Ship fuel consumption abnormal working condition detection method and system based on PELT algorithm

Abstract

The invention provides a ship fuel consumption abnormal working condition detection method and system based on a PELT algorithm, and the method comprises a parameter obtaining and calculation step, a fuel consumption time sequence construction and change point detection step, a change point correction step and a fuel consumption abnormal working condition detection step. The method comprises the following steps: calculating to obtain a difference sequence, smoothing the difference sequence by utilizing a seven-point moving average method, calculating statistical significance correction by adopting an efficient change point detection capability of a PELT algorithm and combining a Bonferroni correction method, and performing linear regression analysis, so that the abnormal oil consumption working conditions of various ship types can be accurately identified under various navigation conditions. And reliable decision support is provided for ship energy efficiency management and cost control.

Description

Technical Field

[0001] The present invention relates to the technical field of ship operation management, and in particular to a method and system for detecting abnormal fuel consumption conditions of a ship based on a PELT algorithm. Background Art

[0002] Identification of abnormal fuel consumption conditions of ships is the basis for green and low-carbon shipping and cost control of shipping companies. As an important part of international trade and supply chain, the shipping industry undertakes more than 80% or more of international trade. According to the International Maritime Organization (IMO), the total greenhouse gas emissions of the shipping industry account for 2%-3% of the world. In order to promote energy conservation and emission reduction in the shipping field, the International Maritime Organization, the European Union and the State Maritime Administration have successively issued a series of policies, which have put forward higher requirements for the monitoring and management of abnormal fuel consumption conditions of ships. Fuel consumption of ships is a direct cost in the operation of ships. It is the goal of shipping companies to control the fuel consumption of ships in a stable and fuel-saving condition. However, due to reasons such as the bottom of the dock, the fuel consumption of ships is often in a state of overconsumption after a period of operation. Therefore, accurate identification of abnormal conditions of ship fuel consumption can not only help reduce operating costs, but also save energy and reduce carbon.

[0003] The current methods for detecting abnormal fuel consumption conditions of ships are mainly divided into two categories: 1) Judging abnormal fuel consumption conditions based on simple statistical analysis or fixed thresholds. This type of method relies on expert knowledge and experience, and it is difficult to accurately capture complex fuel consumption change patterns, especially abnormal conditions under long-term navigation and changing environments. 2) Model-based detection methods that integrate ship-side sensor data. This type of method uses sensors installed on the ship (such as GPS, speed sensors, oil level sensors, etc.) to collect data in real time, and uses ship performance models or machine learning algorithms to predict fuel consumption based on the ship's operating parameters (such as speed, load, weather conditions, etc.) and compare it with the actual fuel consumption. Summary of the invention

[0004] In order to solve the problems that the current abnormal fuel consumption conditions of ships are difficult to accurately detect and the current detection methods based on ship-side sensor data have poor adaptability and high algorithm complexity, the present invention provides a ship fuel consumption abnormal condition detection method based on the PELT algorithm. The method has the characteristics of strong robustness, high accuracy, and low algorithm overhead in complex navigation environments and scenarios with large amounts of data. It can accurately identify the abnormal fuel consumption conditions of various ship types under various navigation conditions, and provide reliable decision support for ship energy efficiency management and cost control. The present invention also relates to a ship fuel consumption abnormal condition detection system based on the PELT algorithm.

[0005] The technical solution of the present invention is as follows:

[0006] A method for detecting abnormal fuel consumption of a ship based on a PELT algorithm, characterized in that it comprises the following steps:

[0007] Parameter acquisition and calculation steps: obtain the MMSI code of a certain ship, and obtain the MRV reported fuel consumption and AI predicted fuel consumption of the ship at each time point in a certain time period in chronological order, and obtain the MRV reported fuel consumption sequence and AI predicted fuel consumption sequence respectively; then calculate the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption at each time point in the time period, and then obtain the difference sequence;

[0008] Steps for constructing the fuel consumption time series and detecting change points: Use the seven-point moving average method to smooth the difference series to obtain the smoothed fuel consumption time series; then use the PELT algorithm to detect change points in the fuel consumption time series and identify multiple change points;

[0009] Change point correction step: conduct hypothesis tests on each identified change point to obtain the significance probability of each change point, set multiple different overall significance levels, calculate multiple corrected significance levels based on each overall significance level and the number of change points, and then compare all significance probabilities with each corrected significance level, retain change points with a significance probability less than the corrected significance level, and screen the change points retained in each comparison process, and use the change points retained in each comparison process as the corrected change points;

[0010] Abnormal fuel consumption condition detection steps: Divide the difference sequence into multiple non-overlapping and continuous intervals based on the corrected change points, and use linear regression method to fit the trend curve in each interval. According to the slope of the trend curve in each interval, the abnormal fuel consumption condition of the ship in the corresponding interval is detected.

