Small reservoir safety monitoring and intelligent scheduling method based on digital twin model

Abstract

The present application belongs to the field of image recognition monitoring, and discloses a small reservoir safety monitoring and intelligent scheduling method based on a digital twin model. By constructing a small reservoir digital twin model, the reservoir topographic evolution parameter index is obtained, and the reservoir key area image parameters obtained based on the key area similarity conversion method are obtained. The reservoir water related data parameters and the corresponding reservoir safety monitoring and intelligent scheduling method are synchronously obtained. The reservoir safety monitoring and intelligent scheduling characteristics are obtained by processing the above data parameters based on the abnormal data interception method. The reservoir safety monitoring and intelligent scheduling neural network model is constructed based on the intelligent scheduling characteristics and the reservoir safety monitoring and intelligent scheduling method. The output of the twenty-four hour reservoir safety monitoring and intelligent scheduling real-time processing method for small reservoirs in different time periods is based on the above model, which realizes the all-day monitoring and early warning of small reservoirs, thereby effectively improving the hidden danger identification accuracy and early warning real-time performance and reducing the artificial operation and maintenance cost.

Description

Technical Field

[0001] This invention belongs to the field of image recognition and monitoring technology, and in particular relates to a method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models. Background Technology

[0002] Reservoirs, as vital water sources for agricultural irrigation and urban water supply, play a crucial role in addressing drought and water scarcity and ensuring the safety of urban and rural water supply. However, many small reservoirs suffer from numerous problems, including low flood control standards, poor dam stability, dam and foundation leakage, aging and damaged structures, and malfunctioning metal structures and electromechanical equipment, hindering their ability to achieve normal flood control, economic, and social benefits. For example, patent CN114998833A employs a digital twin approach for reservoir monitoring and auxiliary management. It not only provides visualized monitoring of reservoir operations but also deeply mines the value of data, enabling functions such as risk level alarms, terrain evolution projection, and equipment energy-saving analysis, thus improving the system's intelligence level. It can also receive interactive requests from remote clients and send control commands to the underlying service modules, achieving remote control of the underlying equipment and better leveraging the advantages of digital twins. However, it only involves simple data collection and processing to construct a three-dimensional model of the reservoir, without utilizing the model to output intelligent monitoring and scheduling methods. Timely understanding of dam operation status allows for early warning of potential disasters, reducing the occurrence of dam accidents, protecting the lives and property of people downstream, and preventing severe social impacts. However, existing manual measurement methods are inconvenient, inaccurate, and inefficient due to limitations in physical measurement techniques. Consequently, real-world early warning systems are inefficient, fail to provide timely warnings, and contribute to the high severity of disasters.

[0003] This invention addresses the challenges of combining and processing multi-source data from terrain, images, and water bodies, and how to construct neural network models for multi-level applications of small reservoirs based on the obtained features. It also proposes a method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models to achieve 24 / 7 monitoring and scheduling of small reservoirs, effectively improving the accuracy of hazard identification and the real-time nature of early warnings, reducing manual operation and maintenance costs, and realizing proactive prevention and scientific scheduling of reservoir safety. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method for safety monitoring and intelligent scheduling of small reservoirs based on a digital twin model, the method comprising: S1. By constructing a digital twin model of a small reservoir, the parameters of reservoir topography evolution are transmitted and obtained, and the relevant data parameters of the reservoir water body and the corresponding reservoir safety monitoring and intelligent scheduling methods are obtained simultaneously. S2. Use satellite imagery and a key area similarity conversion method to obtain the key area image parameters of the corresponding small reservoir. S3. After processing the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the water body data parameters of the reservoir based on the abnormal data extraction method, the reservoir safety monitoring and intelligent scheduling features are obtained. Based on the reservoir safety monitoring and intelligent scheduling features and the reservoir safety monitoring and intelligent scheduling method, a reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation is constructed. S4. Based on the aforementioned reservoir safety monitoring and intelligent scheduling neural network model, output the real-time safety monitoring and intelligent scheduling method for small reservoirs in different time periods, thereby realizing all-day monitoring and early warning of small reservoirs.

[0005] Furthermore, the reservoir topographic evolution parameters are obtained by numerically processing the erosion modulus of the drawdown zone, the reservoir siltation volume, and the earth dam body parameters. The earth dam body parameters are obtained by the ratio of the seepage hydraulic gradient to the critical gradient. The seepage hydraulic gradient is obtained by the ratio of the upstream and downstream head difference of the reservoir seepage to the seepage path length of the reservoir seepage.

[0006] Furthermore, the reservoir scour and sedimentation volume was calculated using the multi-period DEM topographic difference method.

[0007] Furthermore, the relevant data parameters of the reservoir water body include the flow velocity and water level of the discharge tower, rainwater accumulation parameters, reservoir runoff, water flow velocity, and warning water level.

[0008] Furthermore, the method of obtaining key area image parameters of the corresponding small reservoir by adopting the key area similarity conversion method is to calculate the similarity of the reservoir area image with other historical safety risk warning images, and then select the historical safety risk warning image with the highest similarity to fuse with the reservoir area image to obtain the key area image parameters of the reservoir. The key area image parameters of the reservoir obtained by fusion are obtained by calculating the average pixel value of the two images to be fused.

[0009] Furthermore, the reservoir safety monitoring and intelligent scheduling features obtained by processing the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the water body data parameters based on the abnormal data interception method are obtained by calculating the comprehensive average error deviation between the corresponding data values ​​of each dimension of the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the water body data parameters and the corresponding standard data values ​​of each dimension, and then filtering them according to the data error deviation threshold.

[0010] Furthermore, the reservoir safety monitoring and intelligent scheduling neural network model based on the authenticity allocation of image weight data adopts an activation function improved based on image weight data.

