This cloud map enables the dispatching system to achieve adaptive speed limit control for the specific aging state of each meter.

[0049] By mapping individualized predicted friction coefficients and fitted centroid deviation coordinates to a digital twin model and driving it to perform multi-gradient dynamic simulations under multiple operating conditions, a quantitative detection of the performance boundary for sorting decommissioned meters was achieved. Using the extracted critical acceleration values ​​and the projection mapping of the topological layout, abstract mechanical constraints were transformed into intuitive dynamic distribution cloud maps, constructing differentiated safety speed limit benchmarks for each meter entity. This solves the problem of poor adaptability caused by using uniform dynamic limits in traditional scheduling, reduces the probability of sideslip and overturning of meters with centroid offset or surface aging in critical path sections such as turns and diversions, and enhances the operational stability of the automated scheduling system when handling decommissioned equipment with varying physical states.

[0050] Furthermore, the process of obtaining the nonlinear prediction compensation value includes: acquiring the real-time spatial coordinates and offset angle of the smart meter through a visual sensor deployed on the sorting line, as the real-time pose data of the physical entity; calculating the spatiotemporal deviation vector of the real-time pose data relative to the theoretical pose of the virtual entity model, and extracting the variance fluctuation frequency of the sensor signal to evaluate the environmental noise intensity; comparing the environmental noise intensity with the parameter iteration convergence of the virtual entity model to calculate the pose confidence score reflecting the credibility ratio of virtual and real data; using the pose confidence score as a weighting factor to perform weighted fusion of the real-time measurement value and the theoretical prediction value, and calculating the nonlinear prediction compensation value in combination with the current operating speed of the sorting line.

[0051] Specifically, during the sorting operation, real-time pose data of the meters to be sorted, including the three-dimensional coordinates of the geometric center and the yaw angle, is first acquired using industrial-grade high-speed vision sensors (sampling frequency set to 60Hz) deployed on the sorting line. To ensure the accuracy of the calculated deviation vector, the feedback data of the physical entity and the simulation output data of the virtual entity model are timestamped and synchronized based on a unified system millisecond-level clock. The theoretical pose data used for comparison comes from the simulation dynamic response value calculated by the virtual entity model based on the current dynamic constraints, rather than simple instruction values. A first-in-first-out sliding window logic is established in memory to store the sampling sequence of the most recent 15 consecutive frames in real time.

[0052] To assess the reliability of visual data, the system calculates the standard deviation of the pose residual within the sliding window and the zero-point crossover frequency of the signal. In setting the judgment threshold, a physical displacement deviation with a standard deviation exceeding 0.5 mm (which corresponds to approximately 2 pixels at a 1280x720 resolution) is used as the benchmark for triggering high-intensity noise recognition. Subsequently, the above indicators are mapped to the 0-1 interval using the Sigmoid nonlinear normalization function to generate an environmental noise score. Using the Sigmoid function can enhance the system's sensitivity to sudden high-frequency noise such as instantaneous vibration of the conveyor belt, enabling the score to more accurately reflect the true degree of interference to the sensor signal.

[0053] On the virtual space side, the system synchronously monitors the parameter iteration convergence status of the virtual entity model during the dynamic numerical solution process. In this embodiment, the dynamic numerical solution adopts an implicit integration algorithm, and the judgment rule for the parameter iteration residual is as follows: when the residual is less than 0.0001, the model is judged to be fully converged, and the score is 1; when the residual is greater than 0.01, the model is judged to be divergent or severely non-converged, and the score is 0; if it is in the middle range, linear interpolation calculation is performed. This threshold setting based on a specific algorithm background ensures that the convergence score can reflect the physical inference accuracy of the digital twin under the current working conditions in real time and accurately.

[0054] 0.6 and 0.4 are introduced as weight allocation constants for the virtual model and the visual sensor, respectively. By multiplying the convergence score by 0.6 and adding 1 and subtracting the environmental noise score by 0.4, the pose confidence score is finally synthesized. To prevent the confidence score from oscillating violently around 0.5 under complex working conditions, which could cause control jitter in the sorting mechanism, a first-order low-pass filter is applied to the calculated pose confidence score to ensure the smooth evolution of the weight values. This score acts as a dynamic factor in the fusion process: when noise increases or the model is distorted, the score automatically decreases, thereby adaptively reducing the proportion of real-time measurements and increasing the proportion of theoretical predictions.

