An intelligent identification and diagnosis method for abnormal working conditions of cutter suction dredger

Abstract

This invention proposes a method for intelligently identifying and diagnosing abnormal operating conditions of cutter suction dredgers. The method is designed as follows: S1: data collection and screening of key variables; S2: routine data preprocessing; S3: based on the definition of a K-line chart, further preprocessing to obtain K-line chart time series feature data; S4: designing an abnormality diagnosis algorithm based on the K-line chart time series features and annotating construction conditions; S5: establishing an abnormal operating condition database; S6: establishing an XGBoost abnormal operating condition identification model based on the abnormal operating condition database, and outputting the optimal XGBoost classifier through continuous iterative training of the model. This invention applies K-line theory from the financial field to the algorithm design framework for identifying abnormal operating conditions of dredgers. By redefining the sampling period, this method not only greatly improves the speed of time series processing and reduces the time required for anomaly annotation, but also solves the problem of single-point anomalies affecting prediction accuracy. By building an abnormal operating condition identification model using the XGBoost algorithm, the intelligent identification of abnormal operating conditions is achieved.

Description

Technical Field

[0001] The invention relates to the field of safe production of cutter suction dredgers. Background Art

[0002] As "water highways," waterways are vital connections connecting key waterways and serve as a crucial resource for economic development. To ensure smooth water transport, the quality and efficiency of dredging projects must be improved. Cutter suction dredgers are a widely used type of vessel in dredging projects, playing a significant role in improving navigation conditions and ensuring safe navigation. However, dredgers operate underwater, and their construction conditions are complex and varied. This makes them prone to abnormalities such as blocked suction ports, pumps, and pipes during operation, severely impacting construction continuity and reducing efficiency.

[0003] Currently, the identification of abnormal operating conditions on cutter suction dredgers relies primarily on the operator's personal experience, lacking a scientific method for identifying these conditions. This severely impacts the efficiency of cutter suction dredger operations. Manual detection of abnormal operating conditions is subject to subjectivity and lag, hindering timely detection. Furthermore, identification results are often inaccurate due to human factors.

[0004] According to the actual situation of cutter suction dredger, the difficulties in identifying abnormal working conditions are as follows:

[0005] (1) Single-point anomalies often occur in the time series anomaly detection of cutter suction dredgers. Single-point anomalies are usually not accompanied by abnormal working conditions. Abnormal working conditions are usually group anomalies that last for a period of time. However, point anomalies will affect the accuracy of anomaly recognition and need to be identified and processed.

[0006] (2) Establishing an abnormal working condition identification model based on SCADA historical data requires a large amount of abnormal labeling information. However, abnormal labeling relies on the knowledge of experts, which is labor-intensive and time-consuming. There is also the problem of insufficient labeled data. Summary of the Invention

[0007] In response to the defects of the existing technology for identifying abnormal working conditions and the need for improvement, the present invention provides an intelligent identification and diagnosis method for abnormal working conditions of a cutter suction dredger, which aims to solve the technical problems of low efficiency and inaccuracy in identifying abnormal working conditions by existing manual detection methods.

[0008] The technical solutions of the present invention are as follows:

[0009] An intelligent identification and diagnosis method for abnormal working conditions of a cutter suction dredger, comprising:

[0010] S1: Data collection and screening of key variables;

[0011] S2: routine data preprocessing;

[0012] S3: Based on the definition of the candlestick chart, further preprocessing is performed to obtain the time series feature data of the candlestick chart;

[0013] S4: Design an abnormality diagnosis algorithm based on the time series characteristics of the K-line chart and mark the construction conditions;

[0014] S5: Establish abnormal operating condition database;

[0015] S6: Based on the abnormal operating condition database, an XGBoost abnormal operating condition identification model is established. By continuously iteratively training the model, the optimal XGBoost classifier is output.

[0016] Further optimization, the present invention also includes S7, S8:

[0017] Step S7: Input the real-time collected construction data into the optimal classifier output by S6, output the real-time construction working condition information, if an abnormality occurs, initiate an alarm, and store the abnormal working condition data of the alarm in the abnormal working condition database;

[0018] Step S8: Manually perform feedback marking on the alarm information issued by S7. If there is an error, the data in the abnormal operating condition database will be corrected and the abnormal operating condition database of step S5 will be updated.

