Metal mine black light factory intelligent design method and system

Abstract

The application discloses a kind of metal mine black lamp factory intelligent design method and system, it is related to computer aided design technical field, including, acquisition working condition data and process structure data, extract the relationship between crushing, grinding, classification, flotation and thickening dehydration, construct process topology;According to beneficiation knowledge graph and process mechanism element set, mechanism corrector is constructed, mechanism corrector is used to the whole process design prediction execution section projection and constraint solution, output correction prediction, and register inconsistent event, form correction rule log;Corresponding interactive feedback data is obtained with design configuration set, and incremental update is carried out to beneficiation knowledge graph and process mechanism element set, obtain interaction baseline, and the next round of design iteration is completed by interaction baseline.The application carries out section projection and constraint solution to the whole process design prediction by constructing mechanism corrector, and the structured correction and constraint consistent expression of design prediction result are realized.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, and in particular to an intelligent design method and system for a darkroom factory in a metal mine. Background Technology

[0002] With the development of computer-aided design, process industry modeling, data-driven prediction and knowledge graph technology, the digital design method for metal mine beneficiation processes has gradually expanded from single equipment parameter configuration to full-process collaborative modeling of crushing, grinding, classification, flotation and thickening dewatering, and has begun to integrate process structure data, operating condition data and physical mechanism information to carry out design deduction and scheme generation.

[0003] Existing technologies in full-process intelligent design often focus on data fitting or local mechanism analysis, making it difficult to uniformly couple process process topology, knowledge association, mechanism constraints and prediction results. This results in insufficient constraint transfer and inconsistency of segmented results during design reasoning, which in turn affects the stability of scheme optimization and iterative updates. Therefore, there is an urgent need for a constraint consistency calculation method for full-process design. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for intelligent design of darkrooms in metal mines to solve the problem of difficulty in unifying and coordinating prediction results and process mechanism constraints in the intelligent design of the entire metal mine process.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, this invention provides an intelligent design method for a "lights-out" metal mine plant, comprising: collecting operating condition data and process structure data; extracting the relationships between crushing, grinding, classification, flotation, and thickening / dewatering processes to construct a process topology; based on the process topology, annotating equipment constraints and phenomenological patterns to form a mineral processing knowledge graph; forming a set of process mechanism elements based on operating condition data, process structure data, and the physical laws corresponding to each process step; summarizing the process topology, mineral processing knowledge graph, and process mechanism element set to form a calibration baseline; and using the calibration baseline to train and infer a data-driven basic prediction model to obtain the full flow... The process involves designing and predicting the entire process; constructing a mechanism corrector based on the mineral processing knowledge graph and the process mechanism element set; using the mechanism corrector to perform segmented projection and constraint solving on the full-process design prediction, outputting corrected predictions, and registering inconsistency events to form a correction rule log; using the corrected predictions and correction rule logs to trigger full-process design optimization, generating optimized design schemes, performing integrated training and inference processing on the optimized design schemes, and producing a design configuration set; obtaining interactive feedback data corresponding to the design configuration set, incrementally updating the mineral processing knowledge graph and the process mechanism element set to obtain an interactive baseline, and using the interactive baseline to complete the next round of design iteration.

[0008] As a preferred embodiment of the intelligent design method for a darkened metal mine factory according to the present invention, the steps for constructing the process topology are as follows:

[0009] The operating condition data is time-aligned and resampled to form a synchronized operating condition frame;

[0010] Based on the synchronous operating condition frame and process structure data, set the monitoring section corresponding to each process position, and extract the material flow characteristics and energy flow characteristics of each monitoring section;

[0011] Based on the characteristics of material flow and energy flow, determine the set of material flow relationships and the set of energy flow relationships among crushing, grinding, classification, flotation and thickening / dewatering, and construct the process topology.

[0012] As a preferred embodiment of the intelligent design method for a "lights-out" metal mine factory according to the present invention, the steps for forming a mineral processing knowledge graph are as follows:

[0013] Based on the process topology, the synchronous working condition frames are aligned with topology time delay and segmented into states to form a set of topology constraint time sequence segments corresponding to each process step.

[0014] By combining the set of time-series fragments of topological constraints with process structure data, the equipment constraints corresponding to each process step are labeled, and the time delay correlation, change direction and consistency relationship between each process step are extracted within the process connection relationship of the process topology, forming a set of equipment constraint facts and a set of phenomenon regularity facts;

[0015] Based on the set of facts constrained by equipment and the set of facts governing phenomena, and combined with the process connection relationships, material flow relationships and energy flow relationships of the process topology, a mineral processing knowledge graph is constructed.

[0016] As a preferred embodiment of the intelligent design method for a darkened metal mine factory according to the present invention, the steps for forming the process mechanism element set are as follows:

[0017] Based on mineral processing knowledge graph, combined with operating condition data, process structure data and the physical laws corresponding to each process step, a set of process mechanism features is extracted;

[0018] By using the process mechanism feature set, the mechanism features corresponding to each process step are effectively screened and parameters are summarized to form a process mechanism element set.

[0019] As a preferred embodiment of the intelligent design method for a "lights-out" metal mine factory according to the present invention, the steps for obtaining the full-process design prediction are as follows:

[0020] The process topology, mineral processing knowledge graph, and process mechanism element set are structured and summarized to form a calibration baseline container corresponding to each process step;

[0021] By correcting the baseline container, the synchronous operating condition frame is divided into time windows, and the process connection relationship of the process topology is combined to form the baseline input data corresponding to each window;

[0022] The data-driven basic prediction model is invoked to jointly extract and fuse the baseline input data to obtain the local prediction results corresponding to each window.

[0023] The local prediction results are connected in series according to the process connection relationship of the process topology to obtain the whole process design prediction.

[0024] As a preferred embodiment of the intelligent design method for a darkened metal mine factory according to the present invention, the step of forming a correction rule log is as follows:

[0025] The entire process design prediction is segmented according to the process connection relationship of the process topology to form each design prediction segment. Then, a mechanism corrector is constructed by combining the mineral processing knowledge graph and the process mechanism element set.

[0026] A segmented correction context is established for each design prediction segment using a mechanism corrector, and the segmented correction context is converted into a corresponding segmented projection task.

[0027] Based on the segmented projection task, segmented projection and constraint solving are performed sequentially on each design prediction segment to obtain the corrected prediction.

[0028] Based on the conflict information, residual information, displacement information and identification information of each design prediction segment during the correction process, inconsistencies are recorded and summarized to form a correction rule log.

[0029] As a preferred embodiment of the intelligent design method for a darkened metal mine factory according to the present invention, the steps for generating the optimized design scheme are as follows:

[0030] Based on the correction prediction and correction rule logs, target indicators, risk information, resource allocation information and design variable information corresponding to each process step are extracted to form design optimization trigger data;

[0031] Based on the design optimization trigger data, the equipment configuration parameters, connection relationship parameters, operating boundary parameters and collaborative rule parameters corresponding to each process step are coordinated and solved to generate an optimized design scheme.