[0055] Finally, based on the dynamically fused pose and combined with the current operating speed and acceleration of the sorting line, a second-order prediction operator is introduced to calculate the final compensation amount. For the system's preset total delay time of 50 milliseconds, the expected displacement increment within this period is calculated using displacement prediction logic. The specific calculation logic is as follows: the speed is multiplied by 0.05, plus half the acceleration and the square of 0.05, the fused pose is assigned a fixed weight of 0.65, and the expected increment vector is assigned a weight of 0.35 and superimposed to output the final nonlinear prediction compensation value. This mechanism effectively offsets the visual sampling delay and mechanical response lag, ensuring that the sorting instructions and the real-time physical inertia of the meter are dynamically aligned.

[0056] The weighting coefficients mentioned above are based on statistical analysis of error distribution within the Kalman filter framework. Through benchmarking experiments involving over 5000 pose sample points under 135 different operating conditions, the mean square error of the visual sensor measurements and the virtual model predictions was calculated. The ratio of 0.65 to 0.35 was set to enhance the robustness of the visual measurements while ensuring real-time performance, thereby offsetting the prediction bias that may be caused by parameter degradation in the virtual model and minimizing trajectory tracking error.

[0057] By comparing the spatiotemporal deviations between real-time pose data and theoretical pose, and combining sensor variance fluctuations with model convergence for confidence assessment, the confidence score balances physical measurement noise and virtual simulation errors, mitigating pose recognition fluctuations caused by environmental interference and correcting virtual-real synchronization deviations caused by control link delays. Combined with the calculation of predicted compensation values ​​based on operating speed, the system's ability to predict the movement trend of target entities is improved, reducing the risk of gripping point offset under high-speed sorting conditions, and enhancing the closed-loop control stability and pose tracking accuracy of the digital twin scheduling system under complex conditions.

[0058] Furthermore, the process of sending the scheduling control command to the sorting driver includes: extracting the acceleration safety threshold of the current path segment based on the dynamic distribution cloud map, and correcting the theoretical pose by offset in combination with the nonlinear prediction compensation value to obtain the running trajectory correction amount; calculating the dynamic execution torque required for the sorting mechanism to perform actions based on the running trajectory correction amount and in combination with the physical constraints determined by the predicted friction coefficient and the fitted centroid deviation coordinates; summarizing the running trajectory correction amount and the dynamic execution torque to generate the final scheduling control command; and sending the scheduling control command to the sorting driver to complete the sorting operation of the smart meter by adjusting the output and position of the drive motor.

[0059] Specifically, during the implementation phase of sending scheduling control commands to the sorting drive, the dynamic distribution cloud map is first retrieved in real time. When a smart meter to be sorted enters a specific section of the sorting path (such as an arc segment with a turning radius of 500mm), the system extracts the meter's specific acceleration safety threshold from the functional grid corresponding to the cloud map based on the meter's current spatial coordinate index. Subsequently, the system uses the extracted acceleration safety threshold as an upper limit dynamic constraint and superimposes the nonlinear prediction compensation value output from the aforementioned steps to perform real-time pose offset correction on the preset theoretical trajectory. Through this process, a trajectory correction amount that takes into account both the geometric constraints of the path segment and the prediction of individual pose deviations is generated, ensuring that the end effector of the sorting mechanism can always be aligned with the real-time dynamic centroid of the meter.

[0060] Based on the generated trajectory correction, and combined with the physical constraints determined by the predicted friction coefficient and the fitted centroid deviation coordinates, the inverse dynamics operator is invoked to calculate the execution torque online. Specifically, the system uses the Newton-Euler equations to construct a dynamic model of the sorting mechanism, and introduces the fitted centroid deviation coordinates as an additional torque term into the model.

[0061] in, This represents the output torque vector of the sorting mechanism joints. This represents the joint position vector (i.e., joint angle or displacement). Represents the joint velocity vector. Represents the joint acceleration vector. The inertia matrix represents the joint space. Represents the Coriolis force and centripetal force matrix. Represents the gravitational torque vector. This indicates the friction constraint and damping compensation terms. This represents the coefficient of friction.

[0062] During the calculation, the asymmetric gravitational moment generated by the centroid deviation is considered. Compensation is provided to prevent overload or vibration of the robotic arm joints due to center of gravity shift; simultaneously, the system incorporates predicted friction coefficients. The maximum acceleration limit of the end gripper is determined to ensure that the surface of the meter does not slip under the current clamping force. Through this inverse dynamic solution process, the dynamic execution torque required by each drive joint of the sorting mechanism during the execution of the action is calculated.