[0011] Preferably, after the abnormal fuel consumption condition detection step, it also includes an inference mechanism construction and visualization step: obtaining the actual abnormal condition label at each time point that is pre-marked, and extracting the prediction label of each interval from the detected abnormal fuel consumption condition, comparing the actual abnormal condition label with the prediction label, and obtaining the number of predicted labels that are consistent with the abnormal condition label, and then calculating the accuracy to verify the accuracy of the detected abnormal fuel consumption condition; constructing an inference mechanism based on the accuracy and in combination with business scenarios and business logic, and classifying and interpreting the detected abnormal fuel consumption conditions through the inference mechanism, and providing a visual display.

[0012] Preferably, in the abnormal fuel consumption condition detection step, the abnormal fuel consumption condition includes overconsumption and fuel saving; when the slope is greater than zero, it indicates that the ship is in an overconsumption state, and the larger the slope, the more serious the overconsumption; when the slope is less than zero, it indicates that the ship is in a fuel saving state, and the smaller the slope, the better the fuel saving state of the ship.

[0013] Preferably, in the fuel consumption time series construction and change point detection steps, smoothing the difference sequence using the seven-point moving average method includes: based on each difference in the difference sequence, multiple seven-point moving average values ​​of the difference sequence are calculated using the seven-point moving average method, and then the fuel consumption time series is constructed.

[0014] Preferably, in the parameter acquisition and calculation module, the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence are also preprocessed to ensure that the time intervals of the data points are consistent and aligned with the time axis, and the preprocessing includes data cleaning, alignment and interpolation.

[0015] A ship fuel consumption abnormal condition detection system based on PELT algorithm, characterized by comprising a parameter acquisition and calculation module, a fuel consumption time series construction and change point detection module, a change point correction module and a fuel consumption abnormal condition detection module connected in sequence,

[0016] The parameter acquisition and calculation module acquires the MMSI code of a certain ship, and sequentially acquires the MRV reported fuel consumption and AI predicted fuel consumption of the ship at each time point in a certain time period in chronological order, and obtains the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence respectively; and then sequentially calculates the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption at each time point in the time period, and then obtains the difference sequence;

[0017] The fuel consumption time series construction and change point detection module uses the seven-point moving average method to smooth the difference series to obtain the smoothed fuel consumption time series; then uses the PELT algorithm to detect the change points of the fuel consumption time series to identify multiple change points;

[0018] The change point correction module performs hypothesis testing on each identified change point to obtain the significance probability of each change point, and sets multiple different overall significance levels. Based on each overall significance level and the number of change points, multiple corrected significance levels are calculated using the Bonferroni correction method, and then all significance probabilities are compared with each corrected significance level, and change points with a significance probability less than the corrected significance level are retained, and the change points retained in each comparison process are screened, and the change points retained in each comparison process are used as corrected change points;

[0019] The abnormal fuel consumption operating condition detection module divides the difference sequence into multiple non-overlapping and continuous intervals based on the corrected change points, and uses linear regression method to fit a trend curve in each interval, and detects the abnormal fuel consumption operating condition of the ship in the corresponding interval according to the slope of the trend curve in each interval.

[0020] Preferably, it also includes an inference mechanism construction and visualization module, which is connected to the abnormal fuel consumption condition detection module, and is used to obtain the actual abnormal condition label at each time point that is pre-marked, and extract the prediction label of each interval from the detected abnormal fuel consumption condition, compare the actual abnormal condition label with the prediction label, and obtain the number of predicted labels that are consistent with the abnormal condition label, and then calculate the accuracy rate to verify the accuracy of the detected abnormal fuel consumption condition; construct an inference mechanism based on the accuracy rate and in combination with business scenarios and business logic, and classify and interpret the detected abnormal fuel consumption conditions through the inference mechanism, and provide a visual display.

[0021] Preferably, in the abnormal fuel consumption condition detection module, the abnormal fuel consumption conditions include overconsumption and fuel saving; when the slope is greater than zero, it indicates that the ship is in an overconsumption state, and the larger the slope, the more serious the overconsumption; when the slope is less than zero, it indicates that the ship is in a fuel saving state, and the smaller the slope, the better the fuel saving state of the ship.

[0022] Preferably, in the fuel consumption time series construction and change point detection module, smoothing the difference sequence using the seven-point moving average method includes: based on each difference in the difference sequence, multiple seven-point moving average values ​​of the difference sequence are calculated using the seven-point moving average method, and then the fuel consumption time series is constructed.

[0023] Preferably, in the parameter acquisition and calculation module, the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence are also preprocessed to ensure that the time intervals of the data points are consistent and aligned with the time axis, and the preprocessing includes data cleaning, alignment and interpolation.

