A method and device for dynamic optimization of data transmission paths for reservoir dam safety monitoring based on edge computing

Abstract

This invention discloses a method and apparatus for dynamic optimization of data transmission paths for reservoir dam safety monitoring based on edge computing. The method includes: real-time acquisition of dam safety monitoring data via an edge terminal; feature extraction of the dam safety monitoring data using the edge terminal to obtain real-time features, and anomaly detection based on these features to obtain abnormal real-time data; classification of the dam safety monitoring data according to the abnormal real-time data, and construction of a priority labeling system for the dam safety monitoring data; acquisition of network state parameters using the edge terminal and screening of available paths, real-time evaluation of network state parameters using a reinforcement learning algorithm, and dynamic optimization of transmission paths based on the transmission data level and its priority. This invention achieves low-latency transmission of important data and reliable transmission of ordinary data even in environments without stable networks.

Description

Technical Field

[0001] This invention belongs to the technical field of wireless communication, specifically relating to a method and apparatus for dynamic optimization of data transmission paths for reservoir dam safety monitoring based on edge computing. Background Technology

[0002] In the field of reservoir dam safety monitoring, the continuity and stability of data transmission are crucial. Traditional data transmission technologies for reservoir dam safety monitoring primarily rely on wired networks (such as pre-buried optical fibers or cables to establish fixed transmission channels) or single wireless communication networks (such as 4G / 5G mobile networks) for data transmission, suitable for application scenarios with stable networks. However, these single-network link transmission methods often struggle to guarantee the continuity and stability of data transmission when facing complex and ever-changing network environments. Especially in environments without stable networks, such as remote reservoir sites or during natural disasters, the limitations of traditional single-network transmission methods become even more pronounced.

[0003] 1) Fixed transmission strategy, lacking dynamic adaptability

[0004] Existing security monitoring data transmission methods generally employ fixed transmission path strategies, such as prioritizing mobile networks and switching to satellite networks when mobile networks become unavailable. Such static strategies lack a dynamic decision-making mechanism based on real-time network conditions (e.g., transmission rate, transmission latency, available bandwidth, packet loss rate), resulting in poor data transmission continuity, especially prone to packet loss or interruption during network switching.

[0005] (2) Insufficient edge data processing capabilities and high dependence on the cloud.

[0006] In traditional monitoring solutions, data preprocessing (such as anomaly detection and feature extraction) heavily relies on centralized processing by cloud servers. Edge devices only serve as data collection entry points and need to upload all raw data. This can easily lead to delayed early warnings when network latency is high. When the network is unstable or interrupted, edge devices cannot independently complete data preprocessing, and the functions of data validity verification and preliminary analysis are completely ineffective.

[0007] (3) Lack of a data priority-based transmission mechanism and insufficient protection of critical data.

[0008] For critical data reflecting sudden changes in dam safety status, such as reservoir water level, seepage, and displacement, existing technologies lack a priority-adaptive transmission strategy based on data characteristics. When network congestion occurs, using the same transmission queues and scheduling strategies for both emergency and routine data can lead to delays or even loss of critical data transmission due to uneven channel resource allocation, severely impacting the timeliness of engineering safety early warnings. For example, delayed transmission of data on sudden changes in dam seepage may prevent the monitoring system from triggering timely warnings, missing the optimal emergency response window and posing significant safety hazards. Furthermore, the lack of a dynamic assessment mechanism for data timeliness and importance prevents intelligent adjustment of transmission priorities based on the dam's real-time operating status and network environment changes, making it difficult to meet the reliable, low-latency transmission requirements of safety monitoring for critical data.

[0009] With the development of the Internet of Things and intelligent sensing technologies, the real-time and reliability requirements for the transmission of reservoir dam safety monitoring data have been raised. The shortcomings of traditional technologies in multimodal network fusion, edge intelligent processing, and dynamic path optimization need to be addressed. Summary of the Invention

[0010] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and provide a method and apparatus for dynamic optimization of data transmission paths for reservoir dam safety monitoring based on edge computing. By constructing a priority label system through an anomaly judgment algorithm, and then dynamically optimizing the transmission path according to the priority label system, the invention achieves low-latency transmission of important data and reliable transmission of ordinary data in environments without stable networks.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] In a first aspect, the present invention provides a method for dynamic optimization of data transmission paths for reservoir dam safety monitoring based on edge computing, comprising the following steps:

[0013] Dam safety monitoring data is collected in real time via the edge terminal. The dam safety monitoring data includes water level data, seepage data, and displacement data.

[0014] The edge computing platform is used to extract features from dam safety monitoring data to obtain real-time features. Anomalies are identified based on these features to obtain abnormal real-time data. The dam safety monitoring data is then classified into different levels based on the abnormal real-time data, and a priority labeling system is constructed for the data, including multiple priorities. The priorities are labeled according to the magnitude of the impact of the abnormal real-time data on the safety of the reservoir.

[0015] By collecting network status parameters at the edge and filtering available paths, the network status parameters are evaluated in real time through reinforcement learning algorithms, and the transmission path is dynamically optimized based on the transmission data level and its priority.

[0016] As a preferred technical solution, feature extraction and anomaly detection of water level data are performed using edge computing, including:

[0017] The sliding window size is adaptively adjusted according to the current water level data acquisition frequency to obtain multiple real-time water level data and construct the current sliding window.

[0018] Linear regression was used to fit the water level change trend within the sliding window, and the slope was... β h The rate of change is expressed as follows:

[0019] ;

[0020] Among them, t i h represents the time point within the window. i The water level at each point in time. The time within the window, This is the average water level.

[0021] Determine the current water level data when one of the following conditions is met. Exceeding threshold anomaly:

[0022] Water level exceeding threshold judgment: The water level threshold is dynamically selected based on the current time, with the flood season water level threshold h. w =H1, Non-flood season water level threshold h w =H2, where H1 is the reservoir's flood season limiting water level, and H2 is the reservoir's normal storage water level. If h t h w It is determined to exceed the threshold;

[0023] Water level fluctuation feature extraction and judgment: If If the rate exceeds the preset threshold, it is determined to be a rapid rise or fall in water level.

[0024] As a preferred technical solution, feature extraction and anomaly detection of seepage data are performed using the edge end, including:

[0025] The sliding window size is adaptively adjusted based on historical seepage data and prior data to obtain multiple real-time seepage data and construct the current sliding window.