[0019] Collect cutter suction dredger construction data and establish historical database and real-time database.

[0020] To process large amounts of SCADA monitoring data more quickly and accurately, this technology applies K-line theory from the financial field to an algorithmic design framework for identifying abnormal operating conditions on dredging vessels. By redefining the sampling period, this significantly improves the speed of time series processing and reduces the time required to label anomalies. It also addresses the issue of single-point anomalies impacting prediction accuracy. Finally, an abnormal operating condition identification model is constructed using the XGBoost algorithm, enabling intelligent identification of abnormal conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the abnormal operating condition identification algorithm framework.

[0022] Figure 2 Defined for the candlestick chart.

[0023] Figure 3 This is a diagram of the K-line chart conversion process (the sampling period is set to 60 seconds).

[0024] Figure 4 The original production time series diagram (a) and the corresponding production K-line diagram (b). DETAILED DESCRIPTION

[0025] The technical solution provided by this application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of this application will become more apparent with reference to the following description.

[0026] like Figure 1 As shown in the figure, an intelligent identification and diagnosis method for abnormal working conditions of a cutter suction dredger is proposed. The whole algorithm is designed as follows:

[0027] Step S1: Data collection and screening of key variables:

[0028] Collect cutter suction dredger construction data and establish historical database and real-time database;

[0029] The SCADA system on the cutter suction dredger collects more than 1,800 variables. In the algorithm design of the present invention, key variables of the cutter suction dredger construction operation system are selected for monitoring, as shown in Table 1.

[0030]

[0031] Step S2: Data preprocessing:

[0032] Perform normalization and standardization preprocessing operations on historical operating condition data;

[0033] Normalization preprocessing:

[0034] in: is the i-th variable in Table 1, for The minimum value of for The maximum value of .

[0035] Step S3: Based on the definition of the K-line chart, further preprocessing is performed to obtain the K-line chart time series feature data

[0036] Draw a K-line monitoring chart: For the key variables selected in Table 1, calculate their K variables based on the K-line theory and draw a K-line time series monitoring chart.

[0037] In the algorithm design of the present invention, the definition of the K-line chart is: the K-line chart is a columnar line consisting of a shadow and a body. The part of the shadow above the body is called the upper shadow, and the part below is called the lower shadow. The upper shadow and the lower shadow are the maximum and minimum values ​​of the time series of the period, respectively. Figure 2 The definition of the candlestick chart shown is that when the end value is higher than the start value, it is represented by a yang line (red) on the candlestick chart, and when the end value is lower than the start value, it is represented by a yin line (green) on the candlestick chart. Figure 2 In the algorithm, data outside the entity should not exceed 5%, which means: only sampling situations where the number of samples entering the entity can reach more than 95% of the total samples are selected, otherwise they are discarded to ensure the reliability and quality of the samples entering the algorithm.

[0038] In the algorithm design of the present invention, K theory: for a time length of of A time series of variables can be expressed as: , for exist The value of the moment.

[0039] Split it into time lengths Several adjacent time periods of seconds, then There is Time period, any time period Indicates that ,So

[0040]

[0041]

[0042] For any time series data of length L For example, the K variable consists of four variables, namely the starting value of this time series data , minimum , maximum value , end value

[0043]

[0044] For time series The minimum value of For time series The maximum value of .

[0045] See Figure 3 As shown in Figure 2, the original time series monitoring chart with a sampling period of 1 second is converted into a 60-second K-line monitoring chart, which not only retains the key information in the original sequence, but also reduces the impact of single outliers and improves the efficiency of time series data processing. Figure 4 Shown are the original production time series chart and the corresponding production K-line chart.

[0046] Step S4: Design an abnormality diagnosis algorithm based on the time series characteristics of the K-line chart and mark the construction conditions.