[0024] The beneficial effects of the present invention are:

[0025] The present invention provides a method for detecting abnormal operating conditions of ship fuel consumption based on a PELT algorithm, that is, a method for detecting abnormal operating conditions of ship fuel consumption using a PELT (Pruned Exact Linear Time) multiple change point detection algorithm. The method comprises the following steps: firstly, the MRV reported fuel consumption and the AI ​​predicted fuel consumption of the ship at each time point in a certain time period are obtained in chronological order, and an MRV reported fuel consumption sequence and an AI predicted fuel consumption sequence are obtained respectively; then, the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption at each time point in the time period is calculated in turn, and then a difference sequence is obtained. Then, a seven-point moving average method is used to smooth the difference sequence to obtain a fuel consumption time series after smoothing, which can effectively reduce noise and fluctuation in the data, make the change points more obvious, and improve the accuracy of subsequent change point detection; and then, the PELT algorithm is used to perform change point detection on the fuel consumption time series, and multiple change points are identified, so that significant change points in the data can be quickly and accurately identified, and different operating condition intervals can be divided. Then, a hypothesis test is performed on each identified change point to obtain the significance probability of each change point, which can effectively evaluate the statistical significance of each change point, exclude accidental change points, and improve the reliability of the detection results. And set multiple different overall significance levels, based on each overall significance level and the number of change points, and use the Bonferroni correction method and a specific comparison method to obtain the corrected change points. By setting multiple significance levels, the significance of the change points can be evaluated at different levels of rigor, increasing the robustness of the results. The Bonferroni correction method can control the error rate in multiple comparisons, reduce false positive results, and improve the accuracy of change point detection. Through multiple comparisons and screening, it is ensured that the final retained change points have high significance and reliability, reducing misjudgment; finally, based on the corrected change points, the difference sequence is divided into multiple intervals, and the linear regression method is used to fit the trend curve in each interval. According to the slope of the trend curve of each interval, the abnormal fuel consumption of the ship in the corresponding interval is detected. This method can accurately locate the time point when the actual fuel consumption is significantly different from the predicted fuel consumption, providing reliable data support for ship energy efficiency management and abnormal condition identification.

[0026] The present invention uses the PELT multi-change point detection algorithm, multiple test adjustment method and linear regression analysis to process a large amount of historical data, and interprets abnormal fuel consumption conditions through an inference mechanism. The data-driven adaptive PELT algorithm is used, and the algorithm accuracy is not affected by the threshold setting. It can also quickly process large-scale data and improve the efficiency and accuracy of abnormal detection. At the same time, it does not rely on static testing or sensor data. Fuel consumption anomaly detection can be achieved through ship historical data analysis. The algorithm has strong robustness, lower cost and simpler operation. In addition, shorter historical data can provide fuel consumption abnormal condition information for ship fuel consumption management, which is helpful to timely adjust the ship's operating status and maintenance plan. It has the characteristics of strong robustness, high accuracy and low algorithm overhead.

[0027] The present invention also relates to a ship fuel consumption abnormal operating condition detection system based on the PELT algorithm. The system corresponds to the above-mentioned ship fuel consumption abnormal operating condition detection method based on the PELT algorithm, and can be understood as a system for implementing the above-mentioned ship fuel consumption abnormal operating condition detection method based on the PELT algorithm, including a parameter acquisition and calculation module, a fuel consumption time series construction and change point detection module, a change point correction module and a fuel consumption abnormal operating condition detection module connected in sequence. The modules work together, by using AI to predict fuel consumption data and MRV to report fuel consumption data, utilizing the efficient change point detection capability of the PELT algorithm, combined with statistical significance correction, it is possible to accurately identify abnormal fuel consumption conditions of various ship types under various navigation conditions, and provide reliable decision support for ship energy efficiency management and cost control. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a flow chart of the method for detecting abnormal fuel consumption of a ship based on the PELT algorithm of the present invention.

[0029] Figure 2 It is a schematic diagram of the changing trend of the number of changing points when different pen values ​​are set in the present invention.

[0030] Figure 3 This is a schematic diagram of the fuel consumption change point detection and change trend fitting of container ship - Ship A.

[0031] Figure 4 It is a schematic diagram of the detection of fuel consumption change points and fitting of the change trend of liquid bulk carrier-Ship B.

[0032] Figure 5 This is a schematic diagram of the fuel consumption change point detection and change trend fitting of the dry bulk carrier - Ship C.

[0033] Figure 6 This is a schematic diagram of the fuel consumption change point detection and change trend fitting of the special ship - Ship D. DETAILED DESCRIPTION

[0034] The present invention will be described below in conjunction with the accompanying drawings.

[0035] The present invention relates to a method for detecting abnormal operating conditions of ship fuel consumption based on the PELT algorithm. The method is implemented based on the PELT algorithm, the Bonferroni correction method and the linear regression analysis, and mainly utilizes two sets of time series data: AI predicted fuel consumption data and MRV reported fuel consumption data. By comparing these two sets of data, the abnormal operating conditions of ship fuel consumption can be effectively identified and located. The PELT algorithm is used to efficiently detect change points in time series, the Bonferroni correction ensures statistical significance in multiple comparisons, and the linear regression analysis is used to evaluate the trend of fuel consumption changes in each time period. This method can accurately locate the time point when the actual fuel consumption differs significantly from the predicted fuel consumption, and provide reliable data support for ship energy efficiency management and abnormal operating condition identification. The prerequisites for the application of the present invention are as follows:

[0036] 1) It has a continuous and stable AI fuel consumption prediction system that can provide continuous predicted fuel consumption data;

[0037] 2) The MRV system can provide accurate actual fuel consumption data regularly;

[0038] 3) The time resolution of AI predicted fuel consumption data and MRV actual fuel consumption data matches;

[0039] 4) Applicable to all types of commercial vessels participating in the MRV (Monitoring, Reporting, Verification) program.

