Real-time multi-source data fusion risk early warning method and device thereof

Abstract

The application discloses a kind of real-time multi-source data fusion risk early warning method and device thereof, it is related to the technical field of multi-source data fusion risk early warning of hydraulic engineering, comprising: obtaining the multi-source data and working condition data of monitoring area under current working condition;Based on the same working condition historical data, the normalization processing is carried out, the protection appearance index and the floor impact risk index are calculated;The reverse coexistence intensity of each section is calculated, and the longitudinal inversion zone is determined;The interval of the longitudinal inversion zone is used as the control variable to dynamically adjust the semantic weight of each section;According to the weight after adjustment, the above index is weighted and calculated, the risk score is generated, and the in-zone risk hotspot position is located;According to the band level risk score, determine and output the early warning result.The application solves the problem that the surface air entrainment protection appearance covers the real impact risk of floor, effectively avoids the systematic underestimate of actual load risk of floor and the deviation of high risk positioning.

Description

Technical Field

[0001] This invention relates to the field of multi-source data fusion risk early warning technology for water conservancy projects, and in particular to a real-time multi-source data fusion risk early warning method and apparatus. Background Technology

[0002] The flow patterns along the control section to the energy dissipation section of high-head spillways or flood discharge tunnels are extremely complex during operation, generally exhibiting various flow states such as accelerated flow, local low-pressure zones, gas-water two-phase flow, and hydraulic jump pulsation zones. The dominant failure mechanisms in different sections often differ significantly, and the risks show obvious spatial non-uniformity and condition dependence along the flow path. In actual spillway monitoring, a highly deceptive phenomenon of reverse coexistence can easily occur: the surface near the wall region shows enhanced whitening or aeration, but the local pulsating impact and structural response at the bottom of the slab not only fail to decrease synchronously but continue to increase. Faced with this complex situation of reverse coexistence between surface protection and the actual impact risk of the slab, it is necessary to avoid the early warning system directly equating enhanced surface whitening with a reduction in overall risk, thereby preventing a systematic underestimation of the actual load risk on the slab and causing a misalignment of high-risk locations.

[0003] Existing real-time risk warning methods typically employ multi-source data fusion, including video image monitoring of base plate pressure and structural vibration monitoring. Common multi-source fusion methods often involve direct standardization followed by weighted average threshold voting or comprehensive judgment based on fixed rules. These approaches assume that different data sources indicate the same risk direction, severely ignoring the differences in physical layers and dominant risk mechanisms corresponding to different monitoring methods. When water flow exhibits a reverse coexistence phenomenon where surface protection appears enhanced while the actual impact on the base plate intensifies, existing methods struggle to identify semantic conflicts between surface monitoring data and underlying base plate data. This can easily lead to a severe masking effect of surface data on the actual load risk on the base plate. Such simplistic fusion not only causes pseudo-consistency or pseudo-conflict at the data level, prompting the warning system to systematically underestimate the actual impact risk on the base plate, but also leads to incorrect risk positioning biased towards areas of significant surface whitening rather than peak load areas on the base plate. Furthermore, changes in hydrodynamic conditions can easily be misjudged as sensor malfunctions or short-term disturbances, failing to achieve stable and accurate risk warning output. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies where multi-source data fusion methods simply assume that different data sources have the same semantic direction for risk. When there is a reverse coexistence phenomenon where the surface protection of water flow is enhanced while the actual impact on the bottom plate is aggravated, surface data can easily mask the actual risk of the bottom plate, resulting in a systematic underestimation of the actual load risk on the bottom plate and a shift in the location of high-risk locations. Therefore, this invention proposes a real-time multi-source data fusion risk early warning method and device.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: A real-time multi-source data fusion risk early warning method includes: S1. Acquire near-wall area image data, base plate pressure pulsation data, base plate vibration data, and current working condition data for each segment of the monitoring area under the current working conditions. S2. Based on historical data under the same working conditions, normalize the near-wall area image data, base plate pressure pulsation data and base plate vibration data respectively, and calculate the protection appearance index and base plate impact risk index. S3. Based on the protection appearance index and the impact risk index of the base plate, calculate the reverse coexistence intensity of each segment of the monitoring area, and determine the longitudinal reversal zone based on the reverse coexistence intensity. S4. Use the longitudinal reversal zone interval as a control quantity to adjust the semantic weights corresponding to each segment of the monitoring area. S5. Based on the adjusted semantic weights, the protection appearance index and the bottom plate impact risk index are weighted and calculated to generate risk scores for each segment and risk scores for each zone, and the risk hotspots are located within the longitudinal reversal zone. S6. Determine the warning level based on the risk rating and output the warning result.

[0006] Preferably, before acquiring data for each segment of the monitoring area, the control section to the energy dissipation section of the high-head spillway or overflow channel is divided into several segments longitudinally to obtain a longitudinal segment sequence.

[0007] Preferably, based on historical data under the same working conditions, the near-wall region image data, base plate pressure pulsation data, and base plate vibration data are normalized, including: Based on the similarity between the current operating condition data and the historical operating condition data, a sliding sample set with the same operating condition is selected from the historical samples as the adaptive baseline sample set. The operating condition data includes flow rate, gate opening, upstream water level and downstream water level. A robust normalization operator for the same working condition is constructed based on the median and absolute median difference in the sliding sample set under the same working condition.

[0008] Preferably, the calculation of the protection appearance index and the base plate impact risk index includes: Extract the near-wall whitening percentage from the near-wall region image data of each segment of the monitoring area; The near-wall whitening ratio is normalized using a robust normalization operator under the same working conditions to obtain the protection appearance index of each segment of the monitoring area.

[0009] Preferably, the calculation of the impact risk index of the base plate includes: Within the analysis window corresponding to the current moment, extract the pressure pulsation intensity of the pressure pulsation data of each segment of the base plate in the monitoring area, as well as the vibration response intensity of the vibration data of each segment of the base plate. The pressure pulsation intensity and vibration response intensity are normalized based on the robust normalization operator under the same working condition. Based on the fluctuation of each intensity quantity in the sliding sample set under the same working condition, an adaptive combined weight is constructed. The impact risk index of the base plate is obtained by weighting and summing the normalized pressure pulsation intensity and vibration response intensity using adaptive combined weights.

[0010] Preferably, based on the protection appearance index and the base plate impact risk index, the reverse coexistence intensity of each segment of the monitoring area is calculated, and the longitudinal reversal zone is determined based on the reverse coexistence intensity, including: The protection appearance index and the base plate impact risk index are compressed to a unified numerical range to obtain the normalized protection appearance index and the normalized base plate impact risk index. Multiply the normalized protection appearance index of the same segment by the normalized base plate impact risk index to obtain the reverse coexistence intensity of the corresponding segment. The anti-coexistence strength of each segment is vertically centered by subtracting the median of the anti-coexistence strength of all segments from the anti-coexistence strength of each segment. In the vertically segmented sequence, the reverse coexistence intensity after vertical centralization is used as the sequence value, and the maximum cumulative value of the continuous segment interval is searched; the continuous segment interval corresponding to the maximum cumulative value is taken as the vertical reversal zone. The average value of the reverse coexistence intensity of each segment within the longitudinal reversal zone is taken as the reversal zone intensity.

[0011] Preferably, the longitudinal reversal zone is used as a control variable to dynamically adjust the semantic weights corresponding to each segment of the monitoring area, including: The segmented reversal confidence level is constructed based on the relative prominence of the reverse coexistence intensity of each segment in the current process along the monitoring area. Construct a reversal zone indicator function, and assign different Boolean values ​​to the reversal zone indicator function according to whether each segment of the monitoring area is located within the longitudinal reversal zone; The piecewise inversion confidence is corrected based on the inversion band indicator function and the inversion band strength; Semantic weights are constructed using the corrected segmented inverse confidence.

[0012] Preferably, based on the adjusted semantic weights, the protection appearance index and the base plate impact risk index are weighted and calculated to generate risk scores for each segment and risk scores for each zone, and the risk hotspots are located within the longitudinal reversal zone, including: The normalized base plate impact risk index is weighted according to semantic weights, and the complementary value of the normalized protection appearance index is weighted by the complementary weights of the semantic weights. The sum of the two weighting results is used to obtain the segment risk score of each segment of the monitoring area. Calculate the average of the segment risk scores corresponding to all segments within the longitudinal reversal zone, and add the average value to the reversal zone intensity to obtain the zone-level risk score; The segment corresponding to the highest segment risk score within the vertical reversal zone is designated as the risk hotspot location.

