Research on optimization of crane fault predictive control system based on data mining

2.1.1 Introduction to association rules

The dataset to be mined by association rules is denoted as D, D = { t 1 , t 2 , ⋯ , t k ⋯ t n } , where t k = { i 1 , i 2 , ⋯ , i j ⋯ i p } ( 1 , 2 , ⋯ n ) is a transaction. The element i j ( j = 1 , 2 , ⋯ p ) in t k is called Item. I = { i 1 , i 2 , ⋯ , i j , ⋯ i m } is the set of all the terms in D, any subset of I is called the itemset in D, if |X| = k, then the set X is said to be a set of k items. The number of transactions in D that contains itemset X is called the number of items supported by itemset X, denoted as σ . The support degree of entry set X is as follows: sup port ( X ) = σ ∣ D ∣ × 100 % , where ∣ D ∣ is the number of transactions in the dataset D. If sup port ( X ) is not less than the pre-specified minimum support, then X is called a frequent itemset.

If X and Y are itemsets and X ∩ Y = , the implication X ⇒ Y is called association rule, X and Y are, respectively, the premise and conclusion of association rule X ⇒ Y . The support of itemset ( X ∪ Y ) is called the support of association rule X ⇒ Y , remember: sup port ( X ⇒ Y ) , where the confidence degree of association rule X ⇒ Y of sup port ( X ⇒ Y ) = sup port ( X ∪ Y ) is written as follows: confidence ( X ⇒ Y ) , confidence ( X ⇒ Y ) = sup port ( X ∪ Y ) sup port ( X ) × 100 % .

In general, user mining needs to set the minimum confidence as Minsup, support and confidence are important parameters in association rules, the former is used to measure the statistical importance of association rules in D, the latter is used to measure the credibility of association rules, and useful association rules are generally high support and confidence. Mining association rules is divided into two sub-problems. STEP1: find out all frequent itemsets whose support support(X) is not less than minsup given by the user in the transaction database. STEP2: generate association rules according to the minimum itemset and minimum confidence. In this system, the algorithm FP_growth (FP2tree, NULL) is used to mine the FP2tree itemset.

2.1.2 An introduction to decision tree classification

Decision tree classification is one of the core algorithms of data mining, it obtains rules through a large number of training sets and finds out the potential and valuable information for decision making, it is used in some classification models. The core idea of decision tree classification is to select the attributes of decision nodes by using the concepts in information theory and using the information gain as the measure of the ability to classify and discriminate decision attributes. The more influential algorithm in decision tree algorithm is ID3, and its improved algorithm can deal with C4.5 of continuous attributes. Its main purpose is to extract classification rules and make classification prediction. A supervised machine learning algorithm is a decision tree. Both classification and regression methods make use of it. The decision tree resembles a node-based tree. A variety of things influence the branching. It divides data into branches like these until a threshold value is reached. The root nodes, children nodes, and leaf nodes make up a decision tree. A decision tree is a basic diagram for categorizing examples. It is a supervised machine learning in which data are constantly separated according to a parameter. Recursive partitioning is a strategy used to create decision trees. This method is also known as divide and conquer because it divides the data into subsets, which are then divided again into even smaller subsets, and so on, until the algorithm concludes that the data inside the subgroups are sufficiently homogeneous, or another stopping requirement is reached. The following are some of the benefits of using decision trees for classification: constructed at a low cost, extremely quick when it comes to categorizing unfamiliar records. For little trees, it is simple to understand. For many basic datasets, accuracy equivalent to other classification algorithms and unimportant characteristics are left out.

