Power transmission conductor wind deflection angle and deflection angle monitoring method based on improved liquid neural network

Abstract

The invention discloses a power transmission conductor wind deflection angle and deflection angle monitoring method based on an improved liquid neural network, and the method comprises the steps: respectively arranging a tower top and a Beidou III attitude measurement device at a power transmission tower and a conductor suspension point, and constructing a high-precision baseline structure; the method comprises the following steps: receiving a Beidou III satellite carrier signal by using double Beidou III satellite antennas, and extracting a carrier phase, a satellite elevation angle and an azimuth angle; constructing a carrier phase double-difference observation equation based on the extracted information, and establishing an attitude angle mathematical model; a reinforcement learning framework driven by a liquid neural network is adopted to perform high-precision fitting and solving on an attitude angle model, the framework designs a multi-phase fusion mechanism to effectively avoid falling into local optimum, and meanwhile, a proposed multi-path parallel mechanism meets the requirement of real-time monitoring; and calculating a wind deflection angle and a deflection angle of the conductor according to the attitude angle. The method does not need a ground-based augmentation station, directly utilizes Beidou III satellite signals to realize millimeter-level attitude perception, and is suitable for all-weather real-time monitoring of the wind deflection angle and the deflection angle of the power transmission line.

Description

Technical Field

[0001] This invention relates to the field of power transmission safety technology, and in particular to a method for monitoring wind deflection and skew angle of power transmission conductors based on an improved liquid neural network. Background Technology

[0002] During the operation of transmission lines, conductors are prone to galloping due to factors such as wind, temperature changes, and electromagnetic vibrations. This galloping not only causes changes in insulator tension and wear on hardware, but can also lead to serious faults such as flashover discharge and phase-to-phase short circuits between the conductor and the tower, threatening the safe and stable operation of the power grid.

[0003] Conductor galloping typically manifests as the deviation of a conductor in space. The horizontal deflection is called the wind deflection angle, and the vertical deflection is called the skew angle. When either the wind deflection angle or the skew angle exceeds a critical value, it indicates that the conductor has entered a state of severe galloping, requiring immediate warning and intervention. Therefore, there is an urgent need for a high-precision, real-time, and intelligent monitoring system for the wind deflection angle and skew angle of transmission conductors.

[0004] Currently, existing methods for monitoring the wind deflection angle and skew angle of transmission lines have the following main shortcomings:

[0005] 1. Video image method: This method relies on a vision system to identify the trajectory of the conductor's movement, but it is limited by factors such as weather, lighting, and shooting angle, resulting in low accuracy and high cost.

[0006] 2. Accelerometer sensor method: This method requires a battery-powered device to be installed in the wires, resulting in high system maintenance costs and limited sampling frequency.

[0007] 3. Conventional GPS / tilt meter method: can only roughly determine the direction of conductor offset, cannot accurately quantify wind deflection angle and skew angle, and is difficult to meet the requirements of millimeter-level attitude monitoring.

[0008] Therefore, there is an urgent need to change this situation by improving the method for monitoring wind deflection and skew angle of transmission lines based on liquid neural networks. Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings of existing technologies by proposing an improved liquid neural network-based method for monitoring wind deflection and skew angle of transmission lines. Its advantages include eliminating the need for ground-based augmentation stations, directly utilizing BeiDou-3 satellite signals to achieve millimeter-level attitude sensing, and making it suitable for all-weather real-time monitoring of wind deflection and skew angle of transmission lines.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] The improved liquid neural network-based method for monitoring wind deflection and skew angle of transmission lines includes the following steps:

[0012] Step 1: Define BeiDou III satellite antenna A P A Q A P Beidou III attitude measurement device P,A deployed on transmission towers P As the main antenna, and as a fixed point, A Q Deployed at the top Q of the transmission line tower as the suspension point, P and Q form the baseline.

[0013] Step 2: Define the other fixed point of Q as P', the horizontal plane containing the two adjacent fixed points PP' as H, the vertical plane as V, and the projection of Q onto H as Q'. H The projection of V is Q V , ∠Q H PP′ is the wind deflection angle α, ∠Q V PP′ is the deflection angle β;

[0014] Step 3: A Q Receive BeiDou III satellite S i radio frequency carrier signal R i The carrier phase is obtained through analysis. Elevation angle θ i Azimuth Q will use Lora to Send to P; P function: A P Receive BeiDou III satellite S i radio frequency carrier signal R i The carrier phase is obtained through analysis. P receives Q sent via Lora.

[0015] Step 4: Solve for integer ambiguity by performing double difference using two carrier phases, and then calculate the solution based on the integer ambiguity. Attitude angles are used to calculate wind deflection angle and yaw angle, which are then transmitted to a remote network server via 4G signal.