[0032] As a preferred embodiment of the intelligent design method for a darkened metal mine factory according to the present invention, the steps for producing the design configuration set are as follows:

[0033] The integrated training and inference process is invoked to update the optimized design scheme, and the scheme score result that matches the current design goals and design constraints is output during the inference phase;

[0034] Based on the scheme scoring results, a design configuration set is generated according to the equipment configuration parameters, connection relationship parameters, operation boundary parameters, and coordination rule parameters corresponding to each process step.

[0035] As a preferred embodiment of the intelligent design method for a darkened metal mine factory described in this invention, the steps for obtaining the interaction baseline and completing the next round of design iteration based on the interaction baseline are as follows:

[0036] Acquire trial operation feedback information, process change information, result feedback information, and abnormal status information corresponding to the design configuration set, and form interactive feedback data;

[0037] Based on the interactive feedback data, the affected mechanism fragments in the affected facts and process mechanism element set of the mineral processing knowledge graph are incrementally updated to obtain the interactive baseline corresponding to each process link.

[0038] Based on the interaction baseline, the mineral processing knowledge graph, process mechanism element set, correction baseline, mechanism corrector, optimization design scheme and design configuration set are iteratively updated according to the interaction status of each process link.

[0039] Secondly, this invention provides an intelligent design system for a darkroom in a metal mine, comprising: a topology construction module for collecting operating condition data and process structure data, extracting the relationships between crushing, grinding, classification, flotation, and thickening / dewatering, and constructing a process topology; a graph mechanism module for labeling equipment constraints and phenomenological patterns based on the process topology, forming a mineral processing knowledge graph; and forming a set of process mechanism elements based on operating condition data, process structure data, and the physical laws corresponding to each process step; and a predictive design module for summarizing the process topology, mineral processing knowledge graph, and process mechanism element set to form a calibration baseline, and calling the calibration baseline to train and infer a data-driven basic prediction model to obtain the full flow... The system comprises the following modules: a process design prediction module; a correction rule module, which constructs a mechanism corrector based on the mineral processing knowledge graph and the process mechanism element set; a mechanism corrector performs segmented projection and constraint solving on the full-process design prediction; outputs corrected predictions; and registers inconsistencies to form a correction rule log. A design optimization module triggers full-process design optimization using the corrected predictions and correction rule logs, generates optimized design schemes, performs integrated training and inference on the optimized design schemes, and produces a design configuration set. A feedback iteration module obtains interactive feedback data corresponding to the design configuration set, incrementally updates the mineral processing knowledge graph and the process mechanism element set, obtains an interactive baseline, and completes the next round of design iteration using the interactive baseline.

[0040] The beneficial effects of this invention are as follows: by constructing a mechanism corrector to perform segmented projection and constraint solving on the entire process design prediction, the structured correction and constraint consistency expression of the design prediction results are realized; the prediction output of each process link has the processing capabilities of segmented transmission, conflict identification, residual characterization and regular sedimentation, and at the same time forms a traceable record of inconsistencies, providing a stable, standardized and reusable correction basis for design optimization triggering, scheme selection and iterative updates. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 A flowchart for the intelligent design method of a darkroom factory in a metal mine.

[0043] Figure 2 A schematic diagram of an intelligent design system for a lights-out factory in a metal mine.

[0044] Figure 3 A flowchart for constructing the process topology.

[0045] Figure 4 A flowchart for the formation of a mineral processing knowledge graph.

[0046] Figure 5 This is a comparison chart of product purity curves.

[0047] Figure 6 A chart showing the constraint correction data for key flotation indicators.

[0048] Figure 7 This is a comparative data chart showing the average residual information for each stage of the process. Detailed Implementation

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0050] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0051] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0052] Reference Figures 1-7 As one embodiment of the present invention, this embodiment provides a smart design method for a lights-out factory in a metal mine, comprising the following steps:

[0053] S1. Collect operating condition data and process structure data, extract the relationship between crushing, grinding, classification, flotation and thickening / dewatering, and construct the process topology.

[0054] S1.1: Perform time alignment and resampling on the operating condition data to form a synchronized operating condition frame;

[0055] Specifically, the operating condition data is timestamped based on a unified time reference to eliminate the time deviation of multi-source sensor acquisition. The calibrated operating condition data is then resampled and interpolated at fixed time intervals so that operating condition data of different frequencies are mapped to the same time grid. The operating condition data after resampling and interpolation processing forms a synchronous operating condition frame at each time point.

[0056] It should be noted that the operating data includes data reflecting the real-time operating status, equipment load status, process medium status, and product status of each process, including crushing, grinding, classification, flotation, and thickening / dewatering.

[0057] The process structure data includes the flow segment division data, process connection relationship data, material flow relationship data, energy flow relationship data, equipment composition relationship data, instrument configuration relationship data, interlock control relationship data, and automatic control relationship data for each process, including crushing, grinding, classification, flotation, and thickening and dewatering.

[0058] S1.2: Based on the synchronous operating condition frame and process structure data, set the monitoring sections corresponding to each process position, and extract the material flow characteristics and energy flow characteristics of each monitoring section;

[0059] Specifically, based on the process connection relationship data and equipment composition relationship data in the process structure data, monitoring sections are delineated at the material flow nodes of each process, including crushing, grinding, classification, flotation, and thickening and dewatering. Using the instrument configuration relationship data in the process structure data, the sensor values ​​of the corresponding monitoring sections in the synchronous operating condition frame are aggregated to extract the material flow characteristics reflecting the material flow rate, particle size distribution, and grade. At the same time, combined with the energy flow relationship data in the process structure data, the power consumption and medium usage of the corresponding monitoring sections are extracted from the synchronous operating condition frame to form energy flow characteristics.

[0060] S1.3: Determine the set of material flow relationships and the set of energy flow relationships among crushing, grinding, classification, flotation and thickening / dewatering based on material flow characteristics and energy flow characteristics, and construct the process topology;

[0061] Specifically, based on the characteristics of material flow and energy flow, the material flow rate and grade values ​​of each process monitoring section in crushing, grinding, classification, flotation, and thickening / dewatering are compared. Process pairs with direct material transfer relationships are categorized into a material flow relationship set. The power consumption and media usage values ​​of each process monitoring section are analyzed, and process pairs with energy transfer or media sharing relationships are categorized into an energy flow relationship set. The material flow relationship set is used to define directed connections between nodes, and the energy flow relationship set is combined with the energy consumption correlation between nodes to define energy consumption relationships between nodes. Each process in crushing, grinding, classification, flotation, and thickening / dewatering is mapped as a node, generating a process topology that includes material flow direction and energy flow path.

[0062] S2. Based on the process topology, the equipment constraints and phenomena are labeled and accumulated to form a mineral processing knowledge graph; based on the operating condition data, process structure data and the physical laws corresponding to each process link, a set of process mechanism elements is formed.

