A power station oil pillow monitoring method and system based on multi-source data fusion

Abstract

The present application relates to the technical field of power station monitoring, and provides a power station oil pillow monitoring method and system based on multi-source data fusion, comprising: identifying the actual oil level of an oil pillow based on an infrared image of the oil pillow through an image processing algorithm; comparing the readings of an oil level gauge with the theoretical oil level range, the actual oil level of the oil pillow and the historical average value of the oil level respectively to determine whether the readings of the oil level gauge are normal; if the readings of the oil level gauge are normal, diagnosing whether the oil level of the oil pillow is normal; wherein the image processing algorithm comprises: enhancing the infrared image of the oil pillow using a multi-scale Retinex algorithm, calculating the gradient of the enhanced image, retaining the local maximum points in the gradient direction, determining the candidate straight line set under the length constraint and the inclination constraint after the edge classification based on the local maximum points, selecting the straight line with the largest cumulative gradient value in the candidate straight line set as the optimal liquid level line, and calculating the actual oil level of the oil pillow. The present application realizes high-frequency and accurate monitoring of the oil level.

Description

Technical Field

[0001] This invention belongs to the field of power plant monitoring technology, and in particular relates to a method and system for monitoring power plant oil tankers using multi-source data fusion. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The mechanical oil level gauge of the capsule-type transformer oil conservator relies on a float and connecting rod transmission structure. During long-term operation, it is prone to jamming or drifting due to oil aging and dust accumulation, resulting in frequent false oil level phenomena and directly affecting the accuracy of oil level reading.

[0004] Conventional infrared-assisted detection solutions, while capable of identifying false oil level readings, have significant limitations: High reliance on manual labor: Infrared verification requires manual triggering after the oil level gauge malfunctions, necessitating on-site operation of the infrared equipment at night, which is labor-intensive and poses safety hazards. It is also difficult to achieve high-frequency monitoring, resulting in delayed response and increased operational safety risks. Sensitive to environmental interference: The reflective oil film and stains on the surface of the oil conservator can obscure the liquid level boundary, resulting in a high false detection rate for traditional infrared image recognition methods; Static diagnostic defects: It relies solely on a single comparison of dual-source oil level data to determine instantaneous anomalies, without integrating historical oil level data, and cannot capture the gradual downward trend of oil level, resulting in a lack of early warning for slow leakage risks and a lag in early warning for oil leakage risks. Summary of the Invention

[0005] To address the technical problems mentioned above, this invention provides a multi-source data fusion method and system for monitoring power plant oil tanks. It employs a multi-scale Retinex algorithm to decompose the illuminance and reflectance components of infrared images, using specific parameter combinations to eliminate oil film reflection and dirt interference, significantly improving the accuracy of level line identification. Furthermore, it integrates dynamic range verification of the oil temperature-level curve, cross-verification of infrared identification values ​​and oil level gauge readings, and consistency checks between oil level gauge readings and historical trends. This enables multi-dimensional dynamic determination of false oil levels. When the oil level gauge reading is normal, the oil tank level is diagnosed, achieving high-frequency and accurate oil level monitoring and ensuring the safe and stable operation of key power grid equipment.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a method for monitoring power plant oil tankers using multi-source data fusion, comprising: Obtain the oil temperature gauge reading, oil level gauge reading, oil tank infrared image, and historical average oil level under the same working conditions at the current inspection point; Based on the oil temperature gauge reading, the oil temperature-oil level curve is consulted to obtain the theoretical oil level range; based on the oil conservator infrared image, the actual oil level of the oil conservator is identified through image processing algorithms; the oil level gauge reading is compared with the theoretical oil level range, the actual oil level of the oil conservator, and the historical average oil level to determine whether the oil level gauge reading is normal. If the oil level gauge reading is normal, then diagnose whether the oil level in the oil tank is normal. The image processing algorithm includes: using the multi-scale Retinex algorithm to enhance the infrared image of the oil conservator, calculating the gradient of the enhanced image, retaining the local maximum points in the gradient direction, performing edge classification based on the local maximum points, determining a set of candidate lines under length and tilt constraints, selecting the line with the largest cumulative gradient value from the set of candidate lines as the optimal liquid level line, and calculating the actual oil level of the oil conservator.