[0011] A small reservoir safety monitoring and intelligent scheduling system based on a digital twin model is also provided. This system implements the aforementioned method for small reservoir safety monitoring and intelligent scheduling based on a digital twin model. The system includes a reservoir digital twin model construction module, a reservoir water body related data parameter acquisition module, a reservoir key area image parameter acquisition module, an abnormal data interception and processing module, a reservoir safety monitoring and intelligent scheduling neural network model construction module, and a reservoir real-time safety monitoring and intelligent scheduling module. The reservoir digital twin model construction module is used to acquire reservoir topographic evolution parameters of small reservoirs by constructing a small reservoir digital twin model. The reservoir water body related data parameter acquisition module is used to acquire reservoir water body related data parameters and corresponding reservoir safety monitoring and intelligent scheduling methods; The reservoir key area image parameter acquisition module is used to acquire the reservoir key area image parameters of the corresponding small reservoir by using satellite imagery and a key area similarity conversion method. The abnormal data interception and processing module is used to process the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the water body data parameters of the reservoir based on the abnormal data interception method to obtain the reservoir safety monitoring and intelligent scheduling features. The reservoir safety monitoring and intelligent scheduling neural network model construction module is used to construct a reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation, based on the reservoir safety monitoring and intelligent scheduling features and the reservoir safety monitoring and intelligent scheduling method. The reservoir real-time safety monitoring and intelligent scheduling module: Based on the reservoir safety monitoring and intelligent scheduling neural network model, it outputs real-time safety monitoring and intelligent scheduling methods for small reservoirs at different time periods, realizing all-day monitoring and early warning of small reservoirs.

[0012] Furthermore, the reservoir topographic evolution parameters are obtained by numerically processing the erosion modulus of the drawdown zone, the reservoir siltation volume, and the earth dam body parameters. The earth dam body parameters are obtained by the ratio of the seepage hydraulic gradient to the critical gradient. The seepage hydraulic gradient is obtained by the ratio of the upstream and downstream head difference of the reservoir seepage to the seepage path length of the reservoir seepage.

[0013] Furthermore, the reservoir safety monitoring and intelligent scheduling neural network model based on the authenticity allocation of image weight data adopts an activation function improved based on image weight data.

[0014] Therefore, the beneficial effects of this invention are as follows: It obtains reservoir safety monitoring and intelligent scheduling features by collecting and processing multi-scale data based on the constructed reservoir safety monitoring and intelligent scheduling neural network model; it achieves intelligent reservoir monitoring based on the image weight data authenticity allocation neural network model; this invention combines multi-source data of terrain, images, and water bodies; it achieves intelligent feature mining based on the weight authenticity allocation neural network; and it combines digital twins to achieve 24 / 7 monitoring and scheduling of small reservoirs, effectively improving the accuracy of hidden danger identification and the real-time nature of early warning, reducing manual operation and maintenance costs, and realizing proactive prevention and scientific scheduling of reservoir safety. The model can be applied to adaptive training for different reservoirs. This model is a reservoir safety monitoring and intelligent scheduling neural network based on image weight data authenticity allocation. The input is the reservoir safety monitoring and intelligent scheduling features after processing multi-source data, including reservoir terrain evolution parameters, image parameters of key areas of the reservoir, and water body data parameters of the reservoir. The model in this invention relies on the similarity distribution characteristics between training set samples to dynamically correct the mapping threshold and activation slope of the neural network activation function, improving the model's feature fitting effect. First, calculate the multi-dimensional similarity between a single sample and all other samples in the training set to quantify the feature similarity, specificity, and outlier degree of the sample in the global data. For ordinary samples with high similarity, maintain the conventional mapping of the activation function to ensure normal activation and learning of basic features. For samples with low similarity and high specificity, such as samples with potential hazards or abnormal working conditions, adaptively adjust the nonlinear transformation parameters of the activation function to amplify their feature response weights and improve feature activation sensitivity.

[0015] Further embodiments and improvements of the present invention will be described in conjunction with the accompanying drawings and specific examples. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models according to the present invention; Figure 2 This is a schematic diagram of a small reservoir safety monitoring and intelligent scheduling system based on a digital twin model, as described in an embodiment of the present invention. Figure 3 This is a schematic diagram of the principle of image similarity in an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device structure for implementing the method of the present invention in an embodiment of the present invention. Detailed Implementation

[0017] The invention will now be further described in conjunction with the accompanying drawings and specific embodiments. The neural network model used in this invention is an improved application of the neural network model in pattern recognition in the context of this application, which is based on the realism allocation of image weight data in the training dataset. It differs from traditional neural network classification models in the processing of input features and the use of data processing in the model.

[0018] The neural network model of this invention is used for safety monitoring and intelligent scheduling of small reservoirs. It is applied to the field of visual recognition of reservoir dams and belongs to information system integration services such as artificial intelligence systems in the construction field. It also belongs to the emerging software and new information technology service industry.

[0019] In a first aspect of the invention, a method for safety monitoring and intelligent scheduling of small reservoirs based on a digital twin model is provided, the method comprising: S1. By constructing a digital twin model of a small reservoir, the parameters of reservoir topography evolution are transmitted and obtained, and the relevant data parameters of the reservoir water body and the corresponding reservoir safety monitoring and intelligent scheduling methods are obtained simultaneously. S2. Use satellite imagery and a key area similarity conversion method to obtain the key area image parameters of the corresponding small reservoir. S3. After processing the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the water body data parameters of the reservoir based on the abnormal data extraction method, the reservoir safety monitoring and intelligent scheduling features are obtained. Based on the reservoir safety monitoring and intelligent scheduling features and the reservoir safety monitoring and intelligent scheduling method, a reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation is constructed. S4. Based on the aforementioned reservoir safety monitoring and intelligent scheduling neural network model, output the real-time safety monitoring and intelligent scheduling method for small reservoirs in different time periods, thereby realizing all-day monitoring and early warning of small reservoirs.