[0026] Calculate the mean and standard deviation of the data within the current sliding window, obtain the abrupt change point, calculate the rate of change between the current seepage data point and the previous data point, calculate the moving percentile of the historical seepage data as the dynamic threshold, take the k most recent historical seepage data and calculate the 95th percentile to obtain the dynamic threshold;

[0027] Based on the comprehensive characteristics of the seepage data, a threshold judgment is performed on the current seepage data point. The current seepage data is judged to be abnormal if one of the following conditions is met:

[0028] (1) The absolute value of the difference between the current seepage data point and the mean is not less than three standard deviations;

[0029] (2) The rate of change exceeds the set rate of change threshold;

[0030] (3) The current seepage data point exceeds the dynamic threshold.

[0031] As a preferred technical solution, feature extraction and anomaly detection are performed on displacement data using the edge endpoints, including:

[0032] The sliding window size is adaptively adjusted based on historical displacement data and prior data, multiple real-time displacement data are obtained, and the current sliding window is constructed.

[0033] Calculate the magnitude of the three-dimensional composite displacement vector at the current moment, calculate the rate of change of displacement in each direction using the central difference method, and fit the trend slope of the displacement data in each direction within the window using the least squares method.

[0034] Calculate the mean and standard deviation of displacement in each direction within the current sliding window, and calculate the dynamic thresholds of displacement in each direction within the current active window in real time, including the dynamic thresholds of absolute displacement and the dynamic thresholds of displacement change rate; the calculation of the dynamic thresholds of displacement in each direction within the current active window is as follows:

[0035] ;

[0036] ;

[0037] in, for f The dynamic threshold of absolute displacement in direction. for f Dynamic threshold for the rate of change of directional displacement , Each in the current window f Mean and standard deviation of the rate of change of directional displacement, ρ This is the confidence level adjustment coefficient;

[0038] Based on the comprehensive characteristics of the displacement data, the current displacement data point Perform threshold detection. If one of the following conditions is met, the current displacement data is considered abnormal and exceeds the threshold:

[0039] (1) The magnitude of the composite displacement vector is greater than the safety threshold of the dam engineering design, or the absolute displacement in any direction is not less than the dynamic threshold of the absolute displacement in the corresponding direction;

[0040] (2) The magnitude of the rate of change of displacement in any direction is not less than the dynamic threshold of the rate of change of displacement in the corresponding direction;

[0041] (3) The magnitude of the trend slope is not less than the safety threshold of the trend change rate.

[0042] As a preferred technical solution, the step of classifying dam safety monitoring data into levels based on abnormal real-time data and constructing a priority labeling system for the dam safety monitoring data specifically involves:

[0043] Classification: Abnormal real-time data is classified as important data, and other real-time data is marked as normal data;

[0044] Construct a priority labeling system: classify all important data according to data type and data size, and sort them in descending order. Select the data with the greatest impact on reservoir safety among all data types, and select the data with the largest size with the highest priority.

[0045] As a preferred technical solution, the method of collecting network status parameters at the edge and filtering available paths, evaluating network status parameters in real time through reinforcement learning algorithms, and dynamically optimizing transmission paths based on transmission data levels and priorities, specifically involves:

[0046] During communication with the multipath transmission control protocol of the receiving end, the edge end collects multi-dimensional network status information, stores it in a buffer in vector form, calculates real-time path status thresholds based on the multi-dimensional network status information, and filters available paths based on the path status thresholds; the multi-dimensional network status information includes transmission rate, transmission delay, available bandwidth, and packet loss rate.

[0047] The Q-Learning reinforcement learning algorithm is used to process the results of multiple screenings of available paths. A reward function is set according to the level and priority of the transmitted data. The data transmission path is continuously optimized through value update rules until the transmission ends or the strategy converges.

[0048] Preferred technical solutions include network state information modeling, path state threshold calculation, and available path filtering;

[0049] The network state information modeling includes:

[0050] For the i-th transmission path, the network state vector at time t is defined as a four-dimensional real vector, as shown in the following equation:

[0051] , i∈{1,2,…,n};

[0052] Among them, R i (t), D i (t), B i (t), L i(t) represents the transmission rate state vector, transmission delay state vector, available bandwidth state vector, and packet loss rate state vector of path i at time t, respectively. t is the time point for state vector acquisition, used to distinguish different acquisition periods.

[0053] Network status information is collected according to a fixed collection period to generate a path status matrix;

[0054] The path state threshold calculation includes:

[0055] The mean and threshold values ​​of each network state parameter are calculated as follows:

[0056] ;

[0057] ;

[0058] ;

[0059] ;

[0060] in, , , , Let N be the mean of the network state parameters along path i. s R is the number of data collections within a fixed collection period. th,i D th,i B th,i L th,i These are the transmission rate threshold, transmission delay threshold, available bandwidth threshold, and packet loss rate threshold for path i, respectively. φ R φ D φ B and φ L These are the transmission rate coefficient, transmission delay coefficient, available bandwidth coefficient, and packet loss rate coefficient, respectively, and their values ​​are set according to the transmission path type.

[0061] The available path filtering includes:

[0062] The condition that path i is a usable path is as follows:

[0063] ;

[0064] Where R th D th B th L th The global path average state threshold is calculated using the following formula:

[0065] ;

[0066] ;

[0067] ;

[0068] ;

[0069] ;

[0070] Where, ω R ω D ω B ω L These are all global threshold weights for each network state parameter.

[0071] As a preferred technical solution, the reward function is set as follows:

[0072] For critical data, prioritize minimizing latency and packet loss rate; the immediate reward function is:

[0073] ;

[0074] For regular data, we prioritize maximizing rate and bandwidth, and the immediate reward function is:

[0075] ;

[0076] Where, λ R ,λ D ,λ B ,λ L These are weighting coefficients, which are adjusted according to business needs. , , and The new state vectors of each network state parameter obtained after executing action a;

[0077] The value update rules include:

[0078] Q-value updates use the time-series difference update formula of the Q-Learning algorithm:

[0079] ;

[0080] in, For the current state s t Take action a t The expected cumulative reward, where α is the learning rate. r t For the current state s t Take action a t The instant reward obtained, where γ is the discount factor. , Indicates the next state s t+1 The expected cumulative reward for choosing the optimal action 'a'.

[0081] Secondly, the present invention also provides a dynamic optimization system for data transmission paths of reservoir dam safety monitoring based on edge computing, which is applied to the aforementioned dynamic optimization method for data transmission paths of reservoir dam safety monitoring based on edge computing, including a data acquisition module, a data processing module, an environmental perception computing module, a reinforcement learning module, and a path scheduling module;

[0082] The data acquisition module includes a standardized sensor interface for connecting multiple types of sensors and for acquiring dam safety monitoring data in real time, including water level data, seepage data, and displacement data.