[0007] Furthermore, the multi-scale Retinex algorithm includes: extracting the illuminance component from the oil conservator infrared image using multi-scale Gaussian filtering, and then recovering the reflectance component through the logarithmic domain; introducing an adaptively adjustable color restoration factor to process the pseudo-color band of the infrared image to obtain a color restoration image; and calculating an enhanced image based on the reflectance component and the color restoration image.

[0008] Furthermore, the illuminance component is: ;in, As weight, Infrared image of the oil reservoir. is a multi-scale Gaussian kernel function, where x is the pixel coordinate.

[0009] Furthermore, the color-restored image is: ,in, To enhance the strength coefficient, For color balance coefficient, It is a pseudo-color band. This is an infrared image of an oil reservoir.

[0010] Furthermore, it also includes: if the oil level in the oil tank is normal, obtaining multi-dimensional time series data, predicting the rate of oil level decline in future periods through a long short-term memory network, and issuing graded warnings.

[0011] Furthermore, the multi-dimensional time series includes oil level gauge readings, equipment load current, oil temperature gauge readings, oil level drop rate, and oil pressure gradient.

[0012] Furthermore, the loss function of the long short-term memory network is a weighted sum of the prediction error of the oil level drop rate and the absolute error of the oil level.

[0013] A second aspect of the present invention provides a power plant oil tank monitoring system based on multi-source data fusion, comprising: The data acquisition module is configured to acquire the oil temperature gauge reading, oil level gauge reading, oil tank infrared image, and historical average oil level under the same working conditions at the current inspection point. The reading monitoring module is configured to: query the oil temperature-oil level curve based on the oil temperature gauge reading to obtain the theoretical oil level range; identify the actual oil level in the oil conservator based on the infrared image of the oil conservator through an image processing algorithm; and compare the oil level gauge reading with the theoretical oil level range, the actual oil level in the oil conservator, and the historical average oil level to determine whether the oil level gauge reading is normal. The oil level monitoring module is configured to diagnose whether the oil level in the oil tank is normal if the oil level gauge reading is normal. The image processing algorithm includes: using the multi-scale Retinex algorithm to enhance the infrared image of the oil conservator, calculating the gradient of the enhanced image, retaining the local maximum points in the gradient direction, performing edge classification based on the local maximum points, determining a set of candidate lines under length and tilt constraints, selecting the line with the largest cumulative gradient value from the set of candidate lines as the optimal liquid level line, and calculating the actual oil level of the oil conservator.

[0014] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for monitoring power plant oil tankers using multi-source data fusion.

[0015] A fourth aspect of the present invention provides a computer device including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method for monitoring power plant oil tankers using multi-source data fusion.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention innovatively proposes a multi-source data fusion method for monitoring power plant oil tanks. It employs a multi-scale Retinex algorithm to decompose the illuminance and reflectance components of infrared images, and uses specific parameter combinations to eliminate oil film reflection and dirt interference, significantly improving the accuracy of liquid level line identification. Based on this, it integrates dynamic range verification of oil temperature-oil level curve, cross-verification of infrared identification values ​​and oil level gauge readings, and consistency verification of oil level gauge readings with historical trends to achieve multi-dimensional dynamic determination of false oil levels. When the oil level gauge reading is normal, the oil level of the oil tank is diagnosed, realizing high-frequency and accurate oil level monitoring, and ensuring the safe and stable operation of key power grid equipment.

[0017] This invention innovatively proposes an LSTM model that integrates historical data and multiple physical quantities to predict the rate of oil level decline, provide early warning of leakage risks up to 72 hours in advance, achieve early graded warning of leakage, and avoid secondary accidents such as insulation failure. Attached Figure Description

[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0019] Figure 1 This is a flowchart of a power plant oil tank monitoring method based on multi-source data fusion according to Embodiment 1 of the present invention; Figure 2 This is a flowchart of the process for enhancing infrared images of oil confined spaces and extracting liquid level boundary features based on multi-scale Retinex, according to Embodiment 1 of the present invention. Figure 3 This is a schematic diagram of the oil level trend prediction and active early warning technology route of Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device according to Embodiment 4 of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0021] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] Example 1 This embodiment provides a method for monitoring power plant oil tankers using multi-source data fusion.

[0023] This embodiment provides a multi-source data fusion method for monitoring power station oil tanks, which is applicable to substations.

