Harbor district weighbridge anomaly detection method based on multi-source data cross validation

Abstract

The invention discloses a harbor district wagon balance anomaly detection method based on multi-source data cross validation, and the method comprises the steps: monitoring the tare data of the same transport vehicle on different wagon balances, calculating the tare difference of the multiple wagon balances according to the tare data, calculating the transverse anomaly signal intensity when the tare difference of the multiple wagon balances is greater than a transverse dynamic threshold value, and carrying out the cross validation of the transverse anomaly signal intensity. Calculating tare weight data of the same transport vehicle on the same wagon balance, if a deviation of a current prediction error relative to a historical reference exceeds a preset reasonable threshold value, generating a stability score indicating the wagon balance, representing a long-term stability evaluation result of the wagon balance through the stability score, and then constructing a Softmax probability fusion model to obtain a long-term stability evaluation result of the wagon balance. After the stability score is converted into a probability form, the maximum posterior probability principle is adopted to position a fault wagon balance and give an alarm, and by carrying out transverse cross validation and longitudinal trend analysis on the weighing data of the transport vehicle, real-time, automatic, low-cost and positionable anomaly detection on the harbor district multi-wagon balance system is realized.

Description

Technical Field

[0001] This invention relates to the field of port management technology, and in particular to a method for detecting anomalies in port weighbridges based on cross-validation of multi-source data. Background Technology

[0002] As the core weighing equipment used for weighing trucks and cargo in port areas, weighbridges (truck scales) directly affect the accuracy and fairness of trade settlements, and have a crucial impact on the economic interests of both port operators and cargo owners. However, the existing accuracy assurance system for port weighbridges still has significant shortcomings, mainly relying on the following two methods: First, the currently prevalent periodic verification or calibration mechanism is carried out by metrology departments or equipment manufacturers according to fixed cycles. This method is a typical reactive maintenance model, which cannot provide real-time assessment of the equipment's health status within the verification cycle. During long-term, high-frequency use, weighbridges are prone to slow degradation phenomena such as zero-point drift and range drift. Once deviations accumulate within the verification cycle, a large amount of erroneous data may be used for billing and settlement, leading to potential disputes and economic losses.

[0003] Secondly, another common practice in the industry is to only carry out repairs after obvious weighing anomalies occur or customer complaints are received. However, this approach is more reactive, usually initiating repairs only after the anomaly has already caused actual impact or even losses. It lacks the ability to detect and intervene proactively in advance, making it difficult to meet the port's operational needs for high efficiency and accuracy.

[0004] In summary, existing weighbridge operation and maintenance models generally suffer from problems such as insufficient real-time performance, lack of monitoring methods, difficulty in detecting anomalies early, and difficulty in locating fault points in multiple weighbridge systems. Summary of the Invention

[0005] In view of this, the present invention proposes a port weighbridge anomaly detection method based on multi-source data cross-validation, which can realize active monitoring, real-time diagnosis, automatic early warning, and intelligent detection of faulty equipment in multiple weighbridge systems.

[0006] The technical solution of this invention is implemented as follows: A method for detecting anomalies in port area weighbridges based on multi-source data cross-validation includes the following steps: Step S1: Monitor the tare weight data of transport vehicles on different weighbridges within a preset time interval, and obtain the tare weight difference of multiple weighbridges based on the tare weight data; Step S2: Set a lateral dynamic threshold based on the vehicle's natural fluctuations and instrument noise. When the difference in tare weight among multiple weighbridges exceeds the lateral dynamic threshold, calculate the lateral abnormal signal intensity. Step S3: Based on the historical tare weight sequence of transport vehicles on the same weighbridge, construct a time series prediction model implemented by a feedforward neural network and obtain the expected tare weight value; Step S4: Define the current prediction error based on the expected tare weight and the current tare weight data, extract the most recent prediction error sequence of the weighbridge from the system database, and construct a longitudinal dynamic threshold after calculating the mean and standard deviation of the most recent prediction error sequence. Step S5: If the current prediction error is greater than the longitudinal dynamic threshold, construct the stability score of the weighbridge based on the statistical distribution of historical prediction errors. Step S6: Calculate the fusion score based on the stability score of the weighbridge and the intensity of the lateral abnormal signal, and convert the fusion score into a probabilistic form using the Softmax function; Step S7: Using the maximum a posteriori probability principle, the faulty weighbridge is located based on the fusion score in probabilistic form, and an abnormal alarm is triggered.

