A smart monitoring and management system for unloading oil from railway tank cars based on the Internet of Things

Abstract

This invention relates to the field of oil unloading monitoring technology and discloses an IoT-based intelligent oil unloading monitoring and management system for railway tank cars. The system includes: a data acquisition module for acquiring status parameters and image data; a data filtering module for filtering to obtain a steady-state operation window; a data calculation module for calculating hysteresis vectors, endpoint drift vectors, and reverse coupling ratios; a Gini coefficient calculation module for calculating differential pressure and flow rate Gini coefficients; a preliminary judgment module for making a preliminary judgment on safety based on the reverse coordination correlation coefficient; and a state category identification module for obtaining the state category of the railway tank car through a state identification model. This invention obtains a steady-state operation window through data filtering and extracts risk features closely related to the state of the railway tank car for the state identification model to accurately identify the state category of the railway tank car. This achieves intelligent management of the entire process from data acquisition to state identification, improving the accuracy and efficiency of risk identification and ensuring the safety of oil unloading operations.

Description

Technical Field

[0001] This invention relates to the field of oil unloading monitoring technology, and more specifically, to an intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things. Background Technology

[0002] In the unloading of oil from railway tank cars, bottom unloading is often used, and unloading arms and flexible hoses are provided. During the operation, trackside protection such as turnout locking and derailers is required to ensure safety. However, in situations such as jacking of trains, rolling on a small slope, or other cars accidentally entering the train, low-speed micro-displacement may still occur. At this time, the unloading arm and hose first absorb the displacement through geometric compensation, and then enter the stretching and bending stage, which leads to an increase in pipeline resistance. This is manifested as a slow increase in pressure differential, a synchronous decrease in flow rate, and low-frequency step fluctuations.

[0003] With the development of intelligent oil unloading technology, although real-time data such as differential pressure and flow rate are collected through IoT devices, existing monitoring systems have significant shortcomings. Traditional monitoring relies heavily on manual inspections or single parameter thresholds, such as alarms triggered by sudden drops in flow rate or increases in differential pressure. However, normal disturbances during oil unloading from railway tank cars, such as pump pressure fluctuations and changes in fluid viscosity, can easily cause false alarms. Furthermore, the initial parameter changes during slight traction are small and difficult to capture with single thresholds, often triggering alarms only after traction intensifies, posing a risk of missed alarms. Simultaneously, the raw data is a continuous time series containing a large amount of redundant information. Direct modeling is computationally intensive and requires high dimensionality. Traditional monitoring does not specifically extract features, making it difficult to uncover correlations between parameters, such as the inverse relationship between flow rate and differential pressure, or the time lag characteristics of parameter changes. This makes it difficult to distinguish between normal fluctuations and traction risks, resulting in low accuracy in oil unloading monitoring, which not only poses safety hazards but also increases equipment maintenance costs.

[0004] Therefore, there is an urgent need for a railway tank car oil unloading monitoring system to solve the above problems. Summary of the Invention

[0005] This invention provides an Internet of Things-based intelligent oil unloading monitoring and management system for railway tank cars, which solves the technical problems mentioned in the background.

[0006] This invention provides an IoT-based intelligent oil unloading monitoring and management system for railway tank cars, comprising:

[0007] The data acquisition module is used to collect the status parameters and image data of railway tank cars, and preprocess them to obtain differential pressure sequence, flow rate sequence, wheel displacement sequence and unloading arm angle sequence;

[0008] Status parameters include: upstream pressure, downstream pressure, volumetric flow rate, pump speed, inlet valve position, outlet valve position, and unloading arm angle of the main unloading pipeline of the railway tank car at the target unloading station;

[0009] The data filtering module is used to filter and obtain a steady-state operation window based on the discrete state combination of the state parameters of the railway tank car, and retain the pressure difference sequence, flow rate sequence, wheel displacement sequence and unloading arm angle sequence within the window.

[0010] The data calculation module is used to perform mean-removal processing on the pressure difference sequence and flow rate sequence within the steady-state window, thereby calculating the hysteresis vector, endpoint drift vector, and reverse coupling ratio.

