A nielsen reselection process monitoring method based on digital twinning
By constructing a digital twin model and combining non-invasive observation signals, real-time quantitative monitoring of the bed state during the Nielsen reselection process was achieved, solving the problem of difficulty in real-time quantitative perception of the bed state and improving the stability of the reselection process and gold recovery rate.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHANGCHUN GOLD DESIGN INST
- Filing Date
- 2026-03-18
- Publication Date
- 2026-08-04
AI Technical Summary
In the existing Nielsen gravity separation process, it is difficult to quantify and perceive the bed state in real time and to proactively identify stability changes, resulting in a certain degree of passivity in process control. Especially under the conditions of continuous operation in gold mines, bed stability has a significant impact on concentrate grade and recovery rate.
The Nielsen reselection process monitoring method based on digital twins constructs a digital twin model by collecting reselection operation status data in real time. It combines non-invasive observation signals and filtered state estimation to realize online assimilation and inversion of bed state and calculation of stability index, and outputs monitoring conclusions.
This technology enables dynamic characterization of the operating mechanism and working conditions of the reseparation process, enhancing the foresight and effectiveness of bed stability monitoring and improving the stability and gold recovery rate of the reseparation process.
Smart Images
Figure CN122113433B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of gold mine reseparation monitoring technology, and in particular to a Nielsen reseparation process monitoring method based on digital twins. Background Technology
[0002] As gold resource development shifts towards lower-grade and more complex gold mineral distributions, gravity separation technology remains crucial in gold beneficiation processes due to its excellent enrichment effect on coarse-grained and liberated gold minerals. Nelson gravity separation equipment, utilizing a centrifugal enhanced gravity field, achieves efficient collection of high-density gold minerals and is widely used in the gold grinding-gravity separation recovery stage. Current Nelson gravity separation processes typically rely on process parameters such as rotational speed, feed rate, feed concentration, and discharge cycle for operational control, supplemented by online monitoring methods such as pressure and flow rate to indirectly characterize the equipment's operating status. In recent years, with the development of automation and information technology, some gold mines have begun to introduce data acquisition and process monitoring systems to record and analyze key operating data during gravity separation operations, aiming to improve operational stability and gold recovery rates.
[0003] In actual production, due to the complex internal bed structure and highly nonlinear material movement characteristics of the Nelson gravity separation process, existing technologies mostly focus on monitoring surface operating parameters, making it difficult to provide real-time and quantitative characterization of the bed state. When operating conditions fluctuate, changes in the bed state are often reflected in external signals with a lag, resulting in a certain degree of passivity in process control. Especially under continuous operation conditions in gold mines, bed stability has a direct impact on concentrate grade and recovery rate. However, existing technologies still rely mainly on empirical judgment for modeling and predicting the bed state evolution mechanism, lacking dynamic monitoring methods based on the fusion of system models and real-time data. Summary of the Invention
[0004] In view of the aforementioned existing problems, the present invention is proposed.
[0005] Therefore, this invention provides a Nielsen reselection process monitoring method based on digital twins, which solves the problems of difficulty in real-time quantitative perception of bed status and inability to proactively identify stability changes.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a Nielsen re-selection process monitoring method based on digital twins, comprising: real-time acquisition of re-selection operation status data; time synchronization and preprocessing of the re-selection operation status data to form an online operation dataset; construction of a re-selection digital twin model based on historical online operation datasets, and continuous injection of the online operation dataset into the re-selection digital twin model for online synchronization, outputting a set of twin prediction values and twin prediction residuals; setting bed state variables in the re-selection digital twin model based on the set of twin prediction values and twin prediction residuals combined with non-intrusive observation signals, and performing online assimilation and inversion through filtered state estimation to generate a set of bed state parameters; comprehensively evaluating the deviation between the set of bed state parameters and the preset stable operation range to form a bed stability index; using the set of bed state parameters and the bed stability index as inputs, and superimposing the discharge pipeline pressure signal and the set of twin prediction values to establish a discharge fingerprint template, calculating the consistency residual between the real-time discharge pulse characteristics and the discharge fingerprint template, and outputting monitoring conclusions.
[0007] As a preferred embodiment of the Nielsen re-separation process monitoring method based on digital twins described in this invention, the re-separation operation status data includes feed flow rate, feed density, motor current, discharge pipeline pressure signal, and discharge action timestamp.
[0008] As a preferred embodiment of the Nielsen reselection process monitoring method based on digital twins described in this invention, the specific steps for synchronizing and preprocessing the reselection operation status data to form an online operation dataset are as follows. The reselection operation status data is synchronized with the resampling through sampling frequency alignment, and time-aligned operation data is output. Using the timestamp of the ore discharge action as an event marker, the time-aligned operation data is aligned with events and the ore discharge process cycle is segmented to output periodic operation data; The system performs outlier removal, missing data processing, noise suppression, and dimension consistency processing on the periodic operation data, and encapsulates it in a structured manner according to the ore discharge cycle to output the online operation dataset.