[0040] The flowchart of this method is as follows Figure 1 As shown, the following steps are included in sequence:

[0041] 1. Parameter acquisition and calculation steps: Get the MMSI code of a certain ship, and obtain the MRV reported fuel consumption and AI predicted fuel consumption of the ship at each time point in a certain time period in chronological order, and obtain the MRV reported fuel consumption sequence and AI predicted fuel consumption sequence respectively; then calculate the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption at each time point in the time period, and then obtain the difference sequence. Figure 1 Data collection and preprocessing, calculation difference sequence shown.

[0042] Specifically, first obtain the MMSI code of a certain ship from the database, and obtain the MRV reported fuel consumption and AI predicted fuel consumption of the ship at each time point in a certain time period in chronological order, and obtain the MRV reported fuel consumption sequence M and AI predicted fuel consumption sequence A respectively, and perform pre-processing such as cleaning, alignment, and interpolation on the two sets of data of the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence to ensure that the time intervals of the data points are consistent and the time axis can be aligned; delete duplicate data; if there are missing values, perform appropriate interpolation processing to ensure that the data column has the correct data type. Among them, the MRV reported fuel consumption sequence M and the AI ​​predicted fuel consumption sequence A are respectively expressed as follows:

[0043] Ship = {a 1 ,a 2 ,...,a n} (1)

[0044]

[0045] Where Ship is a given list of ships; The AI ​​predicts fuel consumption for the i-th ship at the j-th time point, and It is the fuel consumption reported by MRV at the same time point.

[0046] Then, the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption (actual fuel consumption, i.e., true fuel consumption) at each time point in the time period is calculated in turn, and the difference sequence is obtained. For each time point, the difference between the predicted fuel consumption and the true fuel consumption is calculated, and this difference sequence Δ will reflect the changing trend of the operating conditions. Specifically, first, it is assumed that the MRV reported fuel consumption and the AI ​​predicted fuel consumption show the same development trend under normal operating conditions. When the fuel consumption operating conditions are abnormal, the trends of the two time series curves will change, such as approaching or moving away. When the two curves are close, it represents the occurrence of over-consumption conditions. Conversely, when the two curves are far away, it represents the occurrence of fuel-saving conditions. The trend of the two curves approaching or moving away can be intuitively reflected by the changes in the difference sequence Δ. As shown in the following formulas (4) and (5), Δ is a set of the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption of each ship. is a difference sequence, including the ath i The difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption of the ship at each time point. Difference sequence Δ and difference sequence They are expressed as follows:

[0047]

[0048]

[0049] 2. Steps for constructing fuel consumption time series and detecting change points: Use the seven-point moving average method to smooth the difference series to obtain the smoothed fuel consumption time series; then use the PELT algorithm to detect change points in the fuel consumption time series and identify multiple change points. Figure 1 The seven-point time series shown is smoothed and the PELT algorithm detects change points.

[0050] Seven-point time series smoothing is to apply seven-point moving average to the difference series to smooth the data and reduce the impact of short-term fluctuations. Specifically, firstly, based on each difference in the difference series, the seven-point moving average method is used to calculate multiple seven-point moving averages of the difference series. The seven-point moving average is calculated according to the following formula:

[0051]

[0052] Among them, Δ MA7 [k] is the kth seven-point moving average, which is the fuel consumption time series after smoothing.

[0053] Then, based on the calculated multiple seven-point moving averages, the fuel consumption time series Δ MA7 , as shown below:

[0054] Δ MA7 ={Δ MA7 [1],Δ MA7 [2],...,Δ MA7 [L]} (7)

[0055] Wherein, L is the number of points in the difference sequence Δ that can be used to calculate the moving average (ie, L=N-6, because 7 points are needed to calculate the first moving average).