[0063] S2.1: Based on the process topology, perform topology time delay alignment and state segmentation on the synchronous working condition frame to form a set of topology constraint time sequence segments corresponding to each process step;

[0064] Specifically, the material transfer path from crushing, grinding, classification, flotation to thickening and dewatering is determined based on the directed connections between nodes in the process topology. The physical transmission time required for the material to flow through each process is statistically analyzed as a time delay parameter. The time delay parameter is used to perform time shift correction on the monitoring section data of different processes in the synchronous operating condition frame to achieve topological time delay alignment. Based on the process connection relationship data in the process structure data, the start-up and shutdown times or sudden changes in operating conditions are identified as the dividing points. The data after topological time delay alignment is divided into continuous time periods. The data containing the complete causal chain of crushing, grinding, classification, flotation and thickening and dewatering in each continuous time period constitute a topologically constrained time sequence segment.

[0065] S2.2: Combine the set of time-series fragments of topological constraints with the process structure data, label the equipment constraints corresponding to each process step, and extract the time delay correlation, change direction and consistency relationship between each process step within the process connection relationship of the process topology to form a set of equipment constraint facts and a set of phenomenon regularity facts;

[0066] Specifically, based on the equipment composition relationship data in the process structure data, the range of equipment operating parameters corresponding to each process link is extracted from the topological constraint time sequence fragment set and labeled as equipment constraints. Within the process connection relationship of the process topology, the changing trends of the monitoring section data of the upstream and downstream processes in the topological constraint time sequence fragment set are compared to determine the time delay length as the time lag correlation, the value increase or decrease trend as the change direction, and the fluctuation synchronization degree as the consistency relationship. The labeled equipment constraints are summarized to form an equipment constraint fact set, and the extracted time lag correlation, change direction and consistency relationship are summarized to form a phenomenon pattern fact set.

[0067] It should be noted that the degree of fluctuation synchronization is generated based on the time series of adjacent process key indicators extracted from the historical operating condition database. The time series of adjacent process key indicators originates from the monitoring sections corresponding to the logistics relationship data and instrument configuration relationship data defined in the process structure data. By comparing the numerical change patterns of the output end of the preceding process and the input end of the subsequent process within the same production cycle, the overlap characteristics of the two in terms of the peak occurrence time, the duration of the trough, and the overall rise and fall trend are identified. When the overlap characteristics are greater than the preset similarity threshold, it is determined that there is a consistency relationship. The preset similarity threshold is set based on the distribution of similarity values ​​of adjacent process key indicator time series under statistical historical normal operating conditions. Specifically, the lower limit quantile that can cover the vast majority of stable operating state samples is selected as the similarity threshold. The exemplary value range is 0.75 to 0.90, and the value is based on the empirical range of ensuring that the true process fluctuation transmission characteristics are preserved while eliminating random noise interference.

[0068] S2.3: Based on the set of equipment constraint facts and the set of phenomenon regularity facts, combined with the process connection relationship, material flow relationship and energy flow relationship of the process topology, construct a mineral processing knowledge graph;

[0069] Specifically, nodes in the process topology are used as entities. Directed edges between entities are established based on process connection relationships, material flow relationships, and energy flow relationships. The range of equipment operating parameters in the equipment constraint fact set is assigned as attribute values ​​to the corresponding process nodes. The time delay correlation, change direction, and consistency relationship in the phenomenon law fact set are assigned as attribute values ​​to the directed edges connecting the nodes. Each process node of crushing, grinding, classification, flotation, and thickening dewatering is interconnected through directed edges carrying attribute values, forming a mineral processing knowledge graph containing structural information and operational laws.

[0070] S2.4: Based on the mineral processing knowledge graph, combined with operating condition data, process structure data and the physical laws corresponding to each process step, extract the process mechanism feature set;

[0071] Specifically, process nodes carrying the range of equipment operating parameters and directed edges carrying time-delay correlations, change directions, and consistency relationships are extracted from the mineral processing knowledge graph. Combined with the real-time values ​​of the corresponding monitoring sections in the operating data and the physical law descriptions independent of the process structure data storage, the equipment operating parameter range of the process nodes is directly defined as a static constraint feature, that is, the upper and lower limit value range of the equipment's allowed operation is taken as the value of this feature. The time-delay correlations, change directions, and consistency relationships of the directed edges are combined to define dynamic correlation features, that is, the time delay length, increase / decrease trend judgment, and fluctuation synchronization degree are packaged into a vector description of this feature. When using physical laws to perform causal logic verification on dynamic correlation features, it is necessary to compare whether the change direction in the dynamic correlation features matches the causal mechanism of the physical laws. If the change direction of the upstream process parameters leading to the increase of the downstream process parameters in the dynamic correlation features is consistent with the description of "increased feed rate causing increased processing load" in the physical laws, and the time-delay correlation conforms to the physical time of material transmission, then it is determined to pass the verification. The dynamic correlation features that pass the verification and the static constraint features together constitute the process mechanism features. All process mechanism features are summarized to form a process mechanism feature set.

[0072] S2.5: Effective screening and parameter summarization of the mechanism features corresponding to each process step are carried out through the process mechanism feature set to form a process mechanism element set;

[0073] Specifically, dynamic correlation features that have passed causal logic verification are selected from the set of process mechanism features. Based on the process connection relationship data in the process structure data, the dynamic correlation features and static constraint features are classified into each process link of crushing, grinding, classification, flotation and thickening / dewatering. The static constraint features within each process link are merged into intervals, and the common intersection of the equipment operating parameter ranges is extracted as the summarized constraint parameters. The trend consistency of the dynamic correlation features within each process link is statistically analyzed, and the combination of the time delay correlation length and change direction with the highest frequency is retained as the summarized regularity parameters. The summarized constraint parameters and the summarized regularity parameters are packaged to form the mechanism elements of the process link. The mechanism elements of all process links are summarized to form the process mechanism element set.

[0074] S3. Summarize the process topology, mineral processing knowledge graph and process mechanism elements to form a calibration baseline. Use the calibration baseline to train and infer the data-driven basic prediction model to obtain the full process design prediction.

[0075] S3.1: Structurally summarize the process topology, mineral processing knowledge graph, and process mechanism element set to form a calibration baseline container corresponding to each process step;

[0076] Specifically, based on the process segment division data in the process structure data, each process node and its connected directed edges in the process topology, including crushing, grinding, classification, flotation, and thickening / dewatering, are divided into independent process blocks. The attribute values ​​of the corresponding process nodes and the attribute values ​​of the directed edges in the mineral processing knowledge graph are mapped to their respective independent process blocks to form a structured knowledge subgraph. The constraint parameters and regularity parameters of the process mechanism elements, which are concentrated and attributed to each process link, are filled into the corresponding independent process blocks to form a mechanism parameter package. The structured knowledge subgraph and the mechanism parameter package are encapsulated into a data set to generate a calibration baseline container that corresponds one-to-one with each process link.

[0077] S3.2: The synchronous operating condition frame is divided into time windows by correcting the baseline container, and the baseline input data corresponding to each window is formed by combining the process connection relationship of the process topology;

[0078] Specifically, the mechanism parameter package corresponding to each process step is extracted from the calibration baseline container. The time delay correlation length in the summarized regular parameters is read as the time window width. The synchronous operating condition frame is slid-cut according to the time window width to generate a time window sequence covering the complete causal chain. Combined with the process connection relationship of the process topology, the data segments of each process monitoring section of crushing, grinding, classification, flotation and thickening dewatering in each time window are identified. The process node attributes and directed edge attributes in the structured knowledge subgraph are matched to the corresponding data segments to form baseline input data containing real-time values ​​and structural constraints.