[0011] The hysteresis vector consists of flow hysteresis, wheel displacement hysteresis, and unloading arm angle hysteresis.

[0012] The endpoint drift vector consists of pressure differential drift, flow rate drift, wheel displacement drift, and unloading arm angle drift.

[0013] The Gini coefficient calculation module is used to calculate the differential pressure Gini coefficient and the flow rate Gini coefficient based on the differential pressure sequence and flow rate sequence within the steady-state window.

[0014] The preliminary judgment module calculates the inverse cooperative correlation coefficient based on the pressure difference sequence and flow rate sequence within the steady-state window. If the coefficient is less than the safety threshold, the railway tank car is considered safe. Otherwise, it is concatenated with the hysteresis vector, endpoint drift vector, inverse coupling ratio, pressure difference Gini coefficient, and flow rate Gini coefficient to obtain the feature vector.

[0015] The state category recognition module is used to input feature vectors into the state recognition model and output the state category of the railway tank car;

[0016] The status categories include: safe status, minor traction, severe traction, and mechanical sway.

[0017] Furthermore, each sequence unit of the pressure differential sequence is represented by the difference between the upstream and downstream pressures at a given time point; each sequence unit of the flow rate sequence is represented by the volumetric flow rate at a given time point; each sequence unit of the wheel displacement sequence is represented by the product of the pixel displacement and the calibration coefficient, where the pixel displacement is equal to the difference between the pixel position of the wheel in the image data at the current time point and the pixel position of the wheel in the image data at the initial time point, and the calibration coefficient is the ratio of the known length of the actual object on site to the pixel length of the actual object in the image data; and each sequence unit of the unloading arm angle sequence is represented by the unloading arm angle at a given time point.

[0018] Furthermore, a sliding window is applied to the pump speed, inlet valve position, and outlet valve position at each time point. If the difference between the maximum and minimum pump speed values ​​within the sliding window is less than a first threshold, the difference between the maximum and minimum inlet valve positions within the sliding window is less than a second threshold, and the difference between the maximum and minimum outlet valve positions within the sliding window is less than a third threshold, then that time point is included in the steady-state operation window. The differential pressure sequence, flow rate sequence, wheel displacement sequence, and unloading arm angle sequence within that window are retained. The length of the sliding window, the first threshold, the second threshold, and the third threshold are all user-defined parameters.

[0019] Furthermore, the average values ​​of the pressure difference sequence and flow rate sequence within the steady-state window are first calculated, and then the corresponding average value is subtracted from each sequence unit of the pressure difference sequence and flow rate sequence to complete the mean removal process.

[0020] Furthermore, flow lag The calculation formula is as follows:

[0021] ,in Indicates the first The product of the flow rate This represents the independent variable corresponding to the maximum value of the internal product of the flow;

[0022] No. The product of flow The calculation formula is as follows:

[0023] Where Col represents the set of sequence units in the steady-state window, This represents the k-th sequence unit of the pressure difference sequence after mean removal. Represents the first value of the flow sequence after mean removal processing. Sequence units, Represents integer lag. Its initial value is 0, and it is a custom parameter;

[0024] First, calculate the inner product between the pressure difference sequence and the wheel displacement sequence within the steady-state window to obtain the inner product of wheel displacement. Then, take the independent variable corresponding to the maximum value of the inner product of wheel displacement to obtain the wheel displacement hysteresis.

[0025] First, calculate the inner product between the pressure difference sequence and the unloading arm angle sequence within the steady-state window to obtain the inner product of the unloading arm angle. Then, take the independent variable corresponding to the maximum value of the inner product of the unloading arm angle to obtain the hysteresis of the unloading arm angle.

[0026] Furthermore, the difference between the last sequence unit and the first sequence unit of the pressure difference sequence, flow rate sequence, wheel displacement sequence, and unloading arm angle sequence within the steady-state window is calculated to obtain the pressure difference drift, flow rate drift, wheel displacement drift, and unloading arm angle drift; the negative value of the flow rate endpoint drift is divided by the pressure difference endpoint drift to obtain the reverse coupling ratio.

