CN105787610A - Case-based reasoning method capable of supporting time sequence matching - Google Patents
Case-based reasoning method capable of supporting time sequence matching Download PDFInfo
- Publication number
- CN105787610A CN105787610A CN201410797922.XA CN201410797922A CN105787610A CN 105787610 A CN105787610 A CN 105787610A CN 201410797922 A CN201410797922 A CN 201410797922A CN 105787610 A CN105787610 A CN 105787610A
- Authority
- CN
- China
- Prior art keywords
- case
- dist
- target event
- distance
- current target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012937 correction Methods 0.000 claims abstract description 15
- 230000000694 effects Effects 0.000 claims description 20
- 238000011156 evaluation Methods 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 abstract description 7
- 230000007246 mechanism Effects 0.000 abstract description 3
- 238000009411 base construction Methods 0.000 abstract 1
- 238000004519 manufacturing process Methods 0.000 description 13
- 238000012544 monitoring process Methods 0.000 description 12
- 230000008569 process Effects 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000003745 diagnosis Methods 0.000 description 2
- 238000011112 process operation Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 241000282414 Homo sapiens Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000005452 bending Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 238000004886 process control Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
技术领域technical field
本发明涉及一种案例推理方法，特别是一种支持时间序列匹配的案例推理方法。该方法属于人工智能领域。The invention relates to a case reasoning method, in particular to a case reasoning method supporting time series matching. This method belongs to the field of artificial intelligence.
背景技术Background technique
案例推理(case-basedreasoning，CBR)是人工智能的一个新兴领域。基于案例推理的问题求解方法，非常适用于没有很强理论模型和领域知识不完全、难以定义或定义不一致而经验丰富的决策环境中，目前在医疗诊断、法律咨询、工程规划和故障诊断等领域均得到了广泛的应用。Case-based reasoning (CBR) is an emerging field of artificial intelligence. The problem-solving method based on case reasoning is very suitable for decision-making environments where there is no strong theoretical model and domain knowledge is incomplete, it is difficult to define or the definition is inconsistent, and it is currently in the fields of medical diagnosis, legal consultation, engineering planning, and fault diagnosis. have been widely applied.
目前，案例推理的核心环节是案例检索，即在案例库中找到与问题描述最相近的案例，它主要依赖于案例特征项的匹配，但目前案例特征项大多面向单一数值，没有实现对时间序列的支持，这使得案例推理在某些偏重时间趋势匹配的应用领域相对受限。At present, the core link of case reasoning is case retrieval, that is, to find the case that is most similar to the problem description in the case library. It mainly depends on the matching of case feature items. support, which makes case reasoning relatively limited in some application fields that emphasize time trend matching.
发明内容Contents of the invention
针对上述技术不足，本发明提出一种支持时间序列匹配的案例推理方法，其目的是：将时间序列匹配机制引入到案例推理中，增加案例特征项对时间序列的支持，并基于动态时间弯曲距离方法(Dynamictimewarping，DTW)来求解时间序列的匹配距离；通过采用具有匹配距离比重因子的案例特征项权重调整算法来实现对特征项权重的学习修正，使案例推理更具工程实用性。In view of the above-mentioned technical deficiencies, the present invention proposes a case reasoning method supporting time series matching, the purpose of which is to introduce the time series matching mechanism into case reasoning, increase the support of case feature items for time series, and based on dynamic time warping distance The method (Dynamic timewarping, DTW) is used to solve the matching distance of the time series; the weight adjustment algorithm of the case feature item with the proportion factor of the matching distance is used to realize the learning correction of the feature item weight, so that the case reasoning is more practical in engineering.
本发明解决其技术问题所采用的技术方案是：一种支持时间序列匹配的案例推理方法，包括以下步骤：The technical solution adopted by the present invention to solve the technical problem is: a case reasoning method supporting time series matching, comprising the following steps:
1)构造案例库；1) Construct a case library;
2)案例检索：以当前目标事件的状态信息为输入，基于案例库进行案例检索，得到当前目标事件与案例库中各案例的综合距离；2) Case retrieval: take the state information of the current target event as input, conduct case retrieval based on the case database, and obtain the comprehensive distance between the current target event and each case in the case database;
3)案例重用：如果当前目标事件与案例库中某个案例的综合距离小于设定距离阈值，则当前目标事件与该案例匹配成功，将该案例的案例结果输出用于目标事件，执行下一步骤；否则，匹配失败并结束；3) Case reuse: If the comprehensive distance between the current target event and a case in the case library is less than the set distance threshold, the current target event matches the case successfully, and the case result of this case is output for the target event, and the next step is executed. step; otherwise, the match fails and ends;
4)案例修正：判断匹配成功的案例结果与当前目标事件是否一致；如果一致，则该案例正确，案例推理结束；否则该案例错误，将匹配成功的案例结果根据目标事件的反馈信息进行修正，并将修正后的案例结果提交到案例库，案例推理结束。4) Case correction: judge whether the matching successful case result is consistent with the current target event; if they are consistent, the case is correct and the case reasoning ends; otherwise, the case is wrong, and the matching successful case result is corrected according to the feedback information of the target event, And the revised case results are submitted to the case library, and the case reasoning ends.
所述案例库为多个案例的集合，所述案例包括：案例名称、案例特征项集合、案例结果、案例效果评价。The case library is a collection of multiple cases, and the cases include: case names, case characteristic item sets, case results, and case effect evaluations.