[0083] The data processing module is used to extract features from dam safety monitoring data using the edge terminal, obtain real-time features, and make anomaly judgments based on the real-time features to obtain abnormal real-time data; classify the dam safety monitoring data into levels based on the abnormal real-time data, and construct a priority labeling system for the dam safety monitoring data, including multiple priorities, with the priorities labeled according to the magnitude of the impact of the abnormal real-time data on reservoir safety.

[0084] The environmental awareness computing module is used for network status acquisition and status modeling and filtering. The network status acquisition monitors network status parameters in real time through the MPTCP protocol, including transmission rate, transmission latency, available bandwidth, and packet loss rate. The status modeling and filtering includes defining the status vector of each transmission path, generating the performance baseline of each path based on historical network status data, setting early warning thresholds, and automatically filtering unusable paths. The network includes 4G / 5G mobile networks, satellite communication, and LoRa wireless multi-hop networks.

[0085] The reinforcement learning module uses the Q-Learning reinforcement learning algorithm to achieve dynamic optimization of transmission paths, balancing data priority and network performance.

[0086] The path scheduling module is used to perform path switching based on reinforcement learning results, realize data fragmentation transmission and dynamic rescheduling, including a multi-protocol encapsulation engine and real-time monitoring and rescheduling.

[0087] Thirdly, the present invention provides an edge computing terminal, the edge computing terminal comprising:

[0088] At least one processor; and,

[0089] A memory communicatively connected to the at least one processor; wherein,

[0090] The memory stores computer program instructions that can be executed by the at least one processor, which enables the at least one processor to execute the dynamic optimization method for data transmission paths of reservoir dam safety monitoring based on edge computing.

[0091] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0092] This invention achieves low-latency transmission of important data and reliable transmission of ordinary data in environments without stable networks by using lightweight abnormal data detection and priority classification, multimodal network environment perception and dynamic transmission path optimization.

[0093] Among them, abnormal data detection and priority classification can filter out important or major data by pre-determining priorities, thereby avoiding delays or even loss of important data transmission due to uneven distribution of channel resources, thus further meeting the security monitoring requirements for reliable and low-latency transmission of important data; at the same time, through multimodal network environment perception and dynamic transmission path optimization, the network environment can be monitored in real time, and the data transmission path can be dynamically decided based on the real-time status and characteristics of the transmitted data, thereby avoiding data transmission interruptions, or packet loss or interruptions that may occur during network switching. Attached Figure Description

[0094] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0095] Figure 1 This is a flowchart of a method for dynamic optimization of data transmission paths for reservoir dam safety monitoring based on edge computing, according to an embodiment of the present invention.

[0096] Figure 2 This is a schematic diagram of the dynamic optimization system for data transmission paths of reservoir dam safety monitoring based on edge computing, according to an embodiment of the present invention.

[0097] Figure 3 This is a structural diagram of the edge computing terminal according to an embodiment of the present invention. Detailed Implementation

[0098] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0099] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0100] Please see Figure 1 This embodiment provides a dynamic optimization method for data transmission paths in reservoir dam safety monitoring based on edge computing, including the following steps:

[0101] S1. Real-time collection of dam safety monitoring data via edge devices, including water level data, seepage data, and displacement data.

[0102] In this embodiment, the edge device connects to various types of sensors, including water level gauges, piezometers, flow meters, displacement gauges, and vibration sensors, via a standardized sensor interface (supporting RS485 / Modbus protocols). This allows for real-time acquisition of dam safety monitoring data such as reservoir water level, seepage, and displacement. The acquisition frequency is adjusted according to business needs. By analyzing water level, seepage, and displacement data, the edge device can detect abnormal data and upload this data to the server or cloud.

[0103] To address the issue of channel resource allocation, this embodiment employs a data priority-differentiated transmission mechanism to ensure the transmission of critical data. Please refer to step S2 for details.

[0104] S2. Use the edge to extract features from the dam safety monitoring data, obtain real-time features, and make anomaly judgments based on the real-time features to obtain abnormal real-time data; classify the dam safety monitoring data according to the abnormal real-time data, and build a priority labeling system for the dam safety monitoring data, including multiple priorities, and the priorities are labeled according to the magnitude of the impact of abnormal real-time data on reservoir safety.

[0105] In step S2, the edge device will perform feature extraction and anomaly analysis on the collected water level data, seepage data and displacement data to filter out abnormal values.

[0106] Specifically, you can refer to the following steps:

[0107] S21. Regarding feature extraction and anomaly detection for water level data, the specific content includes the following:

[0108] S211. Construct a sliding window based on real-time water level data.

[0109] Set the sliding window size to N (N is a positive integer, which can be adaptively adjusted according to the water level data acquisition frequency, with a value range of [20, 50]), acquire the latest N water level data, and construct the current sliding window. , where h t This represents the current water level.

[0110] S212, Water level data feature extraction.

[0111] Linear regression was used to fit the water level change trend within the sliding window, with the slope... β h As the rate of change:

[0112] ;

[0113] Among them, t i h represents the time point within the window. i The water level at each point in time. The time within the window, This represents the average water level.

[0114] S213, Water level data exceeding threshold judgment.

[0115] Determine the current water level data when one of the following conditions is met. Exceeding threshold anomaly:

[0116] Water level exceeding threshold judgment: The water level threshold is dynamically selected based on the current time, with the flood season water level threshold h. w =H1, Non-flood season water level threshold h w =H2, where H1 is the reservoir's flood season limiting water level, and H2 is the reservoir's normal storage water level. If h t h w It is determined to exceed the threshold;

[0117] Water level fluctuation feature extraction and judgment: If ( (Based on the preset rate threshold), it is determined to be a rapid rise and fall in water level.

[0118] S22. Seepage data feature extraction and anomaly detection, specifically including the following:

[0119] S221. Construct a sliding window based on real-time seepage data.

[0120] Set the sliding window size to N (N is a positive integer, which can be determined based on the historical fluctuations of reservoir dam seepage data and engineering experience, with a value range of [20, 50]. For example, if the seepage data collection frequency is 1 hour / time, N is the number of seepage data collections corresponding to 24 hours, which is 24). Obtain the latest N seepage data points and construct the current sliding window. ,in This represents the seepage value at the current moment.

[0121] S222, Feature extraction of seepage data.

[0122] Mutation point detection: Calculate the current sliding window Mean μ of internal data t and standard deviation σt.

[0123] Rate of change calculation: Calculate the current seepage data point q t With data point q t-1 rate of change β q The formula is:

[0124] ;

[0125] Historical comparison dynamic threshold: The moving percentile of historical seepage data is calculated as the dynamic threshold. The 95th percentile q of the most recent k historical seepage data points is calculated (this can be set as needed). _95% This refers to the dynamic threshold.