[0020] like Figure 1 The diagram shows the implementation flowchart of a method for safety monitoring and intelligent scheduling of small reservoirs based on a digital twin model.

[0021] The digital twin model of a small reservoir uses the physical reservoir as the mapping object to achieve two-way data linkage between the physical reservoir and the virtual model. On one hand, it receives reservoir topographic evolution parameters collected from the physical reservoir and simultaneously accesses relevant measured data of the reservoir water body, completing the digital replication of the real reservoir topography and water body status in the virtual model, restoring the actual operating conditions such as reservoir topographic changes and water levels. On the other hand, it connects with image parameters of key areas of the reservoir obtained from satellite imagery, mapping multi-source measured data of topography, water body, and images to the twin model space, completing the real-time synchronization of the physical reservoir status to the virtual space.

[0022] The twin model provides a unified digital carrier for subsequent anomaly data extraction and multi-source parameter processing, unifying scattered terrain, water body, and image data into a single virtual reservoir space to support the generation of reservoir safety monitoring and intelligent scheduling features. Based on the standardized features output by the twin model, combined with a preset scheduling method, a neural network model for allocating image weight data accuracy is trained. The virtual twin model continuously synchronizes the reservoir's real-time conditions at different times, inputting real-time mapping data into the neural network to drive the model to output real-time safety monitoring and intelligent scheduling results, ultimately achieving 24 / 7 monitoring and early warning. Simultaneously, the virtual model can reproduce changes in reservoir operating conditions, helping to verify the consistency between the monitoring and scheduling output results and the physical reservoir. In this embodiment, the digital twin model of a small reservoir is used to map the small reservoir in the physical world and to store and digitally display the data that needs to be collected and processed by the subsequent model. This facilitates the reservoir supervisors' safety monitoring and risk identification and scheduling. The digital twin model is generated by software such as Blendeer, QGIS, and CesiumJS according to the needs of technicians regarding the terrain and related conditions of the small reservoir. A detailed description of the generation of this conventional reservoir digital twin model is not provided here.

[0023] Furthermore, the reservoir topographic evolution parameters are obtained by numerically processing the erosion modulus of the drawdown zone, the reservoir sedimentation volume, and the earth dam body parameters. The earth dam body parameters are obtained by the ratio of the seepage hydraulic gradient to the critical gradient. The seepage hydraulic gradient is obtained by the ratio of the upstream and downstream head difference of the reservoir seepage to the seepage path length. The formula for calculating the erosion modulus of the drawdown zone is:

[0024] In the formula, The erosion modulus of the drawdown zone, in units of It is used to characterize the soil erosion intensity per unit area and per unit time in the drawdown zone, and is a core indicator for quantifying the topographic evolution of the reservoir bank drawdown zone. The total eroded sediment mass of the drawdown zone during the monitoring period, expressed in tons (t), can be obtained through topographic differential sedimentation calculation or on-site monitoring. The total area of ​​the exposed area in the reservoir drawdown zone, in meters. 2 This represents the exposed area of ​​the bank slope corresponding to the water level drawdown range. The monitoring time interval is expressed in units of 'a' (months), typically using a monthly monitoring cycle. The data above is obtained by processing historical data since the month of reservoir monitoring, and the calculated values ​​often differ for each different reservoir. This indicator can be directly used for the scour risk rating of the drawdown zone of small reservoirs, and can be adapted to the terrain input parameters of multi-source data processing models for subsequent safety monitoring and intelligent scheduling of small reservoirs.

[0025] In this embodiment, for a small earthen dam reservoir, the critical gradient is set to 0.55. This setting is based on the specific conditions of the dam body by the technicians. If the dam body has a sandy loam foundation, the critical gradient is set to 0.35, which is also based on the relevant settings of the dam body. The earthen dam body parameter data is obtained by using the ratio of seepage hydraulic gradient to critical gradient as the one-dimensional input data for subsequent models because the actual hydraulic gradient is closely related to seepage failure of the dam body. When the ratio of the actual hydraulic gradient to the critical gradient is closer to 1, the safety risk of the dam body increases rapidly.

[0026] Furthermore, the reservoir topographic evolution parameters are obtained by numerically processing the erosion modulus of the drawdown zone, the reservoir siltation volume, and the earth dam body parameters. This involves constructing a spatial feature vector by uniformly removing the dimensions of the aforementioned data indicators. One of the reservoir topographic evolution parameters can be (the characteristic value of the erosion modulus of the drawdown zone, the value of the reservoir siltation volume, and the value of the earth dam body parameters). This is the basic data processing step for subsequent input into the model. This is a general expression of the spatial feature vector representation of the reservoir topographic evolution parameters. Based on this, technicians can obtain the spatial feature data for subsequent input into the model without further description before processing the corresponding data.

[0027] Furthermore, the reservoir scour and sedimentation volume was calculated using the multi-period DEM topographic difference method, and the calculation formula is as follows:

[0028] In the formula, V represents the total scour and deposition volume of the reservoir area. When V > 0, it indicates overall deposition in the entire reservoir area; when V < 0, it indicates overall scour in the entire reservoir area; and when V = 0, it indicates that the reservoir has reached a dynamic equilibrium between scour and deposition. The elevation data represents the water-land topography of various points on the reservoir at historical monitoring times. This data was obtained from historical reservoir data and is expressed in meters (m). It also represents the initial topographic baseline data for the reservoir. The elevation of the reservoir topography at the corresponding monitoring point at the current monitoring time is calculated, in meters. This is the current reservoir topography monitoring data, obtained through the constructed digital twin model. The total area of ​​the reservoir study area is expressed in square meters, and represents the reservoir's year-round water storage area or the designated monitoring zone. , The step size is the planar grid cell size of the DEM digital elevation model, which is the basic unit of measurement for terrain difference calculation.