[0013] Preferably, the warning level is determined based on the risk rating, and the warning result is output, including: On a sliding sample set under the same working conditions, calculate the abnormal intensity of the current zone-level risk score relative to the historical zone-level risk scores within the sample set. Calculate the quantile position of the current zone-level risk score in the sliding sample set under the same working conditions; Generate the corresponding warning level based on the quantile position; The output includes the longitudinal reversal zone, reversal zone intensity, zone-level risk score, anomaly intensity, hotspot location within the zone, and warning level as the warning result.

[0014] To address the aforementioned problems, the present invention also provides a real-time multi-source data fusion risk early warning device, the device comprising: The multi-source data acquisition module is used to acquire near-wall area image data, base plate pressure pulsation data, base plate vibration data, and current working condition data for each segment of the monitoring area under the current working conditions. The risk index calculation module is used to normalize the near-wall area image data, base plate pressure pulsation data and base plate vibration data based on historical data under the same working conditions, and calculate the protection appearance index and the base plate impact risk index. The longitudinal reversal zone module is used to calculate the reverse coexistence intensity of each segment of the monitoring area based on the protection appearance index and the base plate impact risk index, and to determine the longitudinal reversal zone based on the reverse coexistence intensity. The semantic weight module is used to adjust the semantic weights of each segment of the monitoring area by using the vertical reversal zone as a control variable. The risk scoring module is used to perform weighted calculations on the protection appearance index and the bottom plate impact risk index based on the adjusted semantic weights, generate risk scores for each segment and risk scores for each zone, and locate risk hotspots within the longitudinal reversal zone. The early warning classification module is used to determine the early warning level based on the risk score and output the early warning result.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In this invention, by extracting the near-wall whitening ratio from near-wall region image data and the load characteristics from bottom plate pressure and vibration data, a protection appearance index and a bottom plate impact risk index are constructed. Then, the reverse coexistence intensity of each segment of the monitoring area is quantitatively calculated and continuous longitudinal reversal zones are extracted. This directly addresses the complex and specific phenomenon of enhanced surface protection appearance and increased actual bottom plate impact risk during high-head discharge operation. It breaks the limitation of assuming that all monitoring data have completely consistent risk indication directions in multi-source data fusion. It can accurately identify and lock the spatial concentration area where there are differences in physical layer and risk mechanism between near-wall aeration appearance and bottom plate bottom layer data, effectively eliminating serious misleading interference caused by surface protection appearance.

[0016] 2. In this invention, the extracted longitudinal reversal zone intervals and their intensity are used as core control variables. The semantic weights corresponding to each segment of the monitoring area are dynamically adjusted, and the decision-making proportion of the monitoring data of the bottom plate is specifically increased within the longitudinal reversal zone. This can successfully suppress the masking effect of apparent data such as enhanced surface whitening on the actual load risk of the bottom plate, eliminate the problem of false consistency or false conflict at the data level, not only fundamentally prevent the systematic underestimation of the actual impact damage risk of the bottom plate, but also correct the problem of risk hotspot positioning biased towards areas with significant surface whitening. Furthermore, combined with the sliding sample set under the same working conditions, the accurate zone-level risk score warning level and the location of hotspots within the zone are output, ultimately achieving stable, accurate, and highly physically interpretable risk warning under complex hydraulic evolution conditions. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a real-time multi-source data fusion risk warning method according to an embodiment of the present invention. Figure 2 This is a functional block diagram of a real-time multi-source data fusion risk warning device provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example: This example provides a real-time multi-source data fusion risk early warning method. See [link to example]. Figure 1 Specifically, including: S1. Acquire near-wall area image data, base plate pressure pulsation data, base plate vibration data, and current working condition data for each segment of the monitoring area under the current working conditions. S2. Based on historical data under the same working conditions, normalize the near-wall area image data, base plate pressure pulsation data and base plate vibration data respectively, and calculate the protection appearance index and base plate impact risk index. S3. Based on the protection appearance index and the impact risk index of the base plate, calculate the reverse coexistence intensity of each segment of the monitoring area, and determine the longitudinal reversal zone based on the reverse coexistence intensity. S4. Use the longitudinal reversal zone interval as a control quantity to adjust the semantic weights corresponding to each segment of the monitoring area. S5. Based on the adjusted semantic weights, the protection appearance index and the bottom plate impact risk index are weighted and calculated to generate risk scores for each segment and risk scores for each zone, and the risk hotspots are located within the longitudinal reversal zone. S6. Determine the warning level based on the risk rating and output the warning result.

[0020] In embodiments of the present invention, acquiring near-wall region image data, base plate pressure pulsation data, base plate vibration data, and current-moment operating condition data for each segment of the monitoring area under the current operating conditions specifically includes: First, the entire longitudinal region from the inlet section of the control section to the outlet section of the energy dissipation section of a high-head spillway or overflow channel is demarcated. The demarcation is based on the flow regime changes, structural stress characteristics, and monitoring requirements of the region. Priority is given to ensuring relatively uniform flow regime and structural characteristics within each segment, and that each segment can accommodate a complete monitoring unit. The number of segments is reasonably determined based on the total length from the control section to the energy dissipation section and the required monitoring accuracy. Typically, for the control section to energy dissipation section of a high-head spillway or overflow channel with a total length of 80 to 150 meters, it is evenly divided into 8 to 15 segments along the longitudinal flow direction. The inlet section of the control section is used as the starting reference point for the longitudinal segmentation. The flow direction is divided sequentially, with the length of each segment controlled between 5 and 20 meters. The boundaries between adjacent segments are clearly defined by pre-embedded marker stakes or by marking obvious marks on the structural surface to avoid ambiguity in segment boundaries that could lead to deviations in subsequent data acquisition. After the division is completed, all segments are sequentially numbered from the inlet of the control section to the outlet of the energy dissipation section to form a complete longitudinal segment sequence. Each segment corresponds to a unique number, ensuring that the subsequent acquisition of near-wall area image data, base plate pressure pulsation data, and base plate vibration data can accurately correspond to each segment. This provides a clear and definite spatial division basis for the calculation of the protection appearance index and base plate impact risk index of each segment, as well as the extraction of the longitudinal reversal zone.

[0021] Based on the longitudinal segmented sequence formed by the division, a complete set of monitoring equipment is deployed for each segment to achieve accurate and synchronous acquisition of various types of data. Among them, the near-wall area image data is collected by industrial high-definition cameras fixedly installed at preset positions on the near-wall sidewall of each segment. The lens of the industrial high-definition camera is facing the near-wall surface of the corresponding segment, and the lens axis is at an angle of 30 to 45 degrees with the near-wall surface to ensure that the entire range of the near-wall area of ​​the corresponding segment can be captured without any blind spots. The camera resolution is set to no less than 1920×1080, the frame rate is adjusted to 25 frames / second, and the image sequence of the near-wall area is continuously collected to capture the whitening phenomenon and gas entrainment state changes of the near-wall surface in real time, forming the exclusive near-wall area image data for each segment.

[0022] The pressure pulsation data of the base plate is collected by piezoelectric pressure sensors embedded in the center of each segment of the base plate. The sensing surface of the sensor is flush with the surface of the base plate to avoid damage to the sensor caused by water flow impact and to ensure the authenticity of the collected data. The sampling frequency of the sensor is set to no less than 1000Hz, and the pressure pulsation signal of the corresponding segment of the base plate is continuously collected to capture the instantaneous fluctuation characteristics of the pressure and form the base plate pressure pulsation data. The vibration data of the base plate is collected by piezoelectric accelerometers arranged next to the pressure sensors of each segment of the base plate. The sensors are fixed to the surface of the base plate with high-strength bolts to ensure efficient transmission of vibration signals. The sampling frequency is set to no less than 500Hz, and the vibration response signal of the corresponding segment of the base plate is continuously collected to capture the vibration amplitude and frequency changes of the base plate and form the base plate vibration data.

[0023] Current operating data is collected synchronously through corresponding dedicated monitoring instruments. Flow data is collected through electromagnetic flow meters installed at the inlet of the spillway or in front of the spillway gate. Gate opening data is obtained in real time through the position sensor built into the gate control system. Upstream water level data is collected through a static pressure level gauge installed 10 to 15 meters upstream of the gate. Downstream water level data is collected through a static pressure level gauge installed at the end of the energy dissipation section. The collection frequency of various operating data is synchronized with the aforementioned image, pressure, and vibration data, all collected once every 100 milliseconds. All monitoring equipment is electrically connected to a unified time synchronization module and uses the same timestamp for synchronous triggering and collection, ensuring that the near-wall area image data, bottom plate pressure pulsation data, bottom plate vibration data, and current operating data of each segment are accurately correlated in the time dimension. All types of raw data collected are transmitted in real time to the temporary storage module of the data processing unit for caching, with a caching time of not less than 24 hours, to facilitate subsequent data retrieval and anomaly tracing.