[0024] This embodiment provides a multi-source data fusion method for monitoring power plant oil tanks. By fusing Retinex image enhancement, multi-source data dynamic calibration, and LSTM (Long Short-Term Memory) network trend prediction, it ultimately achieves intelligent determination of oil level authenticity and proactive early risk warning. It eliminates the safety risks and response lag associated with manual infrared verification, enabling high-frequency automatic monitoring of the oil tank; overcomes the distortion in level line recognition caused by oil film reflection and dirt obstruction, improving the reliability of infrared image analysis; and breaks through the limitations of single static diagnosis by constructing a time-series oil level change prediction model to identify potential oil leakage risks in advance.

[0025] This embodiment provides a method for monitoring power plant oil tankers using multi-source data fusion, such as... Figure 1 As shown, it includes the following steps: Step 1: Intelligent task scheduling and data collection.

[0026] Step 101, Task Configuration and Triggering: Create a transformer oil level inspection task within the intelligent inspection system, setting parameters such as task execution time, execution cycle, and inspection points. The execution time is typically set during stable ambient temperatures at night, such as 00:00-05:00, to reduce interference from sunlight radiation. The execution cycle supports flexible settings from minutes to several days, with a default of once per day. Inspection points include oil level gauge readings, oil temperature gauge readings, and oil tank level. When the clock reaches the predetermined execution time, the inspection task is automatically executed without manual intervention, resolving the safety risks and response delays associated with manual operation at night.

[0027] Step 102, Multi-source data synchronous acquisition: Based on the selected inspection point, automatically activate the corresponding inspection device (including inspection robot, visible light camera, and infrared camera, etc.) at that inspection point and perform the following operations synchronously: Visible light images of the oil temperature gauge and oil level gauge are acquired, and the readings of the oil temperature gauge and oil level gauge are identified based on image recognition technology, and recorded as follows: , ; Infrared images of the oil conservator are acquired, and the actual oil level (i.e., the actual oil level in the conservator) is identified using image processing algorithms, and denoted as . ; Extract the oil level gauge readings from the database under the same operating conditions (i.e., historical oil level data under the same operating conditions), and take the average oil level over the past 30 days. .

[0028] like Figure 2 As shown, the steps of the image processing algorithm include: (1) Retinex multiscale anti-interference enhancement.

[0029] The surface of the capsule-type oil reservoir is subject to three layers of interference: the reflection of the oil film causes the bright area to obscure the liquid level line, the accumulation of dust and dirt locally blocks the thermal radiation characteristics, and the edge breakage of the arc-shaped end cap due to geometric distortion. These interferences can be specifically suppressed by multi-scale illuminance separation.

[0030] This embodiment uses the multi-scale Retinex algorithm (MSRCR) to decompose infrared images and construct an infrared image decomposition model: ; in: For pixel coordinates, x and y These represent the values ​​for the x and y coordinates, respectively. The original infrared image; For low-frequency illuminance components, focus on oil film reflection and shadow interference; The high-frequency reflection component enhances the thermal gradient characteristics of the liquid level line.

[0031] (a) Illuminance components are extracted using multi-scale Gaussian filtering.

[0032] Constructing a three-scale Gaussian kernel function : ; in: In pixels .

[0033] Calculate the weighted illuminance components: ; Weight allocation: , , .

[0034] At a small scale (k=1), Take 15 pixels (Take 0.4) for precise stripping of oil film reflection, mesoscale (k=2, Take 80 pixels Take 0.3) to eliminate typical stain clumps of 3-5cm, large-scale (k=3) Take 250 pixels (Take 0.3) Cover the oil cushion with a maximum width of 200cm and suppress background interference from ambient thermal radiation.

[0035] (b) Enhanced reflection component.

[0036] Logarithmic domain recovery of reflection components: ; An adaptive color restoration factor is introduced to compensate for color shift caused by enhancement: ; in: To enhance the strength coefficient, this value is chosen to suppress color shift in oily areas while maintaining a 0.5°C temperature difference at the liquid level. This is the color balance coefficient, and its value was determined through training with 200 sets of oil stain samples to optimally balance color distortion. This refers to the pseudo-color bands of infrared images, including R, G, and B.

[0037] Final output enhanced image: .

[0038] (2) Extraction of liquid level boundary line.