[0029] The volume of sediment deposition and scour in a reservoir is a core indicator for safety monitoring, hazard warning, and operational scheduling of small reservoirs. It is directly related to the reservoir's flood control safety, structural safety, and operational safety. Firstly, this data can directly quantify the reduction in reservoir capacity, affecting flood control safety assessments. Continuous sedimentation will cause the effective storage capacity and flood control capacity of the reservoir to decrease year by year. The volume of sediment deposition and scour can be used to accurately calculate the amount and rate of storage capacity loss. The reduction in storage capacity leads to a decrease in the reservoir's flood control capacity, a faster rate of water level rise during the flood season, and an increased risk of exceeding the warning level. It is a key basis for reviewing reservoir flood control safety and predicting flood season risks.

[0030] Secondly, changes in reservoir scour and sedimentation volume can alter the reservoir area's topography and flow conditions, exacerbating safety hazards to the banks and dam. Localized sedimentation in the reservoir area can raise the riverbed, change the original flow pattern and longitudinal gradient, and during the flood season, the water flow scours the dam shoulders and reservoir bank drawdown zones, intensifying bank erosion and the risk of collapse. Simultaneously, uneven scour and sedimentation can alter the reservoir area's stress and seepage environment, indirectly inducing abnormal seepage and localized deformation of the dam, providing a catalyst for seepage-related damage such as piping and soil erosion. This data-adapted multi-source data processing intelligent monitoring model uses scour and sedimentation volume as a core topographic evolution parameter. This data, along with reservoir image features, water level, rainfall, seepage flow, and other water body parameters, can be processed and input into the neural network model to improve the reservoir's multi-dimensional safety feature system, effectively enhancing the accuracy of hazard identification, safety level assessment, and intelligent scheduling.

[0031] Furthermore, the relevant data parameters of the reservoir water body include the flow velocity and water level of the discharge tower, rainwater accumulation parameters, reservoir runoff, water flow velocity, and warning water level.

[0032] The flow velocity and water level of the discharge tower are collected in real time from the inlet and outlet cross-sections of the discharge tower and the operating water level inside the tower. In a specific embodiment of this application, under reservoir safety monitoring and scheduling conditions, the instantaneous flow velocity of the discharge tower (0.8–1.5 m / s) and the water level inside the tower dynamically change with flood discharge scheduling are monitored. If the flow velocity suddenly drops and the water level abnormally rises at the same opening degree, it can be determined that there is siltation and blockage in the discharge tower channel, or a local water obstruction hazard, which is related to the reservoir's flood discharge safety. The accumulated rainfall parameter specifically refers to the hourly rainfall intensity. In one specific embodiment of this application, the hourly rainfall intensity during the flood season is 20 mm / h. The inflow of floodwater rises rapidly, and the reservoir water level continues to rise, which can easily lead to increased dam seepage pressure, bank erosion, and risks of piping and quicksand induction. It is a leading driving parameter for flood emergencies. Reservoir runoff is obtained through real-time monitoring of inflow. During the concentrated confluence period of the rainy season, the inflow surges rapidly, quickly raising the reservoir water level, altering the water flow velocity in the reservoir area, and exacerbating siltation and scouring at the dam toe, directly impacting the reservoir's flood control pressure and the dam's structural safety. Water velocity is obtained by collecting the average velocity of the main flow area, the area in front of the dam, and the near-shore water in the drawdown zone. This is because when the flow velocity in front of the dam exceeds 0.3 m / s, it continuously erodes the surface soil of the dam foundation, while high-velocity near-shore water in the drawdown zone intensifies slope erosion, increasing the risk of collapse and soil loss, directly linking topographic erosion and sedimentation evolution with reservoir bank safety. The average of these three flow velocities is used as the input parameter for monitoring the impact on reservoir dam safety in subsequent models. Furthermore, the warning water level is based on the approved warning water level and the guaranteed water level, with real-time monitoring of the reservoir water level exceeding limits, the duration of exceeding the threshold, and the rate of water level rise and fall. When the reservoir water level exceeds the warning level and continues to rise, with a daily increase exceeding 0.2m, the reservoir is in a high-risk operating state, which can easily induce dangerous situations such as dam seepage damage, overflow, and bank instability. By performing dimensionless annotation processing on the above-mentioned multi-dimensional water monitoring parameters, a unified dimension annotation spatial feature vector is formed. One of the spatial feature vectors of the reservoir water-related data parameters is represented as (water tower flow velocity value, water tower flow velocity and water level value, rainwater accumulation parameter value, reservoir runoff value, water flow velocity value, warning water level value).

[0033] Furthermore, the method of obtaining key area image parameters of a small reservoir using a key area similarity conversion method involves comprehensively calculating the similarity between the reservoir area image and other historical safety risk warning images, then selecting the historical safety risk warning image with the highest similarity and fusing it with the reservoir area image to obtain the key area image parameters. The fusing process obtains the key area image parameters by calculating the mean pixel value of the two images to be fused. The similarity calculation method is a histogram similarity calculation method.

[0034] In the formula, This represents the grayscale histogram similarity, with values ​​ranging from [0, 1]. The closer the value is to 1, the more similar the texture, grayscale, and brightness distribution are between the reservoir area image and the historical safety risk warning image, indicating a higher degree of risk feature matching. This is a grayscale index, representing the number of a specific grayscale range, corresponding to all pixel grayscale levels in the 8-bit grayscale image. Let be the pixel statistical probability distribution value of the reservoir area image at the i-th gray level, representing the texture and brightness features of the reservoir area image. For other historical security risk warning images, the pixel statistical probability distribution value at the i-th gray level represents the texture and brightness features of these images. The pixel statistical probability distribution value at the i-th gray level between two images is a probability value, corresponding to the proportion of pixels in the image that fall within the i-th gray level index interval out of the total pixels in the image. This can be calculated using the principle of Bach distance, and is used by technicians to calculate the similarity between the reservoir area image and other historical security risk warning images.