[0079] S3.3: Call the data-driven basic prediction model to jointly extract and fuse the baseline input data to obtain the local prediction results corresponding to each window;

[0080] It should be noted that the pre-training process of the data-driven basic prediction model uses batch historical baseline input data as training samples. The encoding layer consists of multiple long short-term memory networks, which are used to map time-series data fragments of each process monitoring section into fixed-dimensional process state vectors. The fusion layer aggregates upstream and downstream process state vectors through a graph attention mechanism to generate a fusion feature representation containing topological relationships. The regression head consists of a fully connected neural network, which maps the fusion feature representation into predicted values ​​of key indicators. The loss function uses mean squared error to measure the deviation between the predicted value and the true value. The optimizer uses an adaptive moment estimation algorithm. The initial learning rate (0.001 in the example) decays exponentially with the training rounds. During training, the backpropagation gradient updates the weight parameters of each layer until the loss function converges to the preset convergence threshold, forming a trained data-driven basic prediction model.

[0081] It should be noted that the convergence threshold is set based on the stability criterion that the decrease of the validation set loss function tends to zero over multiple training rounds during the pre-training process. The value ranges from 0.00001 to 0.001. The basis for this value is to prevent the model from overfitting and to ensure that the prediction accuracy meets the allowable error of industrial control.

[0082] The encoding layer maps time-series data fragments from each process monitoring section into fixed-dimensional process state vectors, with dimensions ranging from 64 to 128 to capture temporal dependencies. The fusion layer aggregates upstream and downstream process state vectors through a graph attention mechanism to generate a fusion feature representation containing topological relationships, with dimensions consistent with the process state vectors. The regression head maps the fusion feature representation to the predicted values ​​of key indicators for the corresponding process steps within the corresponding time window and compresses the fusion feature representation into a single-dimensional output.

[0083] Specifically, the real-time values ​​and structural constraints in the baseline input data are fed into a pre-trained data-driven basic prediction model. The coding layer extracts features from the data segments of each process monitoring section, including crushing, grinding, classification, flotation, and thickening and dewatering, to form process state vectors. Based on the process connection relationship of the process topology, the process state vectors of upstream and downstream processes are spliced ​​in time in the fusion layer to generate a fused feature representation that includes the correlation between material flow and energy flow. The fused feature representation is mapped and calculated by the regression head to solve the predicted values ​​of key indicators of each process link within each time window, forming local prediction results.

[0084] By mapping the fused feature representation using a regression head, the predicted values ​​of key indicators for each process step within each time window are calculated. The expression is as follows:

[0085] ;

[0086] In the formula, Indicates the first The process step is in the first... Key indicator forecasts for each time window; Indicates the index of process steps; Indicates the time window index; Indicates the first The transpose of the regression weight vector specific to each process step is obtained through pre-training and optimization using historical data; Represents a linear activation function; Indicates the first The process step is in the first... A fusion feature representation vector for each time window; Indicates the first The bias scalars specific to each process step are obtained through pre-training and optimization using historical data.

[0087] S3.4: Connect the local prediction results according to the process connection relationship of the process topology to obtain the whole process design prediction;

[0088] Specifically, based on the process connection relationship of the process topology, the local prediction results of the preceding process steps are used as the boundary conditions of the subsequent process steps. The local prediction results of crushing, grinding, classification, flotation and thickening are sequentially connected according to the material flow direction. Through the transmission and accumulation of each step, the scattered local prediction results are integrated into a continuous whole process design prediction.

[0089] S4. Construct a mechanism corrector based on the mineral processing knowledge graph and process mechanism element set. Use the mechanism corrector to perform segmented projection and constraint solving on the whole process design prediction, output the corrected prediction, and register inconsistency events to form a correction rule log.

[0090] S4.1: The entire process design prediction is segmented according to the process connection relationship of the process topology to form each design prediction segment, and a mechanism corrector is constructed by combining the mineral processing knowledge graph and the process mechanism element set.

[0091] Specifically, based on the process connection relationships of the process topology, the entire process design prediction is divided into crushing, grinding, classification, flotation, and thickening / dewatering process nodes to form design prediction segments corresponding to each process step. The attribute values ​​of the corresponding process nodes and the attribute values ​​of the directed edges in the mineral processing knowledge graph are extracted. Combined with the constraint parameters and regularity parameters summarized from the process mechanism elements, a mechanism corrector corresponding to each design prediction segment is constructed. The numerical range boundaries of the predicted values ​​of key indicators are set using the summarized constraint parameters, and the changing trends of the predicted values ​​of key indicators are verified using the summarized regularity parameters. When the design prediction segment exceeds the numerical range boundary or violates the changing trend, the mechanism corrector performs deviation correction.

[0092] It should be noted that each design prediction segment refers to a local prediction result formed after dividing the entire process design prediction into segments according to the process connection relationship of the process topology, and corresponding one-to-one with each process link of crushing, grinding, classification, flotation and thickening dewatering.

[0093] S4.2: Establish a segmented correction context for each design prediction segment using the mechanism corrector, and convert the segmented correction context into the corresponding segmented projection task;

[0094] Specifically, the mechanism corrector reads the inductive constraint parameters and inductive regularity parameters corresponding to each design prediction segment, transforms the range of equipment operating parameters into numerical constraints, and transforms the time delay correlation length and change direction into trend matching rules. These are then assembled into a segmented correction context. Based on the numerical constraints in the segmented correction context, the feasible domain boundary of the design prediction segment is set. According to the trend matching rules, the smoothness metric between adjacent time points of the design prediction segment is defined. A segmented projection task is constructed with the feasible domain boundary as the constraint and the smoothness metric as the optimization objective. By mapping the original numerical sequence of the design prediction segment to the interior of the feasible domain boundary and adjusting the shape of the original numerical sequence to minimize the smoothness metric, the transformation from the segmented correction context to the segmented projection task is completed.

[0095] It should be noted that the smoothness metric is defined as the L2 norm of the difference between predicted values ​​at adjacent time points, used to quantify the smoothness of the curve.

[0096] S4.3: Based on the segmented projection task, segmented projection and constraint solving are performed sequentially on each design prediction segment to obtain the corrected prediction;

[0097] Specifically, for each design prediction segment, the corresponding segment-by-segment projection task is invoked to truncate the original predicted numerical sequence within the feasible region boundary, eliminating abnormal values ​​that exceed the range of equipment operating parameters. A smoothness metric is applied to the truncated numerical sequence to determine the deviation value between the numerical changes and trend matching rules at adjacent time points. The shape of the original numerical sequence is iteratively adjusted to minimize the deviation value. The gradient descent algorithm is used to update the numerical sequence. The process stops when the deviation value between two adjacent iterations is less than the convergence criterion or the maximum number of iterations is reached. The numerical sequence after truncation and iterative adjustment forms the corrected prediction, ensuring that the corrected prediction simultaneously meets physical limits and causal logic.