[0063] The trajectory correction (including expected position, velocity, and acceleration components) and dynamic execution torque are then time-sequentially summarized and packaged to generate the final scheduling control command. This command is encapsulated using an industrial-grade real-time Ethernet protocol (such as EtherCAT) to ensure determinism and synchronization in command transmission, and the scheduling control command is sent to each sorting drive on the physical sorting line.

[0064] After receiving instructions, the sorting drive adjusts the output and position of the drive motor through its internal servo control unit. In practice, the drive uses pulse width modulation technology to dynamically adjust the current of the motor stator winding, thereby precisely controlling the motor's output torque to match the calculated dynamic execution torque. At the same time, by reading feedback from the motor's built-in high-precision absolute encoder, the drive adjusts the rotor rotation angle in a closed loop to strictly track the trajectory correction. Through the coordinated adjustment of motor output and position, the sorting mechanism can execute customized motion strategies for the physical differences of each retired meter, ultimately completing the sorting of smart meters with high precision and high stability.

[0065] The process of calculating the dynamic execution torque required for the sorting mechanism to perform its actions further includes: retrieving a preset material stress evolution database to obtain the critical pressure for shell damage corresponding to the number of years of operation; combining the predicted friction coefficient with the instantaneous tangential acceleration in the sorting trajectory to calculate the minimum normal pressure that satisfies the static friction constraint; using the minimum normal pressure as a benchmark and the critical pressure as a safety upper limit, calculating the torque output range of the end gripper drive motor, and adjusting the PWM pulse duty cycle to match the acceleration pulsation of the sorting mechanism in real time to achieve non-destructive anti-slip gripping of the aging shell.

[0066] The system extracts the meter's service life data and retrieves a pre-set material stress evolution database. This database stores the stress bearing limits of typical meter casing materials (such as flame-retardant polycarbonate or ABS) after mechanical performance degradation at different service times. Based on the search results, the critical pressure for casing damage at the current service life is obtained. This value represents the maximum pressure per unit area that the aging material can withstand without structural breakage or pulverization damage. The system multiplies this critical pressure by the effective contact area of ​​the end clamp to calculate the upper limit of the positive pressure for the current gripping action, which serves as the physical boundary to protect the aging casing from being crushed.

[0067] To prevent the meters from slipping during high-speed movement or turns in the sorting process, the system combines the predicted friction coefficient with the instantaneous tangential acceleration in the sorting trajectory for real-time dynamic analysis. Based on the principle of friction balance, it calculates the minimum static friction force required to keep the meters relatively stationary. This friction force must be greater than or equal to the product of the meter's mass and the instantaneous tangential acceleration. Subsequently, this minimum static friction force is divided by the predicted friction coefficient to calculate the minimum normal force that satisfies the anti-slip constraint. This value ensures that under the current acceleration pulsation, the friction force generated by clamping is sufficient to counteract the meter's inertial force.

[0068] The calculated minimum positive pressure is used as the lower limit of the baseline, and the upper limit of the positive pressure safety is used as the protection threshold to determine the torque output range of the gripper drive motor. In actual execution, the scheduling execution module uses pulse width modulation (PWM) technology to perform high-frequency control of the gripper motor: monitoring the motion trajectory of the sorting mechanism at millisecond intervals, and converting the real-time gripping force demand into corresponding PWM duty cycle commands; when the sorting mechanism is in the starting acceleration or sharp turn phase, the PWM duty cycle is automatically increased, and the output torque is increased by increasing the current of the motor stator winding to ensure that the gripping force increases synchronously with the inertial load; when the acceleration decreases, the duty cycle drops accordingly to avoid applying excessive stress to the aging shell. Specifically, the motion trajectory of the sorting mechanism is monitored at a fixed period of 50 milliseconds (i.e., 20 Hz). At the same time, when setting the initial robustness threshold, it is defined as 1.5 times the maximum value of the inherent random deviation (i.e., 1.5 multiplied by the maximum value), thereby ensuring that the control system has a definite stability judgment standard when facing conventional mechanical noise.