[0035] The aforementioned similarity calculation method quantifies the similarity between reservoir area images and historical safety risk warning images of the reservoir through gray-level cross-matching. It can accurately identify structural hidden dangers that are not easily visible to the naked eye, such as seepage spots, muddy water at the dam toe, quicksand accumulation, exposed erosion of the drawdown zone, and abnormal local shadow textures, providing a quantitative basis for image feature fusion.

[0036] Furthermore, the reservoir safety monitoring and intelligent scheduling features obtained after processing the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the reservoir water body data parameters using the abnormal data extraction method are obtained by calculating the comprehensive average error deviation between the corresponding data values ​​of each dimension of the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the reservoir water body data parameters and the corresponding standard data values ​​of each dimension, and then filtering the data according to the data error deviation threshold. The formula for calculating the comprehensive average error deviation is as follows:

[0037] In the formula, Let be the overall average error deviation of the data parameter d, and m be the number of data points in the dimension of the data parameter d. Let d be the j-th spatial feature data value. Let d be the standard value of the j-th spatial feature data of the data parameter d.

[0038] In one embodiment of this application, if the data parameter d is an index of reservoir topographic evolution, then the value of m is 3. This refers to the first-dimensional spatial feature data value of the reservoir topographic evolution parameter index, namely the erosion modulus of the drawdown zone. The representation of other data parameters will not be elaborated here. This method aims to reduce the impact of significantly different data in the training set on the output of the subsequent reservoir safety monitoring and intelligent scheduling method during model training, and to reduce the selection of outlier data. Based on multiple selections of model input data by our technical personnel, we have selected... Training the model with data whose error value is greater than 90% can more accurately obtain the output of the subsequent real-time safety monitoring and intelligent scheduling method for reservoirs. Therefore, in one embodiment of this application, the corresponding reservoir safety monitoring and intelligent scheduling features are obtained after filtering according to the data error deviation threshold, and the data error deviation threshold is set to 0.9.

[0039] Furthermore, the reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation adopts an activation function improved based on image weight data as follows:

[0040] In the formula, For the activation function value, The weighted sum of the input reservoir safety monitoring and intelligent scheduling feature biases, The average similarity between the image parameters of the reservoir area corresponding to the currently input key reservoir area image parameters and the image parameters of other reservoir areas in the training set is calculated. The similarity calculation method can be an image structure similarity calculation method or an image histogram similarity calculation method.

[0041] This model is a reservoir safety monitoring and intelligent scheduling neural network based on image weight data authenticity allocation. The input is reservoir safety monitoring and intelligent scheduling features processed from multi-source data, including reservoir topographic evolution parameters, image parameters of key reservoir areas, and reservoir water body data parameters. This invention relies on the similarity distribution characteristics among training set samples to dynamically adjust the mapping threshold and activation slope of the neural network activation function, improving the model's feature fitting effect. First, the multi-dimensional similarity between a single sample and all other samples in the training set is calculated to quantify the sample's feature similarity, specificity, and outlier degree in the global data. For high-similarity ordinary samples, the activation function maintains its conventional mapping to ensure normal activation and learning of basic features. For low-similarity, high-specificity samples, such as samples containing potential hazards or abnormal operating conditions, the nonlinear transformation parameters of the activation function are adaptively adjusted to amplify their feature response weights and improve feature activation sensitivity.

[0042] In this application, the input layer feature dimension of the neural network model is set to 128 dimensions, which accepts multi-source feature vectors. The image feature input module is weighted, with the weight value range being 0-1. Weighted correction is applied to image features to weaken the interference of low-quality image data on the model.

[0043] The hidden layers consist of three fully connected layers with 256, 128, and 64 neurons respectively. The activation function used is the improved activation function mentioned above. Dropout is applied after each layer with a coefficient of 0.25 to suppress model overfitting and adapt to scenarios with small sample sizes in small reservoirs.

[0044] There are two neurons in the output layer, which include the risk level and the number of people to be assigned for inspection.

[0045] The Adam optimizer was used for training hyperparameters, with an initial learning rate of 0.0001 and a learning rate decay coefficient of 0.95. The loss function combined classification cross-entropy and mean squared error (MSE) to balance risk classification and scheduled value regression. The batch size was set to 32, and the maximum number of epochs was 120. An early stopping mechanism was implemented, terminating training if the validation set loss did not decrease for 12 consecutive epochs. The dataset was divided into training, validation, and test sets in a 7:2:1 ratio. Normalization preprocessing was performed on the samples after outlier removal, and feature standardization was completed before inputting them into the model.

[0046] This invention leverages the similarity distribution characteristics among training set samples to dynamically adjust the mapping threshold and activation slope of the neural network activation function, thereby improving the model's feature fitting performance. First, it calculates the multi-dimensional similarity between a single sample and all other samples in the training set, quantifying the sample's feature similarity, specificity, and outlier degree in the global data. For ordinary samples with high similarity, it maintains the conventional mapping of the activation function to ensure normal activation and learning of basic features. For samples with low similarity and high specificity, such as samples containing potential hazards or abnormal operating conditions, it adaptively adjusts the nonlinear transformation parameters of the activation function to amplify its feature response weights and improve feature activation sensitivity.

[0047] By using sample similarity to drive dynamic correction of the activation function, the problem of fixed activation functions being insufficient for identifying weak features of reservoir risks and niche abnormal conditions can be solved. This also suppresses interference from residual noise samples, enhances the model's ability to learn special risk features of reservoirs such as siltation, seepage, and piping, and improves the generalization accuracy and early warning reliability of the safety monitoring model.

