A beneficiation equipment alarm method and system based on semantic analysis

Abstract

The application provides a beneficiation equipment alarm method and system based on semantic analysis, which comprises the following steps: pre-processing and standardizing collected data; dynamically analyzing and calculating the standardized data through a semantic analysis alarm rule generation algorithm; triggering an alarm when real-time semantic scores meet alarm conditions, and performing abnormal root cause reasoning based on a feature-semantic association library; generating optimized alarm content containing treatment suggestions according to abnormal diagnosis results; associating and verifying alarm logs with production execution data; adjusting feature weights in the semantic analysis alarm rule generation algorithm according to operation feedback and authenticity labels in labeled alarm logs; and taking the adjusted feature weights as initial weights of the alarm rule generation algorithm in the next round to realize continuous self-learning of the alarm system. The application effectively improves the accuracy and practicability of the alarm and overcomes the poor adaptability of traditional fixed threshold rules.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment safety, and in particular to a beneficiation equipment alarm method and system based on semantic analysis. BACKGROUND

[0002] As the core link of mining production, beneficiation industry relies on the stable operation of a large number of equipment (such as flotation machines, grinding machines, and conveyor belts) to achieve efficient processing and resource recovery of ores. The equipment alarm system plays a key role in monitoring equipment status, ensuring production safety, and optimizing process flow. With the development of industrial Internet of Things (IIoT) and sensor technology, beneficiation sites can collect real-time equipment operation data (such as speed, vibration, and temperature), process parameters (such as pulp concentration and recovery rate), and environmental data (such as humidity and air pressure). These data provide a basis for equipment anomaly detection and alarm. However, the complexity of the beneficiation process (such as multivariate coupling and dynamic working conditions) and the high noise characteristics of the data make it difficult for traditional alarm systems to meet the accuracy and adaptability requirements, resulting in frequent false positives and false negatives, affecting operational response efficiency and decision quality.

[0003] Existing technical solutions mainly revolve around device-driven alarm modes. For example, alarm systems based on fixed thresholds trigger warnings by presetting parameter ranges (such as speed exceeding 1200 rpm); alarm schemes based on rule engines combine simple logic (such as temperature rising and vibration anomaly) to make abnormality judgments. These solutions achieve equipment status monitoring to some extent, but have limited performance in complex beneficiation scenarios.

[0004] The deficiencies of existing beneficiation equipment alarm systems mainly lie in the following aspects, and the technical reasons and impacts are as follows:

[0005] (1) Fixed thresholds and simple rules cannot adapt to the complexity and dynamics of beneficiation processes. For example, a rising speed may be a process adjustment rather than a fault, triggering a false alarm; while an implicit anomaly (such as slight vibration caused by wear) is not reported due to not reaching the threshold. The technical reason is that the system only relies on static comparison or Boolean logic, lacking intelligent analysis of the context relationships between equipment status, process flow, and environmental data.

[0006] (2) Alarm information usually only indicates surface information such as "parameter out of range" or "equipment anomaly", without providing the root cause of the problem or solutions, resulting in operators spending time troubleshooting. The technical reason is that the system does not analyze the semantic associations between data, and the alarm generation is superficial.

[0007] (3) In complex scenarios, frequent false positives or ambiguous alarms require manual verification by technical personnel, increasing response time and labor costs. The technical reason is the lack of adaptive optimization capability, which cannot dynamically adjust alarm rules according to production environments.

[0008] (4) The beneficiation process is affected by factors such as ore properties (e.g. hardness variation), equipment aging, etc. Static rules are difficult to cope with changing conditions, resulting in a decline in alarm effectiveness over time. The technical reason is that real-time data analysis and learning mechanisms are not integrated.

[0009] The above shortcomings limit the practicality of the alarm system in beneficiation production, reduce trust through false positives, increase downtime risk through missed reports, and affect production efficiency and equipment safety. SUMMARY

[0010] In view of the above situation, the main purpose of the present application is to propose a beneficiation equipment alarm method and system based on semantic analysis to solve the above technical problems.

[0011] The present application proposes a beneficiation equipment alarm method based on semantic analysis, which comprises the following steps:

[0012] Step 1, real-time collection of operation parameters, process parameters and environmental parameters of beneficiation equipment, and pre-processing and standardization of collected data to obtain standardized data;

[0013] Step 2, semantic feature extraction of standardized data, and construction of semantic relationship between parameters according to pre-constructed beneficiation knowledge graph to form a feature-semantic association library;

[0014] Based on the feature-semantic association library, the standardized data is dynamically analyzed and calculated by a semantic analysis alarm rule generation algorithm to generate dynamic alarm rules and output real-time semantic scores;

[0015] Step 3, according to the dynamic alarm rules, when the real-time semantic score meets the alarm condition, trigger the alarm, and according to the feature-semantic association library, carry out abnormal root cause reasoning to obtain abnormal diagnosis results; according to the abnormal diagnosis results, generate optimization alarm content containing processing suggestions to obtain structured alarm information;

[0016] Step 4, output and record the structured alarm information to generate an alarm log; obtain production execution data, associate and verify the alarm log with the production execution data, and mark the alarm record in the alarm log with a authenticity label to generate a labeled alarm log;

[0017] Step 5, according to the operation feedback and authenticity label in the labeled alarm log, adjust the feature weight in the semantic analysis alarm rule generation algorithm to obtain the adjusted feature weight, and use the adjusted feature weight as the initial weight of the next round of alarm rule generation algorithm, to realize continuous self-learning of the alarm system, for accurate abnormal detection and alarm in the future.