[0027] Furthermore, the difference between adjacent sequence units of the pressure difference sequence within the steady-state window is calculated to obtain the first-order increment sequence of the pressure difference, wherein the first sequence unit of the first-order increment sequence of the pressure difference is equal to the difference between the second sequence unit and the first sequence unit of the pressure difference sequence within the steady-state window. The first-order increment sequence of the flow rate is then calculated in the same way. The absolute values ​​of the sequence units of the first-order increment sequence of the pressure difference are then sorted in ascending order to obtain the ascending order sequence of the pressure difference, and the ascending order sequence of the flow rate is calculated in the same way.

[0028] The differential pressure Gini coefficient is calculated based on the ascending sequence of differential pressure, and the flow rate Gini coefficient is calculated using the same method. The differential pressure Gini coefficient... The calculation formula is as follows:

[0029] Where N represents the length of the pressure differential ascending sequence, and These represent the i-th and j-th sequence units of the pressure difference ascending sequence, respectively.

[0030] Furthermore, the safety threshold is a user-defined parameter; the formula for calculating the reverse cooperative correlation coefficient is as follows:

[0031] Where Col represents the set of sequence units in the steady-state window, and These represent the k-th sequence unit of the pressure difference sequence and flow rate sequence after the mean was removed, respectively.

[0032] Furthermore, the state recognition model is built based on a multilayer perceptron, consisting of an input layer, a hidden layer, and an output layer. The input layer takes in the feature vector, the hidden layer consists of two hidden units, both of which are activated by the GELU activation function, and the output layer is activated by the Softmax activation function. It outputs the probability that the railway tank car is in a safe state, under slight traction, under severe traction, or under mechanical sway, and selects the state category with the highest probability for output.

[0033] Furthermore, by collecting state parameters and image data of railway tank cars under safe, light traction, heavy traction, and mechanical oscillation conditions, and constructing feature vectors, these data are used as training samples for training the state recognition model.

[0034] The beneficial effects of this invention are as follows: First, by obtaining a steady-state operating window through data filtering, this invention avoids data fluctuation interference and provides a reliable data foundation for subsequent analysis. Second, by calculating key parameters such as hysteresis vectors and endpoint drift vectors, risk characteristics are extracted from multiple dimensions such as time correlation and trend changes, eliminating redundancy in the original data and reducing data dimensionality. Third, by combining the preliminary judgment of the reverse cooperative correlation coefficient and safety threshold, the number of calls to the state recognition model is reduced, thereby reducing the risk of false alarms. Finally, the state recognition model based on multilayer sensors accurately identifies states such as safety and slight traction. Thus, it realizes intelligentization of the entire process from data acquisition to state recognition, improves the accuracy and efficiency of risk identification, ensures the safety of oil unloading operations, and reduces equipment maintenance costs and management difficulty. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of an intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things according to the present invention;

[0036] Figure 2 This is a schematic diagram of the state recognition model of the present invention.

[0037] In the diagram: Data acquisition module 101, data filtering module 102, data calculation module 103, Gini coefficient calculation module 104, preliminary judgment module 105, and state category identification module 106. Detailed Implementation

[0038] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0039] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" indicate that the element or object preceding the term encompasses the elements or objects listed following the term and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0040] like Figures 1-2 As shown, an IoT-based intelligent oil unloading monitoring and management system for railway tank cars includes:

[0041] The data acquisition module 101 is used to acquire the status parameters and image data of the railway tank car, and preprocess them to obtain the differential pressure sequence, flow rate sequence, wheel displacement sequence and unloading arm angle sequence.

[0042] Status parameters include: upstream pressure, downstream pressure, volumetric flow rate, pump speed, inlet valve position, outlet valve position, and unloading arm angle of the main unloading pipeline of the railway tank car at the target unloading station;

[0043] The data filtering module 102 is used to filter and obtain a steady-state operation window based on the discrete state combination of the state parameters of the railway tank car, and retain the pressure difference sequence, flow rate sequence, wheel displacement sequence and unloading arm angle sequence within the window.