[0048] Based on the calculation using activation functions, the final result of the neural network is used to output a reservoir safety monitoring and intelligent scheduling method. For example, if the final output value of the neural network is 0.33, the output reservoir safety monitoring and intelligent scheduling method is determined to be of medium risk, and three people are assigned to patrol. This is how small reservoir monitoring personnel determine the probability of the corresponding reservoir safety risk as medium based on the output of the corresponding reservoir safety monitoring and intelligent scheduling method. More corresponding final output values ​​of the neural network and corresponding reservoir safety monitoring and intelligent scheduling methods are not described here, nor is it necessary to exhaustively describe all corresponding reservoir safety monitoring and intelligent scheduling methods. This is only one output of the model's classification monitoring and decision-making method.

[0049] A small reservoir safety monitoring and intelligent scheduling system based on a digital twin model is also provided. This system implements the aforementioned method for small reservoir safety monitoring and intelligent scheduling based on a digital twin model. The system includes a reservoir digital twin model construction module, a reservoir water body related data parameter acquisition module, a reservoir key area image parameter acquisition module, an abnormal data interception and processing module, a reservoir safety monitoring and intelligent scheduling neural network model construction module, and a reservoir real-time safety monitoring and intelligent scheduling module. The reservoir digital twin model construction module is used to acquire reservoir topographic evolution parameters of small reservoirs by constructing a small reservoir digital twin model. The reservoir water body related data parameter acquisition module is used to acquire reservoir water body related data parameters and corresponding reservoir safety monitoring and intelligent scheduling methods; The reservoir key area image parameter acquisition module is used to acquire the reservoir key area image parameters of the corresponding small reservoir by using satellite imagery and a key area similarity conversion method. The abnormal data interception and processing module is used to process the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the water body data parameters of the reservoir based on the abnormal data interception method to obtain the reservoir safety monitoring and intelligent scheduling features. The reservoir safety monitoring and intelligent scheduling neural network model construction module is used to construct a reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation, based on the reservoir safety monitoring and intelligent scheduling features and the reservoir safety monitoring and intelligent scheduling method. The reservoir real-time safety monitoring and intelligent scheduling module: Based on the reservoir safety monitoring and intelligent scheduling neural network model, it outputs real-time safety monitoring and intelligent scheduling methods for small reservoirs at different time periods, realizing all-day monitoring and early warning of small reservoirs.

[0050] like Figure 2The diagram shows a schematic of a small reservoir safety monitoring and intelligent scheduling system based on a digital twin model.

[0051] Furthermore, the reservoir topographic evolution parameters are obtained by numerically processing the erosion modulus of the drawdown zone, the reservoir sedimentation volume, and the earth dam body parameters. The earth dam body parameters are obtained by the ratio of the seepage hydraulic gradient to the critical gradient. The seepage hydraulic gradient is obtained by the ratio of the upstream and downstream head difference of the reservoir seepage to the seepage path length. The formula for calculating the erosion modulus of the drawdown zone is:

[0052] In the formula, Erosion modulus of the drawdown zone, in units of It is used to characterize the soil erosion intensity per unit area and per unit time in the drawdown zone, and is a core indicator for quantifying the topographic evolution of the reservoir bank drawdown zone. The total eroded sediment mass of the drawdown zone during the monitoring period, expressed in tons, can be obtained through topographic differential sedimentation calculation or field monitoring. The total area of ​​exposed areas in the reservoir's drawdown zone is represented by m², while the exposed bank slope area corresponds to the drawdown interval. The monitoring time interval is expressed in units of 'a' (months), typically using a monthly monitoring cycle. The data above is obtained by processing historical data since the month of reservoir monitoring, and the calculated values ​​often differ for each different reservoir. This indicator can be directly used for the scour risk rating of the drawdown zone of small reservoirs, and can be adapted to the terrain input parameters of multi-source data processing models for subsequent safety monitoring and intelligent scheduling of small reservoirs.

[0053] In this embodiment, for a small earthen dam reservoir, the critical gradient is set to 0.55. This setting is based on the specific conditions of the dam body by the technicians. If the dam body has a sandy loam foundation, the critical gradient is set to 0.35, which is also based on the relevant settings of the dam body. The earthen dam body parameter data is obtained by using the ratio of seepage hydraulic gradient to critical gradient as the one-dimensional input data for subsequent models because the actual hydraulic gradient is closely related to seepage failure of the dam body. When the ratio of the actual hydraulic gradient to the critical gradient is closer to 1, the safety risk of the dam body increases rapidly.

[0054] Furthermore, the reservoir topographic evolution parameters are obtained by numerically processing the erosion modulus of the drawdown zone, the reservoir siltation volume, and the earth dam body parameters. This involves constructing a spatial feature vector by uniformly removing the dimensions of the aforementioned data indicators. One of the reservoir topographic evolution parameters can be (the characteristic value of the erosion modulus of the drawdown zone, the value of the reservoir siltation volume, and the value of the earth dam body parameters). This is the basic data processing step for subsequent input into the model. This is a general expression of the spatial feature vector representation of the reservoir topographic evolution parameters. Based on this, technicians can obtain the spatial feature data for subsequent input into the model without further description before processing the corresponding data.

[0055] Furthermore, the reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation adopts an activation function improved based on image weight data as follows:

[0056] In the formula, For the activation function value, The weighted sum of the input reservoir safety monitoring and intelligent scheduling feature biases, The average similarity is calculated between the image parameters of the reservoir area corresponding to the currently input key reservoir area image parameters and the image parameters of other reservoir areas in the training set. The similarity calculation method can be an image structure similarity calculation method or an image histogram similarity calculation method. For example... Figure 3 This is a schematic diagram illustrating the principle of image similarity calculation.