[0018] The application further provides a beneficiation equipment alarm system based on semantic analysis, and the system comprises:

[0019] a data acquisition module, which is used for:

[0020] acquiring running parameters, process parameters and environmental parameters of the beneficiation equipment in real time, and pre-processing and standardizing the acquired data to obtain standardized data;

[0021] a semantic analysis module, which is used for:

[0022] extracting semantic features from the standardized data, and forming a feature-semantic correlation library according to semantic relationships between pre-constructed beneficiation knowledge graphs and construction parameters;

[0023] based on the feature-semantic correlation library, dynamically analyzing and calculating the standardized data through a semantic analysis alarm rule generation algorithm, generating dynamic alarm rules and outputting real-time semantic scores;

[0024] an alarm optimization module, which is used for:

[0025] according to the dynamic alarm rules, triggering an alarm when the real-time semantic scores meet alarm conditions, and obtaining an abnormal diagnosis result by performing abnormal root cause reasoning according to the feature-semantic correlation library; generating optimized alarm content containing processing suggestions according to the abnormal diagnosis result to obtain structured alarm information;

[0026] an alarm output module, which is used for:

[0027] outputting and recording the structured alarm information to generate an alarm log; obtaining production execution data, associating and verifying the alarm log with the production execution data, marking a realness label on alarm records in the alarm log, and generating a labeled alarm log;

[0028] a communication and storage module, which is used for:

[0029] adjusting feature weights in the semantic analysis alarm rule generation algorithm according to operation feedback and the realness label in the labeled alarm log, obtaining adjusted feature weights, taking the adjusted feature weights as initial weights of a next round of alarm rule generation algorithm, realizing continuous self-learning of the alarm system, and performing accurate abnormal detection and alarm in the future.

[0030] Compared with the prior art, the application has the following beneficial effects:

[0031] 1. The application realizes real-time collection of the operation parameters, process parameters and environmental parameters of the beneficiation equipment, pretreatment and standardization, semantic feature extraction of the standardized data, construction of the semantic relationship between the parameters based on the pre-constructed beneficiation knowledge graph to form a feature-semantic association library, dynamic calculation of the real-time semantic score by the semantic analysis alarm rule generation algorithm and generation of the dynamic alarm rule, triggering of the alarm when the score meets the condition, abnormal root cause reasoning based on the feature-semantic association library to generate structured alarm information containing processing suggestions, effectively improving the accuracy and practicality of the alarm, and overcoming the poor adaptability of the traditional fixed threshold rule.

[0032] 2. The application realizes abnormal root cause reasoning based on the feature-semantic association library after triggering the alarm, extracts key features from the abnormal event data packet and traverses the semantic association nodes to generate a candidate root cause set, combines the historical frequency and confidence to calculate the matching score to determine the primary suspected target, and automatically generates targeted processing suggestions, realizes the intelligent closed loop from abnormal detection to root cause analysis, significantly shortens the troubleshooting time and reduces the artificial dependence.

[0033] 3. The application dynamically adjusts the feature weight in the semantic analysis alarm rule generation algorithm according to the operation feedback and authenticity label in the labeled alarm log, calculates the weight adjustment amount by using the weight update formula combined with the learning rate, alarm effect reward and historical accuracy, and uses the adjusted weight as the initial weight of the next round of algorithm, realizes the continuous self-learning and self-adaptive optimization of the alarm system, and continuously improves the accuracy and robustness of the abnormal detection in the complex beneficiation environment.

[0034] Additional aspects and advantages of the application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 A step flow chart of a beneficiation equipment alarm method based on semantic analysis is proposed for the application.

[0036] Figure 2 A beneficiation equipment alarm method based on semantic analysis is proposed for the application.

[0037] Figure 3 A beneficiation equipment alarm system based on semantic analysis is proposed for the application. DETAILED DESCRIPTION

[0038] Embodiments of the present application are described below in detail with reference to the drawings, in which like or similar elements or components are denoted by the same or similar reference numerals throughout the drawings. The embodiments described below are exemplary only, and are not to be construed as limiting the present application.

[0039] These and other aspects of embodiments of the present application will become clear from the following description and the accompanying drawings. In the description and drawings, some specific embodiments of embodiments of the present application are specifically disclosed in order to explain the principles of embodiments of the present application in some ways, but it should be understood that the scope of embodiments of the present application is not limited thereto.

[0040] Referring to Figure 1 The embodiments of the present application propose a beneficiation equipment alarm method based on semantic analysis, which comprises the following steps:

[0041] Step 1, real-time collection of operation parameters, process parameters and environmental parameters of beneficiation equipment, and preprocessing and standardization of collected data to obtain standardized data.