[0044] The data calculation module 103 is used to perform mean-removal processing on the pressure difference sequence and flow rate sequence within the steady-state window, thereby calculating the hysteresis vector, and calculating the endpoint drift vector and the reverse coupling ratio.

[0045] The hysteresis vector consists of flow hysteresis, wheel displacement hysteresis, and unloading arm angle hysteresis.

[0046] The endpoint drift vector consists of pressure differential drift, flow rate drift, wheel displacement drift, and unloading arm angle drift.

[0047] Gini coefficient calculation module 104 is used to calculate the pressure differential Gini coefficient and the flow rate Gini coefficient based on the pressure differential sequence and flow rate sequence within the steady-state window;

[0048] The preliminary judgment module 105 calculates the inverse cooperative correlation coefficient based on the pressure difference sequence and flow rate sequence within the steady-state window. If the coefficient is less than the safety threshold, the railway tank car is safe; otherwise, it is concatenated with the hysteresis vector, endpoint drift vector, inverse coupling ratio, pressure difference Gini coefficient, and flow rate Gini coefficient to obtain the feature vector.

[0049] The state category recognition module 106 is used to input the feature vector into the state recognition model and output the state category of the railway tank car;

[0050] The status categories include: safe status, minor traction, severe traction, and mechanical sway.

[0051] In one embodiment of the present invention, each sequence unit of the pressure difference sequence is represented by the difference between the upstream pressure and the downstream pressure at a given time point; each sequence unit of the flow rate sequence is represented by the volumetric flow rate at a given time point; each sequence unit of the wheel displacement sequence is represented by the product of the pixel displacement and the calibration coefficient, wherein the pixel displacement is equal to the difference between the pixel position of the wheel in the image data at the current time point and the pixel position of the wheel in the image data at the starting time point, and the calibration coefficient is the ratio of the known length of the actual object on site to the pixel length of the actual object in the image data; and each sequence unit of the unloading arm angle sequence is represented by the unloading arm angle at a given time point.

[0052] It should be noted that the upstream pressure is acquired by a diaphragm or strain gauge industrial pressure transmitter installed on the pipe section from the outside of the tank truck connecting flange to the isolation valve; the downstream pressure is acquired by an industrial pressure transmitter installed on the pipe section of the unloading main pipeline near the storage tank or the inlet of the recovery pipeline; the volumetric flow rate is acquired by a flow meter installed on the unloading main pipeline; the pump speed is acquired by a speed sensor installed on the unloading pump drive shaft or motor; the inlet valve position and outlet valve position are acquired by a valve position sensor on the valve stem or actuator; the unloading arm angle is acquired by an absolute angle encoder or industrial potentiometer installed on the main rotating shaft of the unloading arm; the pressure difference sequence, flow rate sequence, wheel displacement sequence, and unloading arm angle sequence have the same length, which can be ensured by unifying the time axis and interpolation alignment. That is, first unify the sampling time interval, for example, set the sampling time interval to 200ms, and then map them to each sampling point on the unified time axis according to their respective timestamps through linear interpolation and other algorithms, which will not be elaborated here.

[0053] In one embodiment of the present invention, a sliding window is applied to the pump speed, inlet valve position, and outlet valve position at each time point. If the difference between the maximum and minimum values ​​of the pump speed within the sliding window is less than a first threshold, and the difference between the maximum and minimum values ​​of the inlet valve position within the sliding window is less than a second threshold, and the difference between the maximum and minimum values ​​of the outlet valve position within the sliding window is less than a third threshold, then the time point is included in the steady-state operation window, and the differential pressure sequence, flow rate sequence, wheel displacement sequence, and unloading arm angle sequence within the window are retained. The length of the sliding window, the first threshold, the second threshold, and the third threshold are all custom parameters. Preferably, the length of the sliding window is set to 10 time points. According to the above, the length of the sliding window is 2 seconds, the first threshold is set to 2 RPM, the second threshold is set to 1%, and the third threshold is set to 1%.