[0098] It should be noted that the maximum number of iterations is set based on the evaluation of the average time taken per iteration and the maximum allowable response delay, by calculating the number of complete update rounds that can be completed before the timeout and reserving a safety margin; the exemplary value range is 50 to 200 times, and the value is based on the engineering experience balance point of preventing the algorithm from getting stuck in local optima or excessively consuming computing resources while ensuring the convergence accuracy of the optimization results.

[0099] S4.4: Based on the conflict information, residual information, displacement information and identification information of each design prediction segment during the correction process, register inconsistencies and summarize them to form a correction rule log;

[0100] Specifically, the overlapping portion between predicted values ​​and physical limits in the design prediction segments is extracted as conflict information; the absolute value of the difference between predicted and measured values ​​is calculated as residual information; the translation distance of the predicted value sequence on the time axis is statistically analyzed as displacement information; the equipment number and operating condition label corresponding to each design prediction segment are recorded as identification information; the magnitude of conflict information and residual information is compared to determine the degree of numerical deviation; the time period and equipment where the deviation occurred are located by combining displacement information and identification information; when the degree of numerical deviation exceeds the allowable range and the time period of deviation occurs matches the equipment successfully, an inconsistency event containing the degree of deviation, the time period of deviation occurrence, and the equipment description is recorded; all inconsistency events generated by the design prediction segments are collected; the inconsistency event sequence is arranged in chronological order of the time period of deviation occurrence; the arranged inconsistency event sequence is associated with the corresponding correction operation action to generate a correction rule log.

[0101] It should be noted that the allowable range is set based on the critical point of process safety operation and the critical point of equipment performance degradation. The example value range is 10% to 20% of the rated operating parameters. The basis for the value is to ensure that inconsistent event recording is triggered only when the deviation is sufficient to affect production stability or cause product quality to be unqualified, so as to avoid the accumulation of invalid logs.

[0102] When the value indicated by the conflict information exceeds the physical limit boundary, and the absolute value of the difference between the prediction and the measured value indicated by the residual information exceeds the preset error tolerance, the degree of deviation of the value is judged as serious deviation; if only one of the conflict information exceeding the limit or only the residual information exceeding the tolerance is met, the degree of deviation of the value is judged as general deviation; if the conflict information does not exceed the limit and the residual information is within the preset error tolerance range, the degree of deviation of the value is judged as normal.

[0103] The preset error tolerance is set based on the statistical fluctuation characteristics of historical measured data and the process control accuracy requirements. The example value ranges from 5% to 15%, and the basis for the value is to ensure that while eliminating normal measurement noise interference, it can sensitively capture substantial deviations that have correction significance.

[0104] S5. Trigger full-process design optimization using correction prediction and correction rule logs, generate optimized design schemes, perform integrated training and inference processing on optimized design schemes, and produce design configuration sets.

[0105] S5.1: Based on the correction prediction and correction rule log, extract the target indicators, risk information, resource allocation information and design variable information corresponding to each process link to form design optimization trigger data;

[0106] Specifically, the calibration rule log is analyzed to extract target indicators, risk information, resource allocation information, and design variable information for each process step. Target indicators include product purity and recovery rate; risk information includes the degree of numerical deviation and the time period of deviation occurrence; resource allocation information includes reagent dosage and energy quota; and design variable information includes equipment speed and reaction temperature. Based on the inconsistency event sequence, the time period of deviation occurrence and equipment description are located, and the degree of numerical deviation is directly mapped to risk information. Combined with the calibration operation actions generated by calibration prediction, the equipment speed and reaction temperature that caused the deviation are deduced as design variable information. The changes in reagent dosage and energy quota associated with the calibration operation actions are recorded as resource allocation information. Using product purity and recovery rate as a benchmark, the difference between the calibration prediction value and the original prediction value is compared to determine the target indicators that need to be optimized. The extracted product purity and recovery rate, degree of numerical deviation and time period of deviation occurrence, reagent dosage and energy quota, equipment speed and reaction temperature are summarized to form design optimization trigger data.

[0107] S5.2: Based on the design optimization trigger data, coordinate and solve the equipment configuration parameters, connection relationship parameters, operating boundary parameters and collaborative rule parameters corresponding to each process step to generate an optimized design scheme;

[0108] Specifically, the design optimization trigger data is analyzed to extract equipment configuration parameters (including motor power and impeller diameter), connection relationship parameters (including pipeline flow direction and valve opening), operating boundary parameters (including upper temperature limit and lower pressure limit), and coordination rule parameters (including start-stop sequence and interlocking logic). Based on the risk information recorded in the design optimization trigger data, the upper temperature limit and lower pressure limit are adjusted to reconstruct the operating boundary parameters. Combining the resource configuration information in the design optimization trigger data, the pipeline flow direction and valve opening are changed to update the connection relationship parameters. Using the design variable information in the design optimization trigger data, the motor power and impeller diameter are corrected to optimize the equipment configuration parameters. Based on the target indicators in the design optimization trigger data, the start-stop sequence and interlocking logic are verified to improve the coordination rule parameters. The adjusted equipment configuration parameters, connection relationship parameters, operating boundary parameters, and coordination rule parameters are coordinated and solved to generate an optimized design scheme.

[0109] S5.3: Invoke the integrated training and inference process to update the optimized design scheme and output the scheme score result that matches the current design goal and design constraints during the inference stage;

[0110] Specifically, the integrated training and inference process is invoked to read the equipment configuration parameters, connection relationship parameters, operating boundary parameters, and collaborative rule parameters from the optimized design scheme. These parameters are mapped to the baseline input data of the data-driven basic prediction model. The equipment configuration parameters and connection relationship parameters are converted into process state vectors using the encoding layer. The process state vectors are aggregated by the fusion layer to form a fusion feature representation containing topological associations. The fusion feature representation is sent to the regression head, and constraints are constructed by combining the operating boundary parameters and collaborative rule parameters to perform inference and obtain the scheme scoring result. The scheme scoring result generated in the current round is compared with the scheme scoring result generated in the previous round. If the absolute value of the difference between the two values ​​is less than the preset error tolerance and this state is maintained for multiple consecutive rounds, the scheme scoring result is determined to be stable.

[0111] It should be noted that the integrated training and inference process refers to the iterative calculation process in which the pre-trained data drives the basic prediction model, maps the optimized design scheme to the baseline input data, performs feature extraction and constraint solving through the encoding layer, fusion layer and regression head, and updates the design parameters in reverse according to the scheme scoring results until convergence.

[0112] The scheme scoring result refers to the numerical evaluation of the degree to which the predicted values ​​of key indicators calculated by the regression head match the current design goals and design constraints, and quantitatively reflects the degree to which the equipment configuration parameters, connection relationship parameters, operating boundary parameters and collaborative rule parameters in the optimized design scheme meet the requirements of the process mechanism element set.