[0007] Preferably, the expression for the tare weight difference among multiple weighbridges in step S1 is: ; in For transport vehicle i in time Tare weight data obtained from the weighbridge at that time. For transport vehicle i in time Tare weight data obtained from the weighbridge at that time. To account for differences in tare weights across multiple weighbridges, the preset time interval is [time interval]. .

[0008] Preferably, the expression for the lateral dynamic threshold in step S2 is: ; in This is a horizontal dynamic threshold. The basic deviation tolerance parameter is set manually according to the weighbridge manufacturing standards. These are the fluctuation response parameters, which are manually configured according to on-site requirements. This represents the standard deviation based on recent tare weight differences. The expression for the intensity of the transverse anomaly signal is: ; in For transverse anomalous signal strength, This is due to differences in tare weight across multiple weighbridges.

[0009] Preferably, the specific steps of step S3 are as follows: Obtain the current tare weight values ​​of transport vehicle i at weighbridge p and weighbridge q. as well as A time series prediction model based on historical tare weight sequences and implemented using a feedforward neural network was constructed. ; in and Let be the expected tare weights of transport vehicle i with respect to weighbridge p and weighbridge q, respectively. This is a nonlinear autoregressive prediction function implemented using a multilayer perceptron structure. This represents the historical tare weight sequence of transport vehicle i on weighbridge p. Let n be the historical tare weight sequence of transport vehicle i on weighbridge q, and n be the length of the historical window. This is an auxiliary feature that includes the number of vehicle axles and the vehicle type.

[0010] Preferably, the expression for the current prediction error in step S4 is: ; in and These represent the current prediction errors of transport vehicle i with respect to weighbridge p and weighbridge q, respectively.

[0011] Preferably, the construction step of the longitudinal dynamic threshold in step S4 is as follows: Extract the most recent prediction error sequences of weighbridge p and weighbridge q from the system database, and calculate their mean values ​​respectively. and The mean and The expression is: ; The standard deviation was calculated based on the most recent prediction error sequence and the mean. and The standard deviation and The expression is: ; in and , respectively, are the most recent prediction error sequences of weighbridge p and weighbridge q extracted from the database, where M is the length of the sliding statistical window. and The expression is: ; The longitudinal dynamic threshold is calculated based on the mean and standard deviation, and the expression for the longitudinal dynamic threshold is pseudo: ; in and are the longitudinal dynamic thresholds for weighbridge p and weighbridge q, respectively, and k is the anomaly sensitivity coefficient.

[0012] Preferably, the expression for the stability score of the weighbridge in step S5 is: ; in and These are the stability scores for weighbridge p and weighbridge q, respectively. To calculate the median, and This indicates that transport vehicle i was recently continuously weighed on weighbridges p and q. The historical prediction error sequence recorded from each weighing.

[0013] Preferably, the expression for the fusion score in step S6 is: ; in and These are the combined scores for weighbridge p and weighbridge q, respectively. and These represent the weights of stability score and lateral anomaly intensity in the judgment.

[0014] Preferably, in step S6, the fusion score is converted into a probabilistic expression using the Softmax function as follows: ; in and Fusion scores and The probability form.

[0015] Preferably, step S7 uses the maximum a posteriori probability principle to filter and locate faulty weighbridges based on the probabilistic fusion score, as expressed in the following expression: ; Where arg max represents the variable that makes the expression reach its maximum value.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. In terms of data utilization, this invention creatively transforms the high-frequency vehicle tare weight data naturally generated during port operations into monitoring signals of the weighbridge's health status, achieving the equipment's self-monitoring capability without the need for additional sensors or additional data collection costs. The data source is stable, has wide coverage, and is updated frequently, giving the system a natural advantage of all-weather, continuous monitoring.

