A material tracking method, system and storage medium based on multi-model data aggregation

Abstract

The present application relates to the technical field of material tracking, and particularly relates to a material tracking method and system based on multi-model data aggregation and a storage medium, a material tracking aggregation model containing multiple parallel sub-models is constructed, diversified tracking bases are provided for process equipment with different material retention characteristics, feedback benchmark data is introduced and performance measurement values of each sub-model are calculated in real time, online evaluation of model tracking accuracy is realized, the system can perceive real-time changes in its own performance, the contribution of each sub-model to the final output result is dynamically determined according to the performance measurement values, the system can adaptively adjust model weights or switch main models according to changes in working conditions, and it is ensured that the optimal tracking strategy is always adopted, finally, the contribution degree is used to aggregate and calculate the outlet material prediction value set, high-precision outlet material tracking data is generated and output to a downstream control system, and accurate and reliable incoming material information is provided for a downstream process.

Description

Technical Field

[0001] This invention relates to the field of material tracking technology, and specifically to a material tracking method, system, and storage medium based on multi-model data aggregation. Background Technology

[0002] In processes such as sintering and pelletizing, materials undergo unique physical transformations, posing challenges to conventional material tracking methods. Take, for example, equipment in the sintering process such as sintering mixers, pelletizers, and mixing silos: For "line-to-line" equipment such as mixers and granulators, materials continuously enter the equipment in a "linear" flow form via belt conveyor, and after undergoing mixing and granulation within the equipment, they exit in the same "linear" flow form. Current technology typically employs the method of installing online monitoring instruments (such as moisture meters and component analyzers) at the equipment inlet and outlet and directly correlating the data. However, materials have residence time within the equipment, and the residence time varies among different particles. Directly correlating inlet and outlet data leads to severe data mismatch; that is, the material data detected at the outlet actually corresponds to material that entered the equipment several minutes ago, rather than material entering at the current moment. This timing misalignment prevents downstream processes from obtaining accurate incoming material information, severely restricting the real-time performance and accuracy of control.

[0003] For "line-pile-line" buffer devices such as mixing silos: after the material enters the silo in a "linear" flow, it forms a "pile," where complex mixing and retention occur before it is discharged in a "linear" flow to the sintering machine. In this case, the material not only experiences a time delay but also undergoes a drastic change in physical form from "line" to "pile" and back to "line," completely disrupting the correspondence between inlet and outlet data. Existing technologies often struggle to effectively handle this data correspondence, either abandoning tracking altogether or employing overly simplified average delay models, resulting in severely distorted tracking data that fails to accurately reflect the characteristics of the discharged material.

[0004] In summary, existing material data tracking technologies have the following shortcomings when dealing with process equipment that involves internal retention, mixing, and morphological changes: 1) They cannot accurately compensate for the residence time of materials within the equipment, resulting in data timing misalignment; 2) For "line-stall-line" type equipment, there is a lack of effective data mapping models, leading to a complete disconnect between inlet and outlet data; 3) The tracking results have poor accuracy and cannot provide a reliable data foundation for advanced process control. Summary of the Invention

[0005] The main objective of this invention is to provide a material tracking method, system, and storage medium based on multi-model data aggregation, in order to solve the technical problems of data timing misalignment, inlet and outlet data disconnection, and poor tracking accuracy in the prior art when dealing with process equipment with internal retention, mixing, and morphological changes.

[0006] To achieve the above objectives, this invention provides a material tracking method based on multi-model data aggregation, applicable to process equipment with characteristics of material internal retention, mixing, and / or morphological changes, comprising the following steps: S1. Obtain a preset sampling period, and continuously collect real-time data of material parameters at the inlet of the target equipment according to the sampling period to generate an inlet data sequence; S2. Obtain a material tracking aggregation model containing multiple parallel sub-models, input the ingress data sequence into each sub-model in the material tracking aggregation model, and each sub-model independently calculates the predicted value of the exit material at the current time to generate a set of predicted values ​​of the exit material. S3. Obtain feedback benchmark data for evaluating the performance of the sub-models, and calculate the performance metric value of each sub-model in real time at the current moment or within a recent time window based on the feedback benchmark data. S4. Determine the contribution of each sub-model to the final output result at the current time based on the performance metric value; wherein the contribution is a continuous value or a discrete value, and the sum of all contributions is 1; S5. Aggregate and calculate the predicted export material value set according to the contribution level to generate export material tracking data at the current moment, and output the export material tracking data for use by the control system or monitoring system of the downstream process.