[0016] The present invention is further configured to define The heading angle ω is the projection PQ H The angle ω between the north direction and true north ranges from 0° to 360°; defined The pitch angle ψ is PQ and PQ H The angle between them The range is -90° to 90°.

[0017] The invention is further configured such that Q is arranged in the middle of the transmission line to ensure that the suspension point is in the maximum position when the line is swinging, and P is arranged at the connection between the transmission line and the transmission tower to ensure that the measured wind deflection angle α and skew angle β are accurate.

[0018] The present invention is further configured to parse A through P and Q. P A Q Received R i Obtain the BeiDou III carrier phase The elevation angle θ of the BeiDou-3 satellite is calculated by interpolating the receiver location, reception time, and ephemeris data. i Beidou III satellite azimuth angle

[0019] The present invention is further configured such that step three includes:

[0020] Establish A respectively P A Q Carrier phase model,

[0021]

[0022] in, It is A P A Q The carrier phase of the i-th Beidou III satellite is received; λ is the wavelength of the radio frequency carrier signal. It is A P A Q The pseudorange relative to the i-th Beidou III satellite; c is the speed of light; δt P δt Q It is the P and Q clock error; δt i This is the clock error of the i-th Beidou III satellite; It is A P A Q Integer ambiguity for the i-th Beidou III satellite; ε represents random observation noise;

[0023] Based on the carrier phase double difference,

[0024]

[0025] in, The integer coefficient;

[0026] Replace ρ with ω, ψ, θ. Obtain the attitude angle model;

[0027] pseudo-range Represented as the projection of the baseline onto the cosine of the satellite direction:

[0028]

[0029] in, b is Length; e(ω,ψ) is Direction vector; u iLet θ be the unit direction vector of the i-th Beidou III satellite; i The i-th satellite pair The altitude angle; The i-th satellite pair The azimuth angle;

[0030] The dot product of two direction vectors is obtained.

[0031]

[0032] Substituting into the formula, we get

[0033]

[0034] Substituting into the formula, we get

[0035]

[0036] The attitude angle model can be obtained from the above formula.

[0037]

[0038] The present invention is further configured such that step four includes:

[0039] The attitude angle mathematical model was solved using a reinforcement learning framework driven by a multiphase liquid neural network to obtain the... The attitude angle;

[0040] The dynamic behavior of the liquid neural network is as follows.

[0041]

[0042] Where t represents the time step; h(t) represents the hidden state vector at time t; γ is the time leakage coefficient; σ is the nonlinear activation function; and W... x Input weight matrix; W h Implicit strong-connection weight matrix; d bias vector; x(t) represents the input feature at time t;

[0043] Modeling the multiphase information field yields:

[0044]

[0045] Where G(t) represents the stealth combination state of the entire network at time t; K represents the number of phase fields; h [k] Represents the state vector of the k-th phase; This represents the flow component of the k-th phase;

[0046] The liquid dynamics model for each phase is as follows:

[0047]

[0048] Among them, h [k] (t), h [l] (t) represents the stealth state vector of phase k and l at time t; γ [k] σ[ represents the time leakage coefficient of the k-th phase; k ] represents the nonlinear activation function of the k-th phase; This represents the input weight matrix for the k-th phase. d represents the hidden strong-linked weight matrix of the k-th phase; [k x represents the bias vector of the k-th phase; [k] This represents the input feature of the k-th phase at time t;

[0049] The overall state fusion multiphase liquid neural network modeling strategy function π(a|s) is used.

[0050]

[0051] Modeling the value function v(s) t+1 ) = MLP(s t ), s t+1 s t Let be the states at times t+1 and t, and MLP(s) be a multilayer vector machine;

[0052] The updated objective is to maximize expected return:

[0053]

[0054] Among them, a t It is the estimate of the dominance function:

[0055] a t =r t +γv(s t+1 )-v(s t );

[0056] clip(s) is the Clip coefficient, a t+1 a t For the actions at times t+1 and t, r t Let γ be the reward at time t, and γ be the weighting coefficient.

[0057] The current network state is defined as the combination of each variable and the current candidate solution (ψ). t ,ω t ),

[0058] State vectors are constructed using relevant physical quantities in the attitude angle model, and a policy network is built to predict attitude angles; where: ψ is a search subspace with a limited range of [0, 90°]; ω is a search subspace with a limited range of [0, 360°];

[0059] The action space is defined based on the ranges defined by ψ and ω. a ω ∈[0,2π];

[0060] The reward function is designed based on the error between the current predicted ambiguity and the integer ambiguity:

[0061]

[0062] Where, |N p -N| 2 Current prediction ambiguity N p The difference between the integer ambiguity N and the reward function weights τ1 and τ2; the change range of the actions |Δψ| and |Δω|; and the auxiliary penalty or reward for action selection, D(ψ,ω).