[0017] 2. Regarding the detection method, this invention proposes a two-level detection framework that combines horizontal cross-validation and vertical trend analysis. Horizontal detection, by comparing the tare weight of the same vehicle on different weighbridges in real time, can promptly detect sudden and significant deviations. Vertical detection, by performing trend modeling and dynamic threshold analysis on the historical tare weight data of the same weighbridge, can effectively capture the slow drift and stability decline of the equipment over time. This combined mechanism can cover the main types of weighbridge anomalies and achieve simultaneous control of sudden and gradual failures, which is an important extension of the traditional single comparison method.

[0018] 3. In terms of fault location, a precise location method based on fusion reasoning mechanism is proposed. When multiple weighbridges have measurement contradictions, the system uses the long-term stability of the weighbridges as a confidence weight and performs comprehensive analysis with the lateral abnormal signals. In this way, the most likely fault point can be accurately identified under the condition of mutual interference between multiple devices. This mechanism breaks through the limitation of traditional weighbridge systems that cannot locate which weighbridge is malfunctioning, and achieves a qualitative improvement from fuzzy judgment to precise location.

[0019] 4. In terms of operation and maintenance mode, this invention utilizes real-time comparison, historical modeling and continuous monitoring technologies to build a 24 / 7 proactive monitoring system, which can provide early warnings before significant deviations occur in equipment, significantly improving the timeliness, effectiveness and intelligence of equipment maintenance, and realizing the transformation from passive handling to proactive prevention. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a port weighbridge anomaly detection method based on multi-source data cross-validation according to the present invention. Detailed Implementation

[0022] To better understand the technical content of this invention, a specific embodiment is provided below, and the invention will be further described in conjunction with the accompanying drawings.

[0023] See Figure 1 The present invention provides a method for detecting anomalies in port weighbridges based on multi-source data cross-validation, comprising the following steps: Step S1: Monitor the tare weight data of transport vehicles on different weighbridges within a preset time interval, and obtain the tare weight difference across multiple weighbridges based on the tare weight data. The expression for the tare weight difference across multiple weighbridges is as follows: ; in For transport vehicle i in time Tare weight data obtained from the weighbridge at that time. For transport vehicle i in time Tare weight data obtained from the weighbridge at that time. To account for differences in tare weights across multiple weighbridges, the preset time interval is [time interval]. .

[0024] Step S2: Based on the vehicle's natural fluctuations and instrument noise, set a lateral dynamic threshold. When the tare weight difference among multiple weighbridges exceeds the lateral dynamic threshold, calculate the lateral abnormal signal intensity. The expression for the lateral dynamic threshold is: ; in This is a horizontal dynamic threshold. The basic deviation tolerance parameter represents a fixed baseline tolerance applicable to all vehicles, set manually according to the weighbridge manufacturing standards. This is the fluctuation response parameter, used to determine the degree to which the lateral dynamic threshold responds to recent tare weight differences. It is usually set to (1, 1.5, 2) and can be manually configured according to on-site requirements. This represents the standard deviation based on recent tare weight differences. when > When this occurs, it indicates that at least one of the weighbridges p and q is abnormal. The strength of the lateral abnormality signal can then be calculated, specifically as follows: ; in For transverse anomalous signal strength, This is due to differences in tare weight across multiple weighbridges.

[0025] Steps S1-S2 are real-time cross-verification of the same vehicle across different weighbridges, used to discover instantaneous significant differences between the same transport vehicle and different weighbridges. The intensity of the generated lateral anomaly signal serves as an important input for subsequent fault location.