[0007] Furthermore, when the contribution rate is a continuous value, the export material tracking data in step S5 is obtained through the following steps: Based on the performance metric, calculate the error statistic for each sub-model within a preset time window; wherein the performance metric is the root mean square error or the mean absolute error, and the error statistic is the performance metric or the monotonic transformation value of the performance metric. Obtain a preset mapping rule, and convert the error statistics into initial weights for each sub-model according to the mapping rule; The initial weights are normalized to obtain the contribution of each sub-model. The export material prediction values ​​are weighted and summed according to the contribution level to generate the export material tracking data for the current moment.

[0008] Furthermore, when the contribution is a discrete value, the export material tracking data in step S5 is obtained by the following steps: Compare the performance metrics of each sub-model, determine which sub-model's performance metrics best meet the preset performance optimization trend, and thus determine the sub-model with the best performance. Wherein, when the performance metric is an error value, the preset performance optimization trend is to reduce the error value, thereby determining the sub-model with the smallest error value among all sub-models as the sub-model with the best performance; Alternatively, when the performance metric is accuracy, the preset performance optimization trend is to increase accuracy, thereby determining the sub-model with the highest accuracy among all sub-models as the sub-model with the best performance. The sub-model with the best performance is used as the main model, and the remaining models are used as secondary models; the contribution of the main model is set to 1, and the contribution of the secondary models is set to 0; the export material prediction value output by the main model is used as the export material tracking data at the current moment.

[0009] Furthermore, the material tracking aggregation model comprising multiple parallel sub-models includes at least two of the following four models: Moving average model based on average dwell time; Convolutional models based on dwell time distribution; Mechanism-based model based on ideal hybrid unit cascade; Exponentially weighted models.

[0010] More preferably, the moving average model based on average dwell time is a simple moving average (SMA) model, which includes the following calculation steps: The average residence time and sampling period of the material in the target equipment are obtained, and the number of sampling points participating in the aggregation operation is determined based on the average residence time and sampling period; where N=T / Δt; where N is the number of sampling points, T is the average residence time, and Δt is the sampling period; Calculate the arithmetic mean of N sampling points back from the current time t in the inlet data sequence, and use it as the predicted value of the outlet material. ; in, The material discharged from the equipment at time t represents the average characteristics of the material that entered the equipment within the past time T after complete mixing.

[0011] More preferably, the convolutional model is a convolutional model RTD based on dwell time distribution, which includes the following calculation steps: Obtain the residence time distribution function E(τ) of the material in the target equipment, where E(τ) is a probability density function that characterizes the probability distribution of the residence time τ of the material particles in the equipment; The ingress data sequence is convolved with the dwell time distribution function, and the calculation formula is as follows: The integration interval covers the non-zero range of E(τ); Alternatively, a summation form can be used in the discrete domain: Where i ranges from 0 to M, and M covers the effective range of E(τ); In the formula, Indicates the ingress data sequence. Describe the phenomenon of materials arriving early or late within the target equipment.

[0012] More preferably, the mechanistic model is a series fully mixed-flow model (CSTR), which includes the following calculation steps: Obtain the ideal number N of the target equipment's fully mixed flow reactors. cstr And the total average stay time T, where N cstr It is an integer greater than or equal to 1; The target device is equivalent to N. cstr A series of ideal continuously stirred-flow reactors of equal volume are connected in series, and the average residence time of each ideal continuously stirred-flow reactor is t. m =T / N cstr ; Establish the material balance differential equation for each fully mixed-flow reactor: Where j represents N cstr The j-th of an ideal fully mixed reactor; ; The Nth differential equation system was solved using numerical methods. cstr Concentration of an ideal fully mixed flow reactor As export material forecast value ; Where, when N cstr When N = 1, the series fully mixed flow model CSTR degenerates into the assumed state of complete mixing. cstr When the value increases, the series fully mixed-flow model CSTR approximates the dwell time distribution behavior of the target device.

[0013] More preferably, the exponentially weighted model is an exponentially weighted moving average (EWMA) model, which includes the following calculation steps: Obtain the preset smoothing factor α, where 0 < α ≤ 1; Using recursive formulas Calculate the predicted export material value at the current time t; in, It not only reflects the impact of current inbound data, but also achieves a forward-looking estimate of the changing trend of inbound materials by using exponential decay weighting of historical data.