[0024] It should be noted that each segment of the monitoring area refers to a spatial unit longitudinally divided along the control section of the high-head spillway or overflow channel to the energy dissipation section. Each segment corresponds to a section of actual water flow channel and bottom structure area, used to unify image observation, pressure observation, and vibration observation to the same spatial location. Specifically, the section from the control section of the high-head spillway or overflow channel to the energy dissipation section refers to a continuous engineering section in the spillway structure that starts from the key control point controlling the discharge and forming the outflow conditions, extends along the main flow direction, and ends at the energy dissipation point used to reduce kinetic energy and mitigate impact. The "high head" indicates that this section has significant upstream and downstream water flow during operation. The difference in elevation and the strong driving force of water flow indicate that the spillway or flood discharge tunnel is responsible for discharging floodwater and transporting high-speed water flow. The control section refers to the area formed by the gate or control section and its adjacent structures, which plays a decisive role in the outflow velocity distribution, pressure state and initial flow state. The energy dissipation section refers to the area where the flow velocity is reduced and energy is dissipated through hydraulic processes such as hydraulic jump, diffusion, aeration, tumbling or stilling basin. Therefore, the control section to the energy dissipation section reflects the actual spatial range corresponding to the continuous hydraulic evolution and structural loading process from the controlled formation of outflow to the completion of energy dissipation. It is also the target monitoring area for segmented monitoring and real-time multi-source data fusion risk early warning.

[0025] It should be noted that near-wall region image data refers to image information collected from the area adjacent to the water flow and the sidewall or bottom plate. It reflects visible phenomena such as near-wall flow regime, aeration whitening, bubble distribution, and apparent morphology of the water flow adhering to the wall, and is used to characterize whether there is enhanced protective appearance or flow regime change in the near-wall region. Bottom plate pressure pulsation data refers to the pressure fluctuation data measured over time by pressure measuring points deployed on the surface or near the surface of the bottom plate. It reflects the dynamic pressure loading process of the bottom plate caused by water turbulence, hydraulic jump impact, air-water two-phase flow, and local flow instability. Bottom plate vibration data refers to the displacement velocity or acceleration data collected by vibration sensors installed on the bottom plate structure as a function of time. It reflects the strength and change of the structural response of the bottom plate under dynamic water flow load. Current operating condition refers to the operating state and boundary conditions of the discharge system at the current moment, reflecting the water flow driving force and structural loading environment at that moment, and is characterized by flow rate, gate opening, upstream water level, and downstream water level.

[0026] In embodiments of the present invention, based on historical data under the same working conditions, the near-wall region image data, base plate pressure pulsation data, and base plate vibration data are normalized, and the protection appearance index and base plate impact risk index are calculated, specifically including: To ensure the accuracy and adaptability of subsequent data normalization processing, it is necessary to first select a sliding sample set of the same operating conditions as the adaptive baseline sample set based on the similarity between the current operating condition data and the historical operating condition data. The operating condition data specifically includes four core parameters: flow rate, gate opening, upstream water level, and downstream water level. These four parameters are key factors affecting the flow pattern and structural stress state along the high-head spillway or floodgate, and directly determine the benchmark level of the monitoring data. Therefore, these four parameters are selected as the core indicators for judging the similarity of operating conditions.

[0027] Historical operating condition data and corresponding monitoring data stored in the data storage unit are retrieved to form a complete historical sample library. The historical operating condition data stored in the historical sample library needs to cover combinations of flow rate, gate opening, upstream water level, and downstream water level under different operating scenarios to ensure the comprehensiveness and representativeness of the samples. Subsequently, the Euclidean distance between the four operating condition parameters at the current time and the four operating condition parameters corresponding to each historical sample in the historical sample library is calculated to quantify the similarity between the current operating condition and each historical operating condition. The formula for calculating the Euclidean distance is: in This represents the similarity distance between the current working condition and the working condition of the k-th historical sample. This represents the current flow rate. This represents the traffic of the k-th historical sample. This indicates the current gate opening degree. This represents the gate opening degree of the k-th historical sample. This indicates the current upstream water level. This represents the upstream water level of the k-th historical sample. This indicates the current downstream water level. This represents the downstream water level of the k-th historical sample. The reason for using Euclidean distance is that it can comprehensively reflect the overall differences of the four operating parameters. The smaller the distance, the more similar the current operating condition is to the operating condition of the historical sample, and the more accurately the historical sample that matches the current operating condition can be selected. All historical samples are sorted in ascending order of Euclidean distance. The top N historical samples with the smallest distances are selected to form a sliding sample set under the same operating conditions. This serves as the adaptive baseline sample set for the current data normalization process. The value of N is determined based on the size of the historical sample database, typically between 30 and 50 samples. This range is chosen because too few samples would result in a lack of representativeness in the baseline samples, failing to reflect the true benchmark level of the current operating conditions. Too many samples would increase the computational load and affect real-time processing efficiency. Furthermore, the sliding sample set under the same operating conditions uses a sliding update method. Whenever new current operating condition data is acquired and sample selection is completed, the oldest historical sample is automatically removed, and new similar historical samples are added to ensure that the baseline sample set can adapt to the dynamic changes in operating conditions in real time.

[0028] After selecting a sliding sample set under the same operating conditions, a robust normalization operator for the same operating conditions is constructed based on the median and absolute median difference in this sample set. The purpose of constructing this operator is to avoid the interference of outliers on the normalization results and ensure the robustness of the normalization process. Because there may be outliers in the monitoring data of water conservancy projects caused by instantaneous sensor failures, instantaneous disturbances in water flow, etc., the median and absolute median difference can effectively suppress the influence of outliers and have stronger anti-interference ability compared with the mean and standard deviation. The calculation formula of the robust normalization operator for the same operating conditions is as follows: in This represents the normalized value. This represents the raw monitoring data to be normalized. Specifically, the raw monitoring data here refers to the near-wall whitening percentage, pressure pulsation intensity, or vibration response intensity. Represents the sliding sample set under the same working conditions All m-th samples correspond to the original monitoring data The median is used because it is not affected by extreme outliers and can accurately reflect the centrality of the sample set. Represents the sliding sample set under the same working conditions All m-th samples correspond to the original monitoring data The absolute median is calculated by first calculating the absolute median for each sample in the sliding sample set under the same working conditions. The absolute median difference is the absolute difference between the number of digits in the sample set and the median of all absolute differences. The reason for using the absolute median difference is that it can reflect the dispersion of the sample data and also has a strong ability to resist outlier interference.

[0029] It should be noted that the near-wall whitening ratio refers to the proportion of the area in the near-wall region image data of each segment of the monitoring area that appears obviously white, shiny, or foamy. It is used to characterize the spatial extent of phenomena such as near-wall bubble aggregation, enhanced aeration, and changes in the apparent morphology of water flow adhering to the wall. The larger the value, the more significant the whitening phenomenon is in the near-wall region of that segment. The pressure pulsation intensity refers to the strength of the pressure fluctuation over time reflected by the pressure pulsation data of the bottom plate within the current analysis window. It is used to characterize the intensity of the dynamic pressure effect on the bottom plate caused by water turbulence, hydraulic jump impact, gas-water two-phase flow fluctuation, and local unstable flow. Its essence is a quantitative description of the amplitude of dynamic water pressure fluctuation of the bottom plate. The vibration response intensity refers to the magnitude or energy level of the vibration response of the bottom plate structure reflected by the vibration data of the bottom plate within the current analysis window. It is used to characterize the strength of the structural response of the bottom plate under dynamic water flow load. The increase of its value usually indicates that the dynamic excitation effect on the bottom plate structure in that segment is enhanced or the response activity is increased.

[0030] It should be noted that the same-condition robust normalization operator refers to using the median of the same-condition sliding sample set as the position benchmark and the absolute median difference as the fluctuation scale to perform dimensionless processing on the corresponding characteristic quantities of each segment at the current moment. It is used to eliminate the dimensional differences and the influence of normal operating condition fluctuations under similar operating conditions such as flow rate, gate opening, upstream water level and downstream water level, so that the near-wall whitening ratio, pressure pulsation intensity and vibration response intensity can be compared and integrated on a unified scale, while reducing the interference of a small number of abnormal historical samples on the normalization results.

