La ricerca sulle regole decisionali per l’estrazione mineraria del processo di produzione delle parti
La tecnologia di estrazione delle regole decisionali di processo mira a estrarre la relazione tra i parametri delle caratteristiche delle parti, i metodi di elaborazione e le risorse di produzione dai dati di processo storici e memorizzarli nel database corrispondente sotto forma di regole decisionali. Nel processo di progettazione del processo, in base ai parametri delle caratteristiche della parte, abbinare i metodi di elaborazione e le risorse di produzione corrispondenti e inviarli all’artigiano per riferimento.
In the field of data mining, commonly used classification methods include support vector machines, neural networks, Bayesian classification, etc. The above algorithms are mainly oriented to irregular data distribution, relying on the support of big data, and mining their potential association relationships through similar measures. It is widely used in fields such as fault diagnosis. However, in the machinery manufacturing industry, the design of feature parameters of parts (such as size, accuracy, etc.) has become standardized, and in actual engineering, each part in the database corresponds to only one process route. Therefore, the repetition rate of the process data is relatively high, and the amount of data is small, which is not suitable for the above-mentioned algorithm processing. Therefore, researchers mostly use rough set theory to guide the mining of process decision rules.
Before mining decision rules, we must first ensure the credibility of the data. This is because in actual engineering, the working conditions are always changing in real time. In order to avoid a small amount of atypical data generated by special working conditions from affecting decision-making, data needs to be pre-predicted. handle. Therefore, the literature generally uses the method of calculating support and confidence to obtain typical process data.
Based on the extended rough set model, the process preference knowledge is mined by the compound relationship of equivalence, similarity, and preference, which verifies that the process preference knowledge can directly guide the designer’s decision-making, and the rough set theory does not require the process rule feasibility evaluation link, which is better than others. The mining method is simpler and more direct.
The rough set theory mining results include the deterministic rules obtained from the lower approximation set and the negative zone, as well as the uncertain rules of the boundary zone. In order to more fully mine the process rules of the boundary zone, Zhang Z. et al. used a variable precision rough set model to pass the accuracy Following the changes in the mining process, the range of the upper approximation set is effectively reduced. The qualitative knowledge is mapped to the association relationship to form a knowledge fusion model, which can effectively mine more decision rules.
The core process of rough set reasoning is to obtain the minimum attribute reduction. Chen Hao et al. analyzed the reduction anomalies caused by the inclusion interval and the positive region. For the variable precision rough set model with constant classification rate and constant positive domain, the content-based Difference matrix and attribute core to obtain the minimum attribute reduction method. Using heuristic reduction algorithm, first obtain the core attribute, and calculate the attribute dependency. According to the ascending order of the dependency, the attribute and the kernel attribute are combined in turn, and finally get Minimal attribute reduction, consider
The inhomogeneity of sample distribution is improved on the basis of neighborhood rough set, and the K-nearest neighbor rough set model is proposed, which effectively removes a large number of attributes. Decision rule mining is mainly divided into two types, one is inductive mining and the other is deduction. Mining method. The main idea of inductive mining is to summarize meaningful decision-making rules in complex data sets. When the target is obtained, match the conditional attributes of the rule set according to the target’s attribute parameters, so as to extract the decision-making rules that meet the matching requirements. The main idea of deductive mining is to split the decision content into a combination of several decision subsets, and use the data set to mine the scope of application of the decision subsets. When the target is obtained, according to the target
The target attribute parameter extracts the appropriate decision-making subset, and reorganizes it into the required decision-making content. In contrast, the decision rules of inferential mining are more diverse and have a wider scope of application, and inductive mining has stricter constraints, which can ensure the reliability of the rules.
In the above-mentioned documents, most of the processing methods are inductive mining. Although the reliability of the decision rules is effectively guaranteed, the strong constraint also leads to the low utilization of data and limits the completeness of the decision rule base. Moreover, although the variable precision rough set can effectively reduce the boundary area, the precision value is mainly set by manual experience, and too many human factors will reduce the reliability of the decision rule. Therefore, how to reduce the boundary area and improve the flexibility of the rules on the basis of ensuring the reliability of the decision-making rules is the main research direction of mining process decision-making rules.