[0014] This invention also provides a material tracking system based on multi-model data aggregation, which applies the material tracking method based on multi-model data aggregation as described above, including: The data acquisition module is used to obtain a preset sampling period, continuously collect real-time data of material parameters at the inlet of the target equipment according to the sampling period, and generate an inlet data sequence; A multi-model parallel computing module, connected to the data acquisition module, is used to acquire a material tracking aggregation model containing multiple parallel sub-models, input the ingress data sequence into each sub-model in the material tracking aggregation model, and each sub-model independently calculates the predicted value of the exit material at the current time to generate a set of predicted values ​​of the exit material. The feedback module is used to acquire feedback benchmark data for evaluating the performance of the sub-model; The performance evaluation module, connected to the multi-model parallel computing module and the feedback module, is used to calculate the performance metric value of each sub-model in real time at the current moment or within a recent time window based on the feedback benchmark data. A contribution determination module, connected to the performance evaluation module, is used to determine the contribution of each sub-model to the final output result at the current time based on the performance metric value; wherein the contribution is a continuous value or a discrete value, and the sum of all contributions is 1; The data aggregation and output module is connected to the multi-model parallel computing module and the contribution determination module. It is used to perform aggregation calculation on the set of predicted export materials based on the contribution, generate export material tracking data at the current moment, and output the export material tracking data for use by the control system or monitoring system of downstream processes.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the material tracking method based on multi-model data aggregation as described above.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs a material tracking aggregation model comprising multiple parallel sub-models, providing a diverse tracking foundation for process equipment with varying material retention characteristics. By introducing feedback benchmark data and calculating the performance metrics of each sub-model in real time, it achieves online evaluation of model tracking accuracy, enabling the system to perceive real-time changes in its own performance. Furthermore, based on the performance metrics, it dynamically determines the contribution of each sub-model to the final output, allowing the system to adaptively adjust model weights or switch the main model according to changes in operating conditions, ensuring the use of the currently optimal tracking strategy. Finally, it aggregates and calculates the predicted output material values ​​based on their contributions, generating high-precision output material tracking data and outputting it to the downstream control system, providing accurate and reliable incoming material information for downstream processes. This invention transforms open-loop material tracking into closed-loop adaptive tracking, significantly improving the system's adaptability to changes in operating conditions and its ability to maintain tracking accuracy. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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 the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a material tracking method based on multi-model data aggregation according to an embodiment of the present invention. Figure 2 This figure shows a comparison of the weight distribution effects of different sub-models in one embodiment of the present invention. In the figure: Horizontal axis: Time backtracking (minutes), 0 is the current time. Vertical axis: Weight coefficients. Weight curve A (SMA): A rectangle of equal height within the [-T, 0] window, 0 outside the window. Weight curve B (EWMA): A smooth curve that decays exponentially from the current time to the past. Weight curve C (RTD): An asymmetric, bell-shaped curve resembling a normal distribution.

[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0023] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0024] Please see Figures 1 to 2 This embodiment provides a material tracking method based on multi-model data aggregation, applicable to process equipment with characteristics of material internal retention, mixing, and / or morphological changes, including the following steps: S1. Obtain a preset sampling period, and continuously collect real-time data of material parameters at the inlet of the target equipment according to the sampling period to generate an inlet data sequence; Specifically, those skilled in the art can pre-set a suitable sampling period Δt based on the characteristics of the tracked material and the response requirements of the downstream control system. For example, in the sintering process, for the key parameter of material moisture content, the sampling period Δt can typically be set to 10 seconds. Subsequently, online detection instruments (such as infrared moisture meters, component analyzers, etc.) are installed at the inlet of the target equipment (such as a sintering mixer, granulator, or mixing silo) to continuously collect the moisture data of the inlet material at the sampling period Δt, thereby generating an inlet data sequence that changes over time, denoted as . Where t represents the current time, such as 10:30:00, 10:30:10, 10:30:20, etc. This step provides the basic data input for all subsequent calculations, ensuring that the material tracking process has a real-time and continuous data source available.

[0025] S2. Obtain a material tracking aggregation model containing multiple parallel sub-models, input the ingress data sequence into each sub-model in the material tracking aggregation model, and each sub-model independently calculates the predicted value of the exit material at the current time to generate a set of predicted values ​​of the exit material. Specifically, this invention pre-configures a material tracking aggregation model, which stores multiple parallel sub-models. These sub-models are constructed based on different physical assumptions or mathematical principles to adapt to different equipment characteristics and operating conditions. In a preferred embodiment, the multiple sub-models include at least two of the following four types of models: Simple Moving Average (SMA); Exponentially Weighted Moving Average (EWMA); Convolutional Model Based on Residence Time Distribution (RTD); and Series Fully Mixed Flow Model (CSTR).

[0026] For the Simple Moving Average (SMA) model: This model assumes that the material is completely mixed within the target equipment, and the characteristics of the outlet material are equal to the arithmetic mean of the inlet material over a past full average residence time. The calculation method is as follows: Based on the average residence time T of the target equipment and the sampling period Δt, determine the number of sampling points N = T / Δt participating in the aggregation operation; then calculate the arithmetic mean of N sampling points back from the current time t in the inlet data sequence, which serves as the predicted value of the outlet material for this model. ).