[0031] The near-wall region image sequences of each segment are preprocessed to remove noise interference and irrelevant background. Specifically, Gaussian filtering algorithm is used to denoise single-frame images. The kernel size of Gaussian filtering is set to 3×3. The reason for using this kernel size is that it can effectively filter out the small noise generated by water flow disturbance and the noise formed by the reflection of impurities in the air, while avoiding the loss of details in the near-wall whitened area due to over-filtering. Subsequently, histogram equalization algorithm is used to enhance image contrast, making the boundary between the whitened and non-whitened parts of the near-wall region clearer, which is convenient for subsequent region segmentation. The processing range of histogram equalization is limited to the effective monitoring area of ​​the near-wall region image, excluding invalid background areas at the image edges.

[0032] After preprocessing, the whitening regions of the single-frame near-wall images in each segment are segmented. The Otsu adaptive threshold segmentation algorithm is used to determine the segmentation threshold. This algorithm can automatically calculate the optimal segmentation threshold based on the grayscale distribution of the current image, adapting to the differences in the degree of near-wall whitening under different discharge conditions, and avoiding the problems of missed or incorrect segmentation caused by fixed threshold segmentation. The algorithm divides the image into whitening and non-whitening regions, where the whitening region corresponds to the bright area formed by near-wall gas mixing, and the non-whitening region corresponds to the normal surface area of ​​the near-wall. After segmentation, the total number of pixels in the effective monitoring area of ​​the near-wall image in each segment and the number of pixels in the whitening region are counted. The effective monitoring area of ​​the near-wall image is determined by pre-calibration, that is, a fixed effective area is defined in the image based on the shooting range of the industrial high-definition camera and the actual position of the near-wall region, ensuring that the area range calculated each time is consistent. The near-wall whitening ratio is calculated by the ratio of the number of pixels in the whitening region to the total number of pixels in the effective monitoring area, which can accurately quantify the relative degree of near-wall whitening.

[0033] After extracting the near-wall whitening percentage of each segment at the current moment, normalization is performed on it based on the robust normalization operator constructed above, yielding the protection appearance index for each segment. The formula for calculating the protection appearance index is as follows: ,in This represents the protection appearance index of the i-th segment at the current time n, used to characterize the relative enhancement of the near-wall gas-incorporation protection appearance of that segment. This indicates a robust normalization operator under the same operating conditions, which can effectively suppress outlier interference and ensure the robustness of the normalization results. This represents the near-wall whitening percentage of the i-th segment at the current time n, which is the ratio of the number of whitened pixels extracted to the total number of pixels in the effective monitoring area. The baseline level of the near-wall whitening percentage varies under different operating conditions. Directly using the original whitening percentage cannot accurately reflect the relative strength of the gas-entrained protection phenomenon. After robust normalization under the same operating conditions, the near-wall whitening percentage of each segment can be converted into a relative value, eliminating the influence of differences in operating conditions and making the protection phenomenon index of different segments and different times comparable. The larger the value, the stronger the near-wall gas entrainment phenomenon in that segment, which perfectly matches the definition of the protection phenomenon index.

[0034] It should be noted that the protection appearance index is a characterizing quantity obtained by normalizing the near-wall whitening ratio of each segment of the monitoring area using a robust normalization operator under the same operating conditions. It describes the degree of enhancement or reduction of the near-wall whitening appearance under the current operating conditions compared to the historical normal level under the same operating conditions. The near-wall whitening ratio reflects the spatial proportion of visible whitening appearance in the observation area, while the protection appearance index further reflects the magnitude and direction of the deviation of this whitening appearance from the typical level of historical samples under the same operating conditions. A larger value indicates that the near-wall whitening phenomenon of the segment is more significant than that under normal operating conditions. This usually corresponds to a more obvious enhancement of near-wall aeration or a more obvious change in the visible flow pattern along the wall. When the protection appearance index value is close to the normal level, it indicates that the near-wall whitening phenomenon of the segment is similar to the historical state under the same operating conditions. When the protection appearance index value is small, it indicates that the near-wall whitening phenomenon of the segment is weaker than the historical normal level under the same operating conditions. Therefore, this index is used to express the near-wall visible protection appearance under different operating conditions in a unified scale and serves as an important input for subsequent calculation of reverse coexistence intensity and fusion risk warning.

[0035] A short-time analysis window corresponding to the current moment is determined. This window selects a continuous data segment that extends 5 seconds backward from the current moment. The reason for choosing a 5-second duration is that the characteristic time scale of pressure pulsations and structural vibrations of the bottom slab of high-head spillways or overflow channels in water conservancy projects is concentrated in the range of several seconds. This duration can fully cover the main periodic characteristics of pressure pulsations and vibration responses while ensuring the real-time nature of data processing and meeting the timeliness requirements of risk warning. The sliding step size of the analysis window is kept consistent with the data acquisition frequency to ensure that the window data is synchronized with the current operating conditions. Within the determined analysis window, the bottom slab pressure pulsation sequence of the i-th segment is first preprocessed. The median within the window is calculated, and the pressure value at each moment is subtracted from the median to remove the DC component from the pressure signal. The DC component corresponds to the static pressure of the bottom slab. Removing it allows for accurate focusing on the dynamic pressure pulsation characteristics, avoiding the static pressure masking the true level of pulsation intensity. Subsequently, the absolute value of the preprocessed pressure value is calculated, and the pressure pulsation intensity is extracted using the 95th percentile method. The formula for calculating the pressure pulsation intensity is as follows: ,in This represents the pressure pulsation intensity of the i-th segment at the current time n. The value within the parentheses is taken as the 95th percentile. The 95th percentile is used because it reflects both the extreme fluctuations in pressure pulsations and avoids interference from individual instantaneous peak outliers, making it more robust than the maximum value and meeting the practical needs of stress risk assessment for hydraulic engineering structures. This represents the absolute difference between the pressure pulsation value of the i-th segment of the bottom plate within the analysis window and the median pressure within that window. This represents the pressure pulsation sequence of the i-th segment of the base plate within the analysis window. The median is used instead of the mean because it has a strong anti-interference ability against abnormal values ​​caused by instantaneous sensor failures and instantaneous water flow impacts, and can more realistically reflect the benchmark level of pressure pulsation.

[0036] The vibration sequence of the base plate in the i-th segment is preprocessed. The median within the analysis window is calculated, and the vibration value at each time step is subtracted from this median to remove the static offset from the vibration signal, thus preventing the static offset from masking the true intensity of the vibration response. Subsequently, the vibration response intensity is extracted using the root mean square method. The formula for calculating the vibration response intensity is as follows: ,in This represents the vibration response intensity of the i-th segment at the current time n. This indicates that the root mean square (RMS) is calculated for the numerical sequence within parentheses. The specific calculation method for the RMS is to divide the sum of the squares of the differences between the vibration values ​​at each time point within the window and the median by the number of data points within the window, and then take the square root of the result. The reason for using the RMS is that it is a classic indicator in the field of vibration engineering for characterizing the energy and intensity of structural vibration. It can comprehensively reflect the vibration amplitude and duration, and conforms to the industry standards for assessing the vibration response of foundation slab structures. This represents the difference between the vibration value of the i-th segment of the base plate within the analysis window and the median vibration value within that window. This represents the vibration sequence of the i-th segment of the base plate within the analysis window. of the median.

[0037] The pressure pulsation intensity of each segment was extracted. and vibration response intensity Then, based on the aforementioned robust normalization operator for the same operating condition, normalization processing is performed on it respectively, and the normalization calculation formula is as follows: and ,in This represents the normalized pressure pulsation intensity of the i-th segment at the current time n. This represents the normalized vibration response intensity of the i-th segment at the current time n. This is a robust normalization operator for the same working conditions. It is constructed based on the median and absolute median difference of the sliding sample set under the same working conditions and has strong resistance to outliers. The reason for using this operator for normalization is that the benchmark strength of the pressure pulsation and vibration response of the base plate under different discharge conditions is significantly different. Directly using the original strength value cannot accurately reflect the relative risk level under the current working condition. After normalization, the influence of the working condition difference can be eliminated, so that the strength index of different segments and different times has a unified comparison benchmark, providing reliable basic data for the construction of the base plate impact risk index.