所述案例检索公式如下：The case retrieval formula is as follows:
上式中，Simk表示案例库中第k个案例Xk与当前目标事件的特征项序列X0的综合距离，ωkj为第k案例的第j个属性在参与案例匹配的属性指标中所占的权重，j＝1,2,...,m，m为第k案例的属性个数；Dist(X0(j),Xk(j))表示第k个案例和当前目标事件的特征项序列X0在第j个属性上的匹配距离。In the above formula, Sim k represents the comprehensive distance between the kth case X k in the case library and the characteristic item sequence X 0 of the current target event, ω kj is the attribute index of the jth attribute of the kth case in the attribute index of the participating case matching The weight accounted for, j=1,2,...,m, m is the number of attributes of the kth case; Dist(X 0 (j),X k (j)) represents the kth case and the current target event The matching distance of the feature item sequence X 0 on the jth attribute.
所述目标事件的特征项在第j个属性上为时间点值，则Dist(X0(j),Xk(j))为曼哈顿距离|X0(j),Xk(j)|；X0(j)为当前目标事件的特征项序列在第j个属性上的时间点值，Xk(j)为第k个案例在第j个属性上的时间点值。The feature item of the target event is a time point value on the j attribute, then Dist(X 0 (j), X k (j)) is the Manhattan distance |X 0 (j), X k (j)|; X 0 (j) is the time point value of the feature item sequence of the current target event on the jth attribute, and X k (j) is the time point value of the kth case on the jth attribute.
所述目标事件的特征项在第j个属性上为时间序列，Dist(X0(j),Xk(j))为动态时间弯曲距离算法中的归整路径距离D(|Xk(j)|,|X0(j)|)；X0(j)为当前目标事件的特征项序列在第j个属性上的时间序列，Xk(j)为第k个案例在第j个属性上的时间序列。The feature item of the target event is a time series on the j attribute, and Dist(X 0 (j), X k (j)) is the rounded path distance D(|X k (j) in the dynamic time warping distance algorithm )|,|X 0 (j)|); X 0 (j) is the time series of the feature item sequence of the current target event on the jth attribute, and X k (j) is the time series of the kth case on the jth attribute time series on .
一种支持时间序列匹配的案例推理方法，还包括对案例特征项的权重进行调整，通过权重调整幅度公式实现：A case reasoning method supporting time series matching, which also includes adjusting the weights of case feature items, which is realized through the weight adjustment range formula:
W'kj＝Wkj(1±Δ×Rkj)(2)W' kj ＝W kj (1±Δ×R kj )(2)
上式中Wkj表示案例k中特征项j在修正前的权重，W'kj表示案例k中特征项j在修正后的权重，Δ表示每次调整的幅度系数，Rkj表示案例k中特征项j的匹配距离比重因子，计算公式如下：In the above formula, W kj represents the weight of feature item j in case k before correction, W' kj represents the weight of feature item j in case k after correction, Δ represents the amplitude coefficient of each adjustment, and R kj represents the feature in case k The matching distance proportion factor of item j, the calculation formula is as follows:
上式中，Simk表示案例库中第k个案例Xk与当前目标事件的特征项序列X0的综合距离，TDist表示特征项匹配距离阈值，Dist(X0(j),Xk(j))表示第k个案例和当前目标事件的特征项序列X0在第j个属性上的匹配距离；ωkj为第k案例的第j个属性在参与案例匹配的属性指标中所占的权重。In the above formula, Sim k represents the comprehensive distance between the kth case X k in the case library and the feature item sequence X 0 of the current target event, T Dist represents the feature item matching distance threshold, Dist(X 0 (j),X k ( j )) represents the matching distance between the kth case and the characteristic item sequence X 0 of the current target event on the jth attribute; Weights.
当案例k为正确且Dist(X0(j),Xk(j)/Simk＞＝TDist时：When case k is correct and Dist(X 0 (j),X k (j)/Sim k ＞=T Dist :
W'kj＝Wkj(1-Δ×ωkjDist(X0(j),Xk(j)/Simk)。W' kj =W kj (1-Δ×ω kj Dist(X 0 (j),X k (j)/Sim k ).
当案例k为正确且Dist(X0(j),Xk(j)/Simk＜TDist时：When case k is correct and Dist(X 0 (j),X k (j)/Sim k <T Dist :
W'kj＝Wkj(1+Δ×(1-ωkjDist(X0(j),Xk(j)/Simk))。W' kj =W kj (1+Δ×(1−ω kj Dist(X 0 (j),X k (j)/Sim k )).
当案例k判断为错误，并且Dist(X0(j),Xk(j)/Simk＞＝TDist时：When case k is judged to be wrong, and Dist(X 0 (j),X k (j)/Sim k ＞=T Dist :
W'kj＝Wkj(1+Δ×ωkjDist(X0(j),Xk(j)/Simk)。W' kj =W kj (1+Δ×ω kj Dist(X 0 (j),X k (j)/Sim k ).
当案例k判断为错误，并且Dist(X0(j),Xk(j)/Simk＜TDist时：When case k is judged to be wrong, and Dist(X 0 (j),X k (j)/Sim k <T Dist :
W'kj＝Wkj(1-Δ×(1-ωkjDist(X0(j),Xk(j)/Simk))。W' kj =W kj (1-Δ×(1-ω kj Dist(X 0 (j),X k (j)/Sim k )).
本发明具有以下有益效果及优点：The present invention has the following beneficial effects and advantages:
1.本发明将时间序列匹配机制引入到案例特征项匹配中，使案例推理具有时间趋势分析能力，可以更好地应用于偏重时间趋势分析的工程领域。1. The present invention introduces the time series matching mechanism into the matching of case feature items, so that case reasoning has the ability of time trend analysis, and can be better applied to engineering fields that emphasize time trend analysis.