[0126] S223, Judgment of seepage data exceeding threshold.

[0127] Based on the comprehensive characteristics of the seepage data, for the current seepage data point q t Perform threshold detection. If one of the following conditions is met, the current seepage data is considered abnormal due to exceeding the threshold:

[0128] (1) The absolute value of the difference between the current seepage data and the mean is not less than three times the standard deviation, as shown in the following formula: ;

[0129] (2) Rate of change Exceeding the set rate of change threshold ;

[0130] (3) Current seepage data q t Exceeding the dynamic threshold q _95% .

[0131] S23. Displacement data feature extraction and anomaly detection, specifically including the following:

[0132] S231. Construct a sliding window based on real-time displacement data;

[0133] Set the sliding window size to N (N is a positive integer, which can be determined based on the historical fluctuations of reservoir dam displacement data and engineering experience, with a value range of [20, 50]. For example, if the displacement data acquisition frequency is 1 hour / time, N is the number of displacement data acquisitions corresponding to 24 hours, which is 24). Obtain the latest N triaxial displacement data sequences and construct the current sliding window. ,in This is the current three-dimensional displacement data sequence.

[0134] S232, Displacement data feature extraction.

[0135] Composite displacement characteristics: Calculate the magnitude of the three-dimensional composite displacement vector at the current moment. This reflects the overall displacement of the dam:

[0136] ;

[0137] Displacement change rate characteristics: The displacement change rate in each direction is calculated using the central difference method.

[0138] ;

[0139] in: For time t f The rate of change of displacement in the direction; This represents the data acquisition time interval.

[0140] Trend term characteristics: Linear trend term of displacement data in each direction within the window is fitted by least squares method.

[0141] ;

[0142] in: a f for f The slope of the directional displacement trend reflects the long-term rate of change; b f The intercept, i.e. f The initial reference value for directional displacement.

[0143] S233, Dynamic threshold calculation.

[0144] Calculate the current sliding window W t Mean value of displacement in all directions and standard deviation .

[0145] Calculate the dynamic thresholds of displacement in each direction within the currently active window in real time:

[0146] ;

[0147] ;

[0148] in, for f The dynamic threshold of absolute displacement in direction. for f Dynamic threshold for the rate of change of directional displacement , Each in the current window f Mean and standard deviation of the rate of change of directional displacement. ρ、 This is the confidence level adjustment coefficient; ( (Configured according to the engineering characteristics of the reservoir dam).

[0149] S234, Displacement data exceeding threshold judgment.

[0150] Based on the comprehensive characteristics of the displacement data, the current displacement data point Perform threshold detection. If one of the following conditions is met, the current displacement data is considered abnormal and exceeds the threshold:

[0151] (1) Composite displacement S t >S th or absolute displacement in any direction S th The safety threshold for dam engineering design (the upper limit of the safety of the composite displacement pre-set according to the dam structural design code);

[0152] (2) Rate of change of displacement in any direction ;

[0153] (3) Trend slope ,in a thresh The safe threshold for the rate of change of trend (set according to the dam deformation stability requirements).

[0154] Next, we need to process the above data in two ways: first, classify the processed data into different levels; second, build a priority labeling system based on the data levels.

[0155] For all dam safety monitoring data, classification can be performed based on the results of anomaly assessment. Therefore, this embodiment classifies abnormal real-time data as "important data" and other real-time data as "normal data." For example: Water level data marking: Water level data determined to exceed flood limits or experience rapid rises and falls in anomaly assessments is marked as important data; other data is marked as normal data. Seepage data marking: Seepage data determined to exceed threshold values ​​in anomaly assessments is marked as important data; other data is marked as normal data. Displacement data marking: Displacement data determined to exceed threshold values ​​in anomaly assessments is marked as important data; other data is marked as normal data.

[0156] Regarding the construction of the priority labeling system, this embodiment provides a simple and efficient solution, as follows:

[0157] Each piece of data marked as "important" is categorized by data type and size, and then sorted in descending order. The data types primarily describe the collected water level, seepage, and displacement data. Generally, reservoir water level data is critical data that must be transmitted; furthermore, seepage data has the most critical impact on dam safety, followed by displacement data. Based on this, the data priority is set as: water level data > seepage data > displacement data. After categorization, this embodiment sorts the "important data" in descending order of their numerical values, thus placing the data with the largest water level value at the beginning of the sequence as the most critical, ensuring its highest priority transmission to the server or cloud.

[0158] After completing the data processing, the next step is to select and optimize the transmission path, as detailed in step S3.

[0159] S3. Collect network status parameters at the edge and filter available paths. Evaluate network status parameters in real time through reinforcement learning algorithm and dynamically optimize transmission paths according to the level and priority of transmitted data.

[0160] In step S3, it is first necessary to clarify that the edge terminal in this embodiment has been configured with three types of communication modules: mobile network, satellite communication, and LoRa-based wireless multi-hop communication, thereby constructing a multimodal transmission path for dam safety monitoring data, wherein:

[0161] (1) Mobile network module (4G / 5G): corresponding to data transmission path 1;

[0162] (2) Satellite communication module: corresponding to data transmission path 2;

[0163] (3) Wireless multi-hop communication module (LoRa technology): contains n neighbor nodes, each node constitutes an independent data transmission path, which are marked as path 3 to path n+2 in sequence.

[0164] The three types of communication modules described above form a heterogeneous network channel combination, which connects to the core processing unit of the edge computing terminal through a standardized interface, supporting adaptive encapsulation and routing selection of monitoring data under different communication protocols. It is worth noting that this multimodal channel fusion design enables the system to adapt to complex network environments and improves transmission fault tolerance.

[0165] Secondly, during the communication process between the edge end and the receiving end via the Multipath Transmission Control Protocol (MPTCP), the environment-aware computing module collects multi-dimensional network status information and stores it in the cache in vector form.

[0166] To clearly represent the "vector form", this embodiment uses the following steps:

[0167] S31. Network state information modeling.

[0168] For the i-th transmission path, the network state vector at time t is defined as a four-dimensional real vector, as shown in the following equation:

[0169] , i∈{1,2,…,n};

[0170] Among them, R i (t), D i (t), B i (t), L i (t) represents the transmission rate state vector, transmission delay state vector, available bandwidth state vector, and packet loss rate state vector of path i at time t, respectively. t is the time point for state vector acquisition, used to distinguish different acquisition periods.