[0113] S5.4: Based on the scheme scoring results, generate a design configuration set according to the equipment configuration parameters, connection relationship parameters, operating boundary parameters and coordination rule parameters corresponding to each process step;

[0114] Specifically, the scheme evaluation results are analyzed to extract optimized design schemes that meet the current design goals and constraints. The included equipment configuration parameters, connection relationship parameters, operational boundary parameters, and coordination rule parameters are identified. Based on the quantitative matching degree reflected in the scheme evaluation results, equipment configuration parameters with numerical evaluations exceeding preset evaluation thresholds are selected. The specific set values ​​for motor power and impeller diameter are retained. Combined with the verified connection relationship parameters in the scheme evaluation results, the final states of pipeline flow direction and valve opening are solidified. Using parameters that meet operational boundary requirements in the scheme evaluation results, the boundary values ​​of the upper temperature limit and lower pressure limit are determined. Based on the verified coordination rule parameters in the scheme evaluation results, the execution sequence of start-up / shutdown order and interlocking logic is established. The selected equipment configuration parameters, connection relationship parameters, operational boundary parameters, and coordination rule parameters are categorized and assembled according to each process stage of crushing, grinding, classification, flotation, and thickening / dewatering to form a design configuration set containing a complete set of parameters.

[0115] It should be noted that the preset evaluation threshold is based on the distribution characteristics of the numerical deviation of historical inconsistencies recorded in the calibration rule log and the process control accuracy requirements. It is set by statistically analyzing the dividing point between normal deviation and general deviation in the historical scheme scoring results. The exemplary value range is 80% to 90% of the full score of the scheme scoring result. The basis for the value is to ensure that the schemes selected for the design configuration set can substantially meet the product purity and recovery rate targets and have no serious equipment operation risks while eliminating normal measurement noise interference.

[0116] S6. Obtain the interactive feedback data corresponding to the design configuration set, incrementally update the mineral processing knowledge graph and process mechanism element set to obtain the interactive baseline, and complete the next round of design iteration based on the interactive baseline.

[0117] S6.1: Obtain trial operation feedback information, process change information, result feedback information, and abnormal status information corresponding to the design configuration set, and form interactive feedback data;

[0118] Specifically, a trial run based on the design configuration set is initiated. Real-time sensor readings, control command responses, and operational fluctuations during equipment operation are collected as process change information. Deviations between final product indicators and intermediate process indicators and expected targets are recorded as result feedback information. Downtime events, parameter limit alarms, process connection anomalies, and coordination mismatches occurring during operation are monitored. Event types, durations, triggering conditions, and associated process stages are compiled into abnormal status information. Evaluations of equipment response speed, control accuracy, process connection smoothness, and scheme execution stability are summarized to form trial run feedback information. Process change information, result feedback information, abnormal status information, and trial run feedback information are aligned with timestamps and packaged to form interactive feedback data. This interactive feedback data is mapped to interactive status inputs corresponding to each process stage. Result feedback information is compared with preset result deviation thresholds, and abnormal status information is compared with preset abnormal trigger thresholds to form the basis for interactive classification judgments for each process stage.

[0119] It should be noted that the steps for setting the result deviation threshold are as follows: based on historical trial operation data, extract the deviation distribution of the final product indicators and intermediate process indicators corresponding to each process step relative to the target value; perform statistical analysis on the deviation distribution and determine the benchmark deviation range in combination with the allowable process fluctuation range; tighten or loosen the boundary of the benchmark deviation range according to the quality control requirements and design tolerance of each process step to form the result deviation threshold.

[0120] The steps for setting the abnormal trigger threshold are as follows: based on historical trial operation data, the frequency, duration and impact of shutdown events, parameter limit alarms, process connection abnormalities and coordination mismatches corresponding to each process link are statistically analyzed. Various abnormalities are graded and evaluated, and the abnormal trigger benchmark range is determined in combination with the requirements of continuous process operation. The abnormal trigger benchmark range is calibrated according to the operational risk level of each process link to form the abnormal trigger threshold.

[0121] S6.2: Based on the interactive feedback data, incrementally update the affected mechanism fragments in the affected facts and process mechanism element set in the mineral processing knowledge graph to obtain the interactive baseline corresponding to each process link;

[0122] Specifically, the interactive feedback data is analyzed to extract abnormal state information, result feedback information, and process change information. Combined with the identification of the relevant process link and the time correlation, the affected facts associated with it in the mineral processing knowledge graph are located. The affected facts are compared with the current process mechanism element set to identify the affected mechanism fragments that match the actual operation deviation.

[0123] If the result feedback information does not exceed the preset result deviation threshold and the abnormal status information does not exceed the preset abnormal trigger threshold, the corresponding process link will be judged as a stable record-type interaction state, and the verified process change information in the interaction feedback data will be written into the attribute field of the affected fact.

[0124] If the result feedback information exceeds the preset result deviation threshold, and the abnormal status information does not exceed the preset abnormal trigger threshold, then the corresponding process link will be judged as a deviation correction interaction state, and the verified process change information in the interaction feedback data will be written into the attribute field of the affected fact. At the same time, the applicable boundary and numerical constraint conditions in the affected mechanism segment will be adjusted.

[0125] If the result feedback information does not exceed the preset result deviation threshold, and the abnormal status information exceeds the preset abnormal trigger threshold, then the corresponding process link will be judged as an abnormal constraint interaction state, and the process connection conditions and collaborative constraint conditions in the affected mechanism segment will be adjusted based on the abnormal status information.

[0126] If the feedback information exceeds the preset result deviation threshold and the abnormal status information exceeds the preset abnormal trigger threshold, the corresponding process link will be judged as a linkage reconfiguration type interactive state, and the attribute fields of the affected facts, the numerical constraints, applicable boundaries, process connection conditions and collaborative constraints in the affected mechanism fragments will be updated simultaneously. At the same time, the corresponding process link will be marked as an object to be propagated and updated.

[0127] After the update is completed, the processes of crushing, grinding, classification, flotation and thickening / dewatering are re-aggregated to generate interactive baselines corresponding to each process step.

[0128] It should be noted that the object to be disseminated and updated refers to the current process step identifier, along with the combination of the affected facts, affected mechanism fragments, and related relationship markers, which are identified as needing to transmit the update impact to subsequent process steps in the linked reconstructive interaction state.

[0129] Verifying correct process change information means that the process change information and the result feedback information, abnormal status information and trial operation feedback information in the corresponding time period meet the time consistency and process logic consistency. Perform timestamp consistency verification, abnormal event alignment verification and process logic matching verification on the process change information. The process change information that passes the verification is considered to be the correct process change information.

[0130] The stable recording interaction state, deviation correction interaction state, abnormal constraint interaction state, and linkage reconfiguration interaction state represent different interaction processing types, namely, performing only record updates, performing deviation corrections, performing abnormal constraint corrections, and performing cross-process linkage updates.

[0131] S6.3: Based on the interaction baseline, the mineral processing knowledge graph, process mechanism element set, correction baseline, mechanism corrector, optimization design scheme and design configuration set are iteratively updated according to the interaction status corresponding to each process link;

[0132] Specifically, based on the interaction status of each process step contained in the interaction baseline, process-by-process synchronous updates are performed according to the process connection relationship in the process topology; based on the interaction status corresponding to each process step and the object to be propagated and updated, it is determined whether the update of the current process step will cause changes in the material flow relationship, energy flow relationship or time delay relationship of the subsequent process steps.