[0069] Through real-time PWM adjustment, the power devices precisely control the on / off time of the motor, ensuring that the output force of the end gripper is always dynamically locked within a safe window that is sufficient to prevent slipping and crushing. This control method allows the sorting mechanism to provide customized gripping torque for each retired meter based on its specific aging condition (how brittle the casing is, how slippery the surface is). Even in complex fork sections or high-speed arc sections, precise real-time matching of acceleration pulsations can effectively counteract the gravitational bias torque generated by the center of mass offset, ultimately completing the sorting operation smoothly and without damage.

[0070] It also includes real-time monitoring of the actual motion trajectory after the physical sorting line performs its actions, and calculating the spatial deviation residual between the actual motion trajectory and the correction amount of the running trajectory; extracting the distribution characteristics of the deviation residual within a continuous operation cycle, and when the residual characteristic value exceeds a preset robust threshold, inputting the residual as a backpropagation correction signal into the state fitting module; and using an incremental learning algorithm to update the neuron connection weights in the material evolution model online, thereby achieving digital twin self-evolutionary calibration of the physical characteristics of a specific batch of retired meters.

[0071] During the sorting mechanism's operation, pose sensors (high-precision laser displacement meters or high-frame-rate visual monitoring cameras) deployed at the end of the robotic arm or on the side of the sorting station collect the actual motion trajectory of the meter in space. The collected actual pose coordinates are time-aligned with the trajectory correction amount (i.e., the expected theoretical target trajectory). By calculating the Euclidean distance and yaw angle deviation between the two in the three-dimensional Cartesian coordinate system, the spatial deviation residual at each moment is extracted. This residual directly reflects the physical execution error caused by inaccurate model parameters (overestimation of friction force or slight shift in the center of mass position).

[0072] The system does not adjust for single random errors, but continuously accumulates deviation data within a continuous operating cycle (e.g., processing 20 meters of the same batch), extracts the distribution characteristics of the deviation residuals, and assesses the systematic tendency of the deviation by calculating the moving average and standard deviation of the residuals within that cycle. To ensure system stability, a preset robust threshold is introduced as a red line to trigger calibration. Before formal operation, a benchmark sorting test is performed using standard samples in an ideal state to record the inherent random deviations caused by mechanical gaps and sensor noise, and 1.5 times the maximum value of these deviations is set as the initial robust threshold. When the residual characteristic value of consecutive batches exceeds this threshold, it is determined that the current material evolution model can no longer accurately represent the physical status of the specific batch of meters (e.g., the batch of meters has undergone unconventional aging due to extreme storage environment), thus automatically activating the self-evolutionary calibration program.

[0073] When calibration is triggered, the out-of-range residual distribution characteristics are transformed into a backpropagation correction signal. This signal serves as the input to the loss function, simulating the error feedback mechanism of a neural network. It reverses the positional deviation at the physical execution level into a cognitive bias in the physical characteristics at the model deduction level. An online incremental learning algorithm is used to fine-tune the material evolution model. Using the gradient descent operator, the connection weights of neurons in the material evolution model are adjusted online based on the feedback residual signal. Through incremental updates of the weights, the model can self-correct for the specific aging characteristics of the current batch of meters (such as atypical powdering on the surface), making the subsequently output predicted friction coefficient more closely match the actual physical state. Through continuous online learning, a high degree of adaptation between the digital twin model and the physical entities of a specific batch is achieved, significantly improving the trajectory tracking accuracy and the stability of the sorting operation under complex working conditions.

[0074] The process of generating the trajectory correction amount also includes: calculating the asymmetric inertial lever vector at the moment of sorting action switching based on the fitted centroid deviation coordinates; adjusting the acceleration and deceleration pulsation coefficient of the sorting mechanism drive motor in the reverse direction according to the magnitude of the lever vector; calling the seven-segment S-shaped motion planning operator to reconstruct the sorting speed curve, and achieving flexible trajectory control of the unbalanced load entity by smoothing the transient oscillating angular momentum caused by centroid offset.

[0075] First, retrieve the fitted centroid deviation coordinates. At the instant of switching sorting actions (e.g., when the robotic arm accelerates from a straight line to a circular flow section, or when it starts grasping from a stationary state), establish a local dynamic three-dimensional coordinate system with the actuator (such as the end joint of the robotic arm or the geometric center of the grasping) as the origin. Based on the fitted centroid deviation value, calculate the asymmetric inertial lever arm vector of the physical entity's true center of gravity relative to the geometric center of the grasping. This vector intuitively quantifies the additional rotational inertia generated by the center of gravity offset during the change of motion state. The longer the lever arm vector, the more asymmetric the mass distribution of the meter, and the greater the transient oscillating angular momentum generated during high-speed start-stop, which also has a stronger interference with sorting stability.