[0171] Network status information is collected at a fixed collection period to generate a path status matrix, as shown in the following formula:

[0172] .

[0173] S32, Path status threshold calculation.

[0174] The mean and threshold values ​​of each network state parameter are calculated as follows:

[0175] ;

[0176] ;

[0177] ;

[0178] ;

[0179] in, , , , Let N be the mean of the network state parameters along path i. s R is the number of data collections within a fixed collection period. th,i D th,i B th,i L th,i These are the transmission rate threshold, transmission delay threshold, available bandwidth threshold, and packet loss rate threshold for path i, respectively. φ R φ D φB and φ L These are the transmission rate coefficient, transmission delay coefficient, available bandwidth coefficient, and packet loss rate coefficient, respectively. Their values ​​are set according to the transmission path type, and n is the total number of paths.

[0180] S33, Filter by available paths.

[0181] The condition that path i is a usable path is as follows:

[0182] ;

[0183] Where R th D th B th L th The global path average state threshold is calculated using the following formula:

[0184] ;

[0185] ;

[0186] ;

[0187] ;

[0188] ;

[0189] Where, ω R ω D ω B ω L These are all global threshold weights for each network state parameter.

[0190] Next, this embodiment employs a reinforcement learning algorithm to learn the optimal path selection strategy through interaction with the transmission environment. The core lies in defining the general elements of reinforcement learning (state, action, policy, reward, and update mechanism), applicable to various reinforcement learning algorithms (including but not limited to Q-Learning, SARSA, deep reinforcement learning, policy gradient algorithms, etc.). The specific element design is as follows:

[0191] S331, Environmental State Definition.

[0192] In reinforcement learning algorithms, the state is used to describe the characteristics of the current transmission scenario, and is defined as a vector containing available path network parameters and data features, i.e.:

[0193] ;

[0194] in:

[0195] k: The number of available paths selected;

[0196] : Transmission rate state vector for each available path;

[0197] : Transmission delay state vector for each available path;

[0198] : Available bandwidth state vector for each available path;

[0199] : Packet loss rate state vector for each available path;

[0200] : Transmitted data characteristics ( =0 indicates normal data. =1 is an important data point.

[0201] S332, Dynamic Space.

[0202] In reinforcement learning algorithms, actions represent the selection of transmission paths, and the action space is defined as the set of indices of available paths.

[0203] ;

[0204] Among them, a t This indicates the selected transmission path number.

[0205] S333, Strategy Function.

[0206] Reinforcement learning algorithms employ strategies to select actions based on the current state, balancing exploration (discovering potentially better paths) with exploitation (selecting the current optimal path), including but not limited to: ε-greedy strategies, upper confidence bound (UCB) strategies, Boltzmann exploration strategies, etc.

[0207] Exploration phase: Randomly select paths with a certain probability to discover potentially better paths;

[0208] Utilization phase: Select the path with the highest current value assessment with a certain probability.

[0209] S334, Reward Function Design.

[0210] The reward function of reinforcement learning algorithms is used to quantify the quality of actions. It is designed differently based on data characteristics. The core is to combine the network performance parameters of the path (speed, latency, bandwidth, packet loss rate) to achieve differentiated incentives.

[0211] For important data ( =1): Prioritize optimization of reliability metrics such as latency and packet loss rate;

[0212] For regular data ( =0): Prioritize optimizing efficiency metrics such as speed and bandwidth.

[0213] The reward function is:

[0214] ;

[0215] in, f(·) The reward calculation function can be adjusted according to business needs (such as linear combination, nonlinear weighting, normalization, etc.); λ R ,λ D ,λ B ,λ L This is a weighting coefficient used to balance the priority of different indicators.

[0216] S335, Value Update Rules.

[0217] The value function update rule of the reinforcement learning algorithm is combined with the current state, action, and immediate reward r. t And the next state, optimize the strategy or value function. The update formula satisfies one of the following forms:

[0218] Temporal differential updates (such as Q-Learning and SARSA value function updates);

[0219] Policy gradient update (e.g., REIN) F (ORCE and PPO policy parameter updates)

[0220] Monte Carlo Update;

[0221] Updates based on value function approximation (such as DQN).

[0222] S336, Algorithm flow in reinforcement learning.

[0223] ① Initialization: Create a value function table V(s,a) or a policy function π(a|s), and set parameters such as learning rate and discount factor;

[0224] ②Environmental awareness: Real-time collection of network status and filtering of available paths;

[0225] ③ State Construction: Based on available path parameters and data characteristics, generate the current state s. t ;

[0226] ④ Action selection: Select actions from the action space based on the exploration-utilization policy function. a t (i.e., the selected transmission path);

[0227] ⑤ Perform the action: by selecting the path a t Transmit data and obtain the next state s t+1and environmental feedback;

[0228] ⑥ Reward calculation: Based on data characteristics and s t+1 The network parameters are used to calculate the immediate reward r. t ;

[0229] ⑦ Value / Strategy Update: Based on the value update rules, utilizing s t ,a t ,r t ,s t+1 Update the value function or policy function;

[0230] ⑧ Loop: Repeat steps ②-⑦ until the transmission ends or the strategy converges.

[0231] To better illustrate the beneficial effects of this embodiment, a more specific example is provided where the Q-Learning algorithm is used to optimize the transmission path. The core elements are designed as follows:

[0232] (1) Definition of environmental state.

[0233] If k usable paths are selected, the state vector at time t is:

[0234] ;

[0235] in:

[0236] : Transmission rate state vector for each available path;

[0237] : Transmission delay state vector for each available path;

[0238] : Available bandwidth state vector for each available path;

[0239] : Packet loss rate state vector for each available path;

[0240] : Transmitted data characteristics ( =0 indicates normal data. =1 is an important data point.

[0241] (2) Action space.

[0242] The action set is an index of available paths: The corresponding transmission path number.

[0243] (3) Policy function (ε-greedy policy).

[0244] In the exploration phase, a path is randomly selected with probability ε to discover a better path; in the utilization phase, the path with the largest current Q value is selected with probability 1-ε. The formula is...

[0245] ;

[0246] (4) Design of reward function.

[0247] Rewards are defined based on differences in data characteristics:

[0248] ① For important data ( Prioritizing minimizing latency and packet loss rate, the immediate reward function is:

[0249] ;

[0250] ②For regular data ( Prioritizing maximizing rate and bandwidth, the immediate reward function is:

[0251] ;

[0252] Where: λ R ,λ D ,λ B ,λ L These are weighting coefficients, which are adjusted according to business needs. , , and This refers to the new state vectors of each network state parameter obtained after executing action a.

[0253] (5) Q value update rules.