[0027] For example, if the average residence time of a mixer is T = 5 minutes and the sampling period is Δt = 10 seconds, then N = 30. The system calculates the average of 30 inlet moisture data over the past 5 minutes as the predicted value of the outlet material at the current moment. SMA can effectively eliminate random fluctuations in inlet data, but it introduces a pure time lag equal to T.

[0028] For the Exponentially Weighted Moving Average (EWMA) model: This model uses a recursive calculation method, giving higher weight to recent data. The core value of EWMA lies in its ability to respond quickly to changes in inlet data: when inlet data changes abruptly, EWMA can immediately and slightly correct the outlet forecast, thereby achieving a forward-looking estimate of material change trends.

[0029] For example, if a sudden increase in inlet moisture is detected at 10:30:00, even though the wet material has just entered the equipment and there are still about 5 minutes before it actually reaches the outlet, the EWMA model will immediately raise the estimated value of the current outlet moisture. This allows the downstream control system to sense the trend of an approaching wet material in advance and take countermeasures in advance, thereby significantly reducing the control lag caused by simply using the moving average method.

[0030] For the residence time distribution-based convolutional model RTD: the residence time distribution function E(τ) of the equipment is pre-obtained through pulse tracer experiments. This function is a probability density function, characterizing the probability distribution of the residence time τ of material particles within the target equipment. Subsequently, a convolution operation is performed on the inlet data sequence and the residence time distribution function. RTD can accurately characterize the "early departure" and "late arrival" phenomena of materials within the equipment and is theoretically the optimal estimation method. For example, after obtaining the RTD curve of a mixing silo through experiments, the system will take all inlet data at 10:30:00, which is approximately twice the residence time of the previous time, multiply it by the probability density value at the corresponding time point on the RTD curve, and then sum them to obtain the final predicted value of the outlet material.

[0031] For the series-connected fully mixed flow (CSTR) model: the actual target equipment is equivalent to M ideal fully mixed flow reactors of equal volume connected in series, with the material in each ideal CSTR reactor being instantaneously and completely mixed. First, the equivalent number of series reactors M and the total average residence time T are obtained. The average residence time of each ideal CSTR reactor is t. m =T / N. Then, establish the material balance differential equation for each ideal continuously stirred-tank reactor. Solve the differential equation using numerical methods (such as the Euler method) to obtain the concentration of the Mth reactor. As the export material forecast value of this model CSTR decomposes the complex homogenization process into a series of multiple ideal mixing units, which retains the mechanistic basis and is easier to implement than the RTD model. When M=1, the model degenerates into the perfect homogenization assumption (approximately SMA); as M increases, its behavior gradually approaches that of the RTD model.

[0032] Through the parallel computation described above, a set of M predicted export material values ​​can be obtained at each sampling time t, providing rich candidate results for subsequent adaptive decision-making.

[0033] S3. Obtain feedback benchmark data for evaluating the performance of the sub-models, and calculate the performance metric value of each sub-model in real time at the current moment or within a recent time window based on the feedback benchmark data. Specifically, this step introduces a closed-loop feedback mechanism. For equipment equipped with online outlet monitoring instruments, the system directly acquires the measured values ​​of the outlet material parameters at the current moment as feedback benchmark data. For example, a moisture meter is installed at the mixer outlet to read the measured outlet moisture value in real time. Then, the system calculates the instantaneous prediction error of each sub-model and further statistically analyzes the error statistics of each sub-model within a sliding time window of a preset length L, as a performance metric. The error statistics can be root mean square error or mean absolute error. For example, setting L=6 (i.e., 6 sampling points within the most recent minute), the system continuously updates the RMSE of each sub-model as an indicator of its recent tracking accuracy.

[0034] For equipment without outlet monitoring instruments, the system acquires indirect performance data reflecting model performance as a feedback benchmark. This indirect performance data includes quality inspection values ​​from downstream processes (such as sintering machine exhaust gas temperature, finished pellet compressive strength, etc.) or equipment operating parameters (such as material level, power, speed, etc.). For example, a sudden change in the material level in the mixing silo indicates a significant change in the material's residence time distribution within the silo, potentially reducing the applicability of the atomic model. The system can use the material level change as a trigger signal or construct an empirical relationship between the material level change rate and the model's expected error to indirectly evaluate the performance of each sub-model. Through this method, regardless of whether an outlet monitoring instrument is available, the system can obtain real-time or near-real-time evaluations of the performance of each sub-model.