[0057] This model is a reservoir safety monitoring and intelligent scheduling neural network based on image weight data authenticity allocation. The input is reservoir safety monitoring and intelligent scheduling features processed from multi-source data, including reservoir topographic evolution parameters, image parameters of key reservoir areas, and reservoir water body data parameters. This invention relies on the similarity distribution characteristics among training set samples to dynamically adjust the mapping threshold and activation slope of the neural network activation function, improving the model's feature fitting effect. First, the multi-dimensional similarity between a single sample and all other samples in the training set is calculated to quantify the sample's feature similarity, specificity, and outlier degree in the global data. For high-similarity ordinary samples, the activation function maintains its conventional mapping to ensure normal activation and learning of basic features. For low-similarity, high-specificity samples, such as samples containing potential hazards or abnormal operating conditions, the nonlinear transformation parameters of the activation function are adaptively adjusted to amplify their feature response weights and improve feature activation sensitivity.

[0058] In this application, the input layer feature dimension of the neural network model is set to 128 dimensions, which accepts multi-source feature vectors. The image feature input module is weighted, with the weight value range being 0-1. Weighted correction is applied to image features to weaken the interference of low-quality image data on the model.

[0059] The hidden layers consist of three fully connected layers with 256, 128, and 64 neurons respectively. The activation function used is the improved activation function mentioned above. Dropout is applied after each layer with a coefficient of 0.25 to suppress model overfitting and adapt to scenarios with small sample sizes in small reservoirs.

[0060] The output layer has two neurons, representing the risk level and the number of inspectors to be assigned. The Adam optimizer is used for training hyperparameters, with an initial learning rate of 0.0001 and a learning rate decay coefficient of 0.95. The loss function combines classification cross-entropy and mean squared error (MSE), balancing risk classification and scheduling value regression. The batch size is set to 32, and the maximum number of epochs is 120. An early stopping mechanism is implemented, terminating training if the validation set loss does not decrease for 12 consecutive epochs. The dataset is divided into training, validation, and test sets in a 7:2:1 ratio. Normalization preprocessing is performed on the samples after anomaly removal, and feature standardization is completed before inputting them into the model.

[0061] This invention leverages the similarity distribution characteristics among training set samples to dynamically adjust the mapping threshold and activation slope of the neural network activation function, thereby improving the model's feature fitting performance. First, it calculates the multi-dimensional similarity between a single sample and all other samples in the training set, quantifying the sample's feature similarity, specificity, and outlier degree in the global data. For ordinary samples with high similarity, it maintains the conventional mapping of the activation function to ensure normal activation and learning of basic features. For samples with low similarity and high specificity, such as samples containing potential hazards or abnormal operating conditions, it adaptively adjusts the nonlinear transformation parameters of the activation function to amplify its feature response weights and improve feature activation sensitivity.

[0062] By using sample similarity to drive dynamic correction of the activation function, the problem of fixed activation functions being insufficient for identifying weak features of reservoir risks and niche abnormal conditions can be solved. This also suppresses interference from residual noise samples, enhances the model's ability to learn special risk features of reservoirs such as siltation, seepage, and piping, and improves the generalization accuracy and early warning reliability of the safety monitoring model.

[0063] Based on the calculation using activation functions, the final result of the neural network is used to output a reservoir safety monitoring and intelligent scheduling method. For example, if the final output value of the neural network is 0.33, the output reservoir safety monitoring and intelligent scheduling method is determined to be of medium risk, and three people are assigned to patrol. This is how small reservoir monitoring personnel determine the probability of the corresponding reservoir safety risk as medium based on the output of the corresponding reservoir safety monitoring and intelligent scheduling method. More corresponding final output values ​​of the neural network and corresponding reservoir safety monitoring and intelligent scheduling methods are not described here, nor is it necessary to exhaustively describe all corresponding reservoir safety monitoring and intelligent scheduling methods. This is only one output of the model's classification monitoring and decision-making method.

[0064] Therefore, the beneficial effects of this invention are as follows: It obtains reservoir safety monitoring and intelligent scheduling features by collecting and processing multi-scale data based on the constructed reservoir safety monitoring and intelligent scheduling neural network model; it achieves intelligent reservoir monitoring based on the image weight data authenticity allocation neural network model; this invention combines multi-source data of terrain, images, and water bodies; it achieves intelligent feature mining based on the weight authenticity allocation neural network; and it combines digital twins to achieve 24 / 7 monitoring and scheduling of small reservoirs, effectively improving the accuracy of hidden danger identification and the real-time nature of early warning, reducing manual operation and maintenance costs, and realizing proactive prevention and scientific scheduling of reservoir safety. The model can be applied to adaptive training for different reservoirs. This model is a reservoir safety monitoring and intelligent scheduling neural network based on image weight data authenticity allocation. The input is the reservoir safety monitoring and intelligent scheduling features after processing multi-source data, including reservoir terrain evolution parameters, image parameters of key areas of the reservoir, and water body data parameters of the reservoir. The model in this invention relies on the similarity distribution characteristics between training set samples to dynamically correct the mapping threshold and activation slope of the neural network activation function, improving the model's feature fitting effect. First, calculate the multi-dimensional similarity between a single sample and all other samples in the training set to quantify the feature similarity, specificity, and outlier degree of the sample in the global data. For ordinary samples with high similarity, maintain the conventional mapping of the activation function to ensure normal activation and learning of basic features. For samples with low similarity and high specificity, such as samples with potential hazards or abnormal working conditions, adaptively adjust the nonlinear transformation parameters of the activation function to amplify their feature response weights and improve feature activation sensitivity.

[0065] like Figure 4 As shown, the method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models of the present invention operates through conventional components of computing devices, including processors, memory, communication interfaces, and buses.