[0042] Referring to Figure 2 In step 1, real-time data collection is performed by a sensor network installed on key beneficiation equipment (such as a flotation machine, a grinding machine). The collected data includes: equipment operation parameters (such as motor speed (unit: rpm), equipment vibration amplitude (unit: mm / s), bearing temperature (unit: ℃)), process parameters (such as pulp concentration (unit: %), concentrate recovery rate (unit: %)) and environmental parameters (such as environmental humidity (unit: %), environmental temperature (unit: ℃)). The raw data collected is subjected to "3σ criterion" to eliminate outliers. That is, the average value and standard deviation of each parameter in the recent period (such as 5 minutes) are calculated, and the data points significantly deviating from the average value (such as more than 3 times the standard deviation range) are regarded as invalid data and eliminated. Then, data standardization is performed, and the "min-max standardization" method is used to convert the parameter values of different dimensions and orders of magnitude to the range of 0 to 1, to obtain the processed standardized data.

[0043] Step 2, semantic feature extraction is performed on the standardized data, and semantic relationships between parameters are constructed according to a pre-constructed beneficiation knowledge graph to form a feature-semantic association library;

[0044] Based on the feature-semantic association library, the standardized data is dynamically analyzed and calculated by a semantic analysis alarm rule generation algorithm to generate dynamic alarm rules and output real-time semantic scores.

[0045] In step 2, semantic feature extraction is performed on the standardized data, and the semantic relationship between the parameters is constructed according to the pre-constructed beneficiation knowledge graph to form a feature-semantic association library, which includes the following steps:

[0046] The time domain features of the standardized data are extracted, and the mean value calculation, standard deviation calculation and root mean square value calculation are performed respectively to obtain the time domain feature set;

[0047] The frequency domain features of the standardized data are extracted, and the target vibration frequency component and amplitude are obtained through fast Fourier transform to obtain the frequency domain feature set;

[0048] The time domain features in the time domain feature set and the frequency domain features in the frequency domain feature set are spliced, and principal component analysis is used for dimension reduction to obtain the reduced comprehensive feature vector;

[0049] The reduced comprehensive feature vector is matched with the pre-stored feature vector of the entity node in the pre-constructed beneficiation knowledge graph by cosine similarity, and the vectors with similarity higher than the preset threshold are selected to form a preliminary semantic association set;

[0050] The current equipment working condition parameters are obtained, and the weight of the associated vector in the preliminary semantic association set is calculated according to the current equipment working condition parameters to obtain a weighted semantic association set;

[0051] The association relationship of the associated vector in the weighted semantic association set is detected to obtain a detection result, and the contradictions detected in the detection result are eliminated according to the weight of the associated vector and the causal relationship chain in the pre-set knowledge graph to obtain a conflict-free semantic association set;

[0052] The semantic vector in the conflict-free semantic association set is bound with the corresponding feature vector and stored in the feature table of the pre-set cloud database to form a feature-semantic association library.

[0053] Based on the feature-semantic association library, the standardized data is dynamically analyzed and calculated by a semantic analysis alarm rule generation algorithm to generate dynamic alarm rules and output real-time semantic scores, which includes the following steps:

[0054] Based on the feature-semantic association library, the core feature list of the current equipment type and the corresponding initial weight value are queried to obtain an initial weight configuration set;

[0055] The real-time environmental sensor data is read, and the influence degree coefficient of the current environment on abnormal judgment is determined according to the pre-defined environment-weight mapping rule to obtain a context factor coefficient;

[0056] Based on the core feature list, the current value of the corresponding feature is extracted from the standardized data to obtain a standardized feature value set;

[0057] For each feature in the core feature list, calculate the Pearson correlation coefficient between each pair of features, take the absolute value of the Pearson correlation coefficient between each pair of features and calculate the average value to obtain the semantic context score;

[0058] The initial weight configuration set is multiplied with the corresponding elements in the standardized feature set, and then the results of the multiplication are summed to obtain the weighted sum.

[0059] The weighted sum, context factor coefficient, and semantic context score are calculated to obtain the real-time semantic score. The judgment logic that the real-time semantic score is greater than the preset dynamic adjustment threshold is defined as an executable alarm rule, and the executable alarm rule is used as the dynamic alarm rule.

[0060] Based on the core feature list, the current value of the corresponding feature is extracted from the standardized data. The relationship in this process is as follows:

[0061] ;

[0062] in, Indicates the first Standardized values ​​of each feature Indicates feature index, This represents the original data parameter value. This represents the minimum value of the parameter when the equipment is operating normally. This indicates the maximum value of the parameter when the equipment is operating normally;

[0063] In the step of calculating the Pearson correlation coefficient between each pair of features in the core feature list, taking the absolute value of the Pearson correlation coefficient between each pair of features and calculating the average value to obtain the semantic context score, the corresponding relationship in the process is as follows:

[0064] ;

[0065] in, Indicates the number of observations. The first feature represents the first characteristic. One observation value, Indicates the index of the observation value. This represents the average of all observations for the first feature. The second feature represents the first One observation value, This represents the average of all observations for the second feature;

[0066] In the step of calculating the weighted sum, context factor coefficient, and semantic context score to obtain the real-time semantic score, the corresponding relationship in the process is as follows:

[0067] ;

[0068] wherein, represents a real-time semantic score, represents a feature number, represents a weight of the th feature, represents a context factor coefficient.