[0039] (a) Edge enhancement based on adaptive Canny.

[0040] Calculate the gradient of the enhanced image : ; in: and For Sobel operator convolution kernels (horizontal / vertical direction); Infrared image enhanced with Retinex.

[0041] Preserve local maxima along the gradient direction To achieve nonmaximum suppression: ; in: It is the normal angle of the line.

[0042] (b) Determine the edge classification.

[0043] Design a dynamic dual threshold: ; ; in: Define the image domain; upper threshold The settings ensure that temperature differences above 0.5°C are typically captured; lower threshold. Avoid interference from weak edges such as oil-filled weld seams.

[0044] Edge detection rules: .

[0045] The edge classification results are processed for connectivity: all strong edge pixels are retained, and only those weak edge pixels connected to the strong edge pixels are retained. All isolated weak edge and non-edge pixels are removed, and finally an optimized binary edge image is generated for subsequent liquid level line positioning, which greatly reduces the amount of data and interference.

[0046] (c) Liquid level line positioning.

[0047] First, based on the generated binary edge image, the Hough Transform is used to compute the line parameter space, in which any line can be represented by parameter pairs. It means that, among them The distance from the line to the origin is the normal distance. Let the angle between the normal and the x-axis be the angle between the normal and the x-axis, thus obtaining the set of candidate lines. Each of them Represents a line composed of parameters The uniquely defined candidate line, and, ; Subsequently, length and tilt constraints are applied to the candidate line set: Length constraints exclude interference edges in the end cap arc region: ; in: This refers to the actual width of the oil cushion.

[0048] Inclination constraints are set according to the horizontal installation specifications for capsule-type oil conservators: ; Interference filtering rules: From the set of candidate lines that meet the constraints Choose the optimal liquid level line, where the cumulative gradient value is significantly higher than other lines: .

[0049] Step 2: Triple verification of false oil level.

[0050] Step 201, Anomaly Detection Logic: Perform three parallel verifications simultaneously. If any verification fails, the oil level gauge is determined to be faulty, and a false oil level alarm is sent. Verification 1: Comparison of oil temperature-oil level curves.

[0051] Based on the current oil temperature, refer to the preset oil temperature-oil level curve to calculate the theoretical oil level range. If the oil level gauge reading If the oil level exceeds the theoretical range, the oil level gauge is considered malfunctioning.

[0052] Verification 2: Cross-verification of infrared images and oil level gauge.

[0053] Calculate the oil level using infrared recognition oil level gauge reading absolute deviation ,like , If a reasonable error threshold is preset, the oil level gauge is determined to be abnormal.

[0054] Verification 3: Historical trend consistency test.

[0055] Calculate the current oil level gauge reading Compared with historical average deviation , if the deviation k, where k is 5mm, is the fluctuation tolerance; if so, the oil level gauge is considered abnormal.

[0056] Step 203, Alarm Generation: When the oil level gauge malfunction is determined, a "false oil level" alarm is automatically generated and pushed, triggering manual review and alarm troubleshooting process.

[0057] Step 3: If the oil level gauge reading is normal, then perform a diagnosis of the actual oil level anomaly.

[0058] Step 301: Dynamic threshold generation.

[0059] Based on the current oil temperature Check the oil temperature-oil level profile to determine the theoretical normal range of oil level. .

[0060] Step 302: Anomaly Diagnosis and Alarms.

[0061] If the oil level gauge reading Automatically generate a "Transformer oil tank level too high" alarm; if It automatically generates a "Transformer oil level too low" alarm; the alarm information will be pushed to the interface in real time, triggering manual review and alarm troubleshooting process.

[0062] Step 4: Visualization and analysis of historical data.

[0063] (1) After the inspection is completed, a historical data trend chart is automatically generated, which includes the following five sets of key curves (i.e., multi-dimensional curves): Curve 1 is the theoretical oil level reference curve, with the data source being the transformer physical model; Curve 2 is the upper limit curve for the oil level threshold. The data source is the transformer physical model. If the transformer physical model does not provide an upper limit curve, the oil temperature-oil level reference curve can usually be used with a 10% upward adjustment. Curve 3 is the lower limit curve for the oil level threshold. The data source is the transformer physical model. If the transformer physical model does not provide a lower limit curve, the oil temperature-oil level reference curve can usually be used with a 10% downward adjustment. Curve 4: Oil level gauge reading sequence; data source: historical data. Dataset; Curve 5 is an infrared identification oil level sequence; the data source is historical data. Dataset.

