Intelligent scheduling control method for tailings resource recycling based on multi-parameter evaluation model

Abstract

The application relates to a tailing resource recycling intelligent scheduling control method based on a multi-parameter evaluation model, and belongs to the technical field of mining resource intelligent scheduling control. The method comprises the following steps: acquiring multi-dimensional original parameter data; calculating the reliability coefficients of each data source and acquiring the standardized characteristic components of each dimension; taking the reliability coefficient of the corresponding data source as the fusion weight of the corresponding dimension standardized characteristic component for weighted combination, generating a fusion feature vector and constructing a time sequence characteristic sequence; inputting the time sequence characteristic sequence into a pre-constructed resource potential evaluation model, extracting the physical property change trend characteristics between batches; splicing the physical property change trend characteristics and the current batch fusion feature vector, inputting the spliced result into the resource potential evaluation model, outputting the resource potential score and the process matching label of the current batch of tailings, and guiding the current batch of tailings into a processing process line corresponding to the process matching label. The application can improve the accuracy of tailing resource potential evaluation.

Description

Technical Field

[0001] This application relates to the technical field of intelligent scheduling and control of mineral resources, and in particular to an intelligent scheduling and control method for tailings resource reuse based on a multi-parameter evaluation model. Background Technology

[0002] In the process of tailings resource utilization, accurately assessing the utilization value of tailings and selecting appropriate treatment processes accordingly is a key step in determining resource recovery efficiency.

[0003] Currently, industrial sites typically employ single-type detection methods for online analysis of tailings. For example, some concentrators rely solely on X-ray fluorescence spectrometry to obtain elemental content information or laser particle size analyzers to acquire particle size distribution data. These single-dimensional parameters are then compared with preset thresholds to determine the tailings' utilization value and allocate appropriate processing procedures. However, the resource potential of tailings is determined by a combination of physicochemical properties, including particle size characteristics, chemical composition, and mineral occurrence state. A single parameter often only reflects one aspect of the tailings' characteristics, failing to comprehensively characterize its overall utilization value. This lack of reliable basis for selecting subsequent processing technologies leads to resource misallocation.

[0004] Even when using multiple types of sensors for joint detection, the signal quality of each data source varies due to environmental interference and other factors in the industrial environment. If these data are fused indiscriminately with equal weights, the data from the lower-quality data source can easily mislead the fusion results. Furthermore, existing methods typically make independent judgments based solely on the detection data of the current batch, without considering the dynamic trends of material properties between batches. This can lead to delayed or overly sensitive responses when material properties undergo trend changes, resulting in frequent process switching and increasing the risk of equipment wear and temporary material mismatch.

[0005] Therefore, there is an urgent need for a scheduling and control method that can comprehensively utilize multi-dimensional physical property parameters to accurately assess the resource potential of tailings and achieve intelligent matching of processing technologies. Summary of the Invention

[0006] To improve the accuracy of tailings resource utilization potential assessment, this application provides an intelligent scheduling and control method for tailings resource reuse based on a multi-parameter evaluation model.

[0007] Firstly, this application provides an intelligent scheduling and control method for tailings resource reuse based on a multi-parameter evaluation model, employing the following technical solution: The current batch of tailings is monitored online by a multi-source sensor array deployed on the tailings conveying channel to obtain multi-dimensional raw parameter data; The data quality of the multidimensional raw parameter data is evaluated, and the credibility coefficient of each data source is calculated based on the signal-to-noise ratio and data integrity of each data source. The multidimensional original parameters are mapped to the standardized feature space to obtain the standardized feature components of each dimension; The credibility coefficient of the corresponding data source is used as the fusion weight of the standardized feature components of the corresponding dimension. The weighted combination is performed to generate a fusion feature vector. Based on the fusion feature vectors of the current batch and the previous N batches, a time series feature sequence is constructed. The time-series feature sequence is input into the time-series coding layer of the pre-constructed resource potential assessment model to extract the trend features of inter-batch physical property changes; The physical property change trend characteristics are concatenated with the current batch fusion feature vector and input into the evaluation decision layer of the resource potential assessment model. The resource potential score and process matching label of the current batch of tailings are output. The evaluation decision layer applies a switching cost penalty term to the process matching label jump between adjacent batches. Based on the process matching tag, a process switching instruction is sent to import the current batch of tailings into the processing process line corresponding to the process matching tag.

[0008] By adopting the above technical solution, multi-source sensors collect multi-dimensional raw parameters online. The reliability coefficient is calculated based on the signal-to-noise ratio and completeness of each data source and used as the fusion weight. The characteristics of a single batch are jointly input into the evaluation model with the time-series characteristics of the previous N batches. At the same time, a switching cost penalty is applied to the process label jump of adjacent batches. This allows the system to consider both the physical property status of the current batch and the physical property evolution trend between batches when making process route allocation decisions. This avoids invalid process switching triggered by occasional score fluctuations, reduces the frequent switching losses of process routes, and improves the accuracy of tailings resource utilization potential assessment.

[0009] Furthermore, the method also includes: When the resource potential score of the current batch of tailings falls into the preset decision fuzzy range, the current batch of tailings is imported into the buffer bin, and the fusion feature vector of the current batch is recorded. After accumulating multiple batches in the buffer silo, the equivalent fusion feature vector of the mixed materials in the buffer silo is calculated by weighted average based on the fusion feature vector of each batch in the buffer silo and the amount of material in each batch. The equivalent fusion feature vector is re-inputted into the resource potential assessment model to obtain the comprehensive resource potential score and process matching label of the mixture. When the comprehensive resource potential score deviates from the decision fuzzy range, the mixture in the buffer bin is imported as a whole into the processing process line corresponding to the process matching label; When the accumulated material quantity in the buffer bin reaches the preset capacity limit and the comprehensive resource potential score is still within the decision fuzzy range, the mixed material in the buffer bin is imported into the default processing line.

[0010] By adopting the above technical solution, for critical batches whose resource potential scores fall into the fuzzy range, materials are temporarily stored in a buffer warehouse, and multi-batch fusion feature vectors are continuously accumulated. The equivalent feature vector of the mixed materials is then used to re-drive the evaluation decision, allowing critical materials to accumulate across batches before being allocated. At the same time, a capacity limit rule is set to ensure that the buffer warehouse does not accumulate indefinitely due to continuous fuzziness, eliminating the risk of scheduling deadlock.