[0133] If none of them change, a partial update of the current process step is performed; if any one of them changes, a synchronous update of the subsequent process steps directly connected to the current process step is triggered; if at least two of them change simultaneously, a synchronous update of the current process step and its multiple subsequent process steps is triggered; if all three of them change and the interaction state corresponding to the current process step is a linkage reconstruction interaction state, a linkage reconstruction update of the mineral processing knowledge graph, process mechanism element set, correction baseline, mechanism corrector, optimization design scheme and design configuration set is triggered.

[0134] By comparing the interaction states of each process step with the existing facts in the mineral processing knowledge graph, conflicting node attributes and association attributes are corrected. The updated node attributes and association attributes are used to verify the process mechanism element set, replace the affected mechanism fragments that no longer match, and correct the corresponding process connection constraints. Based on the updated process mechanism element set, the baseline content corresponding to each process step in the calibration baseline is adjusted to obtain a new calibration baseline. The new calibration baseline is loaded into the mechanism corrector, and the logical judgment rules corresponding to each process step are reset. The updated mechanism corrector is called to perform a process-by-process re-evaluation of the original optimized design scheme, eliminate schemes that do not meet the updated logical judgment rules, and generate a new optimized design scheme. Based on the scoring results in the new optimized design scheme, the equipment configuration parameters, connection relationship parameters, operating boundary parameters, and collaborative rule parameters corresponding to each process step are synchronously corrected to form an updated design configuration set.

[0135] It should be noted that partial updates only update the mineral processing knowledge graph, process mechanism element set, calibration baseline, and design configuration set corresponding to the current process step;

[0136] The synchronous update is performed sequentially on the current process step and subsequent process steps that meet the propagation conditions according to the process connection relationship in the process topology;

[0137] The linkage reconstruction and update process regenerates and updates the mineral processing knowledge graph, process mechanism element set, correction baseline, mechanism corrector, optimized design scheme and design configuration set corresponding to the current process link and the process link within the propagation range.

[0138] Figure 5 This reflects the relationship between prediction, correction, and iterative comparison of product purity across the entire process. The blue line in the figure represents the measured product purity value, the orange line represents the full-process design prediction of Scheme A, the green line represents the correction prediction of Scheme B, and the red line represents the output of Scheme C after iteration. The upper part provides a global comparison, while the lower part magnifies the sensitive area of ​​disturbance. It highlights that near the local peaks and points of maximum difference, Scheme A fluctuates more significantly and deviates more noticeably from the measured value, indicating that relying solely on the full-process design prediction has limited ability to fit disturbances under complex operating conditions. Scheme B converges significantly towards the measured value, indicating that the mechanism corrector has effectively performed structured correction on the original prediction. Scheme C is even closer to the measured value than Scheme B, demonstrating that the iterative update driven by the interactive baseline does indeed improve the output stability and fit.

[0139] Figure 6 This reflects the correction process of key indicators in the flotation process under constrained boundaries. The blue line in the figure represents the measured key indicators in the flotation process, the orange line represents the full-process design prediction of flotation scheme A, the green line represents the corrected prediction of flotation scheme B, and the red line represents the output of flotation scheme C after iteration. The figure also shows two boundary lines: the lower limit and the upper limit of flotation. Scheme A is significantly higher than the measured values ​​in multiple time periods, and locally approaches or even touches the upper limit, indicating that the original full-process design prediction is more likely to exceed the limits in the sensitive flotation process. Scheme B is significantly pushed back between the upper and lower limits, and the curve shape is smoother, indicating that the mechanism corrector has truly implemented the "equipment operating parameter range" and "change trend rules" at the numerical level. Scheme C further approaches the measured values ​​without breaking the boundaries, indicating that after completing the constraint correction, combined with feedback iteration, the key indicators of flotation can both meet the boundary requirements and improve the approximation of the actual operating state.

[0140] Figure 7 This figure reflects a comparison of the stage-averaged residual information of five process stages: crushing, grinding, classification, flotation, and thickening / dewatering. Each process stage is presented with three sets of bar charts, corresponding to the full-process design prediction of Scheme A, the corrected prediction of Scheme B, and the iterative output of Scheme C. The residual information of the flotation process stage is the highest in Scheme A, indicating that flotation is the stage in the entire process most prone to amplifying prediction errors. This is consistent with... Figure 6 The demonstration that "flotation is most sensitive to constraint boundaries" is mutually corroborating; grinding, classification, and thickening / dewatering also show a stepwise decrease in both Scheme B and Scheme C, indicating that the mechanism corrector does not only work on a single process step, but has a uniform deviation compression effect on all process segments; Scheme C has the lowest values ​​in all process steps, which further illustrates that the present invention does not simply "correct once", but rather compresses residual information further through continuous iteration driven by interactive baselines, forming design optimization trigger data and providing a more reliable data foundation.

[0141] This embodiment also provides an intelligent design system for a darkroom in a metal mine, comprising: a topology construction module for collecting operating condition data and process structure data, extracting the relationships between crushing, grinding, classification, flotation, and thickening / dewatering, and constructing a process topology; a graph mechanism module for labeling equipment constraints and phenomenological patterns based on the process topology, forming a mineral processing knowledge graph; and a process mechanism element set based on operating condition data, process structure data, and the physical laws corresponding to each process step; and a predictive design module for summarizing the process topology, mineral processing knowledge graph, and process mechanism element set to form a calibration baseline, and using the calibration baseline to train and infer the data-driven basic predictive model to obtain the full-process design. The system comprises four modules: a prediction module and a correction rule module. The former is used to construct a mechanism corrector based on the mineral processing knowledge graph and the process mechanism element set. The mechanism corrector performs segmented projection and constraint solving on the full-process design prediction, outputs the corrected prediction, and registers inconsistencies to form a correction rule log. The latter is used to trigger full-process design optimization with the corrected prediction and correction rule log, generate optimized design schemes, perform integrated training and inference on the optimized design schemes, and produce a design configuration set. The former is used to obtain interactive feedback data corresponding to the design configuration set, incrementally update the mineral processing knowledge graph and the process mechanism element set, obtain an interactive baseline, and complete the next round of design iteration based on the interactive baseline.

[0142] In summary, this invention achieves structured correction and consistent constraint expression of design prediction results by constructing a mechanism corrector to perform segmented projection and constraint solving of the entire process design prediction. This enables the prediction output of each process step to have the processing capabilities of segmented transmission, conflict identification, residual characterization, and regular sedimentation. At the same time, it forms a traceable record of inconsistencies, providing a stable, standardized, and reusable correction basis for design optimization triggering, scheme selection, and iterative updates.