[0054] It should be noted that by filtering the pump speed, inlet valve position, and outlet valve position to obtain the steady-state operating window, it is possible to indicate that the unloading car is in the unloading state. This can avoid identification and modeling errors caused by data fluctuations, provide a reliable basis for accurately judging the state category of railway tank cars, and eliminate the interference of unstable factors, making the subsequent identification of the state category of railway tank cars more accurate and reducing the risk of false alarms.

[0055] In one embodiment of the present invention, the average values ​​of the pressure difference sequence and the flow rate sequence within the steady-state window are first calculated, and then the corresponding average value is subtracted from each sequence unit of the pressure difference sequence and the flow rate sequence to complete the mean removal process.

[0056] In one embodiment of the present invention, the flow lag amount The calculation formula is as follows:

[0057] ,in Indicates the first The product of the flow rate This represents the independent variable corresponding to the maximum value of the internal product of the flow;

[0058] No. The product of flow The calculation formula is as follows:

[0059] Where Col represents the set of sequence units in the steady-state window, This represents the k-th sequence unit of the pressure difference sequence after mean removal. Represents the first value of the flow sequence after mean removal processing. Sequence units, Represents integer lag. The initial value is 0, and it is a custom parameter, preferably... ,in This represents the size of the sequence unit set within the steady-state window. Indicates rounding down;

[0060] First, calculate the inner product between the pressure difference sequence and the wheel displacement sequence within the steady-state window to obtain the inner product of wheel displacement. Then, take the independent variable corresponding to the maximum value of the inner product of wheel displacement to obtain the wheel displacement hysteresis.

[0061] First, calculate the inner product between the pressure difference sequence and the unloading arm angle sequence within the steady-state window to obtain the inner product of the unloading arm angle. Then, take the independent variable corresponding to the maximum value of the inner product of the unloading arm angle to obtain the hysteresis of the unloading arm angle.

[0062] For example, the mean-removed pressure difference sequence within the steady-state window is [1, 2, 3, 2, 1], and the mean-removed flow rate sequence, when inverted, is [2, 3, 2, 1, 0]. When the integer lag is 0, the corresponding flow rate inner product is 16; when the integer lag is 1, the corresponding flow rate inner product is 18; and when the integer lag is -1, the corresponding flow rate inner product is 10. Therefore, the flow rate lag is 1.

[0063] In one embodiment of the present invention, the difference between the last sequence unit and the first sequence unit of the differential pressure sequence, flow rate sequence, wheel displacement sequence and unloading arm angle sequence within the steady-state window is calculated to obtain the differential pressure drift, flow rate drift, wheel displacement drift and unloading arm angle drift; the negative value of the flow rate endpoint drift is divided by the differential pressure endpoint drift to obtain the reverse coupling ratio.

[0064] In one embodiment of the present invention, the difference between adjacent sequence units of the pressure difference sequence within the steady-state window is calculated to obtain the first-order increment sequence of the pressure difference, wherein the first sequence unit of the first-order increment sequence of the pressure difference is equal to the difference between the second sequence unit and the first sequence unit of the pressure difference sequence within the steady-state window, that is, the length of the first-order increment sequence of the pressure difference is 1 less than the length of the pressure difference sequence within the steady-state window. The first-order increment sequence of the flow rate is calculated in this way. Then, the absolute values ​​of the sequence units of the first-order increment sequence of the pressure difference are sorted in ascending order to obtain the pressure difference ascending sequence. The flow rate ascending sequence is calculated in this way.

[0065] The differential pressure Gini coefficient is calculated based on the ascending sequence of differential pressure, and the flow rate Gini coefficient is calculated using the same method. The differential pressure Gini coefficient... The calculation formula is as follows:

[0066] Where N represents the length of the pressure differential ascending sequence, and These represent the i-th and j-th sequence units of the pressure difference ascending sequence, respectively.