[0009] As a preferred embodiment of the Nielsen reselection process monitoring method based on digital twins described in this invention, the specific steps for constructing the reselection digital twin model based on historical online running datasets are as follows: Based on the historical timestamp of the ore discharge action, a phase identifier is generated for each sampling point within the ore discharge cycle, forming an input-output sample group; The state variable dimension is determined based on the input-output sample set, the ore discharge cycle is used as the state evolution time index, and the state update relationship is constructed for recursive update. The state space twin model structure is formed by establishing the output mapping relationship. Using historical online running datasets, the state update relationship and output mapping relationship of the state-space twin model structure are calibrated with parameters. After meeting the consistency requirements, the parameters are solidified to form a reselected digital twin model.
[0010] As a preferred embodiment of the Nielsen reselection process monitoring method based on digital twins described in this invention, the steps of continuously injecting the online running dataset into the reselection digital twin model for online synchronization, and outputting the twin prediction value set and twin prediction residuals are as follows: The online running dataset is continuously fed into the reselection digital twin model, and the online running dataset is promoted through state update relationships to make online predictions and output a set of twin prediction values. At the same sampling point, read the actual observed output quantity that has the same output dimension as the output mapping relationship, and then combine the difference between the actual observed output quantity and the twin prediction value set to output the twin prediction residual.
[0011] As a preferred embodiment of the Nielsen reselection process monitoring method based on digital twins described in this invention, the specific steps for setting bed state variables in the reselection digital twin model based on the combination of twin predicted value set and twin predicted residuals with non-invasive observation signals are as follows. The twin prediction set and twin prediction residuals are indexed and bound point by point according to the ore discharge cycle and phase identifier, and the twin residuals of the sampling points are output. Non-intrusive observation signals are continuously acquired, and window slicing and fixed-dimensional feature processing are performed according to the ore discharge cycle and phase identifier to output non-intrusive observation features; In the reselection digital twin model, bed state variables are set, and quasi-steady-state evolution constraints are applied to the bed state variables to form an extended twin state structure.
[0012] As a preferred embodiment of the Nielsen reselection process monitoring method based on digital twins described in this invention, the step of generating a set of bed state parameters by performing online assimilation and inversion through filtered state estimation is as follows: An observation mapping terminal is established by using historical online operation data and non-intrusive observation features, and the twin residuals of sampling points and non-intrusive observation features are fused into an assimilated observation vector through the observation mapping terminal. Based on the extended twin state structure, the bed state variables are perturbed and sampled and advanced through the state update relationship to generate the bed state prior set; Using the assimilated observation vector as the observation constraint, online assimilation and inversion of the prior set of bed states are performed through filtered state estimation, and the posterior set of bed states is output. Representative values are extracted from the posterior set of bed state and structured and encapsulated according to the ore discharge cycle to output the set of bed state parameters.
[0013] As a preferred embodiment of the Nielsen reselection process monitoring method based on digital twins described in this invention, the specific steps for forming the bed stability index are as follows: The set of bed state parameters is organized according to the ore discharge cycle, and each bed state parameter is associated with a corresponding preset stable operating range, and the state parameter sequence is output. During the ore discharge cycle, the state parameter sequence is compared with the associated preset stable operating range point by point to calculate the state deviation. The state deviation is then processed and aggregated using a unified scale to output a cycle comprehensive deviation index. The bed stability index is generated through quantitative conversion.
[0014] As a preferred embodiment of the Nielsen reseparation process monitoring method based on digital twins described in this invention, the steps for establishing a discharge fingerprint template by using a set of bed state parameters and a bed stability index as inputs, and superimposing the discharge pipeline pressure signal and a set of twin prediction values, are as follows: The set of bed state parameters and the bed stability index are aligned and organized, and the pressure signal of the ore discharge pipeline and the set of twin prediction values are simultaneously connected to form a fingerprint input sequence. Based on the timestamp and phase identifier of the ore discharge action, the pressure signal of the ore discharge pipeline is aligned for events, and the ore discharge pulse characteristics are extracted within the ore discharge time window; Based on the fingerprint input sequence, the set of bed state parameters and the bed stability index are used as working condition constraints, and the twin prediction value set and the discharge pulse characteristics are integrated to form a structured model, thus forming a discharge fingerprint template.
[0015] As a preferred embodiment of the Nielsen reselection process monitoring method based on digital twins described in this invention, the step of calculating the consistency residual between the real-time discharge pulse features and the discharge fingerprint template, and outputting the monitoring conclusion, refers to extracting the real-time discharge pulse features during online operation, calculating the consistency residual between the real-time discharge pulse features and the discharge pulse features in the discharge fingerprint template, and comparing the consistency residual with the consistency judgment threshold.
[0016] The beneficial effects of this invention are as follows: by constructing and synchronizing the Nielsen reselection digital twin model online, the dynamic characterization of the reselection process operation mechanism and working condition changes can be realized; and further, based on the twin prediction residual and non-invasive observation signal, the bed state is assimilated and inverted online, and the bed stability is transformed into a quantifiable monitoring index, thereby improving the foresight and effectiveness of the reselection process monitoring. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart of a Nielsen reselection process monitoring method based on digital twins.
[0019] Figure 2 This is a flowchart of time synchronization and preprocessing.
[0020] Figure 3 This is a flowchart of the online assimilation and inversion process for bed state.