[0038] An adaptive combined weight is constructed based on the discrete fluctuation degree of pressure pulsation intensity and vibration response intensity in the sliding sample set under the same working condition. The formula for calculating the adaptive combined weight is as follows: in This represents the adaptive combined weight of the i-th segment at the current time n. This represents the absolute median difference in pressure pulsation intensity of the i-th segment within the sliding sample set under the same operating conditions, used to characterize the overall fluctuation of pressure pulsation intensity under the same operating conditions. The absolute median difference represents the vibration response intensity of the i-th segment within the sliding sample set under the same working condition. It is used to characterize the overall fluctuation of the vibration response intensity under the same working condition. The reason for using the absolute median difference to measure the volatility is that it has excellent resistance to outlier interference and can truly reflect the stable fluctuation characteristics of each intensity quantity under the same working condition. Assigning weights based on this volatility can achieve adaptive dynamic adjustment of the weights. When the fluctuation of a certain intensity quantity is more significant, it is given a higher weight, so that the risk index is more in line with the dominant risk characteristics under the working condition and avoids the assessment bias caused by fixed weights.

[0039] After constructing the adaptive combined weights, the normalized pressure pulsation intensity and vibration response intensity are weighted and summed using these weights to obtain the corresponding segment's base plate impact risk index. The formula for calculating the base plate impact risk index is as follows: ,in This represents the impact risk index of the i-th segment at the current time n. The larger the index value, the higher the actual impact risk of the base plate under load. This represents the normalized pressure pulsation intensity of the i-th segment at the current time n. The index represents the vibration response intensity of the i-th segment after normalization at the current time n. The reason for using a weighted summation method is that it can integrate the advantages of both pressure and vibration monitoring data. The introduction of adaptive weights allows the index to adapt to the sensitivity differences of the two types of monitoring signals under different working conditions, and comprehensively and accurately reflect the impact risk level borne by the base plate structure.

[0040] It should be noted that the slab impact risk index is a characterization quantity obtained by extracting the pressure pulsation intensity and vibration response intensity from the slab pressure pulsation data and vibration data of each segment of the monitoring area within the current analysis window, normalizing them using the robust normalization operator under the same working condition, and then weighting and summing them using adaptive combined weights. It is used to describe the comprehensive risk status of the slab segment under the current working condition under dynamic water flow load impact and the degree of structural response activity relative to the historical normal level under the same working condition. The pressure pulsation intensity characterizes the dynamic pressure effect of water flow turbulence, hydraulic jump impact, and air-water two-phase flow fluctuations on the slab. The strength of the vibration response intensity characterizes the response amplitude or energy level of the base plate structure under dynamic excitation. The adaptive combined weight is used to determine the contribution ratio of the two types of information in the comprehensive characterization based on the fluctuation characteristics of each intensity quantity in the sliding sample set under the same working condition. This allows the base plate impact risk index to reflect both the dynamic pressure change on the load side of the base plate and the actual response state on the response side of the structure. When the index value is large, it indicates that the impact risk of the segmented base plate is significantly higher than the historical normal level under the same working condition. When the index is close to the normal level, it indicates that the load and response state of the segmented base plate is basically consistent with the historical state under the same working condition.

[0041] In embodiments of the present invention, the reverse coexistence intensity of each segment of the monitoring area is calculated based on the protection appearance index and the base plate impact risk index, and the longitudinal reversal zone is determined based on the reverse coexistence intensity, specifically including: To eliminate the numerical and dimensional differences between the protection appearance index and the base plate impact risk index, and to provide a unified basis for comparison and calculation between the two indices, a logic function is used to compress both the protection appearance index and the base plate impact risk index to a unified numerical range of zero to one, thereby obtaining the normalized protection appearance index and the normalized base plate impact risk index. The calculation formula for the logic function is as follows: ,in This represents the output of a logic function. This represents the protection appearance index or base plate impact risk index to be compressed. The reason for using this logic function for compression is that it can smoothly map input values ​​within any real number range to the zero-to-one interval, preserving the relative magnitude and trend of the original index while preventing extreme values ​​from affecting subsequent calculation results. It possesses good mapping characteristics and numerical stability. The formula for calculating the normalized protection appearance index is: ,in This represents the normalized protected appearance index of the i-th segment at the current time n. This represents the protection appearance index of the i-th segment at the current time n. The formula for calculating the normalized base plate impact risk index is: ,in This represents the normalized impact risk index of the i-th segment at the current time n. This represents the impact risk index of the i-th segment at the current time n.

[0042] After normalizing and compressing the two types of indices, the normalized protection appearance index and the normalized base plate impact risk index for the same segment are multiplied to obtain the corresponding reverse coexistence strength. The formula for calculating the reverse coexistence strength is as follows: ,in The reverse coexistence intensity of the i-th segment at the current time n is represented by the multiplication operation. The reason for using multiplication to construct the reverse coexistence intensity is that only when the normalized protection appearance index and the normalized base plate impact risk index are both at a high value will their product increase significantly, thereby accurately identifying the segment region where the near-wall gas entrainment protection appearance is enhanced and the base plate impact risk increases synchronously.

[0043] It should be noted that the reverse coexistence intensity refers to a characteristic quantity obtained by combining the normalized protection appearance index and the normalized base plate impact risk index for a certain segment of the monitoring area. This quantity describes whether the segment simultaneously exhibits two opposing but potentially coexisting states in actual engineering: enhanced near-wall aeration protection appearance and increased base plate impact risk, and the degree of their coexistence. Specifically, the normalized protection appearance index characterizes the degree of enhancement of visible whitening in the near-wall area relative to historical normal levels under the same working conditions, while the normalized base plate impact risk index characterizes the degree of increase in dynamic load and structural response of the base plate relative to historical normal levels under the same working conditions. The coexistence strength is characterized by jointly representing both the protection appearance and the impact risk of the base plate. This value increases significantly only when both are significantly enhanced. This highlights segments that appear to have a stronger protective appearance but whose actual load risk on the base plate has not decreased but rather increased. When the reverse coexistence strength is large, it indicates that the segment is more likely to be in a state of reverse coexistence between the near-wall protection appearance and the impact risk of the base plate. When the reverse coexistence strength is small, it indicates that at least one side of the segment is not significant or the two do not form a significant coexistence. Therefore, this value is used for subsequent longitudinal centering, longitudinal reversal zone extraction, and semantic weight adjustment. It is the core characterization indicator that transforms specific characteristics into computable quantities.

[0044] The anti-coexistence intensity of all segments is vertically centered. The process involves first calculating the median of the anti-coexistence intensity of all segments in the entire vertical segment sequence. The median is calculated by arranging the anti-coexistence intensities of all segments in ascending order. If the total number of segments is odd, the value of the middle position is taken; if it is even, the average of the two middle values ​​is taken. The median is used for centering because it is not affected by extreme outliers and can accurately reflect the overall baseline level of the anti-coexistence intensity of all segments, avoiding the impact of overall data offset on subsequent region extraction. Then, the anti-coexistence intensity of each segment is subtracted from the overall median to obtain the vertically centered anti-coexistence intensity of each segment. This process can effectively eliminate the influence of overall baseline fluctuations, highlight the segment regions with relatively high anti-coexistence intensity, and lay the foundation for subsequent continuous interval search.

[0045] After completing the vertical centering process, a search for the maximum cumulative value of consecutive segment intervals is performed on the vertical segmented sequence. The search process employs a segment-by-segment traversal approach. Specifically, starting from the first segment of the vertical segmented sequence, consecutive segment intervals are selected sequentially. The cumulative value of the anti-coexistence intensity after centering of all segments within each consecutive interval is calculated. All possible consecutive segment intervals are traversed, ensuring no consecutive combination is overlooked. The length of each consecutive segment interval is at least two segments to avoid misjudging individual, scattered segments with high anti-coexistence intensity as vertical reversal zones, thus ensuring the continuity of the vertical reversal zone. After the traversal is complete, the consecutive segment interval with the largest cumulative value is selected; this interval is the vertical reversal zone. The reason for searching for the maximum cumulative value is that the vertical reversal zone is a continuous region with relatively prominent anti-coexistence strength. Its cumulative value after centering is necessarily higher than other continuous intervals, which can accurately pinpoint the region where anti-coexistence is most concentrated. After determining the vertical reversal zone, the anti-coexistence strength of all segments within the vertical reversal zone is statistically analyzed. The anti-coexistence strength of these segments is summed, and the sum is divided by the number of segments within the vertical reversal zone. The average value obtained is the reversal zone strength. The reason for using the average value to calculate the reversal zone strength is that it can comprehensively reflect the anti-coexistence strength level of all segments within the vertical reversal zone, quantify the overall strength of the reversal zone, and avoid the influence of outliers of a single segment on the judgment of the overall characteristics of the reversal zone.