[0067] It should be noted that the hysteresis vector is used to capture the optimal temporal alignment of differential pressure with reverse flow, wheel displacement, and unloading boom angle, reflecting the causal temporal relationship between them and helping to distinguish between slight traction and mechanical sway. The endpoint drift vector can measure the net trend within the window, enhancing the sensitivity to slight traction and forming a complementary representation with the hysteresis vector in the time dimension. The reverse coupling ratio is used to quantify the magnitude relationship between the increase in differential pressure and the decrease in flow. The differential pressure Gini coefficient and the flow Gini coefficient are used to describe the imbalance between differential pressure and volumetric flow rate. If the coefficients are abnormal, it indicates that there are abnormal fluctuations in differential pressure or flow rate, which may be related to risk factors such as traction of the unloading boom and hose, and mechanical sway. This helps to extract risk-related features and assists the state recognition model in making accurate judgments. Therefore, the purpose of constructing feature vectors is to extract key features closely related to the operating state of railway tank cars from redundant time series, eliminate invalid redundant information in the original data, reduce the data dimensionality, and thus provide more accurate and targeted input for the state recognition model. To a certain extent, this can improve the model's calculation speed and recognition accuracy.

[0068] In one embodiment of the present invention, the security threshold is a custom parameter, preferably set to 0.4;

[0069] The formula for calculating the inverse cooperative correlation coefficient is as follows:

[0070] Where Col represents the set of sequence units in the steady-state window, and These represent the k-th sequence unit of the pressure difference sequence and flow rate sequence after the mean was removed, respectively.

[0071] It should be noted that unloading booms typically have a rigid boom structure, allowing for flexible rotation and extension for easy docking with tank trucks. The hoses, on the other hand, are flexible pipes connecting the unloading boom and the tank truck for oil transfer. When the unloading boom and hoses are subjected to slight traction, the hoses may deform due to elasticity, causing inner diameter contraction. The unloading boom's trajectory may also experience slight bends, and the connection joints may shift. These geometric changes affect the flow path of the oil within the pipeline. According to the Darcy-Weisbach equation, if traction causes a decrease in the hose's inner diameter and an increase in its effective length, the friction resistance will directly increase. Furthermore, local resistance is caused by local structural elements such as pipe bends and joints; traction-induced pipe bends and joint shifts will increase the local resistance coefficient. As the pressure difference increases, the local resistance also rises. Within the steady-state operating window of oil unloading, the power source (such as the output pressure of the pump and the liquid level difference of the oil storage tank) is relatively stable. According to the law of conservation of energy, the increase in the total pipeline resistance will increase the pressure difference between the inlet and outlet, because the resistance loss will be converted into pressure difference consumption. At the same time, according to the flow formula, when the total pipeline resistance increases while the pressure difference remains unchanged, the volumetric flow rate will inevitably decrease. Therefore, under slight traction, the correlation coefficient between the volumetric flow rate and the pressure difference after reversing is close to 1. However, under normal oil unloading without traction, the correlation between the two is weak, and the reverse correlation coefficient is lower than the safety threshold. This can be used to preliminarily determine whether the railway tank car is in a safe state. In addition, it can reduce the number of times the state recognition model is called and reduce false alarms caused by single-channel noise.

[0072] It should be noted that the safety threshold can be set by long-term monitoring and calculation to obtain the reverse coordination correlation coefficient as described above. Statistical analysis shows that under normal operating conditions of the railway tank car, this coefficient usually fluctuates between -0.2 and 0.3. When the unloading arm and hose are subjected to slight traction, the coefficient is close to 1. Furthermore, as the traction force increases, the deformation of the unloading arm and hose may cause a sudden change in nonlinear flow resistance, breaking the linear relationship between the correlation coefficient and the degree of traction. Statistical analysis shows that under the transitional condition between slight and severe traction of the railway tank car, this coefficient usually fluctuates between 0.4 and 0.8. Therefore, setting the safety threshold to 0.4 can avoid frequent false alarms due to normal small fluctuations. Subsequently, the nonlinear change in flow resistance can be captured by the state recognition model, thereby realizing the identification of the state category of the railway tank car.

[0073] In one embodiment of the present invention, the state recognition model is constructed based on a multilayer perceptron and consists of an input layer, a hidden layer and an output layer. The input layer inputs a feature vector, the hidden layer consists of two hidden units, both of which are activated by the GELU activation function, and the output layer is activated by the Softmax activation function. The model outputs the probability that the railway tank car is in a safe state, under slight traction, under severe traction and under mechanical sway, and selects the state category with the highest probability for output.