[0076] A preset inertial lever arm vector threshold (e.g., 1.5mm in this embodiment) is used to define whether the eccentricity of the meter is sufficient to significantly affect the trajectory accuracy. The logic for setting this threshold is as follows: before formal operation, a standard sample meter with uniform mass distribution is used to conduct no-load and load operation tests at the highest rated acceleration. By monitoring the small vibration displacement of the end of the robotic arm during the start-stop phase in real time, the minimum centroid offset that causes the end swing amplitude to exceed the system's preset position tolerance (e.g., 0.5mm) is determined as the preset threshold. When the lever arm vector length calculated in real time exceeds this threshold, it is determined that the current meter belongs to an unbalanced load entity, and the flexible correction program of the motion curve must be triggered.

[0077] To counteract the inertial impact caused by eccentric loads, the system introduces a preset mapping function to dynamically adjust the acceleration / deceleration pulsation coefficient (i.e., jerk, rate of change of acceleration) of the drive motor. This function establishes a logical relationship where the lever arm length is inversely proportional to the pulsation coefficient. When the lever arm deviation exceeds a preset threshold, the function calculates the appropriate reduction in the pulsation coefficient based on the magnitude of the deviation. The larger the lever arm deviation, the smaller the set pulsation coefficient. This means that when the drive motor changes speed, the acceleration increase or decrease becomes slower and smoother, thus suppressing the oscillation tendency caused by unstable center of gravity during sudden motion changes from a dynamic perspective. In this embodiment, it is specifically set as a linear decay mapping function based on the deviation increment. This function receives the asymmetric inertial lever arm vector length from the state fitting module in real time as the independent variable. In this embodiment, the function divides the sorting state into two intervals using a preset 1.5mm threshold. Normal range (deviation less than or equal to 1.5mm): The mapping function output value is 1, which means that the centroid of the meter deviates within the range allowed by mechanical rigidity. At this time, the system maintains a standard, high-efficiency acceleration and deceleration pulsation coefficient without any attenuation.

[0078] Correction range (deviation greater than 1.5mm): Once the lever arm vector length exceeds 1.5mm, the mapping function is immediately activated and begins to calculate an attenuation coefficient less than 1 based on the offset exceeding 1.5mm. Within the correction range, the mapping function follows a linear negative correlation attenuation logic. Specifically, the function first extracts the difference between the current lever arm vector length and the 1.5mm threshold, and multiplies this difference by a preset flexible attenuation coefficient. In this embodiment, this coefficient represents how much additional acceleration / deceleration pulsation needs to be sacrificed for stability for each additional unit length of center of gravity deviation. The calculated attenuation is subtracted from the standard pulsation coefficient (i.e., 100% intensity). The more severe the center of gravity deviation of the meter (the longer the lever arm vector), the closer the final correction coefficient calculated by the mapping function will be to a smaller value.

[0079] After determining the corrected pulsation coefficient, the seven-segment S-shaped motion planning operator is invoked to reconstruct the sorting speed curve. Compared with the traditional trapezoidal speed curve, this operator subdivides the entire acceleration and deceleration process into: acceleration segment, uniform acceleration segment, deceleration segment, uniform speed segment, acceleration and deceleration segment, uniform deceleration segment, and deceleration and deceleration segment. During the start-stop and reversing phases, the operator automatically extends the running time (i.e., time constant) of the acceleration and deceleration phases based on the calculated eccentric torque. Through this flexible edge processing, the output torque of the motor exhibits a continuous and smooth nonlinear change, rather than an instantaneous change. This processing method smooths out the transient oscillating angular momentum caused by the center of gravity offset, ensuring that even meters with a severely offset center of gravity can maintain posture stability during high-speed sorting.

[0080] Finally, the trajectory correction amount after S-curve reconstruction and the dynamic execution torque calculated based on physical constraints are summarized and encapsulated into the final scheduling control command, which is then sent to the sorting drive. By adjusting the motor's torque output and position step, the drive mechanism can execute a customized flexible motion strategy for the unbalanced characteristics of each retired meter, achieving high-precision sorting. This embodiment effectively solves the problem of operational vibration caused by uneven internal mass distribution in retired meters through the quantization of the lever arm vector and adaptive reconstruction of the motion curve.