[0021] Figure 4 This is a schematic diagram showing the changes in feed flow rate and feed density with the discharge cycle.
[0022] Figure 5 This is a schematic diagram illustrating the variation of twin prediction residuals with the ore discharge cycle. Detailed Implementation
[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0026] Reference Figures 1-5 As one embodiment of the present invention, this embodiment provides a Nielsen reselection process monitoring method based on digital twins, comprising the following steps: S1. Real-time acquisition of reselection operation status data, time synchronization and preprocessing of the reselection operation status data to form an online operation dataset.
[0027] It should be noted that the reselection operation status data includes feed flow rate, feed density, motor current, discharge pipeline pressure signal, and discharge action timestamp.
[0028] S1.1 Synchronize the reselection operation status data with time by aligning the sampling frequency and resampling, and output time-aligned operation data.
[0029] It should be noted that the original sampling frequency of the reselection operation status data is statistically analyzed and a unified target sampling frequency is selected; a target sampling timestamp sequence is generated within a unified time range according to the fixed time interval corresponding to the target sampling frequency; the reselection operation status data is resampled based on the target sampling timestamp sequence, and a complete data record is generated at each target sampling timestamp to form resampled data; the resampled data is encapsulated using the target sampling timestamp sequence as an index to obtain time-aligned operation data.
[0030] S1.2. Using the timestamp of the ore discharge action as an event marker, perform event alignment and ore discharge process cycle segmentation on the time-aligned operation data, and output periodic operation data.
[0031] It should be noted that the timestamp of the ore discharge action in the time-aligned running data is used as the event marker, and an ore discharge event label is generated at the corresponding target sampling timestamp; the start time and end time of the ore discharge cycle are determined by the time interval between adjacent ore discharge event labels, and the time-aligned running data is segmented and encapsulated according to the start time and end time of the ore discharge cycle to output periodic running data.
[0032] S1.3 Perform outlier removal, missing data processing, noise suppression, and dimension consistency processing on the periodic operation data, and encapsulate it in a structured manner according to the ore discharge cycle to output the online operation dataset.
[0033] It should be noted that outlier detection is performed on the periodic operation data for each mining cycle, and outlier sampling points are removed. After the outlier sampling points are removed, missing data marking and missing data imputation are performed on the periodic operation data to form a complete sampling sequence. Noise suppression processing is performed on the complete sampling sequence to obtain a smoothed sampling sequence, and dimension unification processing is performed on the smoothed sampling sequence to obtain a uniform-scale sampling sequence. The uniform-scale sampling sequence is summarized and structured according to the mining cycle number, and the online operation dataset is output.
[0034] like Figure 4 As shown, the Nelson gravity separation process exhibits variations in feed flow rate and feed density over the discharge cycle under continuous operation conditions. The overview diagram depicts the global fluctuation characteristics of feed parameters over a long operating time scale, while the enlarged local diagram displays the subtle changes in feed flow rate and feed density within a selected discharge cycle interval. This more clearly reflects the dynamic changes in feed conditions at different time scales, providing a basic operating condition basis for subsequent analysis of the impact of operating condition changes on bed state and stability.
[0035] S2. Construct a digital twin model for reselection based on the historical online running dataset, and continuously inject the online running dataset into the digital twin model for online synchronization, outputting a set of twin prediction values and twin prediction residuals.
[0036] S2.1. Based on the historical timestamp of the ore discharge action, generate a phase identifier for each sampling point within the ore discharge cycle to form an input-output sample group.
[0037] It should be noted that the start and end times of the ore discharge cycle are determined by the historical timestamp of the ore discharge action. A phase identifier is calculated for each target sampling timestamp within the ore discharge cycle. The phase identifier is obtained by the relative position of the target sampling timestamp between the start and end times of the ore discharge cycle. The phase identifier is combined with the operating status data of the corresponding sampling point to form input data, and the observation output data of the corresponding sampling point is used as output data. The samples are paired and encapsulated point by point to form an input-output sample group.
[0038] S2.2. Determine the dimensions of state variables based on input-output sample groups, use the ore discharge cycle as the state evolution time index, construct state update relationships for recursive updates, and form a state-space twin model structure by establishing output mapping relationships.
[0039] It should be noted that covariance calculation and eigenvalue decomposition are performed on the input-output sample group. Based on the cumulative contribution of eigenvalues, the number of information components that can cover the main dynamic changes are selected as the state variable dimension. The ore discharge cycle is used as the state evolution time index. The input-output sample group is sorted according to the ore discharge cycle number and phase identifier. A recursive update relationship is set for the state vectors corresponding to the state variable dimension between adjacent ore discharge cycles. The state update relationship obtains the state vector sequence by associating the state vector at the end of the previous ore discharge cycle with the input data in the next ore discharge cycle. An output mapping relationship is established to map the state vector sequence to the output data. The output mapping relationship outputs the prediction result with the same dimension as the output data by performing a linear or nonlinear combination on the state vector and the input data. The state update relationship and the output mapping relationship together form the state-space twin model structure.
[0040] S2.3. Using historical online running datasets, the state update relationship and output mapping relationship of the state-space twin model structure are calibrated with parameters, and after meeting the consistency requirements, the parameters are solidified to form a reselected digital twin model.