[0056] Then, the PELT algorithm is used to calculate the smoothed fuel consumption time series Δ MA7 Perform change point detection to identify multiple change points where the statistical characteristics of the data (such as mean, variance, etc.) change significantly. Among them, the PELT algorithm is an algorithm for time series change point detection. The goal is to find a set of change points so that the sum of the costs of all sub-segments (when using the PELT algorithm for change point detection, an attempt is made to find a series of "change points" that divide the original time series into several non-overlapping and continuous sub-sequences, each of which is called a "sub-segment") plus the penalty for the number of change points is minimized. The penalty term pen is used to control the complexity of the model and prevent overfitting. The objective function is actually constructed based on the selected cost function. The goal of the algorithm is to minimize the total cost of the following form:

[0057] Total cost = sum(cost function (each sub-segment)) + penalty term × (number of change points) (8)

[0058] The specific steps of the PELT algorithm are as follows:

[0059] Step 1: Parameter setting

[0060] First, set the cost function model according to the algorithm requirements. The cost function model is set to "rbm" (robust Bayesian model), which is particularly suitable for processing complex time series data containing outliers or non-standard noise. Secondly, set the value range of the initial penalty term pen, which can be set to pen∈[1,20]. The penalty term pen is a positive number used to control the sensitivity of change point detection. It can be understood as the "cost" of adding new change points. A larger pen value will cause the algorithm to tend to detect fewer change points, while a smaller pen value will make it easier for the algorithm to detect more change points. Set the initial value of the parameter pen according to the specific situation.

[0061] Step 2: Adjust the value of the parameter pen (threshold)

[0062] like Figure 2 As shown, the initial pen value is traversed, that is, the pen value is traversed from 1 to 20, and the number of change points under each pen value is calculated (the number of change points at different thresholds shown in A, B, C, and D). It is found that when the pen value is greater than 5, the number of change points tends to be stable, so the pen is set to [1, 2, 3, 4, 5], and the algorithm is re-run to obtain the change points after the parameters are updated. In other words, the pen value range is readjusted according to the development trend of the number of change points, and the algorithm is re-run to obtain the change points after the parameters are updated.

[0063] Step 3: Record the detected change point CP and the number of change points detected under different pen values ​​n_tests.

[0064] 3. Change point correction steps: Conduct hypothesis tests on each identified change point to obtain the significance probability of each change point, and set multiple different overall significance levels. Based on each overall significance level and the number of change points, use the Bonferroni correction method to calculate multiple corrected significance levels, then compare all significance probabilities with each corrected significance level, retain change points with a significance probability less than the corrected significance level, and screen the change points retained in each comparison process, and use the change points retained in each comparison process as the corrected change points. Figure 1 The multiple testing adjustment method shown corrects for change points.

[0065] The multiple testing adjustment method is used to control the false alarm rate, that is, the Bonferroni correction method is used to calculate the corrected significance level of the change point, and the true change points are screened out. Only the corrected significant change points detected by multiple thresholds are retained. Only those points that are identified as change points in multiple tests and whose frequency exceeds the corrected threshold are considered to be statistically significant true change points. Specifically, in change point detection, a position that is not actually a change point may be mistakenly determined as a change point. When multiple hypothesis tests are performed (change point detection is performed using multiple thresholds), the cumulative risk of such errors increases, so multiple testing correction is required. The Bonferroni correction is a statistical method used to control the false positive rate in multiple comparisons. Its basic principle is to reduce the probability of incorrectly rejecting the null hypothesis in multiple hypothesis tests by reducing the significance level of each individual test. The Bonferroni correction obtains the corrected significance level by dividing the original significance level by the number of tests. Points with a corrected significance level less than 0.05 are considered statistically significant. The details are as follows:

[0066] First, a hypothesis test is performed on each identified change point to obtain the significance probability of each change point (which can also be understood as the probability of making an error in each test, or called the p-value). Then, multiple different overall significance levels are set as α (for example, α can be set to 0.05, 0.01, and 0.005), and based on each overall significance level and the number of change points, multiple corrected significance levels are calculated using the Bonferroni correction method. That is, after performing n_tests hypothesis tests (independent tests), the significance level after the Bonferroni correction will become α / n_tests. This means that in the Bonferroni correction, the null hypothesis will only be rejected when the p-value is less than α / n_tests. The corrected significance level corrected_α is calculated as follows:

[0067] corrected_α=α / n_tests (9)

[0068] Data points that still meet the significance level requirements after correction are considered to be independent of the multiple comparison problem. Therefore, the change points with a significance probability (P value) less than the corrected significance level are retained, and the change points retained in each comparison process are screened. The change points retained in each comparison process are used as corrected change points and recorded in the sequence TCP.

[0069] TCP={TCP 1 ,TCP 2 ,...,TCP m} (10)

[0070] Where m is the number of change points.

[0071] For example, suppose there are 10 change points with the following significance probabilities (P values): 1 =0.003, P 2 =0.01, P 3 =0.001, P 4 =0.02, P 5 =0.004, P 6 =0.002, P 7 =0.03, P 8 =0.006, P 9 =0.04, P 10 =0.007; hypothesis tests are performed on these 10 change points respectively, that is, n_tests=10, and then the corrected significance level is calculated (based on the set multiple different overall significance levels and the number of hypothesis tests, multiple corresponding corrected significance levels can be calculated, for example, 0.05, 0.01 and 0.005 respectively).