[0069] It should be noted that the semantic analysis alarm rule generation algorithm of the present application is not intended to completely replace the traditional machine learning algorithm, but in the specific and high requirement scene of industrial alarm, it provides a more explainable, easier to integrate domain knowledge, and can realize intelligent diagnosis of adaptive solution. It effectively solves the "black box problem", "adaptability problem" and "diagnosis missing problem" encountered in the industrial practice of traditional machine learning method, and represents an important development direction of industrial intelligence from "perception" to "cognition".

[0070] Further, the pre-constructed beneficiation knowledge graph relied on by the present application follows a clear and implementable method in its construction process, ensuring the realizability of the system. This process combines domain expert knowledge and historical data-driven methods, first defining the core entities and their relationship types in the knowledge graph based on the beneficiation process and equipment mechanism. Subsequently, it is filled through multi-source knowledge acquisition, including directly extracting equipment properties and fault logic from structured data, using text mining techniques to extract associated information from unstructured documents, and performing association rule mining on historical operation data and fault data to discover potential, not explicitly defined important relationships. All these discoveries need to be confirmed by domain experts to ensure their reliability.

[0071] On this basis, the corresponding feature vectors of the entity nodes representing the equipment state in the knowledge graph are generated, which are encoded by the feature patterns corresponding to the state historical data, serving as templates for subsequent similarity matching. Finally, the complete knowledge graph containing entities, relationships, properties and feature vectors is stored in the database, forming a feature-semantic association library that can be queried and reasoned in real time by the system, and the graph supports regular iterative updates according to new cases and experiences.

[0072] In this embodiment, the information capable of representing the nature of the equipment running state, such as fluctuation, energy distribution and specific fault frequency component, is extracted from the original data through time domain and frequency domain feature extraction. The principal component analysis dimension reduction process aims to retain most of the key information while reducing the data dimension, improving the subsequent calculation efficiency and avoiding overfitting. Subsequently, by matching the real-time feature vector with the pre-stored patterns in the beneficiation knowledge graph, the system can understand the current equipment state in a semantic context composed of expert knowledge and historical experience.

[0073] The innovation of the dynamic alarm rule generation algorithm lies in that the semantic score is not a simple weighted sum, but a comprehensive index integrating the feature itself abnormality degree, environmental influence factors and the health degree of the correlation between features. This makes the alarm threshold not a fixed value, but an intelligent judgment logic that can be dynamically adjusted according to the actual working condition, significantly improving the adaptability of the system to complex production environment.

[0074] It should be noted that the "semantic analysis" referred to in the present application has specific industrial background connotation, and its core lies in understanding the physical meaning of the equipment running parameters in the beneficiation process context and their mutual correlation, which is essentially different from the text semantic analysis in the field of natural language processing. Specifically, the semantic analysis of the present application is a multi-level technical process: first, the original sensor readings are converted into feature indicators with clear engineering significance through time domain and frequency domain feature extraction, realizing the semantic improvement of data at the feature level; then, the pre-constructed beneficiation knowledge graph is used to give these feature indicators context relationships such as causality, linkage and other domain knowledge-based context relationships, forming a structured knowledge network understandable by machines; finally, in the dynamic alarm link, the overall health degree of the Pearson correlation coefficient between features is calculated to quantify the semantic context, thereby capturing the abnormal patterns of multi-parameter synergy. Therefore, the semantic analysis of the present application is essentially a technology system that integrates feature engineering, knowledge graph and statistical correlation analysis, and is specially used for context-related understanding and intelligent reasoning of industrial data.

[0075] In step 3, according to the dynamic alarm rule, when the real-time semantic score meets the alarm condition, an alarm is triggered, and abnormal root cause reasoning is performed according to the feature-semantic association library to obtain an abnormal diagnosis result; according to the abnormal diagnosis result, optimization alarm content containing processing suggestions is generated to obtain structured alarm information.

[0076] In step 3, according to the dynamic alarm rule, when the real-time semantic score meets the alarm condition, an alarm is triggered, and abnormal root cause reasoning is performed according to the feature-semantic association library to obtain an abnormal diagnosis result; according to the abnormal diagnosis result, optimization alarm content containing processing suggestions is generated to obtain structured alarm information, specifically including the following steps:

[0077] When the dynamic alarm rule condition is met, an alarm state is triggered, and snapshot data of all related parameters at the corresponding time is recorded to generate an abnormal event data packet;

[0078] Key abnormal features and corresponding abnormal deviation degrees of the features that cause the semantic score to rise are extracted from the abnormal event data packet to generate a key abnormal feature list;

[0079] Each feature in the key abnormal feature list is taken as a starting point, and priority traversal is performed in the feature-semantic association library to find all potential root cause nodes connected thereto through semantic relationships, a candidate root cause set is generated, and the candidate root cause set is taken as an abnormal diagnosis result;

[0080] According to the confidence of the preset association rule and the abnormal deviation degree corresponding to the feature, a matching degree score of each reason in the abnormal diagnosis result and the key abnormal feature list is calculated to obtain a calculated matching degree score;

[0081] The highest candidate root cause in the calculated matching degree score is selected as a primary suspect target, and a history of the same reason recorded in the preset knowledge graph is retrieved, a historical frequency of the same reason is calculated according to the same reason recorded in the preset knowledge graph, a comprehensive confidence of the highest candidate root cause is calculated according to the historical frequency of the same reason, and a confidence result of the highest candidate root cause is obtained;

[0082] An emergency plan database corresponding to the primary suspect target is queried to obtain a corresponding standard processing flow template text;

[0083] Specific parameter values in the abnormal event data packet are substituted into variable positions in the standard processing flow template text, and instantiation filling is performed to generate an optimal processing suggestion;

[0084] The primary suspect target, the confidence result of the highest candidate root cause, and the optimal processing suggestion are assembled in a predetermined format to generate structured alarm information.