[0041] It should be noted that when calibrating the parameters of the state-space twin model structure using the historical online operation dataset, the historical online operation dataset is sorted by ore discharge cycle number and phase identifier and then input into the state-space twin model structure. The state update relationship is recursively generated to generate the trajectory of state variables under the ore discharge cycle time index, and the output mapping relationship outputs the predicted output at each sampling point. The difference between the observed output and the predicted output in the historical online operation dataset is calculated point by point to form the predicted residual sequence. The parameter calibration uses the sum of squares or the sum of absolute values of the predicted residual sequence as the optimization objective. The state update relationship parameters and the output mapping relationship parameters are updated iteratively until the optimization objective converges. The consistency requirement is achieved by testing the statistical properties of the predicted residual sequence on the historical online operation dataset. The criteria for meeting the consistency requirement include the mean of the predicted residual sequence being close to zero and the fluctuation range of the predicted residual sequence being within the allowable fluctuation range (e.g., several times the standard deviation of the predicted residual sequence during the stable operation phase). After the consistency requirement is met, the state update relationship parameters and the output mapping relationship parameters are written into the fixed parameter configuration and locked for updating, completing the parameter solidification and forming a reselected digital twin model.
[0042] It should also be noted that the training process of the reselected digital twin model is as follows: the historical online running dataset is divided into a parameter calibration subset and a consistency verification subset. The parameter calibration subset is used for iterative optimization of the updated state update relation parameters and output mapping relation parameters. The consistency verification subset is used to calculate the statistics of the predicted residual sequence and perform consistency requirement judgment. If the consistency requirement is not met, the model returns to the parameter calibration subset to continue iterative optimization. If the consistency requirement is met, the output parameter solidification result is output and iterative optimization stops.
[0043] S2.4 Continuously inject the online running dataset into the reselection digital twin model, and advance the online prediction of the online running dataset through the state update relationship, and output the set of twin prediction values.
[0044] It should be noted that the online running dataset is sorted by the ore discharge cycle number and phase identifier, and then input into the reselection digital twin model point by point. The reselection digital twin model reads the input quantity corresponding to the online running dataset at each sampling point and substitutes it into the state update relationship to perform a state recursion. The state variables output by the state recursion are used to calculate the predicted output quantity of the corresponding sampling point through the output mapping relationship. The predicted output quantity is indexed and encapsulated according to the ore discharge cycle number, phase identifier and target sampling timestamp and continuously appended to form a set of twin predicted values.
[0045] The predicted output for the corresponding sampling point is calculated using the output mapping relationship, expressed as follows: ; in: This is the predicted output for the corresponding sampling point; It is the state mapping coefficient matrix in the output mapping relationship, which is used to linearly map the state variable vector to the prediction output space; The first one is obtained by recursion from the state update relation. The state variable vector of each sampling point, the dimension of which is determined by the input-output sample set; It is the input mapping coefficient matrix in the output mapping relationship, used to linearly map the input vector of the online running dataset to the predicted output space; It is the first Each sampling point corresponds to an online running dataset input vector, which consists of a phase identifier and a running state quantity. It is a sampling point index determined by the combined ore discharge cycle and phase identifier.
[0046] S2.5 Read the actual observed output quantity that is consistent with the output dimension of the output mapping relationship at the same sampling point, and combine the difference between the actual observed output quantity and the twin prediction value set to output the twin prediction residual.
[0047] It should be noted that at the same sampling point, the actual observed output vector with the same output dimension as the output mapping relationship is read based on the ore discharge cycle number, phase identifier and target sampling timestamp, and the corresponding predicted output vector is read from the twin prediction value set; the difference operation is performed on the actual observed output vector and the predicted output vector to obtain the twin prediction residual.
[0048] like Figure 5 As shown, the twin prediction residuals between the predicted output of the reseparation digital twin model and the actual operational observation results change with the discharge cycle. The twin prediction residuals reflect the fitting deviation of the digital twin model to the operating state of the reseparation process and its dynamic response characteristics as the operating conditions change. When the operating state of the reseparation process changes, the residual amplitude and distribution characteristics are adjusted accordingly, thus providing sensitive information for identifying operating condition deviations and potential anomalies. This residual sequence provides an important basis for subsequent online assimilation and inversion of bed state based on non-intrusive observation signals.
[0049] S3. Based on the twin prediction set and twin prediction residuals combined with non-intrusive observation signals, bed state variables are set in the reselection digital twin model, and online assimilation inversion is performed through filtered state estimation to generate a set of bed state parameters.
[0050] S3.1. Bind the twin prediction set and twin prediction residuals to each sampling point index according to the ore discharge cycle and phase identifier, and output the sampling point twin residuals.
[0051] It should be noted that the twin prediction set and twin prediction residual are indexed by the ore discharge cycle number and phase identifier, respectively, and the target sampling timestamp is matched point by point under the same ore discharge cycle number and phase identifier. The prediction output corresponding to each target sampling timestamp is bound to the twin prediction residual to form a unified sampling point record, and the sampling point twin residual is output after structured encapsulation.