[0072] Using 0.05 as the corrected significance level, the change points with P values ​​less than 0.05 were retained: P 1 , P 2 , P 3 , P 4 , P 5 , P 6 , P 7 , P 8 , P 9 , P 10 ;

[0073] Using 0.01 as the corrected significance level, the change points with a P value less than 0.01 were retained: P 1 , P 3 , P 5 , P 6 , P 8 , P 10

[0074] Use 0.005 as the corrected significance level: retain the change points with P values ​​less than 0.005: P 1 , P 3 , P 5 , P 6 .

[0075] The change points retained in each comparison process are used as the corrected change points, so P 1 , P 3 , P 5 , P 6 as the corrected change point.

[0076] 4. Abnormal fuel consumption condition detection steps: Divide the difference sequence into multiple non-overlapping and continuous intervals based on the corrected change points, and use linear regression method to fit the trend curve in each interval, and detect the abnormal fuel consumption condition of the ship in the corresponding interval according to the slope of the trend curve in each interval. Figure 1 The linear regression analysis shown shows the fuel consumption trend.

[0077] Use the corrected change point TCP to divide the data interval and perform linear regression on each interval. Apply linear regression analysis to the data around the change point to determine the start and end points of the abnormal fuel consumption. Divide the data into multiple intervals according to the corrected change point, and use linear regression to fit the trend line in each interval to observe the trend of the fuel consumption difference data in each interval. Specifically, divide the difference sequence into multiple non-overlapping and continuous intervals based on the corrected change point. For example, assuming that the time points corresponding to the corrected change points are 10, 20, 30, and 40, respectively, the TCP sequence is: TCP = {10 1 , 20, 30, 40}, then the first interval is from the starting time point to before the first change point, that is, [0,10]; the second interval is from after the first change point to before the second change point, that is, (10,20]; the third interval is from after the second change point to before the third change point, that is, (20,30]; the fourth interval is from after the third change point to before the fourth change point, that is, (30,40]; the last interval is from after the last change point to the last time point, that is, (40, end time point].

[0078] Then, the linear regression method is used in each interval (i.e., linear regression is performed on each interval) to fit the trend curve. The abnormal fuel consumption condition of the ship in the corresponding interval is detected according to the slope of the trend curve of each interval. A slope greater than 0 means that the value of the difference sequence Δ gradually increases (i.e., the values ​​of each difference in the difference sequence gradually increase), and the MRV reported fuel consumption curve and the AI ​​predicted fuel consumption curve gradually move away from each other. Therefore, the ship is in an over-consumption state, and the larger the slope, the more serious the over-consumption situation. On the contrary, when the slope is less than 0, it means that the value of the difference sequence Δ gradually decreases, and the two fuel consumption data curves gradually approach each other, and the ship is in a fuel-saving state. The larger the slope, the better the fuel-saving state of the ship. A slope close to 0 means that the ship's fuel consumption is in a stable state.

[0079] 5. Inference mechanism construction and visualization steps: Figure 1The inference mechanism and visualization shown in the figure are constructed to obtain the actual abnormal operating condition labels at each time point that are pre-marked, and extract the predicted labels for each interval from the detected abnormal fuel consumption conditions. The actual abnormal operating condition labels are compared with the predicted labels to obtain the number of predicted labels that are consistent with the abnormal operating condition labels, and then the accuracy is calculated to verify the accuracy of the detected abnormal fuel consumption conditions. In other words, the fuel saving and over-consumption table is calculated by combining the original AI predicted fuel consumption data and MRV data, and the above algorithm is checked again to accurately identify fuel saving, over-consumption and stable states; the inference mechanism is constructed based on the accuracy and combined with business scenarios and business logic, and the detected abnormal fuel consumption conditions are classified and explained through the inference mechanism, and a visual display is provided. Some results are shown in the figure below. Figure 3-6 shown.

[0080] The present invention also relates to a ship fuel consumption abnormal condition detection system based on the PELT algorithm, which corresponds to the above-mentioned ship fuel consumption abnormal condition detection method based on the PELT algorithm, and can be understood as a system for implementing the above-mentioned method. The system includes a parameter acquisition and calculation module, a fuel consumption time series construction and change point detection module, a change point correction module and a fuel consumption abnormal condition detection module connected in sequence. Specifically,

[0081] The parameter acquisition and calculation module acquires the MMSI code of a certain ship, and sequentially acquires the MRV reported fuel consumption and AI predicted fuel consumption of the ship at each time point in a certain time period in chronological order, and obtains the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence respectively; and then sequentially calculates the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption at each time point in the time period, and then obtains the difference sequence;

[0082] The fuel consumption time series construction and change point detection module uses the seven-point moving average method to smooth the difference series to obtain the smoothed fuel consumption time series; then uses the PELT algorithm to detect the change points of the fuel consumption time series to identify multiple change points;

[0083] The change point correction module performs hypothesis testing on each identified change point to obtain the significance probability of each change point, and sets multiple different overall significance levels. Based on each overall significance level and the number of change points, multiple corrected significance levels are calculated using the Bonferroni correction method, and then all significance probabilities are compared with each corrected significance level, and change points with a significance probability less than the corrected significance level are retained, and the change points retained in each comparison process are screened, and the change points retained in each comparison process are used as corrected change points;

[0084] The abnormal fuel consumption operating condition detection module divides the difference sequence into multiple non-overlapping and continuous intervals based on the corrected change points, and uses linear regression method to fit a trend curve in each interval, and detects the abnormal fuel consumption operating condition of the ship in the corresponding interval according to the slope of the trend curve in each interval.