[0011] Further, the step of calculating the equivalent fusion feature vector of the mixed materials in the buffer silo based on the fusion feature vector of each batch and the material quantity of each batch in the buffer silo using a weighted average method includes: Obtain the fused feature vectors and corresponding material quantities of the K batches currently accumulated in the buffer bin; The proportion of each batch of material to the total amount of material in the buffer bin is used as the quality weight factor. At the same time, a time decay factor is constructed based on the time sequence of each batch entering the buffer bin. The time decay factor assigns a lower time weight to the batch that enters the buffer bin earlier. For each batch, the quality weight factor is multiplied by the time decay factor to obtain the composite weight coefficient of the corresponding batch, and the composite weight coefficients of K batches are normalized. The fusion feature vectors of each batch are weighted and summed using the normalized composite weight coefficients of each batch to obtain the initial equivalent fusion feature vectors. Calculate the feature dispersion among the fused feature vectors of K batches within the buffer, where the feature dispersion is the weighted variance of the distance between the fused feature vectors of each batch and the initial equivalent fused feature vector; The feature dispersion is added as an uncertainty correction amount to the initial equivalent fused feature vector to generate the final equivalent fused feature vector, wherein the uncertainty correction amount is used to expand the decision boundary tolerance of the assessment decision layer when inputting the resource potential assessment model.

[0012] By adopting the above technical solution, when calculating the equivalent fusion feature vector of the buffer bin, a composite weight is introduced by simultaneously introducing the material quantity quality weight and the time decay factor. This makes the contribution of the recent batches entering the buffer bin to the equivalent vector greater, avoiding excessive dilution of the current material state representation by the physical property data of the early batches. Furthermore, the characteristic dispersion between batches is calculated and added to the equivalent vector as an uncertainty correction, driving the evaluation decision layer to adaptively expand the decision boundary tolerance and preventing excessive and incorrect allocation of mixtures with significant differences in internal composition.

[0013] Furthermore, the step of calculating the reliability coefficient of each data source based on the signal-to-noise ratio and data integrity of each data source includes: The initial credibility score of each data source is calculated by weighted summation of signal-to-noise ratio and data integrity. Calculate the parameter correlation coefficient between any two data sources in the current batch, and compare the parameter correlation coefficient with the pre-stored historical correlation benchmark value to obtain the correlation deviation between each data source pair; When the correlation deviation of any data source pair exceeds a preset deviation threshold, the data source with the lower initial confidence score in the data source pair is identified as an abnormal suspected source, and a decay factor is applied to the initial confidence score of the abnormal suspected source. The credibility scores of each data source after attenuation correction are normalized to obtain the credibility coefficient of each data source.

[0014] By adopting the above technical solution, cross-data source correlation deviation verification is introduced on the basis of the initial credibility score. When the correlation of any data source pair deviates from the historical benchmark by more than a threshold, the lower initial score is automatically identified as an abnormal suspected source and attenuation is applied. This makes the credibility coefficient not only reflect the signal quality of a single source, but also reflect the physical consistency constraints between multiple sources. In this way, when the sensor has local drift or anomalies, it can suppress its erroneous dominance on the fusion feature vector and improve the reliability of the fusion result.

[0015] Furthermore, the calculation of the parameter correlation coefficient between any two data sources in the current batch includes: Kernel density estimation is performed on the detection data from each data source in the current batch to obtain the marginal probability density function corresponding to each data source; A joint sample is constructed from the detection data of any two data sources, and the joint kernel density is estimated on the joint sample to obtain the joint probability density function between the two data sources. Based on the edge probability density function and the joint probability density function of the two data sources, the mutual information value between the two data sources is calculated according to the definition of mutual information. The mutual information value is normalized to obtain the parameter correlation coefficient between the two data sources. The normalization method is to divide the mutual information value by the geometric mean of the information entropy of each of the two data sources.

[0016] By adopting the above technical solution, the nonlinear correlation coefficient is used to measure the parameter correlation between data sources, which can capture the nonlinear dependence between data sources and avoid the failure of linear correlation measurement when the material composition relationship is complex. By standardizing the mutual information value by dividing it by the geometric mean of the information entropy of the two data sources, the problem of incomparable mutual information dimensions caused by the difference in marginal distribution of different data sources is eliminated, so that the correlation coefficient of cross-data source pairs has a unified comparison benchmark.

[0017] Furthermore, the mapping of the multidimensional original parameter data to the standardized feature space includes: Local nearest neighbor structure analysis is performed on the original parameter data of each dimension to construct the nearest neighbor relationship graph of the data of each dimension in the original space; Based on the nearest neighbor graph, the manifold mapping method is used to map the data of each dimension to a low-dimensional embedding space of the same dimension, so that the data of each dimension maintains the local topological structure of its original space in the low-dimensional embedding space. In the low-dimensional embedding space, scale alignment is performed based on the distribution range of the mapping results of each dimension to obtain the standardized feature components of each dimension.

[0018] By adopting the above technical solution, a nearest neighbor graph is constructed from the original parameter data of each dimension and projected onto a unified low-dimensional embedding space using a manifold mapping method. This achieves cross-dimensional spatial unification while preserving the local topological structure of the data in each dimension. Compared with direct linear normalization, this method can preserve the local distribution pattern of the parameter data in the original space, avoid the loss of structural information after standardization due to significant differences in the distribution patterns of data in different dimensions, and improve the accuracy of the fused feature vector in representing the true physical properties of materials.

[0019] Furthermore, the method also includes: During multiple consecutive batch processing sessions, the change sequence of the reliability coefficient of each data source with each batch is recorded to form the reliability trend curve of each data source. When the credibility trend curve of any data source shows a monotonically decreasing trend within a preset sliding window and its credibility coefficient at the end of the window is lower than a preset degradation threshold, the corresponding data source is marked as a degradation source. Based on the labeling results of the degradation source, the input connection weights of the dimensions corresponding to the degradation source in the evaluation decision layer of the resource potential assessment model are scaled, so that the resource potential assessment model reduces its decision dependence on the feature dimensions corresponding to the degradation source and increases the input connection weights of the dimensions corresponding to other data sources that have not been labeled as degradation sources.

[0020] By adopting the above technical solution, the reliability trend curves of each data source are continuously monitored through a sliding window. When a monotonically decreasing trend is detected and the end is below the degradation threshold, the corresponding data source is marked as a degradation source. The input connection weights of the corresponding dimensions of the degradation source in the evaluation decision layer are scaled in real time, and the weights of the corresponding dimensions of other data sources are increased simultaneously. This enables the resource potential assessment model to adaptively adjust the decision dependency structure when the sensor undergoes progressive degradation, and maintain the stability of the evaluation results without stopping the model for retraining.