[0052] S3.2 Continuously acquire non-intrusive observation signals, and perform window slicing and fixed-dimensional feature processing according to the ore discharge cycle and phase identifier to output non-intrusive observation features.
[0053] It should be noted that non-intrusive observation signals are continuously acquired and a unified time reference timestamp is added to the non-intrusive observation signals; based on the ore discharge cycle number and phase identifier, the non-intrusive observation signals are aligned to the range of the start time and end time of the ore discharge cycle according to the target sampling timestamp, and window slicing is performed on the non-intrusive observation signals using a fixed time window corresponding to the phase identifier to obtain window signal segments; fixed-dimensional feature processing is performed on each window signal segment. Fixed-dimensional feature processing calculates time-domain statistics and frequency-domain statistics and splices them according to a preset feature order to form a fixed-length feature vector, including time-domain statistics (such as mean, variance, peak value, kurtosis, and skewness) and frequency-domain statistics (such as dominant frequency, spectral entropy, and energy proportion of a specific frequency band); the fixed-length feature vector is structured and encapsulated according to the ore discharge cycle number and phase identifier to output the non-intrusive observation features.
[0054] It should also be noted that the feature order is determined by analyzing the correlation or sensitivity between non-invasive observation signals and bed state parameters (or twin prediction residuals) in the historical online running dataset.
[0055] S3.3. In the reselection digital twin model, set the bed state variables and apply quasi-steady-state evolution constraints to the bed state variables to form an extended twin state structure.
[0056] It should be noted that bed state variables are added to the state variable vector of the reselected digital twin model, and the variable dimensions of the state update relationship and output mapping relationship are updated synchronously. Quasi-steady-state evolution constraints are applied to the bed state variables. By limiting the change amplitude of bed state variables at adjacent target sampling timestamps, the state recursion process is controlled, forming an extended twin state structure. The unmeasurable internal bed state is explicitly incorporated into the model framework, providing a structural basis for filtering and assimilation. At the same time, process physical constraints (quasi-steady state) are integrated, which significantly improves the ability to represent bed dynamics, estimation accuracy and anomaly sensitivity, and supports highly reliable non-destructive online monitoring.
[0057] It should also be noted that the bed state variables are determined by analyzing the continuous shift characteristics of the twin prediction residuals during the discharge cycle, and by combining the variation law of the non-intrusive observation signal under the same discharge cycle and phase identifier. Quasi-steady-state evolution constraints are obtained by statistically analyzing the changes in bed-related quantities during the stable operation phase of historical online operation data between adjacent target sampling timestamps. These constraints are used to constrain the state recursion process, ensuring that the evolution of bed state variables conforms to the physical laws of the slow dynamic changes of the bed during the reselection process. This effectively suppresses estimation noise and non-physical interpretations, and improves the stability and reliability of state inversion.
[0058] S3.4. Establish an observation mapping terminal using historical online operation data and non-intrusive observation features, and fuse the twin residuals of sampling points and non-intrusive observation features into an assimilated observation vector through the observation mapping terminal.
[0059] It should be explained that historical online operation data is aligned to non-intrusive observation features according to the ore discharge cycle number, phase identifier, and target sampling timestamp. Residual samples corresponding to the twin residuals of the sampling points are extracted from the historical online operation data to form calibration samples for the observation mapping end. Parameter calibration is performed on the calibration samples to obtain the mapping parameters that map the non-intrusive observation features to the twin residual space of the sampling points, and the observation mapping end is established. During online operation, the observation mapping end outputs the residual equivalent quantity for the non-intrusive observation features, and the residual equivalent quantity is concatenated with the twin residuals of the sampling points according to the same ore discharge cycle number, phase identifier, and target sampling timestamp to form an assimilated observation vector.
[0060] S3.5. Based on the extended twin state structure, the bed state variables are perturbed and sampled, and the state update relationship is used to generate the bed state prior set.
[0061] It should be noted that, based on the extended twin state structure, the reference value and disturbance range of the bed state variables at the current sampling point are determined, and multiple disturbance samplings are performed on the bed state variables within the disturbance range to obtain a sample group of bed state variables; each bed state variable sample is combined with other state variables in the extended twin state structure to form a bed state sample, and the bed state sample is substituted into the state update relation to perform a recursive advancement on the target sampling timestamp sequence; the bed state samples after the state update relation advancement are encapsulated in a set according to the ore discharge cycle number, phase identifier and target sampling timestamp to generate a bed state prior set.
[0062] S3.6 Using the assimilated observation vector as the observation constraint, online assimilation and inversion are performed on the prior set of bed states through filtered state estimation, and the posterior set of bed states is output.
[0063] It should be noted that the assimilated observation vector is read at the target sampling timestamp, and the assimilated observation vector is subtracted from the observation mapping output obtained by the observation mapping end for each bed state prior sample in the bed state prior set to obtain the observation residual; the bed state prior set is assigned and normalized filter weights according to the size of the observation residuals; the bed state prior set is weighted and updated according to the filter weights to output the bed state posterior set.
[0064] S3.7 Extract representative values from the posterior set of bed state and encapsulate them in a structured manner according to the ore discharge cycle to output the set of bed state parameters.