[0085] Preferably, it also includes an inference mechanism construction and visualization module, which is connected to the above-mentioned abnormal fuel consumption condition detection module: obtaining the actual abnormal condition label at each time point that is pre-marked, and extracting the prediction label of each interval from the detected abnormal fuel consumption condition, comparing the actual abnormal condition label with the prediction label, and obtaining the number of predicted labels that are consistent with the abnormal condition label, and then calculating the accuracy rate to verify the accuracy of the detected abnormal fuel consumption condition; constructing an inference mechanism based on the accuracy rate and in combination with business scenarios and business logic, and classifying and interpreting the detected abnormal fuel consumption condition through the inference mechanism, and providing a visual display.

[0086] Preferably, in the fuel consumption abnormal condition detection module, the abnormal fuel consumption conditions include overconsumption and fuel saving; when the slope is greater than zero, it indicates that the ship is in an overconsumption state, and the larger the slope, the more serious the overconsumption; when the slope is less than zero, it indicates that the ship is in a fuel saving state, and the smaller the slope, the better the fuel saving state of the ship.

[0087] Preferably, in the fuel consumption time series construction and change point detection module, the seven-point moving average method is used to smooth the difference sequence, including: based on each difference in the difference sequence, multiple seven-point moving average values ​​of the difference sequence are calculated using the seven-point moving average method, and then the fuel consumption time series is constructed.

[0088] Preferably, in the parameter acquisition and calculation module, the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence are also preprocessed to ensure that the time intervals of the data points are consistent and aligned with the time axis, and the preprocessing includes data cleaning, alignment and interpolation.

[0089] The present invention provides an objective and scientific method and system for detecting abnormal fuel consumption conditions of ships based on the PELT algorithm. By using AI to predict fuel consumption data and MRV to report fuel consumption data, utilizing the efficient change point detection capability of the PELT algorithm, and combining statistical significance correction, it is possible to accurately identify abnormal fuel consumption conditions of various ship types under various navigation conditions, thereby providing reliable decision support for ship energy efficiency management and cost control.

[0090] It should be noted that the above-described specific implementations can enable those skilled in the art to more fully understand the invention, but do not limit the invention in any way. Therefore, although this specification has described the invention in detail with reference to the drawings and embodiments, those skilled in the art should understand that the invention can still be modified or replaced by equivalents. In short, all technical solutions and improvements that do not deviate from the spirit and scope of the invention should be included in the protection scope of the patent for the invention.

Claims

1. A method for detecting abnormal fuel consumption of a ship based on the PELT algorithm, characterized in that: The following steps are involved: Parameter acquisition and calculation steps: obtain the MMSI code of a certain ship, and obtain the MRV reported fuel consumption and AI predicted fuel consumption of the ship at each time point in a certain time period in chronological order, and obtain the MRV reported fuel consumption sequence and AI predicted fuel consumption sequence respectively; then calculate the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption at each time point in the time period, and then obtain the difference sequence; Steps for constructing the fuel consumption time series and detecting change points: Use the seven-point moving average method to smooth the difference series to obtain the smoothed fuel consumption time series; then use the PELT algorithm to detect change points in the fuel consumption time series and identify multiple change points; Change point correction step: conduct hypothesis tests on each identified change point to obtain the significance probability of each change point, set multiple different overall significance levels, calculate multiple corrected significance levels based on each overall significance level and the number of change points, and then compare all significance probabilities with each corrected significance level, retain change points with a significance probability less than the corrected significance level, and screen the change points retained in each comparison process, and use the change points retained in each comparison process as the corrected change points; Abnormal fuel consumption condition detection steps: Divide the difference sequence into multiple non-overlapping and continuous intervals based on the corrected change points, and use linear regression method to fit the trend curve in each interval. According to the slope of the trend curve in each interval, the abnormal fuel consumption condition of the ship in the corresponding interval is detected.

2. The method for detecting abnormal fuel consumption of a ship based on the PELT algorithm according to claim 1, characterized in that: After the abnormal fuel consumption condition detection step, it also includes the inference mechanism construction and visualization steps: obtaining the actual abnormal condition labels at each time point that are pre-marked, and extracting the predicted labels for each interval from the detected abnormal fuel consumption conditions, comparing the actual abnormal condition labels with the predicted labels, and obtaining the number of predicted labels that are consistent with the abnormal condition labels, and then calculating the accuracy to verify the accuracy of the detected abnormal fuel consumption conditions; constructing an inference mechanism based on the accuracy and in combination with business scenarios and business logic, and classifying and interpreting the detected abnormal fuel consumption conditions through the inference mechanism, and providing a visual display.