Let S be the set composed of S data samples. Suppose that the category identifier attribute has M different values, and then m different classes are defined as C i ( i = 1 , 2 , ⋯ , m ) . Let s i be the number of samples in class C i , and the expected information required for the classification of a given sample is as follows: I ( s 1 , s 2 , ⋯ , s m ) = − ∑ i = 1 m p i log ( p i ) , where p i is the probability that the sample belongs to C i , and it is estimated by s i / s . Let property A i have k different values { a 1 , a 2 , ⋯ , a k } , I can divide A i into k subsets of { S 1 , S 2 , ⋯ S k } , where S j contains such a sample in S, they have the value a j on A i , 1 ≤ j ≤ k if A i is selected as the test property, then these subsets correspond to the branches that grow from the nodes that contain the set S. Let s i be the number of samples of class C i in subset S j , the entropy divided into subsets by A is given as follows: E ( A i ) = ∑ i = 1 k s 1 + s 2 + ⋯ + s m s I ( s 1 , ⋯ , s m ) , where s 1 + s 2 + ⋯ + s m s acts as the weight of the jth subset, and it is equal to the number of samples in the subset divided by the number of samples in S. The smaller the entropy, the higher the purity of the subset. The encoding information that will be obtained by branching on A i is Gain ( A i ) = I ( s 1 , s 2 , ⋯ , s m ) − E ( A i ) , in other words, Gain ( A i ) is the expected compression of entropy due to the value of the property A i . Once you have a more branched tree, you need to selectively prune to get a lean sub tree.

2.3.1 Database design

A good database is the basis for designing the whole system. Because association rules and decision trees use different preprocessed data and obtained results, tables are designed for them, respectively, when designing database tables. The system created a database named QizhongJixie in MYSQL, and designed six tables: qizhongjiXie_admin, xcjy_data, ycl_data, ycljc_data, gz_data, and gzjcs_data, it stores user information, inspection data, association rule preprocessing data, decision tree preprocessing data, rules obtained by association rules, and data obtained by decision tree.

2.3.2 Crane detection data preprocessing

A common feature of raw data in a database is a large number of incomplete, noisy, and inconsistent data. Data preprocessing generally includes noise elimination, deduction, calculation of missing data, elimination of repeated records, etc. The purpose of data preprocessing is to provide clean, accurate, and concise data for data mining process. In the developed system, improving the efficiency and accuracy of data mining is a very important link in data mining. This process mainly follows the following reference principles: Give attribute names explicit meaning whenever possible, unified attribute value encoding for multiple data sources, handle vacant values, etc. For example, in the original data, there are two inspection items, namely, limit failure of large vehicles and brake failure of large vehicles, which are considered to be the same problem in processing, some inspection results that have no impact on the safety of lifting machinery are ignored, and attributes that have little impact on the results are not considered. In preprocessing, the system calls the corresponding algorithm to automatically preprocess the original data according to the parameters submitted on the preprocessing page, and saves the results in the corresponding database table. According to the above principles, ten key attributes are selected: technical documentation, operating environment and appearance, cab inspection, metal structure, track, main components, electrical and control system, hydraulic system, safety protection device, and operation test. For the correlation analysis of the ten attributes, a test is regarded as an event Tid. When there is defect or failure in the test item, it is denoted as 1, otherwise, it is 0. When classifying them, the above ten attributes are used as decision attributes, each of which has a different fault value, the fault category attribute is used as decision attribute, and its value is the most common rectification and recodification within a time limit, which is denoted as two values P1 and P2 in the preprocessing [22].

2.3.3 Realization of two algorithms in the system

The implementation process of the algorithm FP_growth (FP2Tree, nuII) in the system can be simply described as follows: For the given Boolean dataset, minimum support, and minimum confidence, the given algorithm is used to find the frequent itemset, and then the association rules satisfying the minimum confidence are listed [23]. To implement this process, the following classes and interfaces are designed: Dataset (Boolean Dataset), DB Reader (tool for traversing Dataset), Large itemsets Finder (interface to data mining scheme), FP2 Growth (FP2Tree implementation of data mining scheme), and Associations Finder (Association rule mining) [24]. The minimum support degree and minimum confidence degree are randomly set by the user and passed into the class or implementation interface, each class or implementation interface will realize certain functions, and the algorithm function reflects the result of the cooperation between these classes and the implementation interface [25]. The implementation of algorithm C4.5 is basically the same as that of algorithm FP_growth. The classes and interfaces are just different [26].