[0081] By extracting the acceleration threshold from the dynamic distribution cloud map and superimposing nonlinear pose compensation, the predicted friction coefficient and center of mass deviation coordinates are transformed into dynamic torque constraints for the drive motor. This shifts the scheduling command from simple geometric position control to force balance control based on material properties, mitigating the dynamic parameter mismatch problem caused by center of gravity offset or surface aging in retired meters and reducing the risk of slippage and tipping during high-acceleration sorting. Through coordinated adjustment of torque and trajectory, the sorting mechanism's adaptability to individual differentiated states is improved, ensuring the stability and pose tracking accuracy of automated sorting operations.

[0082] By constructing a full-link digital twin architecture from service history data tracing to physical constitutive evolution, this application achieves precise reconstruction and scheduling optimization of the individualized dynamic characteristics of decommissioned meters to be sorted. This scheme transforms a static idealized model into a dynamic property-aware model, using predicted friction coefficients and fitting centroid deviations to correct parameter distortions in traditional modeling, significantly reducing the dynamic response deviation between virtual simulation and physical entities. Guided by stability distribution cloud maps and corrected by nonlinear pose compensation, the motion trajectory and driving torque of the actuator can be adaptively adjusted according to the aging degree and structural evolution state of each meter. This effectively reduces the risks of slippage, overturning, and grasping offset during the sorting process, solving the scheduling mismatch problem caused by the uncertainty of the decommissioned equipment's properties while improving the stability and pose tracking accuracy of automated sorting operations.

[0083] Example 2: To achieve batch and adaptive sorting of retired smart meters with different service backgrounds, this application introduces a warehouse virtual simulation scheduling system based on digital twins, the specific structure of which is as follows: Figure 3 As shown, when this batch of retired electricity meters enters the sorting line, the data acquisition module identifies each meter using a visual scanning unit and extracts corresponding multi-source historical data from the asset management system. For this batch of meters, the module categorizes and obtains data on their years of operation, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current surge data.

[0084] The state fitting module receives the data packets of the batch of meters and performs individualized physical modeling for each meter. The module calls the material evolution model, takes the years of operation data and cumulative ultraviolet radiation dose as input, calculates the degree of aging and pulverization of the casing of each meter in the batch, and outputs the predicted friction coefficient for each meter. The module combines historical thermal load cycle data and determines the loosening displacement of fasteners through the material fatigue damage index table. At the same time, for the load current impact data, it uses the thermal stress-deformation response curve to perform linear interpolation to obtain the thermal deviation of heavy-load components. Through the vector superposition algorithm, the micro displacement is transformed into macroscopic physical deviation to obtain the fitted centroid deviation coordinates of each meter.

[0085] The twin simulation module receives parameters output by the state fitting module and assigns them to specific meter entity models in the virtual space. The system uses the simulation results of meters with extreme characteristics (such as meters with extremely low friction coefficients or large centroid deviations) as the safety baseline for this batch, generating a dynamic distribution cloud map covering the entire line to ensure that all meters in the batch are within the safe operating envelope. Multi-group sorting acceleration gradient simulation is performed in the virtual environment. The system monitors the critical acceleration of sideslip and overturning of the model on straight, curved, and fork paths, and projects these acceleration thresholds onto the topology map of the sorting equipment to generate a dynamic distribution cloud map that guides the entire line operation. This cloud map marks high-risk path areas for this batch of meters in real time, providing a decision basis for subsequent drive speed limits.

[0086] When the batch of meters actually passes through the sorting sensor position, the twin simulation module acquires the real-time pose data (lateral displacement, longitudinal displacement, and yaw angle) of the physical sorting line. The module extracts the variance fluctuation frequency of the sensor signal within the sliding window, assesses the current mechanical vibration noise intensity, and calculates the pose confidence score by combining the parameter iteration convergence of the virtual entity model in the dynamic solution. Using the confidence score as a weighting factor, the module weights and fuses the real-time measurement value with the theoretical prediction value of the virtual model. The higher the confidence score, the greater the weight of the real-time measurement value; conversely, the weight of the virtual theoretical prediction value is increased. The module also calculates the nonlinear prediction compensation value by combining the real-time running speed with the second-order prediction operator to offset the time lag of the physical system in the sensing and transmission process.