[0065] It should be noted that the posterior set of bed state is grouped according to the ore discharge cycle number, phase identifier, and target sampling timestamp. Within each group, the values of the bed state variables are sorted by numerical value, and the values at both ends of the sorted set are discarded. Each value (for example, the example value is...) (That is, removing the two largest and two smallest values), and calculating the representative value of the remaining values using the interval-truncated mean, the expression is: ; in: The ore discharge cycle number is Phase identifier is The target sampling timestamp is The representative values of the bed state variables within the group; This refers to the number of bed state variables that participate in the calculation of the interval truncation mean, representing the values at each end after sorting. The number of valid samples remaining after taking each value; It is the upper limit index position used for summation in the sorted sequence of bed state variable values, corresponding to the removal of the maximum value. The index of the last valid value after all possible values; The posterior set of bed conditions in the ore discharge cycle is numbered as follows: Phase identifier is The target sampling timestamp is Within the group, the first [number] is sorted by the value of the bed state variable. The values of the state variables of each bed layer; The posterior set of bed conditions in the ore discharge cycle is numbered as follows: Phase identifier is The target sampling timestamp is The total number of bed state variable values included in the group; This is the interval truncation parameter, representing the number of values removed from the minimum and maximum ends of the sorted sequence of bed state variable values. For example, the example value is... ; It is the ore discharge cycle number, used to identify the ore discharge process cycle to which the posterior set of bed conditions belongs; It is a phase identifier, used to identify the relative phase position of the posterior set of bed conditions within the corresponding ore discharge cycle; Is it the sorted number? The target sampling timestamp index corresponding to the value of each bed state variable is used to identify the sampling position of that value on the time axis; It is the index number of the sequence of values of the bed state variables after sorting; The representative values corresponding to each target sampling timestamp are concatenated in the order of the ore discharge cycle number to form the ore discharge cycle bed state sequence. The ore discharge cycle bed state sequence is then encapsulated in a structured manner to output a set of bed state parameters.
[0066] S4. A comprehensive evaluation is conducted on the deviation of the set of bed state parameters from the preset stable operating range to form a bed stability index.
[0067] S4.1 Organize the set of bed state parameters according to the ore discharge cycle, associate each bed state parameter with a corresponding preset stable operating range, and output the state parameter sequence.
[0068] It should be noted that the set of bed state parameters is grouped and organized according to the discharge cycle number to form a discharge cycle bed state parameter sequence; each bed state parameter in the discharge cycle bed state parameter sequence is associated with the corresponding preset stable operating range upper and lower limits and encapsulated to output the state parameter sequence.
[0069] It should also be noted that the upper and lower limits of the stable operation range are obtained by statistically analyzing the bed state parameters in the stable operation state of the historical online operation dataset. Specifically, the center value and discrete range of the bed state parameter values are calculated during the stable operation phase, and the upper and lower boundaries covering the main value distribution range are determined as the upper and lower limits of the stable operation range.
[0070] S4.2 During the ore discharge cycle, the state parameter sequence is compared with the associated preset stable operating range point by point to calculate the state deviation. The state deviation is then processed and aggregated using a unified scale to output the cycle comprehensive deviation index. The bed stability index is generated through quantitative conversion.
[0071] It should be noted that the bed state parameter values are read from the target sampling timestamp in the state parameter sequence corresponding to the ore discharge cycle number, and the lower limit and upper limit of the stable operating range corresponding to the bed state parameter are read. When the bed state parameter value falls between the lower limit and the upper limit of the stable operating range, the state deviation is zero. When the bed state parameter value is less than the lower limit of the stable operating range, the state deviation is the difference between the lower limit of the stable operating range and the bed state parameter value. When the bed state parameter value is greater than the upper limit of the stable operating range, the state deviation is the difference between the bed state parameter value and the upper limit of the stable operating range. The difference is calculated; the deviation is processed by uniform scaling according to the width of the stable operating range or the historical statistical scale to obtain the dimensionless deviation, and the dimensionless deviation is summed or averaged within the ore discharge cycle number range to obtain the cycle comprehensive deviation index; the cycle comprehensive deviation index is mapped to the bed stability index through monotonic quantization transformation, which is used to characterize the health status and stability level of the bed operation in real time during the re-separation process, providing a clear criterion for process stability, and serving as the working condition constraint input and abnormal early warning trigger basis for the subsequent ore discharge fingerprint template construction, supporting the hierarchical monitoring decision of "stability-deviation-abnormality".
[0072] S5. Using the set of bed state parameters and the bed stability index as inputs, and superimposing the discharge pipeline pressure signal and twin prediction value set, a discharge fingerprint template is established. The consistency residual between the real-time discharge pulse characteristics and the discharge fingerprint template is calculated, and the monitoring conclusion is output.
[0073] S5.1 Align and organize the set of bed state parameters with the bed stability index, and simultaneously connect the discharge pipeline pressure signal and twin prediction value set to form a fingerprint input sequence.
[0074] It should be noted that the set of bed state parameters and the bed stability index are aligned and organized according to the discharge cycle number, phase identifier, and target sampling timestamp; the discharge pipeline pressure signal and twin prediction value set are synchronously accessed and their fields are spliced according to the same discharge cycle number, phase identifier, and target sampling timestamp to form a fingerprint input sequence.