[0021] Secondly, this application provides an intelligent scheduling and control device for tailings resource reuse based on a multi-parameter evaluation model, which adopts the following technical solution: The data acquisition module is used to perform online detection of the current batch of tailings through a multi-source sensor group deployed on the tailings conveying channel to obtain multi-dimensional raw parameter data; The credibility coefficient calculation module is used to evaluate the data quality of the multidimensional raw parameter data and calculate the credibility coefficient of each data source according to the signal-to-noise ratio and data integrity of each data source. The standardized feature component determination module is used to map the multidimensional original parameters to a standardized feature space to obtain the standardized feature components of each dimension. The fusion feature processing module is used to use the credibility coefficient of the corresponding data source as the fusion weight of the standardized feature components of the corresponding dimension, perform weighted combination, generate fusion feature vector, and construct time series feature sequence based on the fusion feature vector of the current batch and the previous N batches. The physical property change trend analysis module is used to input the time series feature sequence into the time series encoding layer of the pre-constructed resource potential assessment model to extract the physical property change trend features between batches; The potential assessment module is used to concatenate the physical property change trend characteristics with the current batch fusion feature vector, input it into the assessment decision layer of the resource potential assessment model, and output the resource potential score and process matching label of the current batch of tailings. The assessment decision layer applies a switching cost penalty term to the process matching label jump between adjacent batches. The switching instruction sending module is used to send a process switching instruction according to the process matching tag, so as to import the current batch of tailings into the processing process line corresponding to the process matching tag.

[0022] Thirdly, this application provides an electronic device that adopts the following technical solution: An electronic device, comprising: At least one processor; Memory; At least one computer program, wherein the at least one computer program is stored in the memory and configured to be executed by the at least one processor, the at least one computer program being configured to: perform the method as described in any one of the first aspects.

[0023] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and execute the method as described in any one of the first aspects.

[0024] For a detailed description of the second to fourth aspects of this application and their various implementations, please refer to the detailed description in the first aspect and its various implementations; and for a detailed description of the beneficial effects of the second to fourth aspects and their various implementations, please refer to the beneficial effect analysis in the first aspect and its various implementations, which will not be repeated here.

[0025] In summary, this application includes at least one of the following beneficial technical effects: 1. Multi-source sensors collect multi-dimensional raw parameters online. The reliability coefficient is calculated based on the signal-to-noise ratio and completeness of each data source and used as the fusion weight. The single batch features and the time-series features of the previous N batches are jointly input into the evaluation model. At the same time, a switching cost penalty is applied to the process label jump of adjacent batches. This allows the system to consider both the physical property status of the current batch and the physical property evolution trend between batches when making process route allocation decisions. This avoids invalid process switching triggered by occasional score fluctuations and reduces the frequent switching losses of process routes. 2. For critical batches whose resource potential scores fall into the fuzzy range, materials are temporarily stored in a buffer silo and multi-batch fusion feature vectors are continuously accumulated. The equivalent feature vectors of the mixed materials are used to re-drive the evaluation decision, so that the critical materials are accumulated across batches before being allocated. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the intelligent scheduling and control method for tailings resource reuse based on a multi-parameter evaluation model in this application embodiment.

[0027] Figure 2 This is a flowchart illustrating steps S1021 to S1024 in the embodiments of this application.

[0028] Figure 3 This is a structural block diagram of the intelligent scheduling and control device for tailings resource reuse based on a multi-parameter evaluation model in the embodiments of this application.

[0029] Figure 4 This is a structural block diagram of the electronic device in the embodiments of this application. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0032] This application discloses an intelligent scheduling and control method for tailings resource reuse based on a multi-parameter evaluation model, executed by an electronic device. This electronic device can be a server or a terminal device. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet computer, desktop computer, etc., but is not limited to these.

[0033] This method uses multiple sets of sensors deployed on the tailings conveying channel as data entry points, and sequentially goes through five core processing steps: data quality assessment and credibility quantification, multi-dimensional feature standardization mapping, credibility weighted feature fusion and time series construction, resource potential assessment model reasoning, process matching decision and switching instruction issuance. Finally, it imports the current batch of tailings into a processing process line that is compatible with its resource potential.

[0034] The multi-source sensor array includes at least: an X-ray fluorescence spectrometer installed at the cross-section of the conveyor belt to obtain the mass fractions of major elements such as Fe, Cu, Pb, and Zn; a laser particle size analyzer to obtain particle size distribution characteristics such as D10, D50, and D90; and a near-infrared mineral analyzer to obtain the occurrence ratio of major mineral phases. The sensor sampling outputs are synchronously triggered in batches, with each batch corresponding to the amount of material passing through the cross-section within a certain time window on the conveyor channel.

[0035] The sampling results of each sensor are transmitted to the electronic device in real time via industrial Ethernet. The electronic device completes the entire calculation process of the method and outputs the process switching instructions to the corresponding valve or belt branching device via the programmable logic controller (PLC) to realize the physical diversion of materials.

[0036] Reference Figure 1The method includes (steps S101 to S107):

[0037] Step S101: Online detection of the current batch of tailings is performed by a multi-source sensor group deployed on the tailings conveying channel to obtain multi-dimensional raw parameter data.

[0038] Specifically, when the multi-source sensor group samples the current batch, the multi-source sensor group starts detection synchronously. Taking three data sources—X-ray fluorescence spectrometer, laser particle size analyzer, and near-infrared mineral analyzer—as an example, the same batch of material is sampled in parallel by the three sensors within the same time window, and the results are as follows: Data source (X-ray fluorescence spectrometer): Outputs the mass fraction vector of each target element, denoted as the original parameter. ; Data source (Laser particle size analyzer): Outputs a vector of particle size distribution characteristics, denoted as the original parameter. ; Data source (Near-infrared mineral analyzer): Outputs a mineral phase occurrence ratio vector, denoted as the original parameter. .

[0039] Three-channel raw parameter data , , Together, they form the multidimensional raw parameter dataset for the current batch, which then enters the subsequent quality assessment process.

[0040] Step S102: Perform data quality assessment on the multidimensional raw parameter data, and calculate the reliability coefficient of each data source based on the signal-to-noise ratio and data completeness. (Refer to...) Figure 2 This includes (steps S1021 to S1024):

[0041] Step S1021: Calculate the initial credibility score for each data source based on the weighted sum of the signal-to-noise ratio and data integrity of each data source.

[0042] Specifically, the electronic equipment performs quality diagnostics on the raw output signal of each data source. For each data source... Quality indicators are extracted from two dimensions: Signal-to-noise ratio : Calculate the ratio of effective signal power to noise power of the instrument's original output signal within the batch sampling window, take the logarithm and normalize to the [0,1] interval to reflect the current physical working state of the sensor; Data integrity The ratio of the number of valid sampling points within a batch to the total number of theoretical sampling points reflects the integrity of the communication link and instrument response.

[0043] The two indicators mentioned above are weighted and summed according to preset weights α and (1-α) to calculate the data source. Initial credibility score : ; The weight α is preset based on on-site engineering experience, for example, 0.6, and can be configured during the system initialization phase.

[0044] Step S1022: Calculate the parameter correlation coefficient between any two data sources in the current batch, and compare the parameter correlation coefficient with the pre-stored historical correlation benchmark value to obtain the correlation deviation between each data source pair.