[0085] It should be noted that the embodiment realizes the leap from anomaly detection to intelligent diagnosis. When the system triggers an alarm, it does not simply notify the anomaly, but immediately starts the root cause reasoning process. By tracing back the preset causal relationship chain in the feature-semantic association library, the potential upstream cause leading to the abnormal feature can be quickly located, forming a candidate diagnosis list. Further, by comprehensively considering the rule confidence defined in the knowledge base, the severity of the current anomaly, and the historical occurrence frequency of the cause, the comprehensive confidence of each candidate cause is calculated, so as to filter out the most likely root cause. Finally, by querying the emergency plan library, the abstract fault cause is converted into a treatment suggestion containing specific parameter values and operation steps. This process greatly improves the practical value of the alarm information, upgrades the traditional "what" alarm to "why" and "how" decision support, and significantly shortens the troubleshooting and response time.

[0086] In step 4, the structured alarm information is output and recorded to generate an alarm log; production execution data is obtained, the alarm log is associated and verified with the production execution data, and a real-time label is marked on the alarm record in the alarm log to generate a labeled alarm log.

[0087] In step 4, the structured alarm information is output and recorded to generate an alarm log; production execution data is obtained, the alarm log is associated and verified with the production execution data, and a real-time label is marked on the alarm record in the alarm log to generate a labeled alarm log, which includes the following steps:

[0088] The structured alarm information is pushed through a plurality of preset output channels, and the structured alarm information is recorded in the alarm record table of the preset cloud database to generate an alarm log record containing details;

[0089] According to a preset time window rule, the alarm triggering time is taken as a reference point, and a preset time length is extended forward and backward, serving as a time window for association verification;

[0090] Through a preconfigured data interface, all production event records in the association verification time window are obtained to form a production event set;

[0091] The alarm log record containing details and the production events in the production event set are matched according to the time stamp, and production abnormal events associated with the alarm content within the predefined time window are searched to obtain a matching result;

[0092] According to a preset verification rule set, the matching result is logically judged to generate a verification conclusion;

[0093] The alarm records in the verification conclusion that need to be manually reviewed are pushed to a manual review interface, and the manual confirmation results of the operators are recorded.

[0094] According to the preset label assignment rule, the integrated manual confirmation result and verification conclusion of the operation personnel are used to assign a authenticity label to each alarm log record to form a labeled alarm log; wherein, the authenticity label includes a true positive label, a false positive label and a missed alarm label.

[0095] It should be noted that the multi-channel pushing of the alarm information ensures the timely delivery and recording of the information. The key lies in the association and verification of the alarm log and the production execution. This design can effectively distinguish between real device abnormalities and parameter fluctuations caused by planned shutdown, process adjustment and other normal production operations, which is the key link to reduce false positives. By combining automatic logic judgment and necessary manual review, the system marks each alarm record with a "true positive", "false positive" or "missed" authenticity label. These labeled alarm logs constitute a high-quality training and evaluation dataset, providing a crucial supervisory signal for the next step of adaptive adjustment of feature weights, enabling the system to continuously learn from actual application results.

[0096] Step 5, according to the operation feedback and authenticity label in the labeled alarm log, adjusting the feature weight in the semantic analysis alarm rule generation algorithm, obtaining the adjusted feature weight, and taking the adjusted feature weight as the initial weight of the next round of alarm rule generation algorithm, realizing the continuous self-learning of the alarm system, so as to realize accurate abnormal detection and alarm in the subsequent.

[0097] In step 5, according to the operation feedback and authenticity label in the labeled alarm log, adjusting the feature weight in the semantic analysis alarm rule generation algorithm, obtaining the adjusted feature weight, and taking the adjusted feature weight as the initial weight of the next round of alarm rule generation algorithm, realizing the continuous self-learning of the alarm system, so as to realize accurate abnormal detection and alarm in the subsequent.

[0098] According to the preset statistical period, read the labeled alarm log from the cloud database, and for each feature in the labeled alarm log, count the number of true positive labels, false positive labels and missed alarm labels triggered by the alarm, and obtain the statistical result;

[0099] Based on the statistical result, calculate the performance indicators of the accuracy, false positive rate and missed alarm rate of each feature, and identify a set of to-be-optimized features whose performance indicators are lower than the preset threshold;

[0100] For each to-be-optimized feature in the set of to-be-optimized features whose performance indicators are lower than the preset threshold, according to the error type caused by the feature and the preset adjustment strategy, determine the weight adjustment direction; and according to the preset learning rate calculation rule, combine the frequency and severity of the feature error to calculate the weight adjustment amount of each to-be-optimized feature;

[0101] The weight adjustment amount of each feature to be optimized is applied to a current weight value of a corresponding feature in a semantic analysis alarm rule generation algorithm to obtain a new feature weight value;

[0102] The new feature weight value is updated into the feature-semantic association library to generate an adjusted feature weight.