[0254] Q-value updates use the time-series difference update formula of the Q-Learning algorithm:

[0255] ;

[0256] in, For the current state s t Take action a t The expected cumulative reward, where α is the learning rate. r t For the current state s t Take action a t The instant reward obtained, where γ is the discount factor. , Indicates the next state s t+1 The expected cumulative reward for choosing the optimal action 'a'.

[0257] (6) Algorithm flow.

[0258] ① Initialization: The Q-value table Q(s,a) is initialized to 0, and the learning rate α, discount factor γ, and exploration rate ε are set.

[0259] ②Environmental awareness: Collect network status data and filter the set of available paths;

[0260] ③ State Construction: Generate the current state s t The state includes path parameters and data characteristics;

[0261] ④ Action selection: Select a path based on the ε-greedy strategy. a t ;

[0262] ⑤ Perform the action: via the path a t Transmit data and obtain feedback (including new state vectors for each network state parameter).

[0263] ⑥ Reward Calculation: Calculate r based on feedback and data characteristics. t ;

[0264] ⑦ Q-value update: based on the next state s t+1 Update the Q-value table;

[0265] ⑧ Loop: Return to step ② until the termination condition is met (such as the end of transmission).

[0266] Finally, based on the optimal path index output by the reinforcement learning module, the driver interface of the corresponding communication module is called to realize data fragmentation, encapsulation and transmission, and the path status is monitored in real time to trigger rescheduling (such as when the path fails or performance drops sharply).

[0267] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0268] Based on the same idea as the dynamic optimization method for data transmission paths of reservoir dam safety monitoring based on edge computing in the above embodiments, this invention also provides a dynamic optimization system for data transmission paths of reservoir dam safety monitoring based on edge computing. This system can be used to execute the above-described dynamic optimization method for data transmission paths of reservoir dam safety monitoring based on edge computing. For ease of explanation, the structural diagram of the embodiment of the dynamic optimization system for data transmission paths of reservoir dam safety monitoring based on edge computing only shows the parts related to the embodiments of this invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0269] Please see Figure 2 In another embodiment of this application, a dynamic optimization system 10 for data transmission path of reservoir dam safety monitoring based on edge computing is provided. The system includes a data acquisition module 11, a data processing module 12, an environmental perception computing module 13, a reinforcement learning module 14, and a path scheduling module 15.

[0270] The data acquisition module 11 includes a standardized sensor interface for connecting multiple types of sensors and for acquiring dam safety monitoring data in real time, including water level data, seepage data, and displacement data.

[0271] The data processing module 12 is used to extract features from dam safety monitoring data using the edge terminal, obtain real-time features, and make anomaly judgments based on the real-time features to obtain abnormal real-time data; classify the dam safety monitoring data into levels based on the abnormal real-time data, and construct a priority labeling system for the dam safety monitoring data, including multiple priorities, with the priorities labeled according to the magnitude of the impact of the abnormal real-time data on reservoir safety.

[0272] The environmental perception computing module 13 is used for network status acquisition and status modeling and filtering. The network status acquisition monitors network status parameters in real time through the MPTCP protocol, including transmission rate, transmission latency, available bandwidth and packet loss rate. The status modeling and filtering includes defining the status vector of each transmission path, generating the performance baseline of each path based on historical network status data, setting early warning thresholds, and automatically filtering unusable paths. The network includes 4G / 5G mobile networks, satellite communication and LoRa wireless multi-hop networks.

[0273] The reinforcement learning module 14 uses the Q-Learning reinforcement learning algorithm to achieve dynamic optimization of transmission paths and balance data priority with network performance.

[0274] In this module, the state vector of available transmission paths output by the environment perception module is quantized into discrete values ​​as the environment state. The action space is a set of optional paths such as mobile network, satellite, and LoRa multi-hop. A differentiated reward function is designed based on the characteristics of the transmission data. The Q-value table is updated using an ϵ-greedy strategy. The matrix operation is accelerated by the NPU of RK3588 so that the Q-value update latency is ≤10s. The Q-value table is stored in the running memory and can be quickly restored through backup by the storage subsystem after power failure.

[0275] The path scheduling module 15 is used to perform path switching based on reinforcement learning results, realize data fragmentation transmission and dynamic rescheduling, including a multi-protocol encapsulation engine and real-time monitoring and rescheduling; the multi-protocol encapsulation engine adapts transmission protocols for different paths and supports cross-protocol reassembly (completed through collaboration between the edge end and the receiving end); the real-time monitoring and rescheduling: has a built-in path health check thread that periodically collects path status parameters, and if a sudden drop in path performance is detected, it triggers emergency rescheduling.

[0276] It should be noted that the dynamic optimization system for data transmission paths of reservoir dam safety monitoring based on edge computing of the present invention corresponds one-to-one with the dynamic optimization method for data transmission paths of reservoir dam safety monitoring based on edge computing of the present invention. The technical features and beneficial effects described in the embodiments of the dynamic optimization method for data transmission paths of reservoir dam safety monitoring based on edge computing are applicable to the embodiments of the dynamic optimization method for data transmission paths of reservoir dam safety monitoring based on edge computing. For details, please refer to the description in the embodiments of the method of the present invention, which will not be repeated here.

[0277] Furthermore, in the above embodiments of the dynamic optimization system for data transmission paths for reservoir dam safety monitoring based on edge computing, the logical division of each program module is merely illustrative. In actual applications, the above functions can be assigned to different program modules as needed, for example, for the sake of corresponding hardware configuration requirements or the convenience of software implementation. That is, the internal structure of the dynamic optimization system for data transmission paths for reservoir dam safety monitoring based on edge computing can be divided into different program modules to complete all or part of the functions described above.

[0278] Please see Figure 3 In one embodiment, an edge computing terminal is provided to implement a dynamic optimization method for data transmission paths in reservoir dam safety monitoring based on edge computing. This edge computing terminal 20 employs a heterogeneous computing architecture, with an RK3588 microprocessor (8-core 64-bit ARM architecture, 2.4GHz clock speed) as its core, and integrates an NPU acceleration chip (computing power ≥6 TOPS), supporting second-level response for edge-side data processing, network awareness, and path optimization.

[0279] The edge computing terminal adopts a modular design, including a central processing unit 21, a data acquisition unit 22, a multi-mode communication unit 23, a data storage unit 24, and a power supply unit 25. Each functional unit is interconnected through a high-speed internal bus.

[0280] (1) Central processing unit 21 adopts a microservice containerized architecture and is responsible for running tasks such as dam safety monitoring data feature extraction and anomaly judgment algorithm, dynamic allocation of data tags, network environment perception and dynamic optimization model of transmission path.