[0026] Step S3: Based on the historical tare weight sequence of transport vehicles on the same weighbridge, construct a time series prediction model implemented by a feedforward neural network and obtain the expected tare weight value. The specific steps are as follows: Obtain the current tare weight values ​​of transport vehicle i at weighbridge p and weighbridge q. as well as A time series prediction model based on historical tare weight sequences and implemented using a feedforward neural network was constructed. ; in and Let be the expected tare weights of transport vehicle i with respect to weighbridge p and weighbridge q, respectively. This is a nonlinear autoregressive prediction function implemented using a multilayer perceptron structure. This represents the historical tare weight sequence of transport vehicle i on weighbridge p. Let n be the historical tare weight sequence of transport vehicle i on weighbridge q, and n be the length of the historical window. The auxiliary features, including the number of vehicle axles and vehicle type, can be automatically obtained through the port's existing information system. The number of vehicle axles can be automatically identified and recorded by the license plate recognition system, laser scanner, or ground sensor at the port's entry gate when the vehicle passes through the gate. The vehicle type can be obtained by the vehicle classification function in the license plate recognition system or the automatic recognition algorithm of the entrance monitoring image, and bound to the vehicle's unique identifier (license plate number). All features are collected and written into the database when the vehicle enters the port area, and are automatically associated with the tare weight record when the vehicle is weighed later, thus forming a complete auxiliary feature vector to improve the stability and accuracy of the trend prediction model.

[0027] Step S4: Define the current prediction error based on the expected tare weight and the current tare weight data. Extract the most recent prediction error sequence of the weighbridge from the system database, and construct a longitudinal dynamic threshold after calculating the mean and standard deviation of the most recent prediction error sequence. The expression for the current prediction error is: ; in and These represent the current prediction errors of transport vehicle i with respect to weighbridge p and weighbridge q, respectively.

[0028] To achieve adaptive and statistically significant trend detection, this invention further constructs a vertical dynamic threshold based on the most recent prediction error sequence of the weighbridge. The specific construction steps are as follows: Extract the most recent prediction error sequences of weighbridge p and weighbridge q from the system database, and calculate their mean values ​​respectively. and The mean and The expression is: ; The standard deviation was calculated based on the most recent prediction error sequence and the mean. and The standard deviation and The expression is: ; in and These are the most recent prediction error sequences for weighbridge p and weighbridge q extracted from the database, respectively. M is the length of the sliding statistical window, which can be set according to factors such as port business volume and vehicle passage probability. and The expression is: ; The longitudinal dynamic threshold is calculated based on the mean and standard deviation, and the expression for the longitudinal dynamic threshold is pseudo: ; in and These are the longitudinal dynamic thresholds for weighbridge p and weighbridge q, respectively. k is the anomaly sensitivity coefficient, used to adjust the sensitivity of the alarm. It usually ranges from 1.0 to 3.0 and can be configured according to the weighbridge equipment level and business requirements.

[0029] Step S5: If the current prediction error is greater than the longitudinal dynamic threshold, i.e. > as well as > When a longitudinal anomaly alarm is triggered, indicating a potential decrease in the stability of the weighbridge, a stability score for the weighbridge is constructed based on the statistical distribution of historical prediction errors. The expression for the weighbridge's stability score is as follows: ; in and These are the stability scores for weighbridge p and weighbridge q, respectively. To calculate the median, compared to the mean, the median is less sensitive to extreme outliers and more robustly reflects the typical deviation level of the error sequence. The calculated stability score serves as the input for subsequent fault location reliability. and This indicates that transport vehicle i was recently continuously weighed on weighbridges p and q. The historical prediction error sequence recorded from each weighing. , This sequence is used to characterize the error fluctuations of the weighbridge in recent operation and is the basic data for calculating the long-term stability index of the weighbridge.

[0030] Steps S3-S5 are the historical trend analysis process for the same vehicle and the same weighbridge, which is used to identify the slow drift of a single weighbridge over time. By taking the current tare weight value of the same transport vehicle on a single weighbridge, the historical tare weight data of the vehicle on that weighbridge is retrieved to establish a reference benchmark. The current tare weight value is compared with the historical reference benchmark. If the deviation of the current value from the historical reference benchmark exceeds a preset reasonable threshold, an alarm signal indicating that the weighbridge may have an abnormal stability is generated.