[0103] According to a preset learning rate calculation rule, in combination with a frequency and severity of feature error occurrence, a weight adjustment amount of each feature to be optimized is calculated, and a corresponding process exists in a relationship as follows:

[0104] ;

[0105] Among them, denotes a next round weight, denotes a current round weight, denotes a learning rate, denotes an alarm effect reward, denotes a historical alarm accuracy.

[0106] In the embodiment, by periodically analyzing the labeled alarm log, the performance of each feature in the historical alarm can be quantitatively evaluated, and the features that frequently cause false positives or false negatives are identified. Then, according to the preset learning strategy, the weight of these features is adjusted. For example, for the features that often cause false positives, their weights are appropriately reduced; and for the features that can accurately indicate real abnormalities but have been ignored, their weights are improved.

[0107] It should be noted that the technical elements such as dynamic alarm, knowledge graph diagnosis and continuous self-learning integrated by the present application are not simply superimposed in the complex mineral processing production environment, but through deep interaction, a synergistic gain beyond the independent effect of each is generated, and the comprehensive problem of poor adaptability, high false positive rate and missing diagnosis of the traditional system is solved. The real-time semantic score of the dynamic alarm rule generation realizes the improvement of the detection sensitivity of the micro abnormality and the composite fault while effectively suppressing the false positives caused by normal process fluctuations, and achieves the unity of dynamic and accuracy. After the alarm is triggered, the system can immediately perform root cause reasoning based on the causal relationship in the feature-semantic association library, realize seamless connection and functional closed loop from abnormal perception to intelligent diagnosis, greatly shorten the decision chain and time of fault troubleshooting. More importantly, the self-learning mechanism of the system adjusts not only the weight of the alarm rule, but also the practicality and accuracy of the feature-semantic association library in the continuous operation, so that its cognitive ability evolves continuously, and finally forms a virtuous cycle of perception, cognition and optimization that becomes more and more accurate. The depth integration of this technology designed for special working conditions of mineral processing has non-obviousness in the overall technical effect.

[0108] Referring to Figure 3 The embodiment of the present application provides a beneficiation equipment alarm system based on semantic analysis, which comprises:

[0109] A data acquisition module is configured to:

[0110] collect operation parameters, process parameters and environmental parameters of the beneficiation equipment in real time, pre-process and standardize the collected data, and obtain standardized data;

[0111] A semantic analysis module is configured to:

[0112] extract semantic features from the standardized data, and form a feature-semantic association library according to semantic relationships between pre-constructed beneficiation knowledge graphs and construction parameters;

[0113] Based on the feature-semantic association library, the standardized data is dynamically analyzed and calculated by a semantic analysis alarm rule generation algorithm to generate dynamic alarm rules and output real-time semantic scores;

[0114] An alarm optimization module is configured to:

[0115] According to the dynamic alarm rules, when the real-time semantic scores meet the alarm conditions, an alarm is triggered, and abnormal root cause reasoning is performed according to the feature-semantic association library to obtain an abnormal diagnosis result; according to the abnormal diagnosis result, an optimized alarm content containing a processing suggestion is generated to obtain structured alarm information;

[0116] An alarm output module is configured to:

[0117] output and record the structured alarm information to generate an alarm log; obtain production execution data, associate and verify the alarm log with the production execution data, mark a realness label to alarm records in the alarm log, and generate a labeled alarm log;

[0118] A communication and storage module is configured to:

[0119] According to the operation feedback and the realness label in the labeled alarm log, the feature weights in the semantic analysis alarm rule generation algorithm are adjusted to obtain adjusted feature weights, and the adjusted feature weights are used as initial weights of the alarm rule generation algorithm in the next round to realize continuous self-learning of the alarm system, so as to accurately detect and alarm abnormalities in the future.

[0120] It should be appreciated that parts of the present application can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, several steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and as in another embodiment, any of the following technologies, or a combination thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0121] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0122] The above-described embodiments only express several implementation manners of the present application, which are described in a more specific and detailed manner, but cannot be understood as a limitation on the patent scope of the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, which are all within the protection scope of the present application. Therefore, the patent protection scope of the present application should be subject to the appended claims.