[0047] In the algorithm design of the present invention, the K-line chart represents different changes in indicators, and the corresponding judgment criteria are shown in Table 2 below:

[0048]

[0049] The judgment indicator of the big Yang line is ,

[0050] The judgment indicator of the big Yinxian line is ,

[0051] The judgment indicator of the middle Yinxian line is ,

[0052] The judgment indicator of the middle Yang line is ,

[0053] The judgment indicator of the small Yinxian line is ,

[0054] The judgment indicator of the small Yang line is ,

[0055] The judgment indicator of the one-line is ,

[0056] in, is the starting value of the K variable, is the end value of the K variable, is the maximum value of K variable, is the minimum value of the K variable.

[0057] Four basic abnormal working condition judgment rules have been established: suction port blockage, mud pump blockage, mud discharge pipe blockage, and excessive traverse speed. Additional abnormal working condition items can be added during application to further enhance construction stability.

[0058] The abnormal operating condition judgment rules are designed as shown in Table 3:

[0059]

[0060] The abnormal operating condition determination rules include:

[0061] Abnormal condition 1: S1 screening variable X 17 、X 19 、X 23 、X 22 After pre-processing by S2 and S3, if the indicators are simultaneously judged as a "large positive line" for suction vacuum, a "large negative line" for flow rate, a "large negative line" for output, and a "large negative line" for underwater pump discharge pressure, that is, if all four conditions occur simultaneously, it is diagnosed as an abnormality at the mud suction port, and it is determined that the mud suction port of the underwater pump is blocked;

[0062] Abnormal condition 2: S 1 screening variable X 17 、X 19 、X 22 After pre-processing by S2 and S3, when the indicators are judged to be a "large negative line" for suction vacuum, a "large negative line" for flow rate, and a "large negative line" for underwater pump discharge pressure at the same time, that is, when these three conditions occur at the same time, it is diagnosed as an abnormality in the mud pump and is judged to be blocked;

[0063] Abnormal condition 3: S1 screening variable X 17 、X 19 After pre-processing by S2 and S3, when the indicators are judged to be a "large negative line" for suction vacuum and a "large negative line" for flow rate at the same time, that is, when these two situations occur at the same time, it is diagnosed as an abnormality in the mud discharge pipe and is judged to be blocked;

[0064] Abnormal condition 4: S1 screening variable X 22 、X 24 After pre-processing by S2 and S3, when the indicators are judged to be a "large negative line" for flow velocity and a "large positive line" for concentration at the same time, that is, when these two situations occur at the same time, it is diagnosed that the lateral speed is too large;

[0065] Abnormal operating condition 0: Otherwise, the operating condition of the cutter suction dredger is determined to be normal.

[0066] For variables Corresponding time series data Evenly divide it into several adjacent time periods of length T, and discard those with a time period length of less than T seconds. , which correspond one to one with the variables in Table 1. For time series data The kth time period.

[0067] According to the abnormal working condition judgment rules formulated in Table 3, the abnormal working condition categories of the dredging device in any time period are Marking, that is ,in , representing normal, mud suction port blocked, mud pump blocked, mud discharge pipe blocked, and excessive traverse speed. Represents the kth time period, and the abnormal operating state at all times in this time period is set to This operation can greatly improve the speed of identifying and marking abnormal working conditions in time series. The time length should be set reasonably. Setting it too large will reduce the accuracy of judgment, and setting it too small will consume too much manpower cost. The present invention sets it to 60 seconds.

[0068] Step S5: Establishing an abnormal operating condition database: Inputting the abnormal operating condition data diagnosed in S4 into the abnormal operating condition database, and randomly selecting an equal amount of normal operating condition data and inputting them into the abnormal operating condition database, so as to maintain a balance between the abnormal and normal data proportions;

[0069] Step S6: Based on the abnormal operating condition database, an XGBoost abnormal operating condition identification model is established, and the optimal XGBoost classifier is output through continuous iterative training of the model;

[0070] Input: Randomly select 70% of the samples as the training data set, training sample data set D= ,

[0071] The input variables are the variables listed in Table 1, and the output variables are the abnormal operating condition categories.