[0045] Specifically, relying solely on the signal quality indicators of a single source is insufficient to detect systematic drift in sensors. Therefore, cross-validation is performed on the parameter correlation between any two data sources. This includes (steps Sa to Sd):

[0046] Step Sa: For each data source in the current batch Kernel density estimation was performed on the detection data to obtain the marginal probability density function corresponding to each data source. .

[0047] Step Sb: Construct a joint sample from the detection data of any two data sources, and perform joint kernel density estimation on the joint sample to obtain the joint probability density function between the two data sources. .

[0048] Specifically, a joint sample is a set of binary samples formed by pairing and combining two corresponding sensor observations within the same batch and time period.

[0049] Set within a certain batch window, A total of T valid readings were output, denoted as: ; T valid readings were output within the same window (if the two sampling frequencies are different, they are aligned to the same timestamp by interpolation or nearest neighbor), denoted as: ; Pair the two readings one by one in chronological order to obtain T pairs: ; The set of these T pairs represents the joint sample of data sources S1 and S2 in this batch.

[0050] The joint sampling preserves the synchronous observation relationship between the two sensors on the same material segment, while the joint kernel density estimation can estimate the co-occurrence distribution of the two variables under the same material state. This reflects cross-source association information about what value the element content tends to take when the granularity feature takes a certain value.

[0051] Step Sc: Based on the marginal probability density functions and joint probability density functions of the two data sources, calculate the mutual information value between the two data sources according to the definition of mutual information. .

[0052] Specifically, mutual information is defined and calculated as follows: .

[0053] Under normal operating conditions, there is an inherent correlation between the properties measured by the two sensors, determined by mineralogical principles. For example, coarse-grained tailings are often accompanied by a higher degree of monomer liberation, and there is also a stoichiometric relationship between element content and mineral occurrence ratio. This inherent correlation forms a stable mutual information benchmark value in historical data.

[0054] When the mutual information value calculated for a certain batch deviates significantly from the benchmark, it indicates that the correlation structure between the observation results of the two sensors has been abnormally disrupted. The most likely reason is that the reading of one of the sensors is distorted, and the value it outputs no longer reflects the true state of the material, thus changing the cooperative relationship with the normal reading of the other sensor.

[0055] Step Sd: Normalize the mutual information value to obtain the parameter correlation coefficient between the two data sources. The normalization method is to divide the mutual information value by the geometric mean of the information entropy of each of the two data sources.

[0056] Specifically, to eliminate the dimensional influence caused by the differences in information content between different data sources, the mutual information value is divided by the geometric mean of the information entropies H(Sᵢ) and H(Sⱼ) of the two data sources to obtain the normalized parameter correlation coefficient. .

[0057] ; The value range is [0,1]. The closer the value is to 1, the more highly correlated the detection information of the two data sources are.

[0058] Furthermore, electronic device computing The absolute value of the difference between the data source pair and the historical correlation benchmark value stored in the system is used to obtain the data source pair. The degree of correlation deviation.

[0059] Step S1023: When the correlation deviation of any data source pair exceeds the preset deviation threshold, identify the data source with the lower initial confidence score in the data source pair as an abnormal suspected source, and apply a decay factor to the initial confidence score of the abnormal suspected source.

[0060] Specifically, when the relevance deviation exceeds a preset deviation threshold, at least one source in the data source pair is considered an anomaly. In this case, the suspected anomaly source with the lower initial confidence score in the data source pair is identified, and a decay factor β (0 < β < 1, e.g., 0.5) is applied to the initial confidence score, resulting in a corrected score. , k∈{i,j}.

[0061] Therefore, when an abnormal correlation deviation occurs between two sources, the one with the weaker signal quality should be questioned first. If the same data source is identified as a suspected abnormal source in multiple data source pairs, its reliability score is cumulatively reduced by multiplying it by β multiple times.

[0062] Step S1024: Normalize the credibility scores of each data source after attenuation correction to obtain the credibility coefficient of each data source.

[0063] For example, for data sources Data source credibility score after attenuation correction Normalization is performed as follows: .

[0064] Normalization ensures that the sum of the credibility coefficients of all data sources is 1, making them directly usable as the fusion weights in the subsequent weighted fusion process.

[0065] Step S103: Map the multidimensional original parameters to the standardized feature space to obtain the standardized feature components of each dimension.

[0066] Specifically, raw sensor data often exhibits characteristics such as nonlinear distribution, high-dimensional redundancy, and the coexistence of different physical dimensions. Direct fusion can lead to distortion of the geometric structure of the feature space. Therefore, electronic devices employ a manifold mapping method to normalize the raw parameter data across all dimensions. This includes steps S1031 to S1033:

[0067] Step S1031: Perform local nearest neighbor structure analysis on the original parameter data of each dimension, and construct the nearest neighbor relationship graph of the data of each dimension in the original space.

[0068] Specifically, regarding the data source Original parameters of the output Local nearest neighbor structure analysis uses a preset number of nearest neighbors k (e.g., k=10) as a parameter to perform... For each sample point, search its k nearest neighbors and construct a nearest neighbor graph based on the distances between sample points. This nearest neighbor graph depicts the geometric topology of the original parameter data within its local neighborhood.

[0069] Step S1032: Based on the nearest neighbor graph, the manifold mapping method is used to map the data of each dimension to a low-dimensional embedding space of the same dimension, so that the data of each dimension maintains the local topological structure of its original space in the low-dimensional embedding space.

[0070] Specifically, the manifold mapping method can be an isomap or a locally linear embedding (LLE) method, which will... Mapping to a low-dimensional embedding space of uniform dimension d (e.g., taking d=8) yields a low-dimensional embedding representation. .

[0071] Manifold mapping methods aim to preserve the local topological structure during dimensionality reduction, ensuring that local nearest neighbor sample points in the original space maintain their nearest neighbor relationship in the low-dimensional embedding space. This preserves the nonlinear manifold structure information within the data of each dimension while reducing dimensionality, avoiding information loss in nonlinear distribution scenarios as in linear dimensionality reduction methods.

[0072] Step S1033: In the low-dimensional embedding space, scale alignment is performed based on the distribution range of the mapping results of each dimension to obtain the standardized feature components of each dimension.

[0073] Specifically, in obtaining low-dimensional embedding representations from each data source. , , Then, for each dimension's embedding component, min-max normalization is performed using its minimum and maximum values ​​on historical samples as boundaries, thus obtaining the standardized feature components for each dimension. , , At this point, the features from the three data sources are uniformly mapped to the same d-dimensional standardized space, eliminating dimensional differences.

[0074] Step S104: Use the credibility coefficient of the corresponding data source as the fusion weight of the standardized feature components of the corresponding dimension, perform weighted combination to generate a fusion feature vector, and construct a time series feature sequence based on the fusion feature vectors of the current batch and the previous N batches.