2.在案例的表示结构中，加入了用于记录案例重用效果的案例效果评价，可以便于在日常使用中不断地添加评价效果为优的案例，删除评价效果为劣的案例，来完善案例库，使其更具工程实用性。2. In the representation structure of the case, a case effect evaluation for recording the effect of case reuse is added, which can facilitate the continuous addition of cases with excellent evaluation effects and delete cases with poor evaluation effects in daily use to improve the case library , making it more practical for engineering.
3.在案例检索中所采用的支持曼哈顿距离和动态时间弯曲距离的最近相邻法，可以从时间点值和时间序列这两个方面来求解案例特征项匹配距离，最后得出的综合距离更能反映案例与当前目标事件的相似度，并提高了案例检索的准确性。3. The nearest neighbor method that supports Manhattan distance and dynamic time warping distance used in case retrieval can solve the matching distance of case feature items from the two aspects of time point value and time series, and the final comprehensive distance is more accurate. It can reflect the similarity between the case and the current target event, and improve the accuracy of case retrieval.
4.采用的特征项权重调整算法可以基于案例重用情况，来修正案例特征项权重，进而使日后的案例特征项匹配更加符合实际情况，具有重要的现实意义和较高的工程应用价值。4. The feature item weight adjustment algorithm adopted can correct the case feature item weight based on the case reuse situation, so that the future case feature item matching is more in line with the actual situation, which has important practical significance and high engineering application value.
附图说明Description of drawings
图1是本发明的一种支持时间序列匹配的案例推理方法的流程图。FIG. 1 is a flowchart of a case reasoning method supporting time series matching in the present invention.
图2是案例特征项权重调整的流程图。Fig. 2 is a flowchart of weight adjustment of case feature items.
具体实施方式detailed description
下面结合实施例对本发明做进一步的详细说明。The present invention will be further described in detail below in conjunction with the examples.
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合实施例，对本发明进行进一步的详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。In order to make the object, technical solution and advantages of the present invention more clear, the present invention will be further described in detail below in conjunction with the examples. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.
本发明可应用于制造业领域，首先采用编程语言来对本发明提出的支持时间序列匹配的案例推理方法进行代码实现，并封装成软件，然后以工业现场的PC机、服务器或者工控机为载体进行软件安装，最后现场人员可以根据软件显示的案例推理结果来进行工艺调控、故障分析等功能。The present invention can be applied to the field of manufacturing industry. First, the programming language is used to implement the code of the case reasoning method supporting time series matching proposed by the present invention, and it is packaged into software, and then the PC, server or industrial computer on the industrial site is used as the carrier to carry out Software installation, and finally on-site personnel can perform functions such as process control and fault analysis according to the case reasoning results displayed by the software.
具体设计步骤如下：The specific design steps are as follows:
步骤一：构造案例库，具体方法为：Step 1: Construct the case library, the specific method is:
本发明的案例库为制造业领域搜集到的多个案例的集合，案例的表示结构具体包括：案例名称、案例的特征项集合、案例结果、案例效果评价；其中，案例特征项集合、案例效果评价和案例结果为案例的核心组成部分。The case library of the present invention is a collection of multiple cases collected in the field of manufacturing. The representation structure of the case specifically includes: case name, feature item set of the case, case result, and case effect evaluation; wherein, the set of case feature items, case effect Evaluation and case results are the core components of the case.
(1)案例特征项集合中的特征项主要描述制造业现场中，例如电流、电压等监测参数的数值特征，它既可以是时间点值，也可以是时间序列。时间点值描述了监测参数的跳变特征，主要通过捕获跳变发生时对应监测参数的监测数值，来获取特征项的时间点值；时间序列描述了监测参数的趋势变化特征，为了节省案例的存储空间，这里只存储对应时间趋势的起始时间和截止时间，当进行时间序列类型的案例特征项匹配时，可以直接根据起始时间和截止时间来从数据库中提取完整的时间序列。(1) The feature items in the case feature item set mainly describe the numerical characteristics of monitoring parameters such as current and voltage in the manufacturing site, which can be either time point values or time series. The time point value describes the jump characteristics of the monitoring parameters. The time point value of the feature item is mainly obtained by capturing the monitoring value of the corresponding monitoring parameter when the jump occurs; the time series describes the trend change characteristics of the monitoring parameters. In order to save the case Storage space, where only the start time and end time of the corresponding time trend are stored. When matching the case feature items of the time series type, the complete time series can be extracted from the database directly according to the start time and end time.
(2)案例效果评价是基于案例输出到制造业现场目标事件后的现场效果进行追加录入的，制造业现场目标事件主要指针对现场设备或者运行工艺进行的推理分析，例如分析生产线的运行稳定性、分析水泵流量是否正常等，案例效果评价分为优、良、劣三个等级。如果案例结果与当前目标事件一致，则将对应案例的效果评价更新为优；否则将效果评价更新为劣。对于案例库而言，就是要不断地添加评价效果为优的案例，删除评价效果为劣的案例，进而形成较为完备的案例库。(2) The case effect evaluation is based on the on-site effect after the case is exported to the manufacturing site target event for additional entry. The manufacturing site target event mainly refers to the inference analysis of the site equipment or operation process, such as the analysis of the operation stability of the production line , Analyze whether the pump flow rate is normal, etc., the case effect evaluation is divided into three grades: excellent, good, and bad. If the case result is consistent with the current target event, update the effect evaluation of the corresponding case to excellent; otherwise, update the effect evaluation to poor. For the case base, it is necessary to continuously add cases with excellent evaluation effects and delete cases with poor evaluation effects, so as to form a relatively complete case base.