[0143] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart design method for a lights-out factory in a metal mine, characterized in that: include: Collect operating condition data and process structure data, extract the relationship between crushing, grinding, classification, flotation and thickening and dewatering, and construct the process topology; Based on the process topology, equipment constraints and phenomenon patterns are labeled and accumulated to form a mineral processing knowledge graph; based on operating condition data, process structure data and the physical laws corresponding to each process link, a set of process mechanism elements is formed. The process topology, mineral processing knowledge graph and process mechanism elements are summarized to form a calibration baseline. The calibration baseline is then used to train and infer the data-driven basic prediction model to obtain the full process design prediction. A mechanism corrector is constructed based on the mineral processing knowledge graph and the set of process mechanism elements. The mechanism corrector is used to perform segmented projection and constraint solving on the whole process design prediction, output corrected predictions, and register inconsistency events to form a correction rule log. Trigger full-process design optimization using correction prediction and correction rule logs, generate optimized design schemes, perform integrated training and inference processing on optimized design schemes, and produce design configuration sets; The interaction feedback data corresponding to the design configuration set is obtained, and the mineral processing knowledge graph and process mechanism element set are incrementally updated to obtain the interaction baseline. The next round of design iteration is completed by the interaction baseline.

2. The intelligent design method for a darkened metal mine factory as described in claim 1, characterized in that, The construction process topology consists of the following steps: The operating condition data is time-aligned and resampled to form a synchronized operating condition frame; Based on the synchronous operating condition frame and process structure data, set the monitoring section corresponding to each process position, and extract the material flow characteristics and energy flow characteristics of each monitoring section; Based on the characteristics of material flow and energy flow, determine the set of material flow relationships and the set of energy flow relationships among crushing, grinding, classification, flotation and thickening / dewatering, and construct the process topology.

3. The intelligent design method for a darkened metal mine factory as described in claim 2, characterized in that, The steps for forming the mineral processing knowledge graph are as follows: Based on the process topology, the synchronous working condition frames are aligned with topology time delay and segmented into states to form a set of topology constraint time sequence segments corresponding to each process step. By combining the set of time-series fragments of topological constraints with process structure data, the equipment constraints corresponding to each process step are labeled, and the time delay correlation, change direction and consistency relationship between each process step are extracted within the process connection relationship of the process topology, forming a set of equipment constraint facts and a set of phenomenon regularity facts; Based on the set of facts constrained by equipment and the set of facts governing phenomena, and combined with the process connection relationships, material flow relationships and energy flow relationships of the process topology, a mineral processing knowledge graph is constructed.

4. The intelligent design method for a darkened metal mine factory as described in claim 1 or 3, characterized in that, The steps for forming the process mechanism element set are as follows: Based on mineral processing knowledge graph, combined with operating condition data, process structure data and the physical laws corresponding to each process step, a set of process mechanism features is extracted; By using the process mechanism feature set, the mechanism features corresponding to each process step are effectively screened and parameters are summarized to form a process mechanism element set.

5. The intelligent design method for a darkened metal mine factory as described in claim 1, characterized in that, The steps to obtain the full-process design prediction are as follows: The process topology, mineral processing knowledge graph, and process mechanism element set are structured and summarized to form a calibration baseline container corresponding to each process step; By correcting the baseline container, the synchronous operating condition frame is divided into time windows, and the process connection relationship of the process topology is combined to form the baseline input data corresponding to each window; The data-driven basic prediction model is invoked to jointly extract and fuse the baseline input data to obtain the local prediction results corresponding to each window. The local prediction results are connected in series according to the process connection relationship of the process topology to obtain the whole process design prediction.

6. The intelligent design method for a darkened metal mine factory as described in claim 1 or 5, characterized in that, The steps for generating the correction rule log are as follows: The entire process design prediction is segmented according to the process connection relationship of the process topology to form each design prediction segment. Then, a mechanism corrector is constructed by combining the mineral processing knowledge graph and the process mechanism element set. A segmented correction context is established for each design prediction segment using a mechanism corrector, and the segmented correction context is converted into a corresponding segmented projection task. Based on the segmented projection task, segmented projection and constraint solving are performed sequentially on each design prediction segment to obtain the corrected prediction. Based on the conflict information, residual information, displacement information and identification information of each design prediction segment during the correction process, inconsistencies are recorded and summarized to form a correction rule log.

7. The intelligent design method for a darkened metal mine factory as described in claim 6, characterized in that, The steps for generating the optimized design scheme are as follows: Based on the correction prediction and correction rule logs, target indicators, risk information, resource allocation information and design variable information corresponding to each process step are extracted to form design optimization trigger data; Based on the design optimization trigger data, the equipment configuration parameters, connection relationship parameters, operating boundary parameters and collaborative rule parameters corresponding to each process step are coordinated and solved to generate an optimized design scheme.

8. The intelligent design method for a darkened metal mine factory as described in claim 1 or 7, characterized in that, The steps for producing the design configuration set are as follows: The integrated training and inference process is invoked to update the optimized design scheme, and the scheme score result that matches the current design goals and design constraints is output during the inference phase; Based on the scheme scoring results, a design configuration set is generated according to the equipment configuration parameters, connection relationship parameters, operation boundary parameters, and coordination rule parameters corresponding to each process step.

9. The intelligent design method for a darkened metal mine factory as described in claim 8, characterized in that, The interaction baseline is obtained, and the next design iteration is completed based on the interaction baseline. The steps are as follows: Acquire trial operation feedback information, process change information, result feedback information, and abnormal status information corresponding to the design configuration set, and form interactive feedback data; Based on the interactive feedback data, the affected mechanism fragments in the affected facts and process mechanism element set of the mineral processing knowledge graph are incrementally updated to obtain the interactive baseline corresponding to each process link. Based on the interaction baseline, the mineral processing knowledge graph, process mechanism element set, correction baseline, mechanism corrector, optimization design scheme and design configuration set are iteratively updated according to the interaction status of each process link.

10. A smart design system for a darkroom in a metal mine, based on the smart design method for a darkroom in a metal mine according to any one of claims 1 to 9, characterized in that, include: The topology construction module is used to collect operating condition data and process structure data, extract the relationship between crushing, grinding, classification, flotation and thickening and dewatering, and construct the process topology. The graph mechanism module is used to annotate equipment constraints and phenomenological patterns based on process topology to form a mineral processing knowledge graph; and to form a set of process mechanism elements based on operating condition data, process structure data and the physical laws corresponding to each process step. The predictive design module is used to summarize the process topology, mineral processing knowledge graph and process mechanism elements to form a calibration baseline. The calibration baseline is then used to train and infer the data-driven basic prediction model to obtain the full process design prediction. The correction rule module is used to construct a mechanism corrector based on the mineral processing knowledge graph and the set of process mechanism elements. The mechanism corrector is used to perform segmented projection and constraint solving on the whole process design prediction, output the correction prediction, and register inconsistency events to form a correction rule log. The design optimization module is used to correct prediction and correction rule logs to trigger full-process design optimization, generate optimized design schemes, perform integrated training and inference processing on optimized design schemes, and produce design configuration sets. The feedback iteration module is used to obtain interactive feedback data corresponding to the design configuration set, incrementally update the mineral processing knowledge graph and process mechanism element set, obtain the interactive baseline, and complete the next round of design iteration based on the interactive baseline.

Images