Claims

1. A method for alarming mineral processing equipment based on semantic analysis, characterized in that, The method includes the following steps: Step 1: Collect the operating parameters, process parameters and environmental parameters of the mineral processing equipment in real time, and preprocess and standardize the collected data to obtain standardized data; Step 2: Extract semantic features from the standardized data and perform dimensionality reduction. Match the reduced comprehensive feature vectors with the pre-stored feature vectors of entity nodes in the pre-constructed mineral processing knowledge graph using cosine similarity. Select vectors with similarity higher than a preset threshold to form a preliminary semantic association set. Obtain the current equipment operating parameters and calculate the weights of the association vectors in the preliminary semantic association set based on these parameters to obtain a weighted semantic association set. Detect the association relationships among the association vectors in the weighted semantic association set, obtain the detection results, and resolve any contradictions detected in the results based on the weights of the association vectors and the causal relationship chains in the preset knowledge graph to obtain a conflict-free semantic association set. Bind the semantic vectors in the conflict-free semantic association set to their corresponding feature vectors to form a feature-semantic association library. Based on the feature-semantic association library, a semantic analysis alarm rule generation algorithm is used to dynamically analyze and calculate standardized data, generate dynamic alarm rules, and output real-time semantic scores. Step 3: According to the dynamic alarm rules, when the real-time semantic score meets the alarm conditions, an alarm is triggered, an abnormal event data packet is generated, and the key features that cause the semantic score to rise and the degree of abnormal deviation corresponding to the features are extracted from the abnormal event data packet to generate a list of key abnormal features. Starting with each feature in the list of key anomalies, a priority traversal is performed in the feature-semantic association library to find all potential root cause nodes connected to it through semantic relationships, generating a candidate root cause set, which is used as the anomaly diagnosis result; based on the anomaly diagnosis result, optimized alarm content containing processing suggestions is generated to obtain structured alarm information. Step 4: Output and record the structured alarm information to generate an alarm log; Acquire production execution data, associate and verify alarm logs with production execution data, and mark alarm records in alarm logs with authenticity tags to generate tagged alarm logs; Step 5: Based on the operation feedback and authenticity tags in the tagged alarm logs, adjust the feature weights in the semantic analysis alarm rule generation algorithm to obtain the adjusted feature weights, and use the adjusted feature weights as the initial weights for the next round of alarm rule generation algorithm.

2. The mineral processing equipment alarm method based on semantic analysis according to claim 1, characterized in that, In step 2, semantic features are extracted and dimensionality is reduced from the standardized data, specifically including the following steps: Temporal features are extracted from standardized data, and mean, standard deviation and root mean square value are calculated respectively to obtain temporal feature set; Frequency domain features are extracted from standardized data, and the frequency components and amplitudes of the target vibration are obtained through fast Fourier transform to obtain a frequency domain feature set. The time-domain features in the time-domain feature set and the frequency-domain features in the frequency-domain feature set are concatenated, and then the dimensionality is reduced by principal component analysis to obtain the dimensionality-reduced comprehensive feature vector.

3. The mineral processing equipment alarm method based on semantic analysis according to claim 2, characterized in that, Based on a feature-semantic association library, a semantic analysis alarm rule generation algorithm is used to dynamically analyze and calculate standardized data, generate dynamic alarm rules, and output real-time semantic scores. The specific steps include the following: Based on the feature-semantic association library, query the predefined core feature list of the current device type and the corresponding initial weight values ​​to obtain the initial weight configuration set; Read real-time environmental sensor data and determine the influence coefficient of the current environment on anomaly detection according to a predefined environment-weight mapping rule to obtain the context factor coefficient. Based on the core feature list, the current value of the corresponding feature is extracted from the standardized data to obtain a set of standardized feature values; For each feature in the core feature list, calculate the Pearson correlation coefficient between each pair of features, take the absolute value of the Pearson correlation coefficient between each pair of features and calculate the average value to obtain the semantic context score; The initial weight configuration set is multiplied with the corresponding elements in the standardized feature set, and then the results of the multiplication are summed to obtain the weighted sum. The weighted sum, context factor coefficient, and semantic context score are calculated to obtain the real-time semantic score. The judgment logic that the real-time semantic score is greater than the preset dynamic adjustment threshold is defined as an executable alarm rule, and the executable alarm rule is used as the dynamic alarm rule.

4. The mineral processing equipment alarm method based on semantic analysis according to claim 3, characterized in that, Based on the core feature list, the current value of the corresponding feature is extracted from the standardized data. The relationship in this process is as follows: ； in, Indicates the first Standardized values ​​of each feature Indicates feature index, This represents the original data parameter value. This represents the minimum value of the parameter when the equipment is operating normally. This indicates the maximum value of the parameter when the equipment is operating normally; In the step of calculating the Pearson correlation coefficient between each pair of features in the core feature list, taking the absolute value of the Pearson correlation coefficient between each pair of features and calculating the average value to obtain the semantic context score, the corresponding relationship in the process is as follows: ； in, Indicates semantic context score, Indicates the number of observations. The first feature represents the first characteristic. One observation value, Indicates the index of the observation value. This represents the average of all observations for the first feature. The second feature represents the first One observation value, This represents the average of all observations for the second feature; In the step of calculating the weighted sum, context factor coefficient, and semantic context score to obtain the real-time semantic score, the corresponding relationship in the process is as follows: ； in, Represents real-time semantic scoring. Indicates the number of features. Indicates the first The weights of each feature, This represents the context factor coefficient.

5. The mineral processing equipment alarm method based on semantic analysis according to claim 4, characterized in that, In step 3, based on the anomaly diagnosis results, optimized alarm content containing processing suggestions is generated to obtain structured alarm information. This specifically includes the following steps: Based on the confidence level of the preset association rule and the degree of abnormal deviation corresponding to the feature, the matching score between each cause in the abnormal diagnosis result and the list of key abnormal features is calculated to obtain the calculated matching score. The highest candidate root cause among the calculated matching scores is selected as the primary suspected target, and the historical similar causes recorded in the preset knowledge graph are retrieved. The historical frequency of the similar causes is calculated based on the historical frequency of the similar causes, and the comprehensive confidence of the highest candidate root cause is calculated based on the historical frequency of the similar causes, so as to obtain the confidence result of the highest candidate root cause. Search the emergency response plan database for the primary suspected target and obtain the corresponding standard handling procedure template text; The specific parameter values ​​in the abnormal event data packet are substituted into the variable positions in the standard processing flow template text and instantiated to generate the optimal processing suggestion. The confidence results of the primary suspected target, the highest candidate root cause, and the optimal handling recommendations are assembled in a predetermined format to generate structured alarm information.