3. The method for detecting abnormal fuel consumption of a ship based on the PELT algorithm according to claim 1 or 2, characterized in that: In the abnormal fuel consumption condition detection step, the abnormal fuel consumption condition includes overconsumption and fuel saving; when the slope is greater than zero, it means that the ship is in an overconsumption state, and the larger the slope, the more serious the overconsumption; when the slope is less than zero, it means that the ship is in a fuel saving state, and the smaller the slope, the better the fuel saving state of the ship.

4. The method for detecting abnormal fuel consumption of a ship based on the PELT algorithm according to claim 1 or 2, characterized in that: In the fuel consumption time series construction and change point detection steps, the seven-point moving average method is used to smooth the difference sequence, including: based on each difference in the difference sequence, multiple seven-point moving average values ​​of the difference sequence are calculated using the seven-point moving average method, and then the fuel consumption time series is constructed.

5. The method for detecting abnormal fuel consumption of a ship based on the PELT algorithm according to claim 1 or 2, characterized in that: In the parameter acquisition and calculation module, the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence are also preprocessed to ensure that the time intervals of the data points are consistent and aligned with the time axis. The preprocessing includes data cleaning, alignment and interpolation.

6. A ship fuel consumption abnormal condition detection system based on PELT algorithm, characterized in that: It includes a parameter acquisition and calculation module, a fuel consumption time series construction and change point detection module, a change point correction module and a fuel consumption abnormal operating condition detection module, which are connected in sequence. The parameter acquisition and calculation module acquires the MMSI code of a certain ship, and sequentially acquires the MRV reported fuel consumption and AI predicted fuel consumption of the ship at each time point in a certain time period in chronological order, and obtains the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence respectively; and then sequentially calculates the difference between the AI ​​predicted fuel consumption and the MRV reported fuel consumption at each time point in the time period, and then obtains the difference sequence; The fuel consumption time series construction and change point detection module uses the seven-point moving average method to smooth the difference series to obtain the smoothed fuel consumption time series; then uses the PELT algorithm to detect the change points of the fuel consumption time series to identify multiple change points; The change point correction module performs hypothesis testing on each identified change point to obtain the significance probability of each change point, and sets multiple different overall significance levels. Based on each overall significance level and the number of change points, multiple corrected significance levels are calculated using the Bonferroni correction method, and then all significance probabilities are compared with each corrected significance level, and change points with a significance probability less than the corrected significance level are retained, and the change points retained in each comparison process are screened, and the change points retained in each comparison process are used as corrected change points; The abnormal fuel consumption operating condition detection module divides the difference sequence into multiple non-overlapping and continuous intervals based on the corrected change points, and uses linear regression method to fit a trend curve in each interval, and detects the abnormal fuel consumption operating condition of the ship in the corresponding interval according to the slope of the trend curve in each interval.

7. The ship fuel consumption abnormal condition detection system based on PELT algorithm according to claim 6 is characterized in that: It also includes an inference mechanism construction and visualization module, which is connected to the abnormal fuel consumption condition detection module and is used to obtain the actual abnormal condition label at each time point that is pre-marked, and extract the prediction label of each interval from the detected abnormal fuel consumption condition, compare the actual abnormal condition label with the prediction label, and obtain the number of predicted labels that are consistent with the abnormal condition label, and then calculate the accuracy rate to verify the accuracy of the detected abnormal fuel consumption condition; construct an inference mechanism based on the accuracy rate and in combination with business scenarios and business logic, and classify and interpret the detected abnormal fuel consumption conditions through the inference mechanism, and provide a visual display.

8. The ship fuel consumption abnormal condition detection system based on PELT algorithm according to claim 6 or 7, characterized in that: In the abnormal fuel consumption condition detection module, the abnormal fuel consumption conditions include overconsumption and fuel saving; when the slope is greater than zero, it means that the ship is in an overconsumption state, and the larger the slope, the more serious the overconsumption; when the slope is less than zero, it means that the ship is in a fuel saving state, and the smaller the slope, the better the fuel saving state of the ship.

9. The ship fuel consumption abnormal condition detection system based on PELT algorithm according to claim 6 or 7, characterized in that: In the fuel consumption time series construction and change point detection module, the seven-point moving average method is used to smooth the difference sequence, including: based on each difference in the difference sequence, multiple seven-point moving average values ​​of the difference sequence are calculated using the seven-point moving average method, and then the fuel consumption time series is constructed.

10. The ship fuel consumption abnormal condition detection system based on PELT algorithm according to claim 6 or 7, characterized in that: In the parameter acquisition and calculation module, the MRV reported fuel consumption sequence and the AI ​​predicted fuel consumption sequence are also preprocessed to ensure that the time intervals of the data points are consistent and aligned with the time axis. The preprocessing includes data cleaning, alignment and interpolation.

Images