[0031] Step S6: Calculate the fusion score based on the stability score of the weighbridge and the intensity of the lateral abnormal signal, and convert the fusion score into a probabilistic form using the Softmax function. The expression for the fusion score is: ; in and These are the combined scores for weighbridge p and weighbridge q, respectively. and These are the weights of stability score and lateral anomaly intensity in the judgment, respectively, and are preset based on historical testing experience and business needs.

[0032] When a discrepancy in tare weight is indicated between weighbridge p and weighbridge q during steps S1-S2, stability scores are generated for weighbridge p and weighbridge q in steps S3-S5. A lower stability score indicates a larger recent prediction error and a higher long-term drift risk for the weighbridge. Therefore, in step S6... Preferably, in step S6, the fusion score is converted into a probabilistic expression using the Softmax function as follows: ; in and Fusion scores and The probability form.

[0033] Step S7: Using the maximum a posteriori probability principle, the faulty weighbridge is located based on the probabilistic fusion score, and an anomaly alarm is triggered. The expression for using the maximum a posteriori probability principle to locate the faulty weighbridge based on the probabilistic fusion score is as follows: ; Where arg max represents the variable that makes the expression reach its maximum value; specifically, this operation considers two probability values. and By comparing the weighbridges, the system selects the weighbridge with the highest probability as the final result. The weighbridge with the highest probability is marked as the faulty weighbridge, and an alarm for precise location is generated. By jointly modeling the lateral sudden deviation and the longitudinal long-term stability, the system can achieve stable, reliable and quantifiable fault location when multiple weighbridges have contradictory data at the same time, which significantly improves the monitoring accuracy and avoids the problem of false alarms or missed alarms caused by the judgment of a single indicator in traditional methods.

[0034] This invention discloses a port weighbridge anomaly detection method based on multi-source data cross-validation. When anomalies are detected in the tare data of the same transport vehicle on different weighbridges, long-term stability scores are generated for both weighbridges. By comparing the long-term stability of the two weighbridges, the weighbridge with poor long-term stability, i.e., a large deviation from the historical benchmark, is identified as the more likely fault point, thereby generating a fault alarm that is precisely located to the specific weighbridge. By utilizing the massive amount of vehicle weighing data naturally generated in the daily operation of the port area, horizontal cross-validation and vertical trend analysis are performed. Through multi-level data comparison and verification, and through technological innovations such as data reconstruction, intelligent analysis, and fusion reasoning, early detection and precise location of weighbridge anomalies are achieved. This enables real-time, automated, low-cost, and location-based anomaly detection for multi-weighbridge systems in the port area, providing systematic technical support for the safety, reliability, and efficient operation of the port's metering system.

[0035] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting anomalies in port area weighbridges based on multi-source data cross-validation, characterized in that, Includes the following steps: Step S1: Monitor the tare weight data of transport vehicles on different weighbridges within a preset time interval, and obtain the tare weight difference of multiple weighbridges based on the tare weight data; Step S2: Set a lateral dynamic threshold based on the vehicle's natural fluctuations and instrument noise. When the difference in tare weight among multiple weighbridges exceeds the lateral dynamic threshold, calculate the lateral abnormal signal intensity. Step S3: Based on the historical tare weight sequence of transport vehicles on the same weighbridge, construct a time series prediction model implemented by a feedforward neural network and obtain the expected tare weight value; Step S4: Define the current prediction error based on the expected tare weight and the current tare weight data, extract the most recent prediction error sequence of the weighbridge from the system database, and construct a longitudinal dynamic threshold after calculating the mean and standard deviation of the most recent prediction error sequence. Step S5: If the current prediction error is greater than the longitudinal dynamic threshold, construct the stability score of the weighbridge based on the statistical distribution of historical prediction errors. Step S6: Calculate the fusion score based on the stability score of the weighbridge and the intensity of the lateral abnormal signal, and convert the fusion score into a probabilistic form using the Softmax function; Step S7: Using the maximum a posteriori probability principle, the faulty weighbridge is located based on the fusion score in probabilistic form, and an abnormal alarm is triggered.