[0035] S4. Determine the contribution of each sub-model to the final output result at the current time based on the performance metric value; wherein the contribution is a continuous value or a discrete value, and the sum of all contributions is 1; Specifically, based on the performance metric calculated in step S3, the contribution of each sub-model to the final output is dynamically determined. This embodiment supports two working modes to adapt to the needs of different industrial scenarios.

[0036] Continuous Value Contribution Mode (Weighted Fusion Mode): In this mode, the contribution is a continuous value, ranging from 0 to 1, and the sum of the contributions of all models is 1. The system first calculates the initial weights of each model based on the performance metric, following the mapping rule of "the better the performance (the smaller the error), the larger the weight". The specific mapping method can use any of the following to obtain the initial weights: reciprocal transformation, negative exponential transformation, and linear decreasing function. These three mapping methods are existing technologies and will not be elaborated upon here.

[0037] After obtaining the initial weights, the system performs normalization to obtain the final contribution. This mode allows multiple models to contribute to the final output simultaneously, and the contribution changes smoothly with model performance, avoiding the output jumps that may be caused by "hard switching".

[0038] Discrete Value Contribution Mode (Model Switching Mode): In this mode, the contribution is a discrete value. The system selects one of multiple sub-models as the master model, setting its contribution to 1, while the contributions of the remaining models are set to 0. This mode is simple to implement, has low computational overhead, and is suitable for scenarios with extremely high real-time requirements or where there are limitations on multi-model fusion computation. The system can also set an error threshold; switching is only performed when the performance metric of the optimal model is consistently better than other models and falls below this threshold, avoiding frequent switching due to instantaneous fluctuations.

[0039] Through the above steps, the system can adaptively determine the contribution of each sub-model to the final output at each sampling time based on the current operating conditions and the actual performance of each model.

[0040] S5. Aggregate and calculate the predicted export material value set according to the contribution level to generate export material tracking data at the current moment, and output the export material tracking data for use by the control system or monitoring system of the downstream process.

[0041] Specifically, the system performs aggregate calculations based on the contribution determined in step S4 and the set of predicted export material values ​​generated in step S2: If a weighted fusion model is used, the export material tracking data is a weighted sum of the predictions from each model. For example, at a certain moment, the RMSE of the RTD model is 0.2%, the EWMA model is 0.25%, the SMA model is 0.4%, and the CSTR model is 0.3%. After calculation using the softmax function, the RTD model has the highest contribution (e.g., 0.5), followed by EWMA (0.3), while SMA and CSTR are lower. The final output export material tracking data is a weighted average of these four export material predictions, which is closer to the current most accurate RTD model while retaining information from other models, resulting in a smooth output with high robustness.

[0042] If the model switching mode is used, the export material tracking data will directly take the predicted value of the main model.

[0043] After generating the export material tracking data, it is output to the control system (such as the moisture feedforward control system of the sintering machine, or the component adjustment module of the batching system) or monitoring system (such as the human-machine interface of the operator station, or the quality early warning system). Based on this high-precision, real-time adaptive tracking data, the downstream system can adjust process parameters or assess the quality of incoming materials in advance. For example, after receiving the export moisture tracking data, if the sintering machine's thermal state control system detects an upward trend in moisture content, it can increase the ignition temperature or adjust the feeding speed in advance, thereby optimizing parameters before the wet material actually arrives at the sintering machine and avoiding fluctuations in the quality of the sintered ore.

[0044] This embodiment achieves high-precision, adaptive material tracking for process equipment with internal retention, mixing, and morphological change characteristics through the closed-loop process of S1 to S5, providing a reliable data foundation for the refined control and intelligent upgrading of the process industry.

[0045] This embodiment constructs a material tracking aggregation model containing multiple parallel sub-models, providing a diverse tracking foundation for process equipment with different material retention characteristics. By introducing feedback benchmark data and calculating the performance metrics of each sub-model in real time, online evaluation of model tracking accuracy is achieved, enabling the system to perceive real-time changes in its own performance. Furthermore, based on the performance metrics, the contribution of each sub-model to the final output result is dynamically determined, allowing the system to adaptively adjust model weights or switch the main model according to changes in operating conditions, ensuring that the current optimal tracking strategy is always adopted. Finally, the predicted values ​​of the exit material are aggregated and calculated based on the contribution, generating high-precision exit material tracking data and outputting it to the downstream control system, providing accurate and reliable incoming material information for downstream processes. This invention transforms open-loop material tracking into closed-loop adaptive tracking, significantly improving the system's adaptability to changes in operating conditions and its ability to maintain tracking accuracy.