[0075] Specifically, electronic devices will assign credibility coefficients to each data source. As the standardized feature components of the corresponding dimension The fusion weights are calculated by weighting and scaling the feature components of each dimension separately, and then concatenating them to generate the fusion feature vector F for the current batch. ; Where [;] denotes vector concatenation operation. If the standardized feature components of each data source after manifold mapping are all d-dimensional vectors, taking three data sources as an example, then the dimension of the fused feature vector F is 3d.

[0076] Standardized feature components of each data source Multiply by the corresponding credibility coefficient before splicing. Amplitude scaling is performed so that the feature components corresponding to data sources with higher credibility maintain a larger numerical amplitude in the fused feature vector, while the feature components corresponding to data sources with lower credibility are compressed to an amplitude level close to zero.

[0077] When processing the fused feature vectors, the components with larger amplitudes naturally have a more significant impact on the model output, while the components with amplitudes close to zero are close to a silent state. This achieves adaptive suppression of low-quality data sources and can adapt to the dynamic changes in the quality of each data source without changing the model structure.

[0078] Furthermore, the electronic device records the fused feature vectors of N+1 consecutive batches, i.e., from the nNth batch to the nth batch, using the current batch n as a reference. , , …, The time series features are constructed by arranging the data in chronological order. The value of N is determined based on the on-site process response delay and batch processing frequency. For example, if N=9, the time series is constructed using the current batch and the previous 9 batches, totaling 10 batches. The time series features T captures the dynamic changes of tailings mineral properties over time.

[0079] Step S105: Input the time-series feature sequence into the time-series coding layer of the pre-built resource potential assessment model to extract the trend features of material property changes between batches.

[0080] The training process of the resource potential assessment model is as follows: collect the fused feature vectors of historical tailings batches as training samples, and use the actual grade and selectivity indicators obtained by the corresponding batch through chemical analysis as labels; divide the samples into training set and validation set; use a multi-layer fully connected neural network as the model structure, use the mean squared error as the loss function, and iteratively update the model parameters through backpropagation until the validation set loss converges; after training, deploy the model for online inference.

[0081] The temporal coding layer is implemented using a gated recurrent unit (GRU) network. Its hidden state processes the fused feature vectors of each batch in T sequentially, and outputs the hidden state vector at the final time step. As a characteristic of the trend of physical property changes between batches. The model encodes the evolution trajectory information of physical property parameters from batch n to batch n, enabling the model to perceive the upward trend, downward trend or fluctuation pattern of tailings mineral properties, rather than relying solely on the static characteristics of the current batch to make judgments.

[0082] Step S106: Concatenate the physical property change trend characteristics with the current batch fusion feature vector, input it into the evaluation decision layer of the resource potential assessment model, and output the resource potential score and process matching label of the current batch of tailings. The evaluation decision layer applies a switching cost penalty term to the process matching label jump between adjacent batches.

[0083] Specifically, the electronic device will output the trend feature vector from the time-series coding layer. Feature vector fused with the current batch Concatenate along the feature dimension to form a joint feature vector. The input is to the evaluation and decision-making layer. The evaluation and decision-making layer is implemented using a fully connected network and has two outputs: Resource potential score: a continuous scalar value that reflects the resource utilization degree of the current batch of tailings under the comprehensive consideration of particle size characteristics, chemical composition and mineral occurrence state; Process matching tags: These are discrete classification outputs that correspond to multiple predefined processing lines in the system, such as: high-grade beneficiation process, medium-grade flotation process, low-grade heap leaching process, tailings backfilling process, etc.

[0084] On the other hand, process line switching operations in industrial sites involve physical operations such as valve operation and equipment preheating. Frequent switching not only increases equipment wear and tear, but may also cause temporary material mismatch at the moment of switching.

[0085] Specifically, the switching cost penalty term functions differently in the model training and inference phases:

[0086] During the training phase, a switching cost penalty term is included as part of the loss function in gradient backpropagation. In each training batch, the predicted process matching labels from adjacent batches are compared, and when the label... n ≠Label n When ₋1, a penalty value λ·Cswitch is added to the total loss, where λ is the penalty coefficient and Cswitch is the fixed cost of a single jump. Through the constraint of the penalty term, the model parameters are guided during training to: tend to output the same label as the previous batch when the change in physical property parameters is small, and output different labels only when the physical property parameters change significantly.

[0087] During the inference phase, the model has already been trained, and its network parameters implicitly encode a preference for label stability. There is no need to calculate additional penalty terms during inference. The model can output process matching labels that balance accuracy and switching stability through direct forward propagation.

[0088] It should be noted that the current batch's fused feature vector participates in the temporal coding layer processing as the last element of the temporal feature sequence, and is also explicitly concatenated with the trend feature vector after the temporal coding layer output. This is because the hidden state vector output by the temporal coding layer is a compressed representation of the property change pattern within the entire temporal window, and the absolute property state details of the current batch are diluted after being mixed and encoded with the previous N batches. Process matching decisions require both trend information on property evolution and the absolute property level of the current batch itself.

[0089] Therefore, by using a skip connection method, the fused feature vector of the current batch is directly concatenated to the input of the evaluation decision layer, so that the evaluation decision layer can simultaneously obtain macro trend features and micro state features of the current batch. The two work together to support more accurate process matching decisions.

[0090] Step S107: Based on the process matching tag, send a process switching instruction to import the current batch of tailings into the processing process line corresponding to the process matching tag.

[0091] Specifically, the electronic equipment sends a process switching command to the PLC based on the process matching label of the current batch. The PLC then drives the material splitting device at the end of the conveyor channel. This device can be a belt switching mechanism or a pneumatic flap valve, directing the current batch of material into the processing line inlet corresponding to the process matching label. When the process matching label is the same as the previous batch label, the switching command remains in place, and the material splitting device does not activate, avoiding unnecessary mechanical operation.

[0092] In another possible implementation, the above method further includes (steps S11 to S15):

[0093] Step S11: When the resource potential score of the current batch of tailings falls into the preset decision fuzzy range, import the current batch of tailings into the buffer bin and record the fusion feature vector of the current batch.

[0094] Specifically, when the resource potential score of the current batch of tailings falls into the decision ambiguity range, it means that the resource potential score of the current batch of tailings is neither high enough to be judged as having recycling value, nor low enough to be judged as being directly discarded.

[0095] If a judgment is forced at this point, the costs of both errors are significant: If it is mistakenly judged as discarded, tailings that actually have recycling value will be directly stockpiled, resulting in permanent loss of resources. Furthermore, the materials in the tailings dam will be mixed together and cannot be recycled separately afterward. If it is mistakenly judged as recycled, tailings that do not actually have economic value will be sent into the recycling process, wasting the energy consumption, reagents, and equipment wear and tear costs of the beneficiation operation.