[0072] , Represented as the jth sample value of the input variable, is the i-th variable The j-th sample value of

[0073] Output the j-th sample value of the variable.

[0074] Output: The optimal XGBoost tree structure, that is, the optimal classifier

[0075] The XGBoost tree model is represented as:

[0076] (1)

[0077] in is the predicted value of the output variable of the i-th sample, is the i-th sample of the input, is the kth classification tree

[0078] The objective function of XGBoost is:

[0079] (2)

[0080] The objective function (2) consists of two parts. The first part is the loss function , used to represent the difference between the true value and the predicted value, n is the sample size; the second part is the regularization term, which is used to prevent the model from overfitting.

[0081] (3)

[0082] is the regularization parameter of the number of leaf nodes, which inhibits the nodes from continuing to split. is the leaf node weight regularization parameter to prevent the leaf node weight from being too large. Score each leaf node. T is the number of leaf nodes.

[0083] The XGBoost algorithm only adds one tree at a time, and through continuous iterative training, it stops updating after finding a set of parameters that minimizes the objective function and outputs the optimal model structure.

[0084] The remaining 30% of samples are used as test data sets and input into the optimal tree structure of XGBoost to obtain the predicted classification , classify the predictions Classification with measured data Compare and calculate the accuracy of the classifier to evaluate the effect.

[0085]

[0086] TP(1 positives): the number of instances correctly classified as positive examples, that is, the number of instances that are actually positive examples and classified as positive examples by the classifier;

[0087] FP (0 ​​positives): the number of instances that are mistakenly classified as positive examples, that is, the number of instances that are actually negative examples but are classified as positive examples by the classifier;

[0088] FN(0 negatives): The number of instances that are mistakenly classified as negative examples, that is, the number of instances that are actually positive examples but are classified as negative examples by the classifier;

[0089] TN(1 negatives): The number of instances that are correctly classified as negative examples, that is, the number of instances that are actually negative examples and are classified as negative examples by the classifier.

[0090] After the simulation, the model estimation accuracy is shown in Table 4 below:

[0091]

[0092] Step S7: Input the real-time collected construction data into the optimal classifier output by S6, and output the real-time construction working condition information. If an abnormality occurs, an alarm is initiated, and the abnormal working condition data of the alarm is stored in the abnormal working condition database.

[0093] Step S8: Manually perform feedback marking on the alarm information issued by S7. If there is an error, the data in the abnormal operating condition database will be corrected and the abnormal operating condition database of step S5 will be updated.

[0094] The above technical solution adopts a semi-supervised learning method, which can accurately identify abnormal working condition categories. Usually, the abnormal category labels of construction working conditions need to be manually labeled. The present invention automatically labels the abnormal categories of construction data by citing K theory and combining expert knowledge, which solves the problem that manual labeling requires a lot of manpower costs. At the same time, an abnormal working condition identification model based on the XGBoost algorithm is established, which can identify abnormal working conditions of unlabeled real-time data, and continuously correct the abnormal working condition category labels through the manual feedback mode, thereby continuously optimizing the abnormal working condition identification model. The recognition framework of the present invention is scalable, and new abnormal working condition categories can be identified simply by adding new expert experience.

[0095] The above description is only a description of the preferred embodiments of the present application and does not limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical content should be regarded as equivalent valid embodiments and fall within the scope of protection of the technical solution of the present application.