[0046] Furthermore, when the contribution is a discrete value, the export material tracking data in step S5 is obtained through the following steps: Compare the performance metrics of each sub-model, determine which sub-model's performance metrics best meet the preset performance optimization trend, and thus determine the sub-model with the best performance. Wherein, when the performance metric is an error value, the preset performance optimization trend is to reduce the error value, thereby determining the sub-model with the smallest error value among all sub-models as the sub-model with the best performance; Alternatively, when the performance metric is accuracy, the preset performance optimization trend is to increase accuracy, thereby determining the sub-model with the highest accuracy among all sub-models as the sub-model with the best performance. The sub-model with the best performance is used as the main model, and the remaining models are used as secondary models; the contribution of the main model is set to 1, and the contribution of the secondary models is set to 0; the export material prediction value output by the main model is used as the export material tracking data at the current moment.

[0047] Furthermore, the material tracking aggregation model comprising multiple parallel sub-models includes at least two of the following four models: Moving average model based on average dwell time; Convolutional models based on dwell time distribution; Mechanism-based model based on ideal hybrid unit cascade; Exponentially weighted models.

[0048] More preferably, the moving average model based on average dwell time is a simple moving average (SMA) model, which includes the following calculation steps: The average residence time and sampling period of the material in the target equipment are obtained, and the number of sampling points participating in the aggregation operation is determined based on the average residence time and sampling period; where N=T / Δt; where N is the number of sampling points, T is the average residence time, and Δt is the sampling period; Calculate the arithmetic mean of N sampling points back from the current time t in the inlet data sequence, and use it as the predicted value of the outlet material. ; in, The material discharged from the equipment at time t represents the average characteristics of the material that entered the equipment within the past time T after complete mixing.

[0049] More preferably, the convolutional model is a convolutional model RTD based on dwell time distribution, which includes the following calculation steps: Obtain the residence time distribution function E(τ) of the material in the target equipment, where E(τ) is a probability density function that characterizes the probability distribution of the residence time τ of the material particles in the equipment; The ingress data sequence is convolved with the dwell time distribution function, and the calculation formula is as follows: The integration interval covers the non-zero range of E(τ); Alternatively, a summation form can be used in the discrete domain: Where i ranges from 0 to M, and M covers the effective range of E(τ); In the formula, Indicates the ingress data sequence. Describe the phenomenon of materials arriving early or late within the target equipment.

[0050] In one specific embodiment, the residence time distribution function E(τ) is obtained through a pulsed tracer experiment. The specific procedure is as follows: an appropriate amount of tracer (such as colored pigment, fluorescent substance, etc.) is instantaneously injected at the device inlet, while simultaneously the tracer concentration change curve c(t) over time is continuously detected at the device outlet. The detected concentration curve is then normalized to obtain the residence time distribution function. This experiment is usually conducted under stable operating conditions to ensure that the obtained E(τ) accurately reflects the true flow characteristics of the equipment.

[0051] More preferably, the mechanistic model is a series fully mixed-flow model (CSTR), which includes the following calculation steps: Obtain the ideal number N of the target equipment's fully mixed flow reactors. cstr And the total average stay time T, where Ncstr It is an integer greater than or equal to 1; The target device is equivalent to N. cstr A series of ideal continuously stirred-flow reactors of equal volume are connected in series, and the average residence time of each ideal continuously stirred-flow reactor is t. m =T / N cstr ; Establish the material balance differential equation for each fully mixed-flow reactor: Where j represents N cstr The j-th of an ideal fully mixed reactor; ; The Nth differential equation system was solved using numerical methods. cstr Concentration of an ideal fully mixed flow reactor As export material forecast value ; Where, when N cstr When N = 1, the series fully mixed flow model CSTR degenerates into the assumed state of complete mixing. cstr When the value increases, the series fully mixed-flow model CSTR approximates the dwell time distribution behavior of the target device.

[0052] For the above system of differential equations, this embodiment preferably uses the Euler method for discrete solution. Let the sampling period be Δt, then the discrete calculation formula for the k-th sampling time is: Where j=1,2,…,N cstr , This represents the measured value of the inlet material parameters at the previous sampling time. Through iterative calculations, the concentration at the k-th sampling time can be obtained. As the export material forecast value of this model .

[0053] More preferably, the exponentially weighted model is an exponentially weighted moving average (EWMA) model, which includes the following calculation steps: Obtain the preset smoothing factor α, where 0 < α ≤ 1; Using recursive formulas Calculate the predicted export material value at the current time t; in, It not only reflects the impact of current inbound data, but also achieves a forward-looking estimate of the changing trend of inbound materials by using exponential decay weighting of historical data.