[0281] (2) Data acquisition unit 22 realizes standardized access and real-time data acquisition of multi-source heterogeneous sensors, and supports anti-interference transmission and protocol adaptation.

[0282] 1) Hardware interface: It integrates 2 10 / 100 / 1000M Ethernet ports, 4 RS485 ports, 2 RS232 ports, and 2 USB data interfaces, which can be connected to sensor devices such as water level gauges, piezometers, flow meters, displacement gauges, and vibration sensors.

[0283] 2) Protocol processing: Built-in multi-protocol conversion engine supports the parsing of protocols such as Modbus, OPCUA, and MQTT, which can convert raw sensor data into a unified format JSON data stream.

[0284] (3) Multi-mode communication unit 23 enables seamless switching and collaborative operation of heterogeneous networks, and supports low power consumption and high reliability transmission.

[0285] 1) Hardware components: including 4G / 5G communication module, satellite communication module, and LoRa self-organizing network unit.

[0286] 2) Collaborative working mechanism: The mobile network and LoRa link achieve bandwidth aggregation through MPTCP; LoRa nodes adopt time division multiple access (TDMA) scheduling and enter deep sleep during inactive periods.

[0287] (4) Data storage unit 24 adopts an intelligent hierarchical storage architecture design, including a memory subsystem, a storage subsystem, and a power failure protection subsystem, as detailed below:

[0288] 1) Memory Subsystem:

[0289] It consists of 8GB DDR4 dual-channel memory modules and adopts a dual-cache collaborative mechanism, dividing the raw data into a circular buffer and a dynamically allocated processing data buffer pool. Independent addressing and collaborative scheduling of the dual buffers are achieved through a DMA (Direct Memory Access) controller.

[0290] Raw Data Circular Buffer: Used to store unprocessed raw dam safety monitoring data, retaining metadata such as timestamps and device IDs. This buffer employs a circular structure design, using lock-free write technology to achieve concurrent storage of multi-channel sensor data. When the buffer is about to fill, a first-in, first-out (FIFO) approach is used. F I F The O) strategy covers the earliest data and ensures the timely storage of the latest data.

[0291] Data buffer pool: This pool stores feature vectors output by the dam safety monitoring data feature extraction and anomaly detection algorithms, priority-marked data packets to be transmitted (including important and routine data), and multimodal network state parameters. The buffer pool uses a dynamic allocation method, achieving elastic capacity allocation through an intelligent memory management unit.

[0292] 2) Storage subsystem:

[0293] Includes a 32GB industrial-grade eMMC5.1 flash memory module for long-term storage of raw dam safety monitoring data. Raw data is written from the raw data circular buffer to the eMMC5.1 flash memory module via a scheduled task (every 30 minutes by default) or a buffer threshold trigger (remaining capacity ≤ 20%), achieving tamper-proof storage of dam safety monitoring data for more than 36 months.

[0294] 3) Power failure protection subsystem:

[0295] A write-back caching strategy is adopted, recording the write location and status information of the data while it is being written to the memory subsystem. In the event of a power outage, the system uses the recorded information to recover and write the data that has not yet been written to the storage subsystem, ensuring data integrity.

[0296] (5) Power supply unit 25, which is connected to an external power source to supply power to the edge computing terminal or to the monitoring equipment that is wired to the edge computing terminal.

[0297] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0298] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. An edge computing-based reservoir dam safety monitoring data transmission path dynamic optimization method, characterized in that, The method comprises the following steps: Collecting dam safety monitoring data in real time through an edge terminal, the dam safety monitoring data comprising water level data, seepage data and displacement data; Extracting features of the dam safety monitoring data using the edge terminal, obtaining real-time features, and performing abnormality judgment according to the real-time features to obtain abnormal real-time data; Dividing the dam safety monitoring data into grades according to the abnormal real-time data, and constructing a priority label system for the dam safety monitoring data, comprising multiple priorities, which are marked according to the influence of the abnormal real-time data on the safety of the reservoir; Collecting network state parameters and screening available paths using the edge terminal, evaluating the network state parameters in real time through a reinforcement learning algorithm, and dynamically optimizing the transmission path according to the transmission data grade and its priority.

2. The method of claim 1, wherein the method further comprises: Extracting features of the water level data and performing abnormality judgment using the edge terminal, comprising: Adaptively adjusting the size of a sliding window according to the current water level data collection frequency, obtaining multiple real-time water level data, and constructing a current sliding window; The trend of water level change within the sliding window is fitted using linear regression, and the slope β h As the rate of change, the following formula: ； wherein t i is the time point within the window, h i is the water level corresponding to each time point, is the time within the window, is the average water level; When one of the following conditions is met, the current water level data is determined to be abnormal Abnormality over threshold: Threshold water level judgment: dynamically select water level threshold according to current time, flood season water level threshold h w =H1, non-flood season water level threshold h w =H2, wherein H1 is the reservoir flood season limit water level, H2 is the reservoir normal water level, if h t >h w , it is determined as threshold value. Water level fluctuation feature extraction and judgment: if exceeds the preset rate threshold, it is determined that the water level fluctuates rapidly.

3. The method of claim 1, wherein the method further comprises: Extracting features of the seepage data and performing abnormality judgment using the edge terminal, comprising: Adaptively adjusting the size of a sliding window according to historical seepage data and prior data, obtaining multiple real-time seepage data, and constructing a current sliding window; Calculating the mean and standard deviation of the data in the current sliding window to obtain a mutation point, calculating the change rate of the current seepage data and the data at the previous time, calculating the moving percentile of the historical seepage data as a dynamic threshold, taking the latest k historical seepage data and calculating the 95% percentile to obtain the dynamic threshold; Comprehensively considering the features of the seepage data, performing an over-threshold judgment on the current seepage data point, and when one of the following conditions is met, the current seepage data is judged to be over-threshold abnormal: (1) The absolute difference between the current seepage data and the mean is not less than three times the standard deviation; (2) The change rate exceeds the set change rate threshold; (3) The current seepage data point exceeds the dynamic threshold.