6. The mineral processing equipment alarm method based on semantic analysis according to claim 5, characterized in that, In step 4, structured alarm information is output and recorded to generate an alarm log; production execution data is acquired, the alarm log is correlated and verified with the production execution data, and the alarm records in the alarm log are marked with authenticity tags to generate a tagged alarm log. Specifically, this includes the following steps: The structured alarm information is pushed through multiple preset output channels, and the structured alarm information is simultaneously recorded in the alarm record table of the preset cloud database to generate alarm log records with details. According to the preset time window rules, the alarm trigger time is used as the reference point, and the preset time is extended forward and backward as the time window for association verification. All production event records within the time window of the associated verification are obtained through the pre-configured data interface to form a production event set; The alarm log records containing details are matched with production events in the production event set according to timestamps, and production anomaly events associated with alarm content within a predefined time window are retrieved to obtain matching results; Based on a pre-defined set of verification rules, logical judgments are made on the matching results to generate verification conclusions; For alarm records in the verification conclusions that require manual review, push them to the manual review interface and record the manual confirmation results of the operators; Based on the manual confirmation results and verification conclusions of the operators, and in accordance with the preset labeling rules, each alarm log record is assigned a authenticity label to form a labeled alarm log; among which, the authenticity labels include true positive labels, false positive labels, and missed alarm labels.

7. The mineral processing equipment alarm method based on semantic analysis according to claim 6, characterized in that, In step 5, based on the operation feedback and authenticity tags in the tagged alarm logs, the feature weights in the semantic analysis alarm rule generation algorithm are adjusted to obtain the adjusted feature weights. This specifically includes the following steps: According to the preset statistical period, the tagged alarm logs are read from the cloud database. For each feature in the tagged alarm logs, the number of true positive labels, false positive labels and missed alarm labels that triggered the alarm are counted to obtain the statistical results. Based on the statistical results, the performance metrics of accuracy, false positive rate and false negative rate for each feature are calculated, and the set of features with performance metrics below the preset threshold is identified. For each feature in the set of features to be optimized whose performance index is lower than the preset threshold, the direction of weight adjustment is determined according to the error type caused by the feature and the preset adjustment strategy; and the weight adjustment amount of each feature to be optimized is calculated according to the preset learning rate calculation rules, combined with the frequency and severity of feature errors. The weight adjustment amount of each feature to be optimized is applied to the current weight value of the corresponding feature in the semantic analysis alarm rule generation algorithm to obtain a new feature weight value; The new feature weight values ​​are updated in the feature-semantic association library to generate the adjusted feature weights.

8. The mineral processing equipment alarm method based on semantic analysis according to claim 7, characterized in that, Based on the preset learning rate calculation rules, and considering the frequency and severity of feature errors, the weight adjustment for each feature to be optimized is calculated. The corresponding relationship in this process is as follows: ； in, Indicates the weight for the next round. Indicates the weight of the current round. Indicates the learning rate. Indicates a reward for effective alarm response. This indicates the accuracy rate of historical alerts.

9. A mineral processing equipment alarm system based on semantic analysis, characterized in that, The system applies a semantic analysis-based alarm method for mineral processing equipment as described in any one of claims 1 to 8, the system comprising: The data acquisition module is used for: Real-time acquisition of operating parameters, process parameters, and environmental parameters of mineral processing equipment; and preprocessing and standardization of the acquired data to obtain standardized data. The semantic analysis module is used for: Semantic features are extracted from standardized data, and semantic relationships between parameters are constructed based on a pre-built mineral processing knowledge graph to form a feature-semantic association library; Based on the feature-semantic association library, a semantic analysis alarm rule generation algorithm is used to dynamically analyze and calculate standardized data, generate dynamic alarm rules, and output real-time semantic scores. The alarm optimization module is used for: According to the dynamic alarm rules, when the real-time semantic score meets the alarm conditions, an alarm is triggered, and the root cause of the anomaly is inferred based on the feature-semantic association library to obtain the anomaly diagnosis result; based on the anomaly diagnosis result, optimized alarm content containing processing suggestions is generated to obtain structured alarm information. The alarm output module is used for: The system outputs and records structured alarm information to generate alarm logs; it acquires production execution data, associates and verifies the alarm logs with the production execution data, and marks the alarm records in the alarm logs with authenticity tags to generate tagged alarm logs. The communication and storage module is used for: Based on the operation feedback and authenticity labels in the tagged alarm logs, the feature weights in the semantic analysis alarm rule generation algorithm are adjusted to obtain the adjusted feature weights. The adjusted feature weights are then used as the initial weights for the next round of alarm rule generation algorithm, enabling the alarm system to continuously learn and perform accurate anomaly detection and alarms in the future.

Images