[0096] Therefore, the current batch of tailings is introduced into the buffer silo. After collecting relevant information from multiple batches, a more reliable collective decision is made: to send the accumulated materials from several batches in the buffer silo into the recycling process or to transfer them to the stockpile.

[0097] Step S12: After accumulating multiple batches in the buffer silo, calculate the equivalent fusion feature vector of the mixed materials in the buffer silo based on the fusion feature vector of each batch and the material quantity of each batch in the buffer silo using a weighted average method.

[0098] Step S121: Obtain the fusion feature vectors and corresponding material quantities of the K batches currently accumulated in the buffer bin.

[0099] Suppose that the buffer currently contains K batches, and the fused feature vector of each batch is denoted as { , ,…, }, the corresponding material quantity is denoted as { , ,…, The subscripts are arranged in the order of entry into the buffer, with subscript 1 being the earliest entrant and subscript K being the latest entrant.

[0100] Step S122: Use the proportion of each batch of material to the total amount of material in the buffer bin as the quality weight factor, and at the same time construct a time decay factor based on the time sequence of each batch entering the buffer bin. The time decay factor assigns a lower time weight to the batch that enters the buffer bin earlier.

[0101] Specifically, the quality weighting factor is the ratio of the amount of material in each batch to the total amount of material in the buffer silo, reflecting the proportion of each batch's contribution to the overall characteristics of the mixture. .

[0102] Time decay factor: Considering that the earlier a batch enters the buffer, the lower the timeliness of its feature information, an exponential decay function is constructed for the time sequence position k, so that batches entering the buffer earlier are assigned a lower timeliness weight: ; Where γ is the decay rate parameter, for example, γ=0.1, and Kk is the time interval between this batch and the latest batch. The subscript k=K corresponds to the latest batch, and its time decay factor λk =1 (no decay); the subscript k=1 corresponds to the earliest batch, which has the smallest time decay factor.

[0103] Step S123: For each batch, multiply the quality weight factor by the time decay factor to obtain the composite weight coefficient of the corresponding batch, and normalize the composite weight coefficients of K batches.

[0104] Specifically, composite weighting coefficient Then normalize the composite weight coefficients of the K batches: .

[0105] Step S124: The fused feature vectors of each batch are weighted and summed using the normalized composite weight coefficients of each batch to obtain the initial equivalent fused feature vectors. .

[0106] Step S125: Calculate the feature dispersion among the fused feature vectors of K batches in the buffer. The feature dispersion is the weighted variance of the distance between the fused feature vectors of each batch and the initial equivalent fused feature vector.

[0107] Specifically, the electronic device calculates the weighted variance of the K batches of fused feature vectors in the buffer relative to the initial equivalent fused feature vector, which is used as the feature dispersion. :

[0108]

[0109] Characteristic dispersion reflects the degree of internal difference in the mineral properties of different batches of tailings within the buffer silo. When the characteristic dispersion is large, it indicates that the materials in the buffer silo have diverse sources and mixed characteristics, and the representativeness of a single equivalent characteristic vector is uncertain. When the characteristic dispersion is close to zero, it indicates that the materials in different batches are highly homogeneous, and the equivalent characteristic vector has high reliability.

[0110] Step S126: The feature dispersion is added as an uncertainty correction amount to the initial equivalent fused feature vector to generate the final equivalent fused feature vector. The uncertainty correction amount is used to expand the decision boundary tolerance of the assessment decision layer when inputting into the resource potential assessment model.

[0111] .

[0112] Here, ⊕ indicates that the feature dispersion is concatenated to the feature vector as an additional channel.

[0113] When the feature vectors of this batch of tailings are fed into the evaluation model, the model will not only look at the score itself, but also at the feature dispersion of this batch of materials.

[0114] High dispersion indicates that the composition of this batch of material is highly uneven, with significant differences in the data measured at each detection point. In this case, the calculated comprehensive score is actually an average, masking the true differences within the material. If the model still uses the same strict boundaries as for homogeneous materials to distinguish between recycling and disposal, it's easy to treat a vague score as a definitive conclusion. For example, a score that just barely crosses the threshold might be deemed recyclable, when in reality, some of this material may not be worth recycling at all.

[0115] Therefore, when the model detects high dispersion, it will actively widen the decision boundary appropriately. The score that was originally just barely passing will be classified into the fuzzy range under the condition of high dispersion, instead of being allowed to pass directly.

[0116] Step S13: Re-input the equivalent fusion feature vector into the resource potential assessment model to obtain the comprehensive resource potential score and process matching label of the mixture.

[0117] Step S14: When the comprehensive resource potential score deviates from the decision fuzzy range, the mixed materials in the buffer bin are imported into the processing process line corresponding to the process matching label.

[0118] Specifically, materials continuously accumulate in the buffer silo, and each new batch triggers a recalculation of the equivalent characteristics. When the recalculated comprehensive resource potential score falls outside the decision fuzzy range, it indicates that the overall characteristics of the mixed materials can be clearly distinguished by the model. The system then imports all the mixed materials in the buffer silo into the corresponding process line and clears the buffer silo ledger.

[0119] Step S15: When the accumulated material quantity in the buffer bin reaches the preset capacity limit and the comprehensive resource potential score is still in the decision fuzzy range, the mixed material in the buffer bin is imported into the default processing line.

[0120] Specifically, when the accumulated material in the buffer bin reaches the preset capacity limit and the comprehensive resource potential score is still in the decision fuzzy range, the mixed materials in the buffer bin are introduced into the preset default processing line, which is usually the most versatile processing line with the broadest requirements for material characteristics, and the buffer bin is emptied to ensure that materials do not accumulate and the production process is not blocked.

[0121] By mixing materials, single batches of materials that were originally near the decision boundary are transformed into statistically representative mixed samples. Then, based on the equivalent characteristics of the mixed samples, more robust process allocation decisions are made, thereby reducing misallocation caused by single-batch detection noise.

[0122] In another possible implementation, the above method also includes a sensor degradation detection and model weight adaptive adjustment method, including (steps S21 to S23):

[0123] Step S21: During the processing of multiple consecutive batches, record the sequence of changes in the reliability coefficient of each data source with each batch to form a reliability trend curve for each data source.

[0124] Specifically, the electronic device maintains a credibility trend record table for each data source, records the credibility coefficient calculated for each batch, and updates it on a rolling basis with each batch to form a credibility trend curve for each data source.

[0125] Step S22: When the credibility trend curve of any data source shows a monotonically decreasing trend within a preset sliding window and its credibility coefficient at the end of the window is lower than the preset degradation threshold, the corresponding data source is marked as a degradation source.