[0064] (2) By superimposing and comparing five sets of curves, the evolution of oil level status is presented intuitively, which helps maintenance personnel to locate the root cause of abnormality.

[0065] Step 5: Oil level trend prediction and proactive early warning. For example... Figure 3 As shown, it includes the following steps: Step 501: Multi-source time series data fusion and feature extraction.

[0066] To achieve early warning of oil tank leakage, a multi-dimensional time-series database was first constructed, and calibrated oil level data was synchronously integrated. (i.e., the oil level gauge reading when it is normal), equipment load current. Ambient temperature Real-time oil temperature Hydraulic And auxiliary parameters (humidity, cooler status).

[0067] Data is continuously collected in 15-minute intervals and then... The principle is to clean up abnormal mutation values.

[0068] Specifically designed for oil tank characteristics, the following oil leakage sensitive factors are constructed: The rate of oil level drop is used to quantify the change in oil level per unit time. ; This indicates a time interval, representing the oil level data within that time interval. The change in; The load-oil-temperature correlation coefficient is used to assess the impact of load changes on oil temperature. ; Oil pressure gradient, used to reflect the trend of sealing degradation: .

[0069] The "load-oil temperature correlation coefficient" is mainly used for offline analysis and model interpretation. The input of the LSTM prediction model directly uses the load current. With real-time oil temperature The original time-series features are automatically learned by the model to understand the complex nonlinear relationships during the process.

[0070] Step 502: LSTM prediction engine construction and oil leak risk assessment.

[0071] The deep learning-based LSTM spatiotemporal model is used as the core of the prediction. This model takes a 72-hour sliding window (3 days of historical data) as input and learns the complex mapping relationship between oil level sequence and equipment parameters.

[0072] Input feature vector: ; After entering the LSTM memory unit, the network captures the long-term oil level drift pattern, the transmission effect of load fluctuation on oil temperature, and the spatiotemporal pattern of oil pressure decay through a gating mechanism. The model outputs predicted rates of oil level decline over the next 24 / 48 / 72 hours. , , .

[0073] The loss function fuses the rate prediction error (weighted at 70%) and the absolute oil level error (weighted at 30%) to ensure sensitive detection of gradual leakage. ; Where: the predicted value is the sequence value at multiple future time points, and N is the total length of the predicted sequence. and Represent the actual and predicted rates of oil level decline at the i-th future time point, respectively; weight values , ; This is the oil level absolute error compensation term.

[0074] The calculation method for MAE(V) is as follows: ; in: The actual oil level at the i-th sampling time; is the oil level value predicted by the LSTM model at the same time; N is the total length of the prediction sequence.

[0075] Early warning rules are deeply tied to power standards, when the predicted rate When (corresponding to the leakage risk threshold specified in DL / T572), a general-level warning is triggered; if This indicates an accelerating leak, triggering a severe level warning; once... This indicates a serious risk of leakage; an emergency alert should be activated immediately. This indicates the percentage of the hourly oil level drop relative to the maximum oil level height of the conservator. For example, consider the maximum oil level height of a 220kV transformer's conservator. =2000mm, warning rule 0.3% This means that the warning height decreases by 0.3% per hour = 2000mm = 6mm. In other words, if the prediction model indicates that the rate of decrease exceeds 6 millimeters per hour in the next 72 hours, a general level warning will be triggered.

[0076] The threshold design takes into account both the safety margin of the transformer and the physical limits of the oil conservator. For example, 0.3% / h is equivalent to the critical risk point of a daily oil level drop of >5cm for a 220kV transformer.

[0077] Step 503: Hierarchical early warning linkage and dynamic optimization closed loop.

[0078] The early warning signal automatically triggers a multi-level response mechanism: general-level early warnings push messages to the monitoring system, prompting manual review within 24 hours; severe-level early warnings automatically generate enhanced patrol tasks, linking inspection robots and cameras to re-examine the target oil tank, automatically capturing high-definition visible light / infrared images to capture visual evidence of oil stains or sealing defects, and mandating handling within 12 hours; critical-level early warnings will automatically generate and execute high-frequency enhanced patrol tasks, closely monitoring the development of risks, and simultaneously sending early warning information to remind maintenance personnel to handle the situation immediately.