[0054] The value of the smoothing factor α can be determined based on the dynamic characteristics and control requirements of the target equipment. In a specific embodiment, if it is desired that the response speed of the exponentially weighted moving average (EWMA) model is equivalent to that of a simple moving average with N points, α can be approximately calculated using the following formula: Where, N eq The equivalent window length can be estimated based on the device's average dwell time T and sampling period Δt, i.e., N. eq =T / Δt. In practical applications, an appropriate α value can also be selected by balancing response speed and the stability of the estimated value through on-site debugging.

[0055] It should be noted that the physical quantities involved in the above models have specific dimensions in practical engineering applications: the average residence time T is in seconds (s); the sampling period Δt is in seconds (s); and the residence time distribution function E(τ) is in seconds. - ¹(s - ¹), satisfying ∫E(τ)dτ=1; the smoothing factor α is a dimensionless number.

[0056] This invention also provides a material tracking system based on multi-model data aggregation, which applies the material tracking method based on multi-model data aggregation as described above, including: The data acquisition module is used to obtain a preset sampling period, continuously collect real-time data of material parameters at the inlet of the target equipment according to the sampling period, and generate an inlet data sequence; A multi-model parallel computing module, connected to the data acquisition module, is used to acquire a material tracking aggregation model containing multiple parallel sub-models, input the ingress data sequence into each sub-model in the material tracking aggregation model, and each sub-model independently calculates the predicted value of the exit material at the current time to generate a set of predicted values ​​of the exit material. The feedback module is used to acquire feedback benchmark data for evaluating the performance of the sub-model; The performance evaluation module, connected to the multi-model parallel computing module and the feedback module, is used to calculate the performance metric value of each sub-model in real time at the current moment or within a recent time window based on the feedback benchmark data. A contribution determination module, connected to the performance evaluation module, is used to determine the contribution of each sub-model to the final output result at the current time based on the performance metric value; wherein the contribution is a continuous value or a discrete value, and the sum of all contributions is 1; The data aggregation and output module is connected to the multi-model parallel computing module and the contribution determination module. It is used to perform aggregation calculation on the set of predicted export materials based on the contribution, generate export material tracking data at the current moment, and output the export material tracking data for use by the control system or monitoring system of downstream processes.

[0057] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the material tracking method based on multi-model data aggregation as described above.

[0058] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A material tracking method based on multi-model data aggregation, applied to process equipment with characteristics of material internal retention, mixing, and / or morphological changes, characterized in that, Includes the following steps: S1. Obtain a preset sampling period, and continuously collect real-time data of material parameters at the inlet of the target equipment according to the sampling period to generate an inlet data sequence; S2. Obtain a material tracking aggregation model containing multiple parallel sub-models, input the ingress data sequence into each sub-model in the material tracking aggregation model, and each sub-model independently calculates the predicted value of the exit material at the current time to generate a set of predicted values ​​of the exit material. S3. Obtain feedback benchmark data for evaluating the performance of the sub-models, and calculate the performance metric value of each sub-model in real time at the current moment or within a recent time window based on the feedback benchmark data. S4. Determine the contribution of each sub-model to the final output result at the current time based on the performance metric value; wherein the contribution is a continuous value or a discrete value, and the sum of all contributions is 1; S5. Aggregate and calculate the predicted export material value set according to the contribution level to generate export material tracking data at the current moment, and output the export material tracking data for use by the control system or monitoring system of the downstream process.

2. The material tracking method based on multi-model data aggregation according to claim 1, characterized in that, When the contribution rate is a continuous value, the export material tracking data in step S5 is obtained by the following steps: Based on the performance metric, calculate the error statistic for each sub-model within a preset time window; wherein the performance metric is the root mean square error or the mean absolute error, and the error statistic is the performance metric or the monotonic transformation value of the performance metric. Obtain a preset mapping rule, and convert the error statistics into initial weights for each sub-model according to the mapping rule; The initial weights are normalized to obtain the contribution of each sub-model. The export material prediction values ​​are weighted and summed according to the contribution level to generate the export material tracking data for the current moment.

3. The material tracking method based on multi-model data aggregation according to claim 1, characterized in that, When the contribution is a discrete value, the export material tracking data in step S5 is obtained by the following steps: Compare the performance metrics of each sub-model, determine which sub-model's performance metrics best meet the preset performance optimization trend, and thus determine the sub-model with the best performance. Wherein, when the performance metric is an error value, the preset performance optimization trend is to reduce the error value, thereby determining the sub-model with the smallest error value among all sub-models as the sub-model with the best performance; Alternatively, when the performance metric is accuracy, the preset performance optimization trend is to increase accuracy, thereby determining the sub-model with the highest accuracy among all sub-models as the sub-model with the best performance. The sub-model with the best performance is used as the main model, and the remaining models are used as secondary models; the contribution of the main model is set to 1, and the contribution of the secondary models is set to 0; the export material prediction value output by the main model is used as the export material tracking data at the current moment.