[0075] S5.2. Based on the timestamp and phase identifier of the ore discharge action, perform event alignment on the ore discharge pipeline pressure signal and extract the ore discharge pulse characteristics within the ore discharge time window.
[0076] It should be explained that, based on the timestamp of the discharge action, the discharge event is located in the pressure signal of the discharge pipeline and time alignment is completed according to the phase identifier; the pressure signal of the discharge pipeline within the discharge time window is extracted with the timestamp of the discharge action as the center to form a discharge pressure pulse segment; the pressure peak value, pulse width and pulse area and other features of the discharge pressure pulse segment are extracted and encapsulated to obtain the discharge pulse feature, which is used to construct the discharge fingerprint template under historical working conditions, and is compared with the pulse feature extracted in real time to accurately identify abnormal states in the discharge process (such as pipeline blockage, abnormal slurry concentration and valve response lag), and provide direct and quantifiable criteria for monitoring conclusions.
[0077] S5.3 Based on the fingerprint input sequence, the set of bed state parameters and the bed stability index are used as working condition constraints, and the twin prediction value set and the discharge pulse characteristics are integrated to form a structured model, thus forming a discharge fingerprint template.
[0078] It should be noted that the fingerprint input sequence is grouped according to the discharge cycle number and phase identifier, and samples are screened within each group by limiting the working condition range based on the set of bed state parameters and the bed stability index. The twin prediction value set and discharge pulse characteristics in the screened samples are concatenated and aligned according to the phase identifier to form a fixed structure record. The feature statistical description of the fixed structure record is calculated and the value range of the template field is determined. The fixed structure record and the value range of the template field are structurally encapsulated according to the discharge cycle number and phase identifier to form the discharge fingerprint template.
[0079] S5.4 Extract real-time ore discharge pulse features from the online operation process, calculate the consistency residual between the real-time ore discharge pulse features and the ore discharge pulse features in the ore discharge fingerprint template, and compare the consistency residual with the consistency judgment threshold to generate monitoring conclusions.
[0080] It should be noted that during online operation, the discharge time window is extracted from the discharge pipeline pressure signal based on the discharge action timestamp, and the real-time discharge pulse characteristics are extracted; the corresponding discharge pulse characteristic template value range or template value vector is read from the discharge fingerprint template according to the discharge cycle number and phase identifier; the consistency residual is calculated between the real-time discharge pulse characteristics and the discharge pulse characteristic template values. The consistency residual is calculated using feature-by-feature difference or normalized distance and aggregated into a single-value residual; the consistency residual is compared with the interval of the consistency judgment threshold one by one. When the consistency residual is within the normal threshold interval of the consistency judgment threshold, a normal discharge monitoring conclusion is output; when the consistency residual exceeds the normal threshold interval of the consistency judgment threshold, an abnormal discharge monitoring conclusion is output.
[0081] It should also be noted that the consistency judgment threshold is set based on the statistical results of the consistency residuals of the ore discharge pulse characteristics in the stable operation phase of the historical online operation data. Specifically, the central tendency and dispersion of the consistency residual sequence in the stable operation phase are calculated, and the upper and lower boundaries covering the main stable fluctuation range are determined as the consistency judgment threshold. The consistency judgment threshold is determined based on the statistical distribution range of consistency residuals during the stable operation phase of historical online operation datasets, using dimensionless consistency residuals as the object. For example, the example value is that the consistency residuals are located in the interval... This indicates that the ore discharge operation is normal, and the consistency residual is within the range. This indicates a slight deviation in the ore discharge conditions, with a consistency residual greater than [value missing]. This indicates an abnormal ore discharge condition.
[0082] In summary, this invention achieves dynamic characterization of the reselection process's operating mechanism and condition changes by constructing and synchronizing a Nielsen reselection digital twin model online; and further, it performs online assimilation and inversion of the bed state based on the twin prediction residuals and non-invasive observation signals, transforming bed stability into quantifiable monitoring indicators, thereby improving the foresight and effectiveness of reselection process monitoring.