[0126] Specifically, if the credibility coefficient of a data source shows a monotonically decreasing trend among batches within the window, that is, the credibility coefficient does not increase between any two adjacent batches, and the credibility coefficient at the end of the window is lower than the preset degradation threshold τ, for example, τ=0.15, then the data source is marked as a degradation source.

[0127] The combined effect of a monotonically declining trend and an absolute threshold can effectively distinguish between short-term, occasional fluctuations and genuine, gradual degradation, thus avoiding mislabeling.

[0128] Step S23: Based on the labeling results of degradation sources, scale the input connection weights of the dimensions corresponding to degradation sources in the evaluation decision layer of the resource potential assessment model, so that the resource potential assessment model reduces its decision dependence on the feature dimensions corresponding to degradation sources, and increases the input connection weights of the dimensions corresponding to other data sources that have not been labeled as degradation sources.

[0129] Specifically, once a data source is marked as a degraded source, the system immediately scales the input connection weights of the feature dimension corresponding to the data source in the resource potential assessment model's assessment decision layer by multiplying them by a scaling factor less than 1, such as 0.3, so that the model reduces its decision dependence on that feature dimension.

[0130] At the same time, the input connection weights of the corresponding dimensions of other data sources that are not marked as degradation sources are increased proportionally, for example by multiplying by a compensation factor greater than 1, to ensure that the sum of the input weights of all dimensions remains unchanged, thereby maintaining the robustness of model inference during sensor degradation and giving on-site maintenance personnel time to repair or replace the sensor.

[0131] When the credibility coefficient of a degraded source recovers to above the degradation threshold within W subsequent batches after maintenance, the system automatically removes the degradation mark and restores the connection weight of the corresponding dimension to the normal level.

[0132] To better implement the above method, this application also provides an intelligent scheduling and control device for tailings resource reuse based on a multi-parameter evaluation model, referring to... Figure 3 The intelligent scheduling and control device 200 for tailings resource reuse based on a multi-parameter evaluation model includes: The data acquisition module 201 is used to perform online detection of the current batch of tailings through a multi-source sensor group deployed on the tailings conveying channel to acquire multi-dimensional raw parameter data; The credibility coefficient calculation module 202 is used to evaluate the data quality of multidimensional raw parameter data and calculate the credibility coefficient of each data source according to the signal-to-noise ratio and data integrity of each data source. The standardized feature component determination module 203 is used to map the multidimensional original parameters to the standardized feature space to obtain the standardized feature components of each dimension. The fusion feature processing module 204 is used to use the credibility coefficient of the corresponding data source as the fusion weight of the standardized feature components of the corresponding dimension, perform weighted combination, generate fusion feature vector, and construct time series feature sequence based on the fusion feature vector of the current batch and the previous N batches. The property change trend analysis module 205 is used to input the time series feature sequence into the time series encoding layer of the pre-built resource potential assessment model to extract the property change trend features between batches; The potential assessment module 206 is used to concatenate the physical property change trend characteristics with the current batch fusion feature vector, input it into the assessment decision layer of the resource potential assessment model, and output the resource potential score and process matching label of the current batch of tailings. The assessment decision layer applies a switching cost penalty term to the process matching label jump between adjacent batches. The switching instruction sending module 207 is used to send a process switching instruction according to the process matching tag, so as to import the current batch of tailings into the processing process line corresponding to the process matching tag.

[0133] The various variations and specific examples in the methods of the foregoing embodiments are also applicable to the intelligent scheduling and control device for tailings resource reuse based on the multi-parameter evaluation model in this embodiment. Through the detailed description of the intelligent scheduling and control method for tailings resource reuse based on the multi-parameter evaluation model, those skilled in the art can clearly understand the implementation method of the intelligent scheduling and control device for tailings resource reuse based on the multi-parameter evaluation model in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0134] To better implement the above methods, embodiments of this application provide an electronic device, referring to... Figure 4The electronic device 300 includes a processor 301, a memory 303, and a display screen 305. The memory 303 and the display screen 305 are both connected to the processor 301, such as via a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of this electronic device 300 does not constitute a limitation on the embodiments of this application.

[0135] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0136] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 may be divided into address bus, data bus, control bus, etc.

[0137] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0138] The memory 303 is used to store application code that executes the solution of this application, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0139] Figure 4 The electronic device 300 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0140] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the program implements the intelligent scheduling and control method for tailings resource reuse based on a multi-parameter evaluation model provided in the above embodiments. Multi-source sensors collect multi-dimensional raw parameters online, calculate reliability coefficients based on the signal-to-noise ratio and completeness of each data source, and use these coefficients as fusion weights. The single-batch characteristics are jointly input into the evaluation model along with the time-series characteristics of the previous N batches. Simultaneously, a switching cost penalty is applied to process label jumps in adjacent batches. This ensures that the system considers both the physical property status of the current batch and the evolution trend of physical properties between batches when making process route allocation decisions. This avoids invalid process switching triggered by occasional score fluctuations, reduces frequent switching losses in process routes, and improves the accuracy of tailings resource utilization potential assessment.

[0141] In this embodiment, the computer-readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device. The computer-readable storage medium can be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, the computer-readable storage medium can be a portable computer disk, a hard disk, a USB flash drive, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory stick, floppy disk, optical disk, magnetic disk, mechanical encoding device, or any combination thereof.

[0142] The computer program in this embodiment includes program code for performing all the aforementioned methods. The program code may include instructions corresponding to the method steps provided in the above embodiments. The computer program can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded to an external computer or external storage device via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network). The computer program can be executed entirely on the user's computer as a standalone software package.

[0143] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

[0144] Additionally, it should be understood that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

Claims

1. A smart scheduling and control method for tailings resource reuse based on a multi-parameter evaluation model, characterized in that, include: The current batch of tailings is monitored online by a multi-source sensor array deployed on the tailings conveying channel to obtain multi-dimensional raw parameter data; The data quality of the multidimensional raw parameter data is evaluated, and the credibility coefficient of each data source is calculated based on the signal-to-noise ratio and data integrity of each data source. The multidimensional original parameters are mapped to the standardized feature space to obtain the standardized feature components of each dimension; The credibility coefficient of the corresponding data source is used as the fusion weight of the standardized feature components of the corresponding dimension, and a weighted combination is performed to generate a fusion feature vector. Based on the fusion feature vectors of the current batch and the previous N batches, a time-series feature sequence is constructed. The time-series feature sequence is input into the time-series coding layer of the pre-constructed resource potential assessment model to extract the trend features of inter-batch physical property changes; The physical property change trend characteristics are concatenated with the current batch fusion feature vector and input into the evaluation decision layer of the resource potential assessment model. The resource potential score and process matching label of the current batch of tailings are output. The evaluation decision layer applies a switching cost penalty term to the process matching label jump between adjacent batches. Based on the process matching tag, a process switching instruction is sent to import the current batch of tailings into the processing process line corresponding to the process matching tag.