[0079] A dynamic learning loop is built synchronously: the sample library is automatically expanded by the annotation feedback of maintenance personnel on false alarm events, driving monthly model iterations; the warning threshold is adaptively adjusted based on the distribution characteristics of historical oil level decline rates. ; in: The standard deviation of the historical rate of oil level decline; This represents the average rate of historical oil level decline. The baseline warning threshold is used.

[0080] For example, the natural volatility of oil levels at a certain station When the threshold reaches 0.15, the threshold is dynamically increased by 15% to reduce false triggering caused by seasonal changes.

[0081] This embodiment provides a power plant oil tank monitoring method based on multi-source data fusion, achieving a breakthrough in detection accuracy: by using Retinex multi-scale image enhancement technology, it effectively suppresses oil film reflection and dirt interference, significantly improving the accuracy of liquid level line recognition and solving the problem of false oil level misjudgment.

[0082] This embodiment provides a multi-source data fusion method for monitoring power plant oil tanks, which revolutionizes operation and maintenance efficiency: the fully automated nighttime execution mechanism replaces manual inspections, the monitoring frequency can be increased to the minute level, labor costs are significantly reduced, and the safety risks of nighttime operations are eliminated.

[0083] This embodiment provides a multi-source data fusion method for monitoring power plant oil tanks, which significantly enhances early warning capabilities: by fusing historical data with an LSTM model of multiple physical quantities, it can predict the rate of oil level decline, provide early warning of leakage risks up to 72 hours in advance, achieve early graded early warning of leakage, and avoid secondary accidents such as insulation failures.

[0084] This embodiment provides a multi-source data fusion method for monitoring power plant oil tanks, with optimized decision-making closed loop: hierarchical early warning linkage with inspection robots for active re-verification, forming an intelligent closed loop of "prediction-verification-disposal", reducing the fault response speed from hours to minutes.

[0085] Example 2 This embodiment provides a power plant oil tank monitoring system based on multi-source data fusion, comprising: The data acquisition module is configured to acquire the oil temperature gauge reading, oil level gauge reading, oil tank infrared image, and historical average oil level under the same working conditions at the current inspection point. The reading monitoring module is configured to: query the oil temperature-oil level curve based on the oil temperature gauge reading to obtain the theoretical oil level range; identify the actual oil level in the oil conservator based on the infrared image of the oil conservator through an image processing algorithm; and compare the oil level gauge reading with the theoretical oil level range, the actual oil level in the oil conservator, and the historical average oil level to determine whether the oil level gauge reading is normal. The oil level monitoring module is configured to diagnose whether the oil level in the oil tank is normal if the oil level gauge reading is normal. The image processing algorithm includes: using the multi-scale Retinex algorithm to enhance the infrared image of the oil conservator, calculating the gradient of the enhanced image, retaining the local maximum points in the gradient direction, performing edge classification based on the local maximum points, determining a set of candidate lines under length and tilt constraints, selecting the line with the largest cumulative gradient value from the set of candidate lines as the optimal liquid level line, and calculating the actual oil level of the oil conservator.

[0086] It should be noted that each module in this embodiment corresponds one-to-one with each step in Embodiment 1, and their specific implementation processes are the same, so they will not be repeated here.

[0087] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a power plant oil tank monitoring method based on multi-source data fusion as described in Embodiment 1 above.