(3)案例结果是指在对应的监测参数下，制造业现场发生的工艺运行结果，例如电机转速异常、生产线故障等。对于如下的一个案例：在转速很低，机壳温度很高的状态下，电机已经烧坏。则在该案例中，“电机已经烧坏”即为案例结果。(3) The case results refer to the process operation results that occur on the manufacturing site under the corresponding monitoring parameters, such as abnormal motor speed, production line failure, etc. For the following case: the motor has burned out when the speed is very low and the casing temperature is high. Then in this case, "the motor has burned out" is the case result.
步骤二：案例检索，具体方法为：Step 2: Case retrieval, the specific method is:
以制造业现场当前目标事件的监测参数信息为输入，基于案例库进行案例检索，得到当前目标事件与案例库中各案例的综合距离。Taking the monitoring parameter information of the current target event on the manufacturing site as input, the case retrieval is carried out based on the case database, and the comprehensive distance between the current target event and each case in the case database is obtained.
本发明设计的案例检索方法将时间趋势匹配融入到了最近相邻法中，支持属性间的曼哈顿距离和动态时间弯曲距离，案例检索算法公式如下：The case retrieval method designed by the present invention integrates time trend matching into the nearest neighbor method, and supports Manhattan distance and dynamic time bending distance between attributes. The case retrieval algorithm formula is as follows:
上式中，Simk表示案例库中第k个案例Xk与当前目标事件的特征项序列X0的综合距离，案例Xk和特征项序列X0均由一系列现场监测参数值组成，ωkj为第k案例的第j个属性在参与案例匹配的属性指标中所占的权重，j＝1,2,...,m，m为第k案例的属性个数；Dist(X0(j),Xk(j))表示第k个案例和当前目标事件的特征项序列X0在第j个属性上的匹配距离，其中，第j个属性也就是第j个现场监测参数。In the above formula, Sim k represents the comprehensive distance between the kth case X k in the case library and the characteristic item sequence X 0 of the current target event, and both the case X k and the characteristic item sequence X 0 are composed of a series of on-site monitoring parameter values, ω kj is the weight of the jth attribute of the kth case in the attribute index of the participating case matching, j=1,2,...,m, m is the number of attributes of the kth case; Dist(X 0 ( j), X k (j)) represents the matching distance between the kth case and the characteristic item sequence X 0 of the current target event on the jth attribute, where the jth attribute is also the jth on-site monitoring parameter.
特征项序列X0的第j个现场监测参数如果为时间点值的话，Dist(X0(j),Xk(j))即为曼哈顿距离|X0(j),Xk(j)|；Xk(j)为第k个案例在第j个属性上的时间点值，X0(j)为当前目标事件的特征项序列在第j个属性上的时间点值。If the jth field monitoring parameter of the feature item sequence X 0 is a time point value, Dist(X 0 (j),X k (j)) is the Manhattan distance |X 0 (j),X k (j)| ; X k (j) is the time-point value of the k-th case on the j-th attribute, and X 0 (j) is the time-point value of the characteristic item sequence of the current target event on the j-th attribute.
特征项序列X0的第j个现场监测参数如果为时间序列的话，Dist(X0(j),Xk(j))即为动态时间弯曲距离算法中的归整路径距离D(|Xk(j)|,|X0(j)|)，其中X0(j)为当前目标事件的特征项在第j个属性上的时间序列，|X0(j)|为其长度，Xk(j)为第k个案例在第j个属性上的时间序列，|Xk(j)|为其长度。If the jth on-site monitoring parameter of the feature item sequence X 0 is a time series, Dist(X 0 (j),X k (j)) is the normalized path distance D(|X k (j)|,|X 0 (j)|), where X 0 (j) is the time series of the feature item of the current target event on the j attribute, |X 0 (j)| is its length, X k (j) is the time series of the kth case on the jth attribute, and |X k (j)| is its length.
曼哈顿距离和归整路径距离属于两种不同的距离计算方法，一般情况下，归整路径距离可能会大于曼哈顿距离，但它们同时被包含在了综合距离Simk中。针对上述情况，可以通过对案例特征项权重的调整，来使这两种距离进行累加时能更好地将各个特征项的匹配情况反映在综合距离中。The Manhattan distance and the normalized path distance belong to two different distance calculation methods. Generally, the normalized path distance may be greater than the Manhattan distance, but they are included in the comprehensive distance Sim k at the same time. In view of the above situation, it is possible to adjust the weights of the case feature items so that when the two distances are accumulated, the matching of each feature item can be better reflected in the comprehensive distance.
步骤三：案例重用，具体方法为：Step 3: Case reuse, the specific method is:
案例重用包括判断是否匹配成功和案例结果输出两个步骤。如果制造业现场当前目标事件与案例库中某个案例的综合距离小于设定距离阈值，则当前目标事件与该案例匹配成功，两者具有相近的特征，并且可能发生相同的结果，将该案例的案例结果输出用于目标事件，认为当前制造业现场发生了案例结果相一致的运行情况，然后执行下一步骤；否则，匹配失败并结束；Case reuse includes two steps: judging whether the matching is successful and outputting case results. If the comprehensive distance between the current target event on the manufacturing site and a case in the case library is less than the set distance threshold, the current target event is successfully matched with the case, the two have similar characteristics, and the same result may occur. The output of the case results is used for the target event, and it is considered that the current manufacturing site has an operation situation consistent with the case results, and then executes the next step; otherwise, the matching fails and ends;
步骤四：案例修正，具体方法为：Step 4: Case correction, the specific method is:
案例修正包括案例结果与制造业现场当前目标事件的一致性判断和案例结果修正两个步骤。案例结果与当前目标事件的一致性判断是指将案例结果输出到目标事件后，判断目标事件是否确实发生了对应的案例结果，如果目标事件确实发生了对应的案例结果，则案例结果与当前目标事件一致；否则案例结果与当前目标事件不一致。例如将“电机损坏”这个案例结果应用于“电机状态判断”这个目标事件中，即认为当前电机已经损坏并采取对应的解决措施，如果事后查明电机确实已经损坏，则案例结果与当前目标事件一致；否则案例结果与当前目标事件不一致。The case revision includes two steps: the consistency judgment between the case result and the current target event at the manufacturing site and the case result revision. The consistency judgment between the case result and the current target event means that after the case result is output to the target event, it is judged whether the corresponding case result has indeed occurred in the target event. event is consistent; otherwise the case result is inconsistent with the current target event. For example, apply the case result of "motor damage" to the target event of "motor status judgment", that is, it is considered that the current motor is damaged and corresponding solutions are taken. Consistent; otherwise the case result is inconsistent with the current target event.