[0066] Of course, it is understood that each embodiment of the present invention can achieve one of the effects individually, and the combination of multiple embodiments of the present invention can achieve all the above effects. However, it is not required that each embodiment of the present invention achieve all the above advantages and effects, because each embodiment of the present invention can constitute a separate technical solution and make one or more contributions to the prior art. Each embodiment of the present invention is not affected, and the solution can still be implemented even if one embodiment is deleted.

[0067] For any module structures not specifically defined in this invention, the existing technical descriptions shall prevail. The prior art mentioned in the foregoing background and specific embodiments sections can be considered part of this invention and used to understand the meaning of certain technical features or parameters.

Claims

1. A method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models, characterized in that, The method includes: S1. By constructing a digital twin model of a small reservoir, the reservoir topography evolution parameters of the small reservoir are transmitted and obtained, and the relevant data parameters of the reservoir water body and the corresponding reservoir safety monitoring and intelligent scheduling methods are obtained simultaneously. S2. Use satellite imagery and a key area similarity conversion method to obtain the key area image parameters of the corresponding small reservoir. S3. After processing the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the water body data parameters of the reservoir based on the abnormal data extraction method, the reservoir safety monitoring and intelligent scheduling features are obtained. Based on the reservoir safety monitoring and intelligent scheduling features and the reservoir safety monitoring and intelligent scheduling method, a reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation is constructed. S4. Based on the aforementioned reservoir safety monitoring and intelligent scheduling neural network model, output the real-time safety monitoring and intelligent scheduling method for small reservoirs in different time periods, thereby realizing all-day monitoring and early warning of small reservoirs.

2. The method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models as described in claim 1, characterized in that: The reservoir topographic evolution parameters are obtained by numerically processing the erosion modulus of the drawdown zone, the reservoir siltation volume, and the earth dam body parameters. The earth dam body parameters are obtained by the ratio of the seepage hydraulic gradient to the critical gradient. The seepage hydraulic gradient is obtained by the ratio of the upstream and downstream head difference of the reservoir seepage to the seepage path length of the reservoir seepage.

3. The method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models as described in claim 2, characterized in that: The reservoir scour and sedimentation volume was calculated using the multi-phase DEM topographic difference method.

4. The method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models as described in claim 1 or 3, characterized in that: The relevant data parameters of the reservoir water body include the flow velocity and water level of the discharge tower, rainwater accumulation parameters, reservoir runoff, water flow velocity, and warning water level.

5. The method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models as described in claim 1, 2, or 3, characterized in that: The method of obtaining key area image parameters of a small reservoir based on key area similarity conversion involves calculating the similarity between the reservoir area image and other historical safety risk warning images, then selecting the historical safety risk warning image with the highest similarity and fusing it with the reservoir area image to obtain the key area image parameters. The key area image parameters obtained by fusing the images are obtained by calculating the average pixel value of the two images to be fused.

6. The method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models as described in claim 5, characterized in that: The method for processing the reservoir topographic evolution parameters, key area image parameters, and water body data parameters based on abnormal data extraction to obtain reservoir safety monitoring and intelligent scheduling features involves calculating the comprehensive average error deviation between the corresponding data values ​​of each dimension of the reservoir topographic evolution parameters, key area image parameters, and water body data parameters and the corresponding standard data values ​​of each dimension, and then filtering the data based on the data error deviation threshold.

7. The method for safety monitoring and intelligent scheduling of small reservoirs based on digital twin models as described in claim 6, characterized in that: The reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation adopts an activation function improved based on image weight data.

8. A small reservoir safety monitoring and intelligent scheduling system based on a digital twin model, the system implementing the small reservoir safety monitoring and intelligent scheduling method based on a digital twin model as described in any one of claims 1-7, the system comprising a reservoir digital twin model construction module, a reservoir water body related data parameter acquisition module, a reservoir key area image parameter acquisition module, an abnormal data interception and processing module, a reservoir safety monitoring and intelligent scheduling neural network model construction module, and a reservoir real-time safety monitoring and intelligent scheduling module, characterized in that: The reservoir digital twin model construction module is used to acquire reservoir topographic evolution parameters of small reservoirs by constructing a small reservoir digital twin model. The reservoir water body related data parameter acquisition module is used to acquire reservoir water body related data parameters and corresponding reservoir safety monitoring and intelligent scheduling methods; The reservoir key area image parameter acquisition module is used to acquire the reservoir key area image parameters of the corresponding small reservoir by using satellite imagery and a key area similarity conversion method. The abnormal data interception and processing module is used to process the reservoir topographic evolution parameters, the key area image parameters of the reservoir, and the water body data parameters of the reservoir based on the abnormal data interception method to obtain the reservoir safety monitoring and intelligent scheduling features. The reservoir safety monitoring and intelligent scheduling neural network model construction module is used to construct a reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation, based on the reservoir safety monitoring and intelligent scheduling features and the reservoir safety monitoring and intelligent scheduling method. The reservoir real-time safety monitoring and intelligent scheduling module: Based on the reservoir safety monitoring and intelligent scheduling neural network model, it outputs real-time safety monitoring and intelligent scheduling methods for small reservoirs at different time periods, realizing all-day monitoring and early warning of small reservoirs.

9. The small reservoir safety monitoring and intelligent scheduling system based on a digital twin model as described in claim 8, characterized in that: The reservoir topographic evolution parameters are obtained by numerically processing the erosion modulus of the drawdown zone, the reservoir siltation volume, and the earth dam body parameters. The earth dam body parameters are obtained by the ratio of the seepage hydraulic gradient to the critical gradient. The seepage hydraulic gradient is obtained by the ratio of the upstream and downstream head difference of the reservoir seepage to the seepage path length of the reservoir seepage.

10. The small reservoir safety monitoring and intelligent scheduling system based on a digital twin model as described in claim 8, characterized in that: The reservoir safety monitoring and intelligent scheduling neural network model based on image weight data authenticity allocation adopts an activation function improved based on image weight data.

Images