Claims

1. A method for intelligent identification and diagnosis of abnormal working conditions of a cutter suction dredger, characterized by: The method is designed to: S1: Data collection and screening of key variables; S2: routine data preprocessing; S3: Based on the definition of the candlestick chart, further preprocessing is performed to obtain the time series feature data of the candlestick chart; S4: Design an abnormality diagnosis algorithm based on the time series characteristics of the K-line chart and mark the construction conditions; S5: Establish abnormal operating condition database; S6: Based on the abnormal operating condition database, an XGBoost abnormal operating condition identification model is established. The optimal XGBoost classifier is output through continuous iterative training of the model. The step S1: Collect cutter suction dredger construction data and establish historical database and real-time database; Key variable screening: Includes tide level (TIDE) , trunnion draft (DRAFT_TRNION) , bridge angle (BRIDGE_ANGLE) , belongs to the bridge system; Including traverse speed (TRNS_SPD) , wire rope tension (ROPE_TENSION) , traverse angle (TRNS_ANGLE) , traverse rope speed (TRNS_ROPE_SPD) , Transverse hydraulic system pressure (TH_PR) , belongs to the transverse movement system; Including trolley travel (CR_STRK) , double pile landing (DOUBLE_PILE_LANDING) , belongs to the steel pile trolley system; Including cutter speed (CUTTER_SPD) , CUTTER_DEEP , Cutter power (CUTTER_POWER) , Cutter hydraulic system pressure (CUTTER_CH_PR) , belongs to the reamer system; Including submersible pump discharge pressure (LP_OUTLET_PR) , submersible pump power (LP_POWER) , Suction vacuum (SUCK_VAC) , belongs to the underwater pump system; Including mud pump speed (DP_SPD) , Mud pump discharge pressure (DP_OUTLET_SPD) , mud pump power (DP_POWER) , belongs to the mud pump system; Includes density (SLURRY_SPD) , flow rate (SLURRY_FLOW) , instantaneous output (OUTPUT) , concentration (SLURRY_CON) , which belongs to the yield measurement system; The step S4 comprises: S4.1 The criteria for determining the performance of different indicators on the K-line chart include: The judgment indicator of the big Yang line is , the judgment indicator of the big Yinxian is , the judgment indicator of the middle Yinxian is , the judgment indicator of the middle Yang line is , the judgment indicator of the small Yinxian is , the judgment indicator of the small Yang line is , the judgment indicator of the one-line is ,in, is the starting value of the K variable, is the end value of the K variable, is the maximum value of K variable, is the minimum value of the K variable; S4.2 Establish four basic abnormal operating condition determination rules: mud suction port blockage, mud pump blockage, mud discharge pipe blockage, and excessive traverse speed; the abnormal operating condition determination rules include: Abnormal condition 1: S1 screening variable X 17 、X 19 、X 23 、X 22 After pre-processing by S2 and S3, if the indicators are simultaneously judged as a "large positive line" for suction vacuum, a "large negative line" for flow rate, a "large negative line" for output, and a "large negative line" for underwater pump discharge pressure, that is, if all four conditions occur simultaneously, it is diagnosed as an abnormality at the mud suction port and is determined to be blocked at the mud suction port of the underwater pump; Abnormal condition 2: S 1 screening variable X 17 、X 19 、X 22 After pre-processing by S2 and S3, if the indicators are simultaneously judged as "large negative lines" for suction vacuum, "large negative lines" for flow rate, and "large negative lines" for underwater pump discharge pressure, that is, if these three conditions occur simultaneously, it is diagnosed as an abnormality in the mud pump and is judged to be blocked; Abnormal condition 3: S1 screening variable X 17 、X 19 After pre-processing by S2 and S3, if the indicators are judged to be a "large negative line" for suction vacuum and a "large negative line" for flow rate at the same time, that is, if these two situations occur at the same time, it is diagnosed as an abnormality in the mud discharge pipe and is judged to be blocked; Abnormal condition 4: S1 screening variable X 22 、X 24 After pre-processing by S2 and S3, if the indicators are simultaneously judged as "large negative line" for flow velocity and "large positive line" for concentration, that is, when these two situations occur at the same time, it is diagnosed that the lateral speed is too high; Abnormal operating condition 0: Otherwise, the cutter suction dredger operating condition is determined to be normal; S4.3 According to the established abnormal operating condition judgment rules, the abnormal operating condition category of the dredging device in any time period Marking, that is ,in , representing normal, mud suction port blocked, mud pump blocked, mud discharge pipe blocked, and excessive traverse speed. Represents the kth time period, and the abnormal operating state at all times in this time period is set to .

2. The method for intelligent identification and diagnosis of abnormal working conditions of a cutter suction dredger according to claim 1, characterized in that: Described step S2: After the key variables are screened in step 1, these historical operating condition data are normalized and pre-processed; Normalization preprocessing: in: is the preset i-th variable, for The minimum value of for The maximum value of .