如果案例结果与当前目标事件一致，则该案例正确，案例推理结束；否则该案例错误，将匹配成功的案例结果根据目标事件的反馈信息进行修正，并将修正后的案例结果提交到案例库，案例推理结束。If the case result is consistent with the current target event, the case is correct, and the case reasoning ends; otherwise, the case is wrong, and the matching successful case result is corrected according to the feedback information of the target event, and the corrected case result is submitted to the case library, Case reasoning ends.
步骤五：案例特征项权重调整，具体方法为：Step 5: Adjust the weight of case feature items, the specific method is:
案例特征项的初始权重，即案例对应的各个监测参数的初始权重，是由制造业领域专家直接给出的，具有一定的主观经验性，所以需要在日常工艺运行中不断地对特征项权重进行调整，使其更加具有工程实用性。The initial weight of the characteristic item of the case, that is, the initial weight of each monitoring parameter corresponding to the case, is directly given by experts in the manufacturing field and has a certain degree of subjective experience. Therefore, it is necessary to continuously adjust the weight of the characteristic item in daily process operation. Adjust to make it more practical for engineering.
本发明设计的案例特征项权重的调整流程如图2所示，本发明在权重调整幅度公式中加入了匹配距离比重因子R，比重因子R会根据特征项匹配距离占综合距离的比重的大小来采用不同的表达式，主要用于保证如下两点：The adjustment process of the case feature item weight designed by the present invention is shown in Figure 2. The present invention adds a matching distance proportion factor R to the weight adjustment range formula, and the proportion factor R will be determined according to the proportion of the feature item matching distance in the comprehensive distance. Using different expressions is mainly used to ensure the following two points:
(1)当特征项匹配距离占综合距离的比重大于或等于特征项阈值TDist时，特征项更倾向于降低当前目标事件和案例库中案例的相似度，则权重调整幅度与特征项匹配距离为正比关系；(1) When the proportion of the feature item matching distance to the comprehensive distance is greater than or equal to the feature item threshold T Dist , the feature item is more inclined to reduce the similarity between the current target event and the case in the case library, and the weight adjustment range is equal to the feature item matching distance is a proportional relationship;
(2)当特征项匹配距离占综合距离的比重小于特征项阈值TDist时，特征项更倾向于增加当前目标事件和案例库中案例的相似度，则权重调整幅度与特征项匹配距离为反比关系。(2) When the proportion of the feature item matching distance to the comprehensive distance is less than the feature item threshold T Dist , the feature item is more likely to increase the similarity between the current target event and the case in the case library, and the weight adjustment range is inversely proportional to the feature item matching distance relation.
匹配距离比重因子R的设定一方面可以使得权重的修正更加符合人类的认知规律，另一方面匹配距离比重因子R的加入也解决动态时间弯曲距离和曼哈顿距离的数值差异性问题。On the one hand, the setting of the proportion factor R of the matching distance can make the correction of the weight more in line with the cognitive law of human beings. On the other hand, the addition of the proportion factor R of the matching distance can also solve the problem of numerical differences between the dynamic time warping distance and the Manhattan distance.