[0046] It should be noted that the longitudinal reversal zone refers to the continuous segmented intervals identified based on the distribution characteristics of the reverse coexistence intensity of each segment after longitudinal centering processing, on the longitudinally segmented sequence formed by dividing the control section to the energy dissipation section of the high-head spillway or overflow channel. This zone characterizes the spatial region where the reverse coexistence state of enhanced near-wall aeration protection and increased risk of bottom impact is concentrated along the flow direction. The longitudinal centering process eliminates the overall baseline influence of the current moment's overall reverse coexistence intensity across all segments and highlights relatively prominent sections. The continuous segmented intervals maintain the spatial continuity of this state during actual hydraulic processes and structural loading. This longitudinal reversal band is not an anomaly marker of a single discrete measuring point or a single segment, but rather reflects the banded distribution range of specific characteristics along the route from the control section to the energy dissipation section. When the length of the longitudinal reversal band is large or the overall reverse coexistence intensity of the corresponding segment is high, it usually indicates that the semantic contrast between the surface protection appearance and the actual risk of the base plate within the continuous section is more significant, which may be more misleading for real-time multi-source data fusion risk warning. Therefore, the longitudinal reversal band is used to construct the reversal band indicator function, adjust the fusion semantic weight, generate the band-level risk score, and locate the risk hotspot. It is a key result in elevating the segment-level reverse coexistence state to a spatial continuous risk judgment unit.

[0047] It should be noted that the reverse zone strength refers to the characterization quantity obtained by statistically summarizing the reverse coexistence strength of each segment within the longitudinal reverse zone. It is preferably the average level of the reverse coexistence strength of each segment within the longitudinal reverse zone. It is used to describe the overall significance and concentration of the reverse coexistence state within the longitudinal reverse zone. The greater the reverse zone strength, the more obvious the coexistence of enhanced near-wall protection and increased impact risk of the base plate within the continuous section.

[0048] In an embodiment of the present invention, the longitudinal reversal zone is used as the semantic weight corresponding to each segment of the control quantity adjustment monitoring area, specifically including: The piecewise reversal confidence level is constructed based on the relative prominence of the reversal coexistence intensity of each segment within the entire longitudinal segmented sequence. The formula for calculating the piecewise reversal confidence level is as follows: ,in This represents the confidence level of segment reversal for the i-th segment at the current time n. The larger the value, the higher the confidence level that the segment exhibits reversal coexistence. Representing a logistic function, it can smoothly map the calculation results to the interval between zero and one, ensuring the standardization and comparability of confidence scores. This represents the reverse coexistence strength of the i-th segment at the current time n. This represents the median intensity of the opposite coexistence across all longitudinal segments along the entire route, used to eliminate the impact of overall baseline fluctuations along the route. The absolute median difference represents the total negative coexistence intensity of all longitudinal segments along the entire path. It is used to objectively measure the overall dispersion of the negative coexistence intensity along the path. The reason for using this formula to construct the segmented reversal confidence score is that it can accurately quantify the relative prominence of the negative coexistence intensity of each segment relative to the overall level along the path. At the same time, by taking advantage of the strong anti-anomaly characteristics of the median and absolute median difference, it avoids the interference of extreme values ​​on the confidence score calculation. The mapping processing of the logistic function can keep the confidence score within a reasonable range, which is convenient for subsequent weighted calculation and numerical matching.

[0049] Construct the inversion band indicator function. The formula for calculating the inversion band indicator function is as follows: in This represents the inversion zone indicator function for the i-th segment at the current time n. When the i-th segment is within the extracted vertical inversion zone range, the inversion zone indicator function takes the value of one; when the i-th segment is outside the vertical inversion zone range, the inversion zone indicator function takes the value of zero. The clear distinction made by this Boolean value can accurately identify the spatial boundary of the vertical inversion zone, providing a clear spatial judgment basis for subsequent correction of segment inversion confidence and dynamic adjustment of semantic weights.

[0050] The confidence level of the piecewise reversal is corrected, and the corrected calculation formula is as follows: ,in This represents the segment reversal confidence level of the i-th segment after correction at the current time n, used to more accurately characterize the true confidence level of the segment exhibiting reverse coexistence. This represents the confidence level of the i-th segment in the original segment reversal at the current time n, before any correction. This represents the inversion band indicator function for the i-th segment at the current time n. Its value is either one or zero, and it is used to indicate whether the segment is located within the vertical inversion band. This represents the inversion band intensity at the current time n, used to quantify the overall anti-coexistence intensity level of the longitudinal inversion band. The reason for using this formula for correction is that it can inject the spatial location information and overall intensity information of the inversion band into the segmented inversion confidence. When the segment is located within the longitudinal inversion band, the inversion band indicator function takes a value of one. In this case, the original segmented inversion confidence is improved by the inversion band intensity. The higher the inversion band intensity, the more significant the improvement in the corrected confidence, making the confidence of segments within the inversion band more closely match their true anti-coexistence characteristics. When the segment is located outside the longitudinal inversion band, the inversion band indicator function takes a value of zero. The corrected confidence remains consistent with the original confidence, avoiding unreasonable interference with the confidence of non-inversion band segments and ensuring the rationality and relevance of the correction results.

[0051] The corrected piecewise inversion confidence scores for each segment are obtained. Then, semantic weights are constructed using this confidence level. Specifically, these semantic weights are the base class semantic weights, calculated using the following formula: ,in Let represent the semantic weight of the i-th segment at the current time n, which is used to adjust the proportion of base plate monitoring data in the risk score. The reason for using this formula to construct the semantic weight is that the value range of the base plate semantic weight is limited to between 0.5 and 1. When the segment is not located in the longitudinal reversal zone and the corrected segment reversal confidence is low, the base plate semantic weight is close to 0.5. At this time, the proportion of base plate and surface monitoring data in the risk score is equivalent, which meets the risk assessment requirements of non-reversal zone areas. When the segment is located in the longitudinal reversal zone and the corrected segment reversal confidence is high, the base plate semantic weight approaches 1. At this time, the proportion of base plate monitoring data is significantly increased, which can effectively suppress the masking effect of surface data on the actual impact risk of the base plate.

[0052] It should be noted that the segmented reversal confidence level refers to a characterization quantity constructed based on the relative prominence of the reversal coexistence intensity of a specific segment within the monitoring area compared to all segments along the current path. It describes the confidence level of whether this segment is in a state of reversal coexistence between enhanced near-wall aeration protection and increased risk of floor impact. This quantity does not directly represent the magnitude of the risk. A higher segmented reversal confidence level indicates that the reversal coexistence intensity of this segment is more prominent than other segments along the current path, and therefore it is more likely to belong to or be close to a longitudinal reversal zone. A lower segmented reversal confidence level indicates that the reversal coexistence state of this segment is not obvious or has little difference from the normal state along the path. Semantic weight refers to the weight assigned to reflect the difference in the credible contribution of the current physical state of the segment to the two types of information when performing the integrated calculation of the protection appearance index and the bottom plate impact risk index. It is used to adjust the proportion of the two types of indices in the risk score. In the longitudinal reversal zone or when the segment reversal confidence is high, the semantic weight is used to improve the contribution of the information corresponding to the bottom plate impact risk index and suppress the misleading influence of the near-wall protection appearance on risk reduction. In the segment where the reversal coexistence state is not significant, the relative balance of the two types of information is maintained or the weight is allocated according to the rules, so that the risk score is more consistent with the actual hydraulic evolution and the load state of the bottom plate structure along the control section to the energy dissipation section.

[0053] In an embodiment of the present invention, the protection appearance index and the base plate impact risk index are weighted and calculated according to the adjusted semantic weights to generate risk scores for each segment and risk scores for each zone, and risk hotspots are located within the longitudinal reversal zone, specifically including: Calculate the segmented risk score for each segment of the monitoring area. The formula for calculating the segmented risk score is as follows: in This represents the segment risk score of the i-th segment at the current time n. The larger the value, the higher the overall risk level of that segment. This represents the semantic weight of the i-th segment at the current time n, which is used to adjust the proportion of base plate monitoring data in the risk score. This represents the normalized impact risk index of the i-th segment at the current time n, which directly reflects the actual impact risk level of the base plate under load. This represents the weight that complements the semantic weights of the base plate class, corresponding to the weights of the surface class monitoring data. The complementary value represents the Normalized Protection Appearance Index (NPEI), used to adjust the semantic direction of surface data to be consistent with that of base plate data. That is, the higher the NPEI, the lower its complementary value, avoiding misleading risk assessments due to surface gas entrainment. It can dynamically adjust the weights of the two types of monitoring data based on whether the segment is located in the longitudinal reversal zone. Within the longitudinal reversal zone, the base plate semantic weight is higher, strengthening the influence of the base plate impact risk index and suppressing interference from surface protection appearances. Within the non-reversal zone, the weights of the two types of data are balanced, comprehensively reflecting the overall risk level and ensuring the relevance and accuracy of risk scoring.