2. The method according to claim 1, characterized in that, The method further includes: When the resource potential score of the current batch of tailings falls into the preset decision fuzzy range, the current batch of tailings is imported into the buffer bin, and the fusion feature vector of the current batch is recorded. After accumulating multiple batches in the buffer silo, the equivalent fusion feature vector of the mixed materials in the buffer silo is calculated by weighted average based on the fusion feature vector of each batch in the buffer silo and the amount of material in each batch. The equivalent fusion feature vector is re-inputted into the resource potential assessment model to obtain the comprehensive resource potential score and process matching label of the mixture. When the comprehensive resource potential score deviates from the decision fuzzy range, the mixture in the buffer bin is imported as a whole into the processing process line corresponding to the process matching label; When the accumulated material quantity in the buffer bin reaches the preset capacity limit and the comprehensive resource potential score is still within the decision fuzzy range, the mixed material in the buffer bin is imported into the default processing line.

3. The method according to claim 2, characterized in that, The step of calculating the equivalent fusion feature vector of the mixed materials in the buffer silo using a weighted average method, based on the fusion feature vector of each batch and the material quantity of each batch, includes: Obtain the fused feature vectors and corresponding material quantities of the K batches currently accumulated in the buffer bin; The proportion of each batch of material to the total amount of material in the buffer bin is used as the quality weight factor. At the same time, a time decay factor is constructed based on the time sequence of each batch entering the buffer bin. The time decay factor assigns a lower time weight to the batch that enters the buffer bin earlier. For each batch, the quality weight factor is multiplied by the time decay factor to obtain the composite weight coefficient of the corresponding batch, and the composite weight coefficients of K batches are normalized. The fusion feature vectors of each batch are weighted and summed using the normalized composite weight coefficients of each batch to obtain the initial equivalent fusion feature vectors. Calculate the feature dispersion among the fused feature vectors of K batches within the buffer, where the feature dispersion is the weighted variance of the distance between the fused feature vectors of each batch and the initial equivalent fused feature vector; The feature dispersion is added as an uncertainty correction amount to the initial equivalent fused feature vector to generate the final equivalent fused feature vector, wherein the uncertainty correction amount is used to expand the decision boundary tolerance of the assessment decision layer when inputting the resource potential assessment model.

4. The method according to claim 1, characterized in that, The calculation of the reliability coefficient of each data source based on the signal-to-noise ratio and data integrity of each data source includes: The initial credibility score of each data source is calculated by weighted summation of signal-to-noise ratio and data integrity. Calculate the parameter correlation coefficient between any two data sources in the current batch, and compare the parameter correlation coefficient with the pre-stored historical correlation benchmark value to obtain the correlation deviation between each data source pair; When the correlation deviation of any data source pair exceeds a preset deviation threshold, the data source with the lower initial confidence score in the data source pair is identified as an abnormal suspected source, and a decay factor is applied to the initial confidence score of the abnormal suspected source. The credibility scores of each data source after attenuation correction are normalized to obtain the credibility coefficient of each data source.

5. The method according to claim 4, characterized in that, The calculation of the parameter correlation coefficient between any two data sources in the current batch includes: Kernel density estimation is performed on the detection data from each data source in the current batch to obtain the marginal probability density function corresponding to each data source; A joint sample is constructed from the detection data of any two data sources, and the joint kernel density is estimated on the joint sample to obtain the joint probability density function between the two data sources. Based on the edge probability density function and the joint probability density function of the two data sources, the mutual information value between the two data sources is calculated according to the definition of mutual information. The mutual information value is normalized to obtain the parameter correlation coefficient between the two data sources. The normalization method is to divide the mutual information value by the geometric mean of the information entropy of each of the two data sources.

6. The method according to claim 1, characterized in that, The process of mapping multidimensional raw parameter data to a standardized feature space includes: Local nearest neighbor structure analysis is performed on the original parameter data of each dimension to construct the nearest neighbor relationship graph of the data of each dimension in the original space; Based on the nearest neighbor graph, the manifold mapping method is used to map the data of each dimension to a low-dimensional embedding space of the same dimension, so that the data of each dimension maintains the local topological structure of its original space in the low-dimensional embedding space. In the low-dimensional embedding space, scale alignment is performed based on the distribution range of the mapping results of each dimension to obtain the standardized feature components of each dimension.

7. The method according to claim 1, characterized in that, The method further includes: During multiple consecutive batch processing sessions, the change sequence of the reliability coefficient of each data source with each batch is recorded to form the reliability trend curve of each data source. When the credibility trend curve of any data source shows a monotonically decreasing trend within a preset sliding window and its credibility coefficient at the end of the window is lower than a preset degradation threshold, the corresponding data source is marked as a degradation source. Based on the labeling results of the degradation source, the input connection weights of the dimensions corresponding to the degradation source in the evaluation decision layer of the resource potential assessment model are scaled, so that the resource potential assessment model reduces its decision dependence on the feature dimensions corresponding to the degradation source and increases the input connection weights of the dimensions corresponding to other data sources that have not been labeled as degradation sources.

8. A smart scheduling and control device for tailings resource reuse based on a multi-parameter evaluation model, characterized in that, include: The data acquisition module is used to perform online detection of the current batch of tailings through a multi-source sensor group deployed on the tailings conveying channel to obtain multi-dimensional raw parameter data; The credibility coefficient calculation module is used to evaluate the data quality of the multidimensional raw parameter data and calculate the credibility coefficient of each data source according to the signal-to-noise ratio and data integrity of each data source. The standardized feature component determination module is used to map the multidimensional original parameters to a standardized feature space to obtain the standardized feature components of each dimension. The fusion feature processing module is used to use the credibility coefficient of the corresponding data source as the fusion weight of the standardized feature components of the corresponding dimension, perform weighted combination, generate fusion feature vector, and construct time series feature sequence based on the fusion feature vector of the current batch and the previous N batches. The physical property change trend analysis module is used to input the time series feature sequence into the time series encoding layer of the pre-constructed resource potential assessment model to extract the physical property change trend features between batches; The potential assessment module is used to concatenate the physical property change trend characteristics with the current batch fusion feature vector, input it into the assessment decision layer of the resource potential assessment model, and output the resource potential score and process matching label of the current batch of tailings. The assessment decision layer applies a switching cost penalty term to the process matching label jump between adjacent batches. The switching instruction sending module is used to send a process switching instruction according to the process matching tag, so as to import the current batch of tailings into the processing process line corresponding to the process matching tag.

9. An electronic device, characterized in that, include: At least one processor; Memory; At least one computer program, wherein the at least one computer program is stored in the memory and configured to be executed by the at least one processor, the at least one computer program being configured to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and execute the method as described in any one of claims 1 to 7.

Images