[0088] Example 4 This embodiment provides a computer device, such as... Figure 4 As shown, the system includes a computer-readable storage medium 1003, a processor 1001, a communication interface 1002, and a computer program stored on the computer-readable storage medium 1003 and executable on the processor 1001. The processor 1001, communication interface 1002, and computer-readable storage medium 1003 can be connected via a bus or other means. The communication interface 1002 is used to receive and transmit data. When the processor 1001 executes the program, it implements the steps in the multi-source data fusion method for monitoring power plant oil tankers described in Embodiment 1 above.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring power plant oil tankers using multi-source data fusion, characterized in that, include: Obtain the oil temperature gauge reading, oil level gauge reading, oil tank infrared image, and historical average oil level under the same working conditions at the current inspection point; Based on the oil temperature gauge reading, the oil temperature-oil level curve is consulted to obtain the theoretical oil level range; based on the oil conservator infrared image, the actual oil level of the oil conservator is identified through image processing algorithms; the oil level gauge reading is compared with the theoretical oil level range, the actual oil level of the oil conservator, and the historical average oil level to determine whether the oil level gauge reading is normal. If the oil level gauge reading is normal, then diagnose whether the oil level in the oil tank is normal. The image processing algorithm includes: using the multi-scale Retinex algorithm to enhance the infrared image of the oil conservator, calculating the gradient of the enhanced image, retaining the local maximum points in the gradient direction, performing edge classification based on the local maximum points, determining a set of candidate lines under length and tilt constraints, selecting the line with the largest cumulative gradient value from the set of candidate lines as the optimal liquid level line, and calculating the actual oil level of the oil conservator. The multi-scale Retinex algorithm includes: extracting the illuminance component from the oil conservator infrared image using multi-scale Gaussian filtering, and then recovering the reflectance component in the logarithmic domain; introducing an adaptively adjustable color restoration factor to process the pseudo-color band of the infrared image to obtain a color restoration image; and calculating an enhanced image based on the reflectance component and the color restoration image. The illuminance component is: ;in, As weight, Infrared image of the oil reservoir. The function is a multi-scale Gaussian kernel, where x is the pixel coordinate. The restored color image is: ,in, To enhance the strength coefficient, For color balance coefficient, It is a pseudo-color band. This is an infrared image of an oil reservoir.

2. The method for monitoring power plant oil tankers using multi-source data fusion as described in claim 1, characterized in that, Also includes: If the oil level in the oil reservoir is normal, multi-dimensional time series data is obtained, and the rate of oil level decline in future periods is predicted through a long short-term memory network, with graded early warnings issued.

3. The method for monitoring power plant oil tankers using multi-source data fusion as described in claim 2, characterized in that, The multi-dimensional time series includes oil level gauge readings, equipment load current, oil temperature gauge readings, oil level drop rate, and oil pressure gradient.

4. The method for monitoring power plant oil tankers using multi-source data fusion as described in claim 2, characterized in that, The loss function of the Long Short-Term Memory network is the weighted sum of the prediction error of the oil level decline rate and the absolute error of the oil level.

5. A power plant oil tank monitoring system with multi-source data fusion, characterized in that, include: The data acquisition module is configured to acquire the oil temperature gauge reading, oil level gauge reading, oil tank infrared image, and historical average oil level under the same working conditions at the current inspection point. The reading monitoring module is configured to: query the oil temperature-oil level curve based on the oil temperature gauge reading to obtain the theoretical oil level range; identify the actual oil level in the oil conservator based on the infrared image of the oil conservator through an image processing algorithm; and compare the oil level gauge reading with the theoretical oil level range, the actual oil level in the oil conservator, and the historical average oil level to determine whether the oil level gauge reading is normal. The oil level monitoring module is configured to diagnose whether the oil level in the oil tank is normal if the oil level gauge reading is normal. The image processing algorithm includes: using the multi-scale Retinex algorithm to enhance the infrared image of the oil conservator, calculating the gradient of the enhanced image, retaining the local maximum points in the gradient direction, performing edge classification based on the local maximum points, determining a set of candidate lines under length and tilt constraints, selecting the line with the largest cumulative gradient value from the set of candidate lines as the optimal liquid level line, and calculating the actual oil level of the oil conservator. The multi-scale Retinex algorithm includes: extracting the illuminance component from the oil conservator infrared image using multi-scale Gaussian filtering, and then recovering the reflectance component in the logarithmic domain; introducing an adaptively adjustable color restoration factor to process the pseudo-color band of the infrared image to obtain a color restoration image; and calculating an enhanced image based on the reflectance component and the color restoration image. The illuminance component is: ;in, As weight, Infrared image of the oil reservoir. The function is a multi-scale Gaussian kernel, where x is the pixel coordinate. The restored color image is: ,in, To enhance the strength coefficient, For color balance coefficient, It is a pseudo-color band. This is an infrared image of an oil reservoir.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the power plant oil tank monitoring method of multi-source data fusion as described in any one of claims 1-4.

7. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the power plant oil tank monitoring method of multi-source data fusion as described in any one of claims 1-4.

Images