[0054] After calculating the risk scores for each segment, the zone-level risk score is calculated. The formula for calculating the zone-level risk score is as follows: ,in This represents the zone-level risk score at current time n, used to quantify the overall risk level of the vertical reversal zone. Indicates the number of segments within the longitudinal reversal band. This represents the sum of the segmental risk scores for all segments within the longitudinal reversal zone. This represents the average risk score for each segment within the longitudinal reversal zone, reflecting the average risk level of each segment within the reversal zone. It represents the intensity of the reversal zone at the current time n, reflecting the overall reverse coexistence intensity of the longitudinal reversal zone; it can comprehensively consider the average risk level within the reversal zone and the intensity of the reversal zone itself, reflecting both the actual risk status of each segment within the reversal zone and the degree of concentration of the reverse coexistence phenomenon, so that the zone-level risk score can comprehensively and objectively reflect the overall risk level of the longitudinal reversal zone.

[0055] By traversing all segments within the longitudinal reversal zone and comparing the segment risk scores of each segment one by one, the segment with the highest segment risk score is selected as the risk hotspot. The risk hotspot is the area with the highest risk level within the longitudinal reversal zone and is also the location where the risk of impact on the bottom plate is most concentrated. Accurately locating this location can provide a clear target for subsequent inspections and scheduling control, thereby improving the practicality and operability of risk warning.

[0056] It should be noted that the segmented risk score refers to a comprehensive representation of a specific segment within a monitoring area. This score is obtained by weighting and fusing the complementary values ​​of the normalized base plate impact risk index and the normalized protection appearance index based on the semantic weights of that segment. It describes the overall risk level of that segment under the combined effects of the actual load risk on the base plate and the near-wall apparent condition. A higher segmented risk score generally indicates that the segment is more likely to experience enhanced base plate impact at the current moment and requires close monitoring. The zone-level risk score, on the other hand, refers to a comprehensive risk representation of a continuous segmented interval called the longitudinal reversal zone. This score is obtained by statistically summarizing the segmented risk scores of each segment within the reversal zone and combining them with the reversal zone intensity. The value is used to describe the comprehensive significance of the specific characteristics and risk status within the entire longitudinal reversal zone. The larger the value, the more prominent the reversal coexistence state and the higher the overall risk level within the continuous segment. It is suitable as a basis for determining the warning level and assessing the zone-level anomaly. The risk hotspot location refers to the spatial location corresponding to the segment with the highest segment risk score within the longitudinal reversal zone. It is used to characterize the area with the most concentrated risk or the area that needs priority inspection and attention within the reversal zone at the current moment. This location can be represented by the corresponding segment number, the segment center mileage, or the longitudinal positioning information of the segment from the control section to the energy dissipation section, thereby providing a clear spatial orientation for on-site inspection layout, intensified monitoring, and operation scheduling.

[0057] In an embodiment of the present invention, the warning level is determined based on the risk rating and the warning result is output, specifically including: Calculate the anomaly intensity of the current zone-level risk score relative to historical zone-level risk scores within the same working condition sliding sample set. The formula for calculating the anomaly intensity is as follows: ,in This indicates the intensity of the risk score anomaly at the current time n. The larger the value, the higher the degree of anomaly of the current risk score relative to the historical level under the same working conditions, and the better it reflects the anomaly of the overall risk of the current vertical reversal zone. This represents the zone-level risk score at the current time n. This represents the median of the zone-level risk scores at the corresponding time point for all historical samples within the sliding sample set under the same working conditions. The median is used because it has a strong ability to resist outlier interference, accurately reflects the normal baseline level of the zone-level risk scores under the same working conditions, and avoids the impact of extreme outliers in historical samples on the accuracy of anomaly intensity calculation. This represents the absolute median difference of the zone-level risk scores at corresponding times across all historical samples within the sliding sample set under the same operating conditions. It is used to quantify the normal discrete fluctuation range of the zone-level risk scores under the same operating conditions, enabling the calculation of anomaly intensity to eliminate interference from normal fluctuations and accurately capture true risk anomalies. To prevent the denominator from being zero, a very small constant is set to a value of 10 to the power of negative 6. The reason for setting this constant is that when all historical zone-level risk scores in the sliding sample set under the same working conditions are exactly the same, the absolute median difference will be zero. Adding a very small constant can avoid the situation where the denominator is zero and the calculation cannot be performed. The core purpose of using this formula to calculate the abnormal intensity is to eliminate the influence of normal fluctuations under the same working conditions, accurately quantify the degree of abnormality of the current zone-level risk score relative to the historical benchmark, and provide a quantitative basis for subsequent warning level determination.

[0058] After completing the anomaly intensity calculation, calculate the quantile position of the current risk level score within the sliding sample set under the same working conditions. The formula for calculating the quantile position is as follows: ,in This represents the quantile position of the current zone-level risk score at time n within the sliding sample set of the same working conditions. The value ranges from zero to one. A larger value indicates that the current zone-level risk score ranks higher in the historical samples of the same working conditions, and the risk level is relatively higher. This indicates that among all historical zone-level risk scores within the statistical sliding sample set under the same working conditions, the zone-level risk score at the current moment is less than or equal to the current time level. The number of samples, This is an indicator function that takes a value of one when the condition within the parentheses is true and a value of zero when the condition is false. This indicator function enables accurate statistical analysis of samples that meet the conditions. The formula adds one to both the numerator and denominator to avoid extreme values ​​of zero or one at the quantile position, ensuring that the quantile positions are evenly distributed in the range of zero to one. This makes the subsequent warning level classification based on the quantile position more reasonable and accurate. The reason for using this formula to calculate the quantile position is that it can intuitively reflect the relative position of the current risk level score in the historical data of the same working condition. It takes into account the distribution characteristics of all historical samples, avoids the judgment bias caused by a single benchmark value, and allows the quantile position to accurately represent the relative level of the current risk under the same working condition.

[0059] Based on the quantile position, a corresponding early warning level is generated. The quantile position ranges from zero to one. Considering the control requirements for the impact risk of high-head spillway or floodway floor slabs, a three-level early warning classification rule is preset. This classification rule conforms to industry standards for safety monitoring of water conservancy projects and can accurately distinguish risk levels. Specifically, when the quantile position is less than one-third, a Level 1 early warning is generated. A Level 1 early warning corresponds to a low risk level, indicating that the overall risk of the current longitudinal reversal zone is within the normal fluctuation range of historical conditions under similar working conditions, the floor impact risk is low, and no special control measures are required; only routine monitoring is needed. When the quantile position is greater than or equal to one-third but less than one-third... At the second time, a Level II warning is generated, corresponding to a medium-risk level. This indicates that the overall risk of the current longitudinal reversal zone is higher than the historical average level under the same working conditions, and the risk of impact on the bottom plate has increased. It is necessary to strengthen the monitoring frequency of the corresponding area, arrange staff to conduct regular inspections, and closely monitor the trend of risk changes. When the quantile position is greater than or equal to two-thirds, a Level III warning is generated, corresponding to a high-risk level. This indicates that the overall risk of the current longitudinal reversal zone is significantly higher than the historical normal level under the same working conditions, and the bottom plate has a high risk of impact damage. It is necessary to immediately activate emergency control measures, stop or adjust the flood discharge conditions, organize professional personnel to conduct emergency inspections, and promptly investigate hidden dangers and take protective measures.

[0060] After the warning level is generated, all relevant parameters are output as a complete warning result. The output process combines real-time transmission with local storage. On the one hand, the warning result is transmitted in real time to the terminal equipment of the water conservancy project operation monitoring center. The monitoring interface clearly displays the specific range of the longitudinal reversal zone, the specific value of the reversal zone intensity, the quantitative result of the zone-level risk score, the magnitude of the anomaly intensity, the specific segment number corresponding to the hot spot location within the zone, and the generated warning level, so that monitoring personnel can intuitively grasp the current risk status and key information. On the other hand, the warning result is synchronously stored in the storage module of the data processing unit, which is convenient for subsequent risk tracing, data review, and algorithm optimization. Among them, the longitudinal reversal zone clearly defines the spatial area where the reverse coexistence phenomenon is concentrated, providing a target range for risk control; the reversal zone intensity quantifies the overall intensity of the reverse coexistence phenomenon, assisting in judging the severity of the risk; the zone-level risk score intuitively reflects the overall risk level of the longitudinal reversal zone; the abnormal intensity reflects the degree of abnormality of the current risk relative to the historical level of the same working conditions; the hot spot location within the zone pinpoints the specific segment with the highest risk within the reversal zone, providing a clear target for inspection work; and the early warning level provides a direct basis for the formulation of control measures. All output parameters complement each other and are logically complete, forming a comprehensive set of risk early warning information.