[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for monitoring a Nielsen reselection process based on digital twinning, the method comprising: include, Real-time acquisition of re-selection operation status data, time synchronization and preprocessing of the re-selection operation status data to form an online operation dataset; the re-selection operation status data includes feed flow rate, feed density, motor current, discharge pipeline pressure signal and discharge action timestamp; A digital twin model for reselection is constructed based on historical online running datasets, and the online running datasets are continuously injected into the digital twin model for online synchronization, outputting a set of twin prediction values and twin prediction residuals; Based on the twin prediction set and twin prediction residual combined with non-invasive observation signals, bed state variables are set in the reselection digital twin model, and online assimilation inversion is performed through filtered state estimation to generate a set of bed state parameters. The specific steps for setting bed state variables in the reselection digital twin model, based on the twin prediction value set and twin prediction residuals combined with non-invasive observation signals, are as follows: The twin prediction set and twin prediction residuals are indexed and bound point by point according to the ore discharge cycle and phase identifier, and the twin residuals of the sampling points are output. Non-intrusive observation signals are continuously acquired, and window slicing and fixed-dimensional feature processing are performed according to the ore discharge cycle and phase identifier to output non-intrusive observation features; In the reselection digital twin model, bed state variables are set, and quasi-steady-state evolution constraints are applied to the bed state variables to form an extended twin state structure; A comprehensive evaluation is conducted on the deviation of the set of bed state parameters from the preset stable operating range to form a bed stability index. The specific steps are as follows: The set of bed state parameters is organized according to the ore discharge cycle, and each bed state parameter is associated with a corresponding preset stable operating range, and the state parameter sequence is output. During the ore discharge cycle, the state parameter sequence is compared with the associated preset stable operating range point by point to calculate the state deviation. The state deviation is then processed and aggregated using a unified scale to output the cycle comprehensive deviation index. The bed stability index is generated through quantitative conversion. Using the set of bed state parameters and the bed stability index as inputs, and superimposing the discharge pipeline pressure signal and twin prediction value set, a discharge fingerprint template is established. The consistency residual between the real-time discharge pulse characteristics and the discharge fingerprint template is calculated, and the monitoring conclusion is output.
2. The Nielsen reselection process monitoring method based on digital twin as described in claim 1, characterized in that: The process of synchronizing and preprocessing the reselection operation status data to form an online operation dataset involves the following steps: The reselection operation status data is synchronized with the resampling through sampling frequency alignment, and time-aligned operation data is output. Using the timestamp of the ore discharge action as an event marker, the time-aligned operation data is aligned with events and the ore discharge process cycle is segmented to output periodic operation data; The system performs outlier removal, missing data processing, noise suppression, and dimension consistency processing on the periodic operation data, and encapsulates it in a structured manner according to the ore discharge cycle to output the online operation dataset.
3. The Nielsen reselection process monitoring method based on digital twin as described in claim 1, characterized in that: The specific steps for constructing the digital twin model for reselection based on historical online operation datasets are as follows. Based on the historical timestamp of the ore discharge action, a phase identifier is generated for each sampling point within the ore discharge cycle, forming an input-output sample group; The state variable dimension is determined based on the input-output sample set, the ore discharge cycle is used as the state evolution time index, and the state update relationship is constructed for recursive update. The state space twin model structure is formed by establishing the output mapping relationship. Using historical online running datasets, the state update relationship and output mapping relationship of the state-space twin model structure are calibrated with parameters. After meeting the consistency requirements, the parameters are solidified to form a reselected digital twin model.
4. The Nielsen reselection process monitoring method based on digital twin as described in claim 3, characterized in that: The process of continuously injecting the online running dataset into the reselected digital twin model for online synchronization, and outputting the twin prediction value set and twin prediction residuals, is detailed below. The online running dataset is continuously fed into the reselection digital twin model, and the online running dataset is promoted through state update relationships to make online predictions and output a set of twin prediction values. At the same sampling point, read the actual observed output quantity that has the same output dimension as the output mapping relationship, and then combine the difference between the actual observed output quantity and the twin prediction value set to output the twin prediction residual.
5. The Nielsen reselection process monitoring method based on digital twin as described in claim 1, characterized in that: The process of performing online assimilation and inversion through filtered state estimation to generate a set of bed state parameters involves the following specific steps. An observation mapping terminal is established by using historical online operation data and non-intrusive observation features, and the twin residuals of sampling points and non-intrusive observation features are fused into an assimilated observation vector through the observation mapping terminal. Based on the extended twin state structure, the bed state variables are perturbed and sampled and advanced through the state update relationship to generate the bed state prior set; Using the assimilated observation vector as the observation constraint, online assimilation and inversion of the prior set of bed states are performed through filtered state estimation, and the posterior set of bed states is output. Representative values are extracted from the posterior set of bed state and structured and encapsulated according to the ore discharge cycle to output the set of bed state parameters.
6. The Nielsen reselection process monitoring method based on digital twin as described in claim 1, characterized in that: The process of establishing a discharge fingerprint template by using a set of bed state parameters and a bed stability index as inputs, and superimposing the discharge pipeline pressure signal and a set of twin prediction values, is as follows: The set of bed state parameters and the bed stability index are aligned and organized, and the pressure signal of the ore discharge pipeline and the set of twin prediction values are simultaneously connected to form a fingerprint input sequence. Based on the timestamp and phase identifier of the ore discharge action, the pressure signal of the ore discharge pipeline is aligned for events, and the ore discharge pulse characteristics are extracted within the ore discharge time window; Based on the fingerprint input sequence, the set of bed state parameters and the bed stability index are used as working condition constraints, and the twin prediction value set and the discharge pulse characteristics are integrated to form a structured model, thus forming a discharge fingerprint template.
7. The Nielsen reselection process monitoring method based on digital twin as described in claim 6, characterized in that: The calculation of the consistency residual between the real-time ore discharge pulse features and the ore discharge fingerprint template, and the output monitoring conclusion, refers to extracting the real-time ore discharge pulse features during online operation, calculating the consistency residual between the real-time ore discharge pulse features and the ore discharge pulse features in the ore discharge fingerprint template, and comparing the consistency residual with the consistency judgment threshold.