[0087] The scheduling and execution module generates final action commands based on real-time data provided by the twin simulation module. It extracts the acceleration safety threshold of the current path segment from the dynamic distribution cloud map and, combined with nonlinear prediction compensation values, corrects the theoretical operating trajectory of the meter by offset, obtaining the trajectory correction amount. Based on this correction amount, and considering the meter's unique predicted friction coefficient (determining the upper limit of the gripping force) and fitted centroid deviation coordinates (determining the gravitational torque offset), the module calls the inverse dynamic equation to calculate the dynamic execution torque required for each joint of the sorting mechanism. The module summarizes the correction amount and execution torque, encapsulates them into scheduling control commands, and sends them to the sorting driver. The driver precisely adjusts the output torque and position by adjusting the pulse width modulation duty cycle of the drive motor and controlling the on / off time of the power devices. This process ensures that every meter in the batch, regardless of its surface friction or centroid offset, can be accurately sorted within the set dynamic safety boundaries.

[0088] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A warehouse virtual simulation scheduling method based on digital twins, characterized in that, include: Acquire data on the operating years of smart meters, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current surge data; Based on the years of operation data and cumulative ultraviolet radiation dose, the predicted friction coefficient of the smart meter casing is calculated through a material evolution model. Combined with historical thermal load cycle data and cumulative load current impact data, fatigue matching and thermal displacement correction are performed to obtain the fitted centroid deviation coordinates. The predicted friction coefficient and the fitted centroid deviation coordinates are assigned to the virtual entity model. By performing multi-group sorting acceleration gradient simulation, a dynamic distribution cloud map is generated. The real-time pose data of the physical sorting line is obtained. The deviation between the real-time pose data and the theoretical pose data synchronously output by the virtual entity model is compared and fused to obtain the nonlinear prediction compensation value. Acceleration safety thresholds are extracted from dynamic distribution cloud maps, and nonlinear prediction compensation values ​​are combined to generate trajectory correction values. The dynamic execution torque of the sorting mechanism is calculated based on the predicted friction coefficient and the fitted centroid deviation coordinates. The trajectory correction values ​​and dynamic execution torque are summarized to generate scheduling control commands, which are then sent to the sorting driver to complete the sorting operation of smart meters.

2. The warehouse virtual simulation scheduling method based on digital twins according to claim 1, characterized in that, The process of acquiring the operating years data, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current impact data of smart meters includes: scanning the barcode on the smart meter casing using an industrial vision sensor deployed at the front end of the sorting line to obtain a unique identifier; matching the unique identifier with the historical installation latitude and longitude coordinates and the daily solar radiation intensity in the historical meteorological database, and performing cumulative calculations in conjunction with a preset environmental shading coefficient to obtain the cumulative ultraviolet radiation dose; retrieving the historical current operating curve from the electricity information collection system and using a peak detection algorithm to extract the frequency of pulse events exceeding a preset rated current threshold to obtain the cumulative load current impact data; statistically analyzing the number of ambient temperature temperature difference cycles and the Joule thermal stress cycle characteristics caused by the load current during the smart meter's service life to obtain the historical heat load cycle data; and obtaining the operating years data by counting the cumulative number of days between the commissioning date and the dismantling date corresponding to the unique identifier recorded in the management system.

3. The warehouse virtual simulation scheduling method based on digital twins according to claim 1, characterized in that, The material evolution model includes: a baseline aging encoder, which receives the operating years data as input, processes it using a first multilayer perceptron, and generates a baseline aging feature vector; The stress intensity encoder receives the cumulative ultraviolet radiation dose and operating years data as input, calculates the derivative mapping of cumulative radiation dose with operating years, and generates a stress intensity feature vector that reflects the pulverization rate of the material surface. The response encoder receives the preset material attributes associated with the unique identifier as input, processes them using an embedded lookup table, transforms discrete material categories into continuous spatial constitutive vectors, and generates material constitutive feature vectors. The recalibration unit concatenates the stress intensity feature vector with the material constitutive feature vector and inputs it into the attention mechanism network to calculate the damage gain factor. The damage gain factor is then applied to the reference aging feature vector for dynamic recalibration, and the predicted friction coefficient of the smart meter casing is output.