在调整特征项权重时，将匹配距离比重因子R引入到公式中，每次的权重调整幅度公式如下：When adjusting the weight of feature items, the matching distance proportion factor R is introduced into the formula, and the formula for each weight adjustment range is as follows:
W'kj＝Wkj(1±Δ×Rkj)(3)W' kj ＝W kj (1±Δ×R kj )(3)
上式中Wkj表示案例k中特征项j在修正前的权重，W'kj表示案例k中特征项j在修正后的权重，Δ表示每次调整的幅度系数，Rkj表示案例k中特征项j的匹配距离比重因子，计算公式如下：In the above formula, W kj represents the weight of feature item j in case k before correction, W' kj represents the weight of feature item j in case k after correction, Δ represents the amplitude coefficient of each adjustment, and R kj represents the feature in case k The matching distance proportion factor of item j, the calculation formula is as follows:
式(4)中，Simk表示案例库中第k个案例与当前目标事件的综合距离，TDist表示特征项阈值，Dist(X0(j),Xk(j))表示第k个案例和当前目标序列在第j个属性上的距离，根据公式(4)中Rkj的计算公式，可以得出公式(3)一共有如下四种情况：In formula (4), Sim k represents the comprehensive distance between the kth case in the case library and the current target event, T Dist represents the feature item threshold, and Dist(X 0 (j), X k (j)) represents the kth case According to the calculation formula of R kj in formula (4), it can be concluded that formula (3) has the following four situations:
(1)当案例k为正确且Dist(X0(j),Xk(j)/Simk＞＝TDist时，降低案例k中案例项j的权重，W'kj＝Wkj(1-Δ×ωkjDist(X0(j),Xk(j)/Simk)，并且Dist(X0(j),Xk(j)越大，降低幅度越大；(1) When case k is correct and Dist(X 0 (j),X k (j)/Sim k ＞=T Dist , reduce the weight of case item j in case k, W' kj ＝W kj (1- Δ×ω kj Dist(X 0 (j),X k (j)/Sim k ), and the larger Dist(X 0 (j),X k (j), the greater the reduction;
(2)当案例k为正确且Dist(X0(j),Xk(j)/Simk＜TDist时，增加案例k中案例项j的权重，W'kj＝Wkj(1+Δ×(1-ωkjDist(X0(j),Xk(j)/Simk))，并且Dist(X0(j),Xk(j)越小，增加的幅度越大；(2) When case k is correct and Dist(X 0 (j),X k (j)/Sim k <T Dist , increase the weight of case item j in case k, W' kj ＝W kj (1+Δ ×(1-ω kj Dist(X 0 (j),X k (j)/Sim k )), and the smaller Dist(X 0 (j),X k (j), the greater the increase;
(3)当案例k判断为错误，并且Dist(X0(j),Xk(j)/Simk＞＝TDist时，增加案例k中案例项j的权重，W'kj＝Wkj(1+Δ×ωkjDist(X0(j),Xk(j)/Simk)，并且Dist(X0(j),Xk(j)越大，增加的幅度越大；(3) When case k is judged to be wrong, and Dist(X 0 (j),X k (j)/Sim k ＞=T Dist , increase the weight of case item j in case k, W' kj ＝W kj ( 1+Δ×ω kj Dist(X 0 (j),X k (j)/Sim k ), and the larger Dist(X 0 (j),X k (j), the greater the increase;
(4)当案例k判断为错误，并且Dist(X0(j),Xk(j)/Simk＜TDist时，降低案例k中案例项j的权重，W'kj＝Wkj(1-Δ×(1-ωkjDist(X0(j),Xk(j)/Simk))，并且Dist(X0(j),Xk(j)越小，降低的幅度越大。(4) When case k is judged as wrong, and Dist(X 0 (j),X k (j)/Sim k <T Dist , reduce the weight of case item j in case k, W' kj ＝W kj (1 -Δ×(1-ω kj Dist(X 0 (j),X k (j)/Sim k )), and the smaller Dist(X 0 (j),X k (j), the greater the reduction.
图1为本发明实施例提供的支持时间序列匹配的案例推理方法的流程，具体的设计方法包括如下五个步骤：Fig. 1 is the process flow of the case reasoning method supporting time series matching provided by the embodiment of the present invention, and the specific design method includes the following five steps:
1)案例构造1) Case structure
以发动机为例，建立如下三组案例，每组案例均包括案例名称、案例结果、案例效果评价和案例特征项，如表1至表3所示。Taking the engine as an example, the following three groups of cases are established, and each group of cases includes the case name, case results, case effect evaluation and case feature items, as shown in Table 1 to Table 3.
选取电压、电流和温度来作为案例特征项，其中温度特征项同时支持时间点值类型和时间序列类型，电压和电流只支持时间点值类型。Select voltage, current, and temperature as case feature items. The temperature feature item supports both time point value types and time series types, and voltage and current only support time point value types.
案例效果评价是基于事后的案例重用效果进行追加录入的，所以构造案例时先将其留空。The case effect evaluation is based on the subsequent case reuse effect for additional entry, so leave it blank when constructing the case.
表1发动机案例1Table 1 Engine Case 1
表2发动机案例2Table 2 Engine Case 2
表3发动机案例3Table 3 Engine Case 3
2)案例检索2) Case retrieval
输入当前时刻的电压为190，电流为32，温度为180，其中温度在当前时刻之前的历史趋势为50、55、80、110、140、155、163、170，得益于动态时间弯曲距离算法，可以支持当前时间序列与历史案例特征项时间序列之间的不等长度距离匹配。案例检索结果如表4所示。Input the voltage at the current moment as 190, the current as 32, and the temperature as 180, where the historical trend of the temperature before the current moment is 50, 55, 80, 110, 140, 155, 163, 170, thanks to the dynamic time warping distance algorithm , which can support unequal-length distance matching between the current time series and the feature time series of historical cases. The case retrieval results are shown in Table 4.
表4案例检索结果Table 4 Case search results
在表4中，发动机案例1和发动机案例3均计算的距离为曼哈顿距离距离，发动机案例2计算的距离为动态时间弯曲距离。In Table 4, the distance calculated by engine case 1 and engine case 3 is the Manhattan distance, and the distance calculated by engine case 2 is the dynamic time warping distance.
按照与当前目标案例的相似度从高到低排序，依次为发动机案例1、发动机案例2和发动机案例3。发动机案例1中的电压、电流和温度这三个特征项数值与当前目标案例对应的数值很相近，因此相似度匹配距离很小，相似度很高；发动机案例2中的温度时间序列与当前目标案例的时间序列的长度虽然不一致，但是趋势相近，均为剧增的过程，因此相似度匹配距离较小，相似度较高；发动机案例3中的各项特征项数值与当前目标案例对应的数值相差较大，因此相似度匹配距离较大，相似度较低。According to the similarity with the current target case from high to low, they are engine case 1, engine case 2 and engine case 3. The values of the three characteristic items of voltage, current and temperature in engine case 1 are very similar to the values corresponding to the current target case, so the similarity matching distance is small and the similarity is high; the temperature time series in engine case 2 is consistent with the current target case Although the lengths of the time series of the cases are not consistent, the trends are similar, and they are all in the process of sharp increase, so the similarity matching distance is small and the similarity is high; the value of each feature item in the engine case 3 corresponds to the value of the current target case The difference is large, so the similarity matching distance is large and the similarity is low.