[0063] Based on the current state s t And the policy function π(a|s) generates multiple candidate inference paths a={a1,a2,…,a… k},a i ∈[-0.1,0.1] 2 Let the current state be s. t This includes GNSS observation data, attitude angle estimates, and the strategy model π. θ (a|s) Output an action distribution a t ~π(a|s);

[0064] Each action corresponds to an attitude angle adjustment [Δθ, Δφ], thus generating k different inference paths:

[0065] trajector i =f(s) t ,a i ), i = 1, ..., k;

[0066] Where f is the environmental dynamics function, representing the effect of the action on the state;

[0067] The environment executes all actions and generates a trajectory, based on the candidate action 'a' for each step. i and current state s t Perform the operation and update the state through a physical model or simulation mechanism to form a complete trajectory s t+1 =f(s) t ,a t ), S={s1,s2,…,s k},s i =f(s) t ,a i For each a i Calculate the new attitude angle:

[0068]

[0069] The ambiguity solution N is calculated based on the new attitude angle. The above information is encapsulated into structured text for subsequent scoring. The PRM scores the trajectory: scores = PRM(S). The trajectory with the highest score is selected as the current Step action.

[0070]

[0071] Update the relevant parameters of the policy function π(a|a) based on the current action;

[0072] The search process terminates when any of the following conditions are met, and the final determined attitude angle and integer ambiguity are output as the final solution for calculating the wind deflection angle and skew angle of the transmission line: The ambiguity prediction error meets the predetermined accuracy requirement, i.e., |N... p -N| 2 The predetermined convergence threshold has been reached; the predetermined maximum number of iterations has been reached.

[0073] The present invention is further configured such that the formula for calculating the wind deflection angle is: α=arctan2(cosψ·sinω,cosψ·cosω); and the formula for calculating the skew angle is: β=arcsin(sinψ).

[0074] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0075] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0076] The beneficial effects of this invention are as follows:

[0077] 1. This invention offers high measurement accuracy. By acquiring radio frequency carrier signals from two BeiDou-3 satellites, this invention establishes a mathematical model of the conductor-tower baseline attitude angle. Combining the satellite elevation and azimuth angles, it performs dual-satellite difference calculations to accurately determine the conductor's wind deflection and skew angles, effectively reflecting the conductor's real-time galloping state.

[0078] 2. This invention has a high degree of intelligence. The system integrates a Beidou attitude measurement device and a LoRa wireless communication module. The receiver at the conductor end sends the collected carrier phase information to the gateway host at the tower end. Based on artificial intelligence algorithms, the attitude angle is automatically calculated, realizing the automatic identification and monitoring of wind deflection angle and yaw angle.

[0079] 3. The invention has low deployment costs. It only requires a pair of BeiDou-3 antennas (tower top, BeiDou-3 attitude measurement device) to be deployed at both the tower and the conductor suspension point to establish a high-precision attitude baseline, eliminating the need for expensive CORS ground stations. The system equipment is lightweight and can be quickly installed on existing power lines, facilitating large-scale deployment and application. Attached Figure Description

[0080] Figure 1 This is a schematic diagram of the mathematical model of the baseline attitude angle on the transmission tower based on the improved liquid neural network transmission conductor wind deflection and skew angle monitoring method proposed in this invention.

[0081] Figure 2 This is a schematic diagram of the wind deflection angle of the transmission line based on the improved liquid neural network method for monitoring wind deflection and skew angle proposed in this invention.

[0082] Figure 3 This is a schematic diagram of the deflection angle of the transmission line based on the improved liquid neural network method for monitoring wind deflection and skew angle proposed in this invention.

[0083] Figure 4 This is a flowchart of the transmission line galloping monitoring method based on the improved liquid neural network transmission conductor wind deflection and skew angle monitoring method proposed in this invention.

[0084] Figure 5 This is a schematic diagram of a multiphase liquid neural network based on an improved liquid neural network method for monitoring wind deflection and skew angle of transmission lines, as proposed in this invention. Detailed Implementation

[0085] The technical solution of this patent will be further described in detail below with reference to specific embodiments.

[0086] The embodiments of this patent are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this patent, and should not be construed as limiting this patent.

[0087] Reference Figure 1-5 Based on an improved liquid neural network method for monitoring wind deflection and skew angle of transmission lines, including tower top, BeiDou III satellite antenna, and BeiDou III attitude measurement device;

[0088] Define BeiDou III satellite antenna A P A Q A P Beidou III attitude measurement device P,A deployed on transmission towers P As the main antenna, and as a fixed