4. The method of claim 1, wherein the method further comprises: Extracting features of the displacement data and performing abnormality judgment using the edge terminal, comprising: Adaptively adjusting the size of a sliding window according to historical displacement data and prior data, obtaining multiple real-time displacement data, and constructing a current sliding window; Calculating the vector length of the three-direction combined displacement at the current time, calculating the displacement change rate in each direction using the central difference method, and fitting the trend slope of the displacement data in each direction in the window through the least squares method; the three-direction combined displacement vector is a sequence of three-direction displacement data at the current time; Calculating the mean and standard deviation of each direction displacement in the current sliding window, and calculating the dynamic threshold of each direction displacement in the current active window in real time, including the dynamic threshold of the absolute displacement and the dynamic threshold of the displacement change rate; the calculation of the dynamic threshold of each direction displacement in the current active window is as follows: ； ； wherein, is f a dynamic threshold of absolute displacement of the direction, is f a dynamic threshold of the rate of change of displacement of the direction, , are respectively the mean and the standard deviation of the rate of change of displacement of the direction within the current window, f are respectively the mean and the standard deviation of the rate of change of displacement of the direction within the current window, ρ、 is a confidence adjustment coefficient; According to the characteristics of the displacement data, the current displacement data point is judged to be an abnormal threshold value when one of the following conditions is met: (1) The vector length of the combined displacement is greater than the dam engineering design safety threshold, or the absolute displacement in any direction is not less than the dynamic threshold of the absolute displacement in the corresponding direction; (2) The modulus of the displacement change rate in any direction is not less than the dynamic threshold of the displacement change rate in the corresponding direction; (3) The modulus of the trend slope is not less than the trend change rate safety threshold.

5. The method of claim 1, wherein the method further comprises: The abnormal real-time data is divided into important data, and other real-time data is marked as normal data. The priority label system is constructed by classifying each important data according to the data type and data size and sorting in descending order, and the highest priority is selected in each data type, and the highest priority is selected in each data type. The network state parameters are collected and the available paths are screened, the network state parameters are evaluated in real time by using the reinforcement learning algorithm, and the transmission path is dynamically optimized according to the transmission data level and its priority, and the specific process is as follows:

6. The method of claim 1, wherein the method further comprises: In the process of communicating with the receiving end of the multi-path transmission control protocol, the edge end collects multi-dimensional network state information, which is stored in the cache in the form of a vector, calculates the real-time path state threshold according to the multi-dimensional network state information, and screens the available paths according to the path state threshold; the multi-dimensional network state information includes transmission rate, transmission delay, available bandwidth and packet loss rate; The results of screening the available paths for multiple times are processed by using the Q-Learning reinforcement learning algorithm, the reward function is set according to the transmission data level and its priority, and the data transmission path is continuously optimized through the value update rule until the transmission is completed or the strategy converges. It includes network state information modeling, path state threshold calculation and available path screening.

7. The method of claim 6, wherein the method further comprises: The network state information modeling includes: For the ith transmission path, the network state vector at time t is defined as a four-dimensional real number vector, as follows: The network state information is collected according to a fixed collection period to generate a path state matrix; , i e {1,2,...,n}; wherein R i (t), D i (t), B i (t), and L i (t) are the transmission rate state vector, the transmission delay state vector, the available bandwidth state vector, and the packet loss rate state vector of path i at time t, respectively, and t is the state vector collection time point, used to distinguish different collection periods. The path state threshold calculation includes: The mean value and threshold value of each network state parameter are calculated, as follows: The available path screening includes: ； ； ； ； wherein, , , , is the average value of each network state parameter on path i, N s is the number of collection times within a fixed collection period, R th,i , D th,i , B th,i , L th,i are respectively the transmission rate threshold, the transmission delay threshold, the available bandwidth threshold and the packet loss rate threshold on path i, The condition that path i is an available path is as follows: R The setting of the reward function includes: D For important data, the delay and packet loss rate are minimized first, and the immediate reward function is as follows: B and For regular data, the rate and bandwidth are maximized first, and the immediate reward function is as follows: L are respectively the transmission rate coefficient, the transmission delay coefficient, the available bandwidth coefficient and the packet loss rate coefficient, the values of which are set according to the type of transmission path, and n is the total number of paths. The value update rule includes: The Q value update adopts the time difference update formula of the Q-Learning algorithm: ； where R th , D th , B th , L th are global path average state thresholds, and the calculation formula is as follows: ； ； ； ； ； where ω R , ω D , ω B , ω L are global threshold weights of each network state parameter.

8. The method of claim 6, wherein the method further comprises: The edge computing-based reservoir dam safety monitoring data transmission path dynamic optimization method according to any one of claims 1-8 includes a data acquisition module, a data processing module, an environment perception computing module, a reinforcement learning module and a path scheduling module. The data acquisition module includes a standardized sensor interface for connecting multiple types of sensors and collecting real-time dam safety monitoring data, including water level data, seepage data and displacement data. ； The data processing module is used for feature extraction of dam safety monitoring data by edge, obtaining real-time features, and performing abnormality judgment according to real-time features to obtain abnormal real-time data. ； wherein λ R , λ D , λ B , λ L are weight coefficients, which are tuned according to the service requirements, , , and are the new state vectors of the network state parameters acquired after the execution of the action a. ​ ​ ； where, is the expected cumulative reward for being in the current state s t take action a t the expected cumulative reward for taking action a in state s , r t is the expected cumulative reward for being in the current state s t take action a t the immediate reward obtained, ​ is a discount factor, , represents the expected cumulative reward for choosing the optimal action a in the next state s t+1 .

9. An edge computing-based reservoir dam safety monitoring data transmission path dynamic optimization system, characterized in that, ​ ​ ​ The dam safety monitoring data is graded according to the abnormal real-time data, and a priority label system is constructed for the dam safety monitoring data, including multiple priorities, which are marked according to the influence of abnormal real-time data on the safety of the reservoir; The environment perception computing module is used for network state acquisition and state modeling and screening. The network state acquisition is used for monitoring network state parameters in real time through an MPTCP protocol, including transmission rate, transmission delay, available bandwidth and packet loss rate. The state modeling and screening include defining a state vector of each transmission path, generating a performance baseline of each path according to historical network state data statistics, setting an early warning threshold and automatically filtering unusable paths. The network includes a 4G / 5G mobile network, a satellite communication and a LoRa wireless multi-hop network. The reinforcement learning module is used for realizing dynamic optimization of a transmission path based on a Q-Learning reinforcement learning algorithm, and balancing data priority and network performance. The path scheduling module is used for executing path switching according to the reinforcement learning result, realizing data fragment transmission and dynamic rescheduling, including a multi-protocol encapsulation engine and real-time monitoring and rescheduling.

10. An edge computing terminal, characterized by, The edge computing terminal comprises: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores computer program instructions executable by the at least one processor, and the computer program instructions are executed by the at least one processor to enable the at least one processor to execute the method for dynamic optimization of a transmission path of reservoir dam safety monitoring data based on edge computing according to any one of claims 1-8.

Images