4. The warehouse virtual simulation scheduling method based on digital twins according to claim 1, characterized in that, The process of obtaining the fitted centroid deviation coordinates includes: matching the number of cycles in the historical thermal load cycle data with a preset material fatigue damage index table to obtain the loosening displacement of the internal fasteners; performing linear interpolation calculation on the cumulative load current impact data to obtain the thermal deviation correction value of the internal heavy-duty components of the smart meter under thermal stress conditions; vector superimposing the loosening displacement and the thermal deviation correction value to obtain the actual offset vector of the internal heavy-duty components relative to the original installation position; compensating the actual offset vector to the original coordinates of the heavy-duty components retrieved from the standard three-dimensional model, and calculating the static moment disturbance deviation in combination with the mass of each component to obtain the fitted centroid deviation coordinates.

5. The warehouse virtual simulation scheduling method based on digital twins according to claim 1, characterized in that, The process of generating the dynamic distribution cloud map includes: initializing the digital twin attributes of a specific meter entity by assigning the predicted friction coefficient and the fitted centroid deviation coordinates to the virtual entity model; constructing a standard sorting path model consisting of straight segments, circular turning segments, and diversion fork segments in the virtual environment, and driving the virtual entity model to perform dynamic simulations under multiple speed gradients; determining the sorting performance boundaries of the meter under different sorting actions by extracting the critical acceleration values ​​of the virtual entity model when it experiences sideslip, overturning, and derailment in each path segment; and generating the dynamic distribution cloud map that guides actual scheduling speed limits by projecting the critical acceleration values ​​corresponding to each path segment onto the topology layout diagram of the sorting equipment and marking the safe speed limits for different areas.

6. The warehouse virtual simulation scheduling method based on digital twin according to claim 1, characterized in that, The process of obtaining the nonlinear prediction compensation value includes: acquiring the real-time spatial coordinates and offset angle of the smart meter through a visual sensor deployed on the sorting line, as the real-time pose data of the physical entity; calculating the spatiotemporal deviation vector of the real-time pose data relative to the theoretical pose of the virtual entity model, and extracting the variance fluctuation frequency of the sensor signal to evaluate the environmental noise intensity; comparing the environmental noise intensity with the parameter iteration convergence of the virtual entity model to calculate the pose confidence score reflecting the credibility ratio of virtual and real data; using the pose confidence score as a weighting factor to perform weighted fusion of the real-time measurement value and the theoretical prediction value, and calculating the nonlinear prediction compensation value in combination with the current running speed of the sorting line.

7. The warehouse virtual simulation scheduling method based on digital twin according to claim 1, characterized in that, The process of sending the scheduling control command to the sorting driver includes: extracting the acceleration safety threshold of the current path segment based on the dynamic distribution cloud map, and correcting the theoretical pose by offset in combination with the nonlinear prediction compensation value to obtain the running trajectory correction amount; calculating the dynamic execution torque required for the sorting mechanism to perform actions based on the running trajectory correction amount and in combination with the physical constraints determined by the predicted friction coefficient and the fitted centroid deviation coordinates; summarizing the running trajectory correction amount and the dynamic execution torque to generate the final scheduling control command; and sending the scheduling control command to the sorting driver to complete the sorting operation of the smart meter by adjusting the output and position of the drive motor.

8. A warehouse virtual simulation scheduling system based on digital twins, characterized in that, include: Data acquisition module: acquires data on the operating years of smart meters, cumulative ultraviolet radiation dose, historical heat load cycle data, and cumulative load current surge data; State fitting module: Based on the years of operation data and cumulative ultraviolet radiation dose, the predicted friction coefficient of the smart meter casing is calculated through the material evolution model. Combined with historical heat load cycle data and cumulative load current impact data, fatigue matching and thermal displacement correction are performed to obtain the fitted centroid deviation coordinates. Twin simulation module: Assigns the predicted friction coefficient and the fitted centroid deviation coordinates to the virtual entity model, generates a dynamic distribution cloud map by performing multi-group sorting acceleration gradient simulation; acquires the real-time pose data of the physical sorting line, compares the deviation between the real-time pose data and the theoretical pose data synchronously output by the virtual entity model and fuses them to obtain the nonlinear prediction compensation value. The scheduling and execution module extracts the acceleration safety threshold based on the dynamic distribution cloud map, generates the running trajectory correction amount by combining the nonlinear prediction compensation value, calculates the dynamic execution torque of the sorting mechanism based on the predicted friction coefficient and the fitted centroid deviation coordinates, summarizes the running trajectory correction amount and the dynamic execution torque to generate scheduling control instructions, and sends the scheduling control instructions to the sorting driver to complete the sorting operation of smart meters.

Images