3)案例重用3) Case reuse
设定案例推理的相似度匹配距离阈值为80，则基于步骤二的案例检索结果，考虑针对发动机案例1和发动机案例2进行案例重用，认为由于电压过高，电流过低，温度持续增高等原因，发动机目前已经烧坏，需要对发动机进行查修。Set the similarity matching distance threshold of case reasoning to 80, then consider case reuse for engine case 1 and engine case 2 based on the case retrieval results in step 2. It is considered that the voltage is too high, the current is too low, the temperature continues to rise, etc. , the engine is currently burnt out, and the engine needs to be inspected and repaired.
4)案例修正4) Case correction
经专家现场核实发现，发动机确实已经烧坏，即对应的案例结果与目标事件一致，因此将发动机案例1和发动机案例2的案例效果评价更新为优，并且无需对重用的案例进行修正。After on-site verification by experts, it was found that the engine had indeed burned out, that is, the corresponding case results were consistent with the target event. Therefore, the case effect evaluations of engine case 1 and engine case 2 were updated to excellent, and there was no need to revise the reused cases.
5)案例特征项权重调整5) Weight adjustment of case feature items
经专家判断，发动机案例1重用效果最好，因此针对案例1进行权重调整，将Δ设为0.1，TDist设为0.5。发动机案例1中，电压、电流和温度这三个特征项的权重调整过程如下所示。According to expert judgment, the reuse effect of engine case 1 is the best, so the weight adjustment is carried out for case 1, and Δ is set to 0.1, and T Dist is set to 0.5. In engine case 1, the weight adjustment process of the three characteristic items of voltage, current and temperature is as follows.
(1)电压这个特征项的匹配距离在综合距离中所占的比重较大，为0.9×(220-190)/32.6＝0.828，大于TDist，因此需要降低其特征项权重，修正后的权重为0.9×(1-0.1×27/32.6)＝0.825。(1) The matching distance of the characteristic item of voltage accounts for a large proportion in the comprehensive distance, which is 0.9×(220-190)/32.6=0.828, which is greater than T Dist , so it is necessary to reduce the weight of its characteristic item, and the corrected weight It is 0.9×(1−0.1×27/32.6)=0.825.
(2)电流这个特征项的匹配距离在综合距离中所占的比重为0.7×(40-32)/32.6＝0.172，小于TDist，所以提高其特征项权重，修正后的权重为0.7×(1+0.1×(1-0.172))＝0.758。(2) The proportion of the matching distance of the feature item current in the comprehensive distance is 0.7×(40-32)/32.6=0.172, which is smaller than T Dist , so the weight of its feature item is increased, and the weight after correction is 0.7×( 1+0.1×(1-0.172))=0.758.
(3)温度这个特征项的匹配距离为0，在综合距离中所占的比重小于TDist，所以提高其特征项权重，修正后的权重为0.8×(1+0.1×1)＝0.88。(3) The matching distance of the characteristic item of temperature is 0, and its proportion in the comprehensive distance is less than T Dist , so the weight of its characteristic item is increased, and the corrected weight is 0.8×(1+0.1×1)=0.88.
经过上述权重修正后，可以使得案例特征项的权重更加符合实际情况，提高了日后案例推理的准确性。After the above weight correction, the weight of the case characteristic item can be made more in line with the actual situation, and the accuracy of case reasoning in the future can be improved.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410797922.XA CN105787610A (en) | 2014-12-18 | 2014-12-18 | Case-based reasoning method capable of supporting time sequence matching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410797922.XA CN105787610A (en) | 2014-12-18 | 2014-12-18 | Case-based reasoning method capable of supporting time sequence matching |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105787610A true CN105787610A (en) | 2016-07-20 |
Family
ID=56384993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410797922.XA Pending CN105787610A (en) | 2014-12-18 | 2014-12-18 | Case-based reasoning method capable of supporting time sequence matching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105787610A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106599264A (en) * | 2016-12-22 | 2017-04-26 | 国家行政学院 | Big data-based emergency event evolution reasoning method and system |
| CN106897378A (en) * | 2017-01-20 | 2017-06-27 | 浙江大学 | A kind of semantic parameter search dynamic regulating method towards three-dimensional CAD model |
| CN107358348A (en) * | 2017-07-05 | 2017-11-17 | 成都电科智联科技有限公司 | A kind of method that plan fault-tolerance is improved based on OEE |
| CN108090567A (en) * | 2018-01-19 | 2018-05-29 | 国家电网公司 | Power communication system method for diagnosing faults and device |
| CN108231137A (en) * | 2016-12-15 | 2018-06-29 | 童综合医疗社团法人童综合医院 | Medical system with feedback learning function |
| CN108305007A (en) * | 2018-02-05 | 2018-07-20 | 深圳市瑞捷建筑工程咨询有限公司 | Engineering construction report-generating method, device and its system, server |
| CN108764805A (en) * | 2018-06-11 | 2018-11-06 | 河南理工大学 | A kind of multi-model self-adapting recommendation method and system of collaborative logistics Services Composition |
| CN108830306A (en) * | 2018-05-30 | 2018-11-16 | 东软集团股份有限公司 | The workflow method for diagnosing faults and device, medium and electronic equipment of business datum |
| CN109556861A (en) * | 2019-01-23 | 2019-04-02 | 昆明理工大学 | A kind of bearing real-time fault diagnosis system of case-based reasioning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101383027A (en) * | 2008-09-12 | 2009-03-11 | 同济大学 | A method and system for generating an emergency plan for environmental emergencies |
| CN102176239A (en) * | 2011-02-23 | 2011-09-07 | 电子科技大学 | Kitchenware fault diagnosing method based on case-based reasoning |
| CN103135014A (en) * | 2012-12-14 | 2013-06-05 | 西安电子科技大学 | Transformer fault diagnosis method based on case-based reasoning |