3. The method for intelligently identifying and diagnosing abnormal operating conditions of a cutter suction dredger according to claim 1, wherein: Step S3: For the preprocessed data in step 2, calculate its K variable based on K-line theory, draw a K-line time series monitoring chart, and convert the original time series monitoring chart with a sampling period of 1 second into a K-line monitoring chart with a time length of T seconds to obtain the K-line chart time series feature data, namely the original production time series chart and the corresponding production K-line chart; In the K-line time series monitoring chart, the K-line chart is a columnar line consisting of a shadow and a body. The part of the shadow above the body is called the upper shadow, and the part below is called the lower shadow. The upper shadow and the lower shadow are the maximum and minimum values ​​of the corresponding time series respectively. The end value is higher than the start value and is represented by a positive line of the K-line chart. The end value is lower than the start value and is represented by a negative line of the K-line chart. The K-line theory: for a time length of of A time series of variables, expressed as: , for exist The value of the moment; Split it into time lengths Several adjacent time periods of seconds, then have Time period, any time period Indicates that ,So For any time series data of length L For example, the K variable consists of four variables, namely the starting value of this time series data , minimum , maximum value , end value , For time series The minimum value of For time series The maximum value of .

4. The method for intelligently identifying and diagnosing abnormal operating conditions of a cutter suction dredger according to claim 3, wherein: For variables Corresponding time series data Evenly split it into several adjacent time periods of length T, and discard those with a length of less than T seconds; , which correspond one to one with the key variables; For time series data The kth time period.

5. The method for intelligently identifying and diagnosing abnormal operating conditions of a cutter suction dredger according to claim 1, wherein: Step S5: The abnormal operating condition data diagnosed in S4 are input into the abnormal operating condition database, and an equal amount of normal operating condition data are randomly selected and input into the abnormal operating condition database to maintain a balance between the abnormal and normal data proportions.

6. The method for intelligently identifying and diagnosing abnormal operating conditions of a cutter suction dredger according to claim 1, wherein: The step S6 is specifically as follows: Input: Randomly select a part of the samples as the training data set, the training sample data set D= , Output: The optimal XGBoost tree structure, that is, the optimal classifier.

7. The method for intelligently identifying and diagnosing abnormal operating conditions of a cutter suction dredger according to claim 6, wherein: The input variables are the selected key variables, and the output variables are the abnormal operating condition categories: , Represented as the jth sample value of the input variable, is the i-th variable The j-th sample value of Output the j-th sample value of the variable; The XGBoost tree model is represented as: （1） in is the predicted value of the output variable of the i-th sample, is the i-th sample of the input, is the kth classification tree; The objective function of XGBoost is: （2） The objective function (2) consists of two parts. The first part is the loss function , used to represent the difference between the true value and the predicted value, n is the sample size; the second part is the regularization term, which is used to prevent the model from overfitting; （3） is the regularization parameter of the number of leaf nodes, which inhibits the nodes from continuing to split. is the leaf node weight regularization parameter to prevent the leaf node weight from being too large. Score each leaf node; T is the number of leaf nodes; The XGBoost algorithm only adds one tree at a time, and through continuous iterative training, it stops updating after finding a set of parameters that minimize the objective function and outputs the optimal model structure; The remaining samples are used as test data sets and input into the optimal tree structure of XGBoost to obtain the predicted classification , classify the predictions Classification with measured data Compare and calculate the accuracy of the classifier to evaluate the effect.

8. The method for intelligently identifying and diagnosing abnormal operating conditions of a cutter suction dredger according to claim 1, wherein: The following steps are also included: S7: Input the real-time collected construction data into the optimal classifier output by S6, output the real-time construction working condition information, and if an abnormality occurs, initiate an alarm and store the abnormal working condition data in the abnormal working condition database; S8: Manually mark the alarm information sent by S7. If there is an error, the data in the abnormal operating condition database will be corrected and the S5 abnormal operating condition database will be updated.

Images