2. The method for detecting anomalies in port area weighbridges based on multi-source data cross-validation as described in claim 1, characterized in that, The expression for the tare weight difference among multiple weighbridges in step S1 is: ； in For transport vehicle i in time Tare weight data obtained from the weighbridge at that time. For transport vehicle i in time Tare weight data obtained from the weighbridge at that time. To account for differences in tare weight across multiple weighbridges, the preset time interval is [time interval]. .

3. The method for detecting anomalies in port area weighbridges based on multi-source data cross-validation according to claim 2, characterized in that, The expression for the lateral dynamic threshold in step S2 is: ； in This is a horizontal dynamic threshold. The basic deviation tolerance parameter is set manually according to the weighbridge manufacturing standards. These are the fluctuation response parameters, which are manually configured according to on-site requirements. This represents the standard deviation based on recent tare weight differences. The expression for the intensity of the transverse anomaly signal is: ； in For transverse anomalous signal strength, This is due to differences in tare weight across multiple weighbridges.

4. The method for detecting anomalies in port area weighbridges based on multi-source data cross-validation according to claim 3, characterized in that, The specific steps of step S3 are as follows: Obtain the current tare weight values ​​of transport vehicle i at weighbridge p and weighbridge q. as well as A time series prediction model based on historical tare weight sequences and implemented using a feedforward neural network was constructed. ； in and Let be the expected tare weights of transport vehicle i with respect to weighbridge p and weighbridge q, respectively. This is a nonlinear autoregressive prediction function implemented using a multilayer perceptron structure. This represents the historical tare weight sequence of transport vehicle i on weighbridge p. Let n be the historical tare weight sequence of transport vehicle i on weighbridge q, and n be the length of the historical window. This is an auxiliary feature that includes the number of vehicle axles and the vehicle type.

5. The method for detecting anomalies in port area weighbridges based on multi-source data cross-validation as described in claim 4, characterized in that, The expression for the current prediction error in step S4 is: ； in and These represent the current prediction errors of transport vehicle i with respect to weighbridge p and weighbridge q, respectively.

6. The method for detecting anomalies in port area weighbridges based on multi-source data cross-validation according to claim 5, characterized in that, The steps for constructing the longitudinal dynamic threshold in step S4 are as follows: Extract the most recent prediction error sequences of weighbridge p and weighbridge q from the system database, and calculate their mean values ​​respectively. and The mean and The expression is: ； The standard deviation was calculated based on the most recent prediction error sequence and the mean. and The standard deviation and The expression is: ； in and , respectively, are the most recent prediction error sequences of weighbridge p and weighbridge q extracted from the database, where M is the length of the sliding statistical window. and The expression is: ； The longitudinal dynamic threshold is calculated based on the mean and standard deviation, and the expression for the longitudinal dynamic threshold is pseudo: ； in and are the longitudinal dynamic thresholds for weighbridge p and weighbridge q, respectively, and k is the anomaly sensitivity coefficient.

7. The method for detecting anomalies in port area weighbridges based on multi-source data cross-validation as described in claim 6, characterized in that, The expression for the stability score of the weighbridge in step S5 is: ； in and These are the stability scores for weighbridge p and weighbridge q, respectively. To calculate the median, and This indicates that transport vehicle i was recently continuously weighed on weighbridges p and q. The historical prediction error sequence recorded from each weighing.

8. The method for detecting anomalies in port area weighbridges based on multi-source data cross-validation as described in claim 7, characterized in that, The expression for the fusion score in step S6 is: ； in and These are the combined scores for weighbridge p and weighbridge q, respectively. and These represent the weights of stability score and lateral anomaly intensity in the judgment.

9. The method for detecting anomalies in port area weighbridges based on multi-source data cross-validation as described in claim 8, characterized in that, In step S6, the fusion score is converted into a probabilistic expression using the Softmax function as follows: ； in and Fusion scores and The probability form.

10. A method for detecting anomalies in port area weighbridges based on multi-source data cross-validation as described in claim 9, characterized in that, The expression for filtering and locating faulty weighbridges in step S7 based on the maximum a posteriori probability principle and the fusion score in probabilistic form is as follows: ； Where arg max represents the variable that makes the expression reach its maximum value.

Images