[0061] like Figure 2 The diagram shown is a functional block diagram of a real-time multi-source data fusion risk warning device provided in an embodiment of the present invention.

[0062] In this embodiment, the functions of each module / unit are as follows: The multi-source data acquisition module is used to acquire near-wall area image data, base plate pressure pulsation data, base plate vibration data, and current working condition data for each segment of the monitoring area under the current working conditions. The risk index calculation module is used to normalize the near-wall area image data, base plate pressure pulsation data and base plate vibration data based on historical data under the same working conditions, and calculate the protection appearance index and the base plate impact risk index. The longitudinal reversal zone module is used to calculate the reverse coexistence intensity of each segment of the monitoring area based on the protection appearance index and the base plate impact risk index, and to determine the longitudinal reversal zone based on the reverse coexistence intensity. The semantic weight module is used to adjust the semantic weights of each segment of the monitoring area by using the vertical reversal zone as a control variable. The risk scoring module is used to perform weighted calculations on the protection appearance index and the bottom plate impact risk index based on the adjusted semantic weights, generate risk scores for each segment and risk scores for each zone, and locate risk hotspots within the longitudinal reversal zone. The early warning classification module is used to determine the early warning level based on the risk score and output the early warning result.

[0063] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time multi-source data fusion risk early warning method, characterized in that, include: S1. Acquire near-wall area image data, base plate pressure pulsation data, base plate vibration data, and current working condition data for each segment of the monitoring area under the current working conditions. S2. Based on historical data under the same working conditions, normalize the near-wall area image data, base plate pressure pulsation data and base plate vibration data respectively, and calculate the protection appearance index and base plate impact risk index. S3. Based on the protection appearance index and the impact risk index of the base plate, calculate the reverse coexistence intensity of each segment of the monitoring area, and determine the longitudinal reversal zone based on the reverse coexistence intensity. S4. Use the longitudinal reversal zone interval as a control quantity to adjust the semantic weights corresponding to each segment of the monitoring area. S5. Based on the adjusted semantic weights, the protection appearance index and the bottom plate impact risk index are weighted and calculated to generate risk scores for each segment and risk scores for each zone, and the risk hotspots are located within the longitudinal reversal zone. S6. Determine the warning level based on the risk rating and output the warning result.

2. The real-time multi-source data fusion risk early warning method according to claim 1, characterized in that, Before acquiring data for each segment of the monitoring area, the control section to the energy dissipation section of the high-head spillway or overflow channel is divided into several segments longitudinally to obtain a longitudinal segment sequence.

3. The real-time multi-source data fusion risk early warning method according to claim 1, characterized in that, Based on historical data under the same working conditions, the near-wall region image data, base plate pressure pulsation data, and base plate vibration data were normalized, including: Based on the similarity between the current operating condition data and the historical operating condition data, a sliding sample set with the same operating condition is selected from the historical samples as the adaptive baseline sample set. The operating condition data includes flow rate, gate opening, upstream water level and downstream water level. A robust normalization operator for the same working condition is constructed based on the median and absolute median difference in the sliding sample set under the same working condition.

4. The real-time multi-source data fusion risk early warning method according to claim 3, characterized in that, The calculation of the protection appearance index and the impact risk index of the base plate includes: Extract the near-wall whitening percentage from the near-wall region image data of each segment of the monitoring area; The near-wall whitening ratio is normalized using a robust normalization operator under the same working conditions to obtain the protection appearance index of each segment of the monitoring area.

5. The real-time multi-source data fusion risk early warning method according to claim 3, characterized in that, The calculation of the impact risk index of the base plate includes: Within the analysis window corresponding to the current moment, extract the pressure pulsation intensity of the pressure pulsation data of each segment of the base plate in the monitoring area, as well as the vibration response intensity of the vibration data of each segment of the base plate. The pressure pulsation intensity and vibration response intensity are normalized based on the robust normalization operator under the same working condition. Based on the fluctuation of each intensity quantity in the sliding sample set under the same working condition, an adaptive combined weight is constructed. The impact risk index of the base plate is obtained by weighting and summing the normalized pressure pulsation intensity and vibration response intensity using adaptive combined weights.

6. The real-time multi-source data fusion risk early warning method according to claim 2, characterized in that, Based on the protection appearance index and the base plate impact risk index, the reverse coexistence intensity of each segment of the monitoring area is calculated, and the longitudinal reversal zone is determined based on the reverse coexistence intensity, including: The protection appearance index and the base plate impact risk index are compressed to a unified numerical range to obtain the normalized protection appearance index and the normalized base plate impact risk index. Multiply the normalized protection appearance index of the same segment by the normalized base plate impact risk index to obtain the reverse coexistence intensity of the corresponding segment. The anti-coexistence strength of each segment is vertically centered by subtracting the median of the anti-coexistence strength of all segments from the anti-coexistence strength of each segment. In the vertically segmented sequence, the reverse coexistence intensity after vertical centering is used as the sequence value, and the maximum cumulative value of the continuous segment interval is searched; the continuous segment interval corresponding to the maximum cumulative value is taken as the vertical reversal zone. The average value of the reverse coexistence intensity of each segment within the longitudinal reversal zone is taken as the reversal zone intensity.

7. The real-time multi-source data fusion risk early warning method according to claim 6, characterized in that, Using the vertical reversal zone interval as a control variable, the semantic weights corresponding to each segment of the monitoring area are dynamically adjusted, including: The segmented reversal confidence level is constructed based on the relative prominence of the reverse coexistence intensity of each segment in the current process along the monitoring area. Construct a reversal zone indicator function, and assign different Boolean values ​​to the reversal zone indicator function according to whether each segment of the monitoring area is located within the longitudinal reversal zone; The piecewise inversion confidence is corrected based on the inversion band indicator function and the inversion band strength; Semantic weights are constructed using the corrected segmented inverse confidence.

8. The real-time multi-source data fusion risk early warning method according to claim 7, characterized in that, Based on the adjusted semantic weights, the protection appearance index and the base plate impact risk index are weighted and calculated to generate risk scores for each segment and zone level. Risk hotspots are then located within the longitudinal reversal zone, including: The normalized base plate impact risk index is weighted according to semantic weights, and the complementary value of the normalized protection appearance index is weighted by the complementary weights of the semantic weights. The sum of the two weighting results is used to obtain the segment risk score of each segment of the monitoring area. Calculate the average of the segment risk scores corresponding to all segments within the longitudinal reversal zone, and add the average value to the reversal zone intensity to obtain the zone-level risk score; The segment corresponding to the highest segment risk score within the vertical reversal zone is designated as the risk hotspot location.

9. The real-time multi-source data fusion risk early warning method according to claim 8, characterized in that, The warning level is determined based on the risk rating, and the warning results are output, including: On a sliding sample set under the same working conditions, calculate the abnormal intensity of the current zone-level risk score relative to the historical zone-level risk scores within the sample set. Calculate the quantile position of the current zone-level risk score in the sliding sample set under the same working conditions; Generate the corresponding warning level based on the quantile position; The output includes the longitudinal reversal zone, reversal zone intensity, zone-level risk score, anomaly intensity, hotspot location within the zone, and warning level as the warning result.

10. A real-time multi-source data fusion risk early warning device, applied in the real-time multi-source data fusion risk early warning method according to any one of claims 1-9, characterized in that, The device includes: The multi-source data acquisition module is used to acquire near-wall area image data, base plate pressure pulsation data, base plate vibration data, and current working condition data for each segment of the monitoring area under the current working conditions. The risk index calculation module is used to normalize the near-wall area image data, base plate pressure pulsation data and base plate vibration data based on historical data under the same working conditions, and calculate the protection appearance index and the base plate impact risk index. The longitudinal reversal zone module is used to calculate the reverse coexistence intensity of each segment of the monitoring area based on the protection appearance index and the base plate impact risk index, and to determine the longitudinal reversal zone based on the reverse coexistence intensity. The semantic weight module is used to adjust the semantic weights of each segment of the monitoring area by using the vertical reversal zone as a control variable. The risk scoring module is used to perform weighted calculations on the protection appearance index and the bottom plate impact risk index based on the adjusted semantic weights, generate risk scores for each segment and risk scores for each zone, and locate risk hotspots within the longitudinal reversal zone. The early warning classification module is used to determine the early warning level based on the risk score and output the early warning result.

Images