4. The material tracking method based on multi-model data aggregation according to claim 1, characterized in that, The material tracking aggregation model comprises multiple parallel sub-models, wherein the sub-models include at least two of the following four models: Moving average model based on average dwell time; Convolutional models based on dwell time distribution; Mechanism-based model based on ideal hybrid unit cascade; Exponentially weighted models.

5. The material tracking method based on multi-model data aggregation according to claim 4, characterized in that, The moving average model based on average dwell time is a simple moving average (SMA) model, which includes the following calculation steps: The average residence time and sampling period of the material in the target equipment are obtained, and the number of sampling points participating in the aggregation operation is determined based on the average residence time and sampling period; where N=T / Δt; where N is the number of sampling points, T is the average residence time, and Δt is the sampling period; Calculate the arithmetic mean of N sampling points back from the current time t in the inlet data sequence, and use it as the predicted value of the outlet material. ; in, The material discharged from the equipment at time t represents the average characteristics of the material that entered the equipment within the past time T after complete mixing.

6. The material tracking method based on multi-model data aggregation according to claim 4, characterized in that, The convolutional model is a real-time diff (RTD) convolutional model based on dwell time distribution, which includes the following calculation steps: Obtain the residence time distribution function E(τ) of the material in the target equipment, where E(τ) is a probability density function that characterizes the probability distribution of the residence time τ of the material particles in the equipment; The ingress data sequence is convolved with the dwell time distribution function, and the calculation formula is as follows: The integration interval covers the non-zero range of E(τ); Alternatively, a summation form can be used in the discrete domain: Where i ranges from 0 to M, and M covers the effective range of E(τ); In the formula, Indicates the ingress data sequence. Describe the phenomenon of materials arriving early or late within the target equipment.

7. The material tracking method based on multi-model data aggregation according to claim 4, characterized in that, The mechanistic model is a series fully mixed-flow model (CSTR), which includes the following calculation steps: Obtain the ideal number N of the target equipment's fully mixed flow reactors. cstr And the total average stay time T, where N cstr It is an integer greater than or equal to 1; The target device is equivalent to N. cstr A series of ideal continuously stirred-flow reactors of equal volume are connected in series, and the average residence time of each ideal continuously stirred-flow reactor is t. m =T / N cstr ; Establish the material balance differential equation for each fully mixed-flow reactor: Where j represents N cstr The j-th of an ideal perfectly mixed reactor; C0(t) = C in (t); The Nth differential equation system was solved using numerical methods. cstr The concentration C of an ideal mixed-flow reactor Ncstr (t) is the forecast value of exported materials. ; Where, when N cstr When N = 1, the series fully mixed flow model CSTR degenerates into the assumed state of complete mixing. cstr When the value increases, the series fully mixed-flow model CSTR approximates the dwell time distribution behavior of the target device.

8. The material tracking method based on multi-model data aggregation according to claim 4, characterized in that, The exponentially weighted model is the exponentially weighted moving average (EWMA) model, which includes the following calculation steps: Obtain the preset smoothing factor α, where 0 < α ≤ 1.

9. A material tracking system based on multi-model data aggregation, employing the material tracking method based on multi-model data aggregation as described in any one of claims 1-8, characterized in that, include: The data acquisition module is used to obtain a preset sampling period, continuously collect real-time data of material parameters at the inlet of the target equipment according to the sampling period, and generate an inlet data sequence; A multi-model parallel computing module, connected to the data acquisition module, is used to acquire a material tracking aggregation model containing multiple parallel sub-models, input the ingress data sequence into each sub-model in the material tracking aggregation model, and each sub-model independently calculates the predicted value of the exit material at the current time to generate a set of predicted values ​​of the exit material. The feedback module is used to obtain feedback benchmark data for evaluating the performance of the sub-model; The performance evaluation module, connected to the multi-model parallel computing module and the feedback module, is used to calculate the performance metric value of each sub-model in real time at the current moment or within a recent time window based on the feedback benchmark data. A contribution determination module, connected to the performance evaluation module, is used to determine the contribution of each sub-model to the final output result at the current time based on the performance metric value; wherein the contribution is a continuous value or a discrete value, and the sum of all contributions is 1; The data aggregation and output module is connected to the multi-model parallel computing module and the contribution determination module. It is used to perform aggregation calculation on the set of predicted export materials based on the contribution, generate export material tracking data at the current moment, and output the export material tracking data for use by the control system or monitoring system of downstream processes.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the material tracking method based on multi-model data aggregation as described in any one of claims 1 to 8.

Images