[0074] It should be noted that during the training process of the state recognition model, the cross-entropy loss function is selected to measure the difference between the predicted probability distribution and the true label distribution, and the Adam optimizer is selected to adaptively adjust the learning rate and accelerate the convergence speed of the model.

[0075] In one embodiment of the present invention, state parameters and image data of railway tank cars under safe, light traction, heavy traction and mechanical sway conditions are collected and constructed into feature vectors as training samples for training the state recognition model.

[0076] It should be noted that at least 1000 valid samples should be prepared (the more samples, the better the model training effect is usually). The samples should be divided into training set (70%), validation set (20%) and test set (10%) in a ratio of 7:2:1. The training set is used for learning model parameters, the validation set is used to adjust model hyperparameters (such as the number of hidden layer neurons, learning rate, etc.), and the test set is used to evaluate the final generalization ability of the model.

[0077] It should be noted that by first comparing the inverse correlation coefficient with the safety threshold, it is possible to quickly screen whether the railway tank car is in a safe state, reducing the number of calls to the state recognition model, lowering computational costs, and reducing false alarms caused by single noise. Then, the state recognition model combines multi-dimensional features to accurately identify the specific state category. This not only avoids the limitations of single parameter judgment, but also makes up for the deficiencies of the initial judgment through in-depth analysis of the model, achieving a combination of efficient screening and accurate identification, and ensuring the reliability of oil unloading operation monitoring.

[0078] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0079] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A smart monitoring and management system for unloading oil from railway tank cars based on the Internet of Things, characterized in that, include: The data acquisition module is used to collect the status parameters and image data of railway tank cars, and preprocess them to obtain differential pressure sequence, flow rate sequence, wheel displacement sequence and unloading arm angle sequence; Status parameters include: upstream pressure, downstream pressure, volumetric flow rate, pump speed, inlet valve position, outlet valve position, and unloading arm angle of the main unloading pipeline of the railway tank car at the target unloading station; The data filtering module is used to filter and obtain a steady-state operation window based on the discrete state combination of the state parameters of the railway tank car, and retain the pressure difference sequence, flow rate sequence, wheel displacement sequence and unloading arm angle sequence within the window. The data calculation module is used to perform mean-removal processing on the pressure difference sequence and flow rate sequence within the steady-state window, thereby calculating the hysteresis vector, endpoint drift vector, and reverse coupling ratio. The hysteresis vector consists of flow hysteresis, wheel displacement hysteresis, and unloading arm angle hysteresis. The endpoint drift vector consists of pressure differential drift, flow rate drift, wheel displacement drift, and unloading arm angle drift. The Gini coefficient calculation module is used to calculate the differential pressure Gini coefficient and the flow rate Gini coefficient based on the differential pressure sequence and flow rate sequence within the steady-state window. The preliminary judgment module calculates the inverse cooperative correlation coefficient based on the pressure difference sequence and flow rate sequence within the steady-state window. If the coefficient is less than the safety threshold, the railway tank car is considered safe. Otherwise, it is concatenated with the hysteresis vector, endpoint drift vector, inverse coupling ratio, pressure difference Gini coefficient, and flow rate Gini coefficient to obtain the feature vector. The state category recognition module is used to input feature vectors into the state recognition model and output the state category of the railway tank car; The status categories include: safe status, minor traction, severe traction, and mechanical sway.

2. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, Each unit of the differential pressure sequence is represented by the difference between the upstream and downstream pressures at a given time point; each unit of the flow rate sequence is represented by the volumetric flow rate at a given time point; each unit of the wheel displacement sequence is represented by the product of the pixel displacement and the calibration coefficient. The pixel displacement is equal to the difference between the pixel position of the wheel in the image data at the current time point and the pixel position of the wheel in the image data at the starting time point. The calibration coefficient is the ratio of the known length of the actual object on site to the pixel length of the actual object in the image data; each unit of the unloading arm angle sequence is represented by the unloading arm angle at a given time point.

3. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, A sliding window is applied to the pump speed, inlet valve position, and outlet valve position at each time point. If the difference between the maximum and minimum pump speed values ​​within the sliding window is less than a first threshold, the difference between the maximum and minimum inlet valve positions within the sliding window is less than a second threshold, and the difference between the maximum and minimum outlet valve positions within the sliding window is less than a third threshold, then that time point is included in the steady-state operation window. The differential pressure sequence, flow rate sequence, wheel displacement sequence, and unloading arm angle sequence within that window are retained. The length of the sliding window, the first threshold, the second threshold, and the third threshold are all user-defined parameters.

4. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, First, calculate the average value of the pressure difference sequence and flow rate sequence within the steady-state window. Then, subtract the corresponding average value from each sequence unit of the pressure difference sequence and flow rate sequence to complete the mean removal process.

5. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, Flow lag The calculation formula is as follows: ,in Indicates the first The product of the flow, This represents the independent variable corresponding to the maximum value of the internal product of the flow; No. Intra-flow product The calculation formula is as follows: Where Col represents the set of sequence units in the steady-state window, This represents the k-th sequence unit of the pressure difference sequence after mean removal. Represents the first value of the flow sequence after mean removal processing. Sequence units, Represents integer lag. Its initial value is 0, and it is a custom parameter; First, calculate the inner product between the pressure difference sequence and the wheel displacement sequence within the steady-state window to obtain the inner product of wheel displacement. Then, take the independent variable corresponding to the maximum value of the inner product of wheel displacement to obtain the wheel displacement hysteresis. First, calculate the inner product between the pressure difference sequence and the unloading arm angle sequence within the steady-state window to obtain the inner product of the unloading arm angle. Then, take the independent variable corresponding to the maximum value of the inner product of the unloading arm angle to obtain the hysteresis of the unloading arm angle.

6. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, The pressure difference drift, flow rate drift, wheel displacement drift, and unloading arm angle drift are obtained by calculating the difference between the last sequence unit and the first sequence unit of the pressure difference sequence, flow rate sequence, wheel displacement sequence, and unloading arm angle sequence within the steady-state window, respectively. The negative value of the flow rate endpoint drift is divided by the pressure difference endpoint drift to obtain the reverse coupling ratio.

7. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, The difference between adjacent sequence units of the pressure difference sequence within the steady-state window is calculated to obtain the first-order pressure difference increment sequence. The first sequence unit of the first-order pressure difference increment sequence is equal to the difference between the second and first sequence units of the pressure difference sequence within the steady-state window. The first-order flow rate increment sequence is then calculated using the same method. The absolute values ​​of the sequence units of the first-order pressure difference increment sequence are then sorted in ascending order to obtain the ascending pressure difference sequence. The ascending flow rate sequence is then calculated using the same method. The differential pressure Gini coefficient is calculated based on the ascending sequence of differential pressure, and the flow rate Gini coefficient is calculated using the same method. The differential pressure Gini coefficient... The calculation formula is as follows: Where N represents the length of the pressure differential ascending sequence, and These represent the i-th and j-th sequence units of the pressure difference ascending sequence, respectively.

8. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, The safety threshold is a user-defined parameter; the formula for calculating the inverse cooperative correlation coefficient is as follows: Where Col represents the set of sequence units in the steady-state window, and These represent the k-th sequence unit of the pressure difference sequence and flow rate sequence after the mean was removed, respectively.

9. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, The state recognition model is built on a multilayer perceptron and consists of an input layer, a hidden layer, and an output layer. The input layer takes in the feature vector, the hidden layer consists of two hidden units, both of which are activated by the GELU activation function, and the output layer is activated by the Softmax activation function. It outputs the probability that the railway tank car is in a safe state, under slight traction, under severe traction, or under mechanical sway, and selects the state category with the highest probability for output.

10. The intelligent oil unloading monitoring and management system for railway tank cars based on the Internet of Things as described in claim 1, characterized in that, By collecting state parameters and image data of railway tank cars under safe, light traction, heavy traction and mechanical sway conditions, and constructing feature vectors, these data are used as training samples for training the state recognition model.

Images