-
2014
- 2014-12-18 CN CN201410797922.XA patent/CN105787610A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101383027A (en) * | 2008-09-12 | 2009-03-11 | 同济大学 | A method and system for generating an emergency plan for environmental emergencies |
| CN102176239A (en) * | 2011-02-23 | 2011-09-07 | 电子科技大学 | Kitchenware fault diagnosing method based on case-based reasoning |
| CN103135014A (en) * | 2012-12-14 | 2013-06-05 | 西安电子科技大学 | Transformer fault diagnosis method based on case-based reasoning |
Non-Patent Citations (3)
| Title |
|---|
| 周荣良: "装备车辆故障的远程诊断", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
| 徐小力，洪孝安主编: "《设备与维修工程现代技术及应用》", 31 January 2009 * |
| 程文聪: "面向大规模网络安全态势分析的时序数据挖掘关键技术研究", 《中国博士学位论文全文数据库信息科技辑》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108231137A (en) * | 2016-12-15 | 2018-06-29 | 童综合医疗社团法人童综合医院 | Medical system with feedback learning function |
| CN106599264A (en) * | 2016-12-22 | 2017-04-26 | 国家行政学院 | Big data-based emergency event evolution reasoning method and system |
| CN106599264B (en) * | 2016-12-22 | 2019-07-19 | 国家行政学院 | A method and system for evolutionary reasoning of emergencies based on big data |
| CN106897378A (en) * | 2017-01-20 | 2017-06-27 | 浙江大学 | A kind of semantic parameter search dynamic regulating method towards three-dimensional CAD model |
| CN107358348A (en) * | 2017-07-05 | 2017-11-17 | 成都电科智联科技有限公司 | A kind of method that plan fault-tolerance is improved based on OEE |
| CN108090567A (en) * | 2018-01-19 | 2018-05-29 | 国家电网公司 | Power communication system method for diagnosing faults and device |
| CN108305007A (en) * | 2018-02-05 | 2018-07-20 | 深圳市瑞捷建筑工程咨询有限公司 | Engineering construction report-generating method, device and its system, server |
| CN108830306A (en) * | 2018-05-30 | 2018-11-16 | 东软集团股份有限公司 | The workflow method for diagnosing faults and device, medium and electronic equipment of business datum |
| CN108764805A (en) * | 2018-06-11 | 2018-11-06 | 河南理工大学 | A kind of multi-model self-adapting recommendation method and system of collaborative logistics Services Composition |
| CN109556861A (en) * | 2019-01-23 | 2019-04-02 | 昆明理工大学 | A kind of bearing real-time fault diagnosis system of case-based reasioning |
| CN109556861B (en) * | 2019-01-23 | 2020-08-07 | 昆明理工大学 | A Bearing Real-time Fault Diagnosis System Based on Case Reasoning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105787610A (en) | Case-based reasoning method capable of supporting time sequence matching | |
| US20220260981A1 (en) | Optimization decision-making method of industrial process fusing domain knowledge and multi-source data | |
| Khelif et al. | RUL prediction based on a new similarity-instance based approach | |
| CN106598791B (en) | A preventive identification method for industrial equipment faults based on machine learning | |
| KR102275126B1 (en) | Electrical facility safety evaluation system using artificial intelligence technique | |
| CN105787563A (en) | Self-learning mechanism-base fast matching fuzzy reasoning method | |
| CN103246801B (en) | A kind of shaft furnace working of a furnace failure prediction method based on improving reasoning by cases | |
| CN103926490A (en) | Power transformer comprehensive diagnosis method with self-learning function | |
| CN104569666A (en) | Power transformer fault prediction method based on electricity-graph model | |
| CN113935440A (en) | Iterative evaluation method and system for error state of voltage transformer | |
| CN104715342A (en) | Electrical equipment fault handling and aid decision making method based on case-based reasoning | |
| CN111142488B (en) | Process industrial optimization method based on working condition comparison driving | |
| CN114595943B (en) | Mechanical equipment image generation method | |
| CN106682159A (en) | Threshold configuration method | |
| CN115221318A (en) | Construction method and system of equipment health archives based on unified semantic representation of data | |
| CN117034169A (en) | Power grid main transformer equipment abnormal state prediction method based on time sequence causality network | |
| Xie et al. | Transformer Maintenance Policies Selection Based on an Improved Fuzzy Analytic Hierarchy Process. | |
| CN115953252A (en) | Method for determining construction safety liability insurance premium | |
| CN105741184B (en) | Transformer state evaluation method and device | |
| CN110766248A (en) | Workshop human factor reliability evaluation method based on SHEL and interval intuition fuzzy evaluation | |
| CN118260437A (en) | Public opinion information display method and system based on knowledge graph | |
| CN108363738B (en) | Recommendation method for industrial equipment data analysis algorithm | |
| She et al. | Health indicator construction of rolling bearings based on deep convolutional neural network considering phase degradation | |
| Mukhtar et al. | Forecasting Covid-19 time series data using the long short-term memory (LSTM) | |
| CN116224950A (en) | Intelligent fault diagnosis method and system for self-organization reconstruction of unmanned production line |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160720 |
|
| RJ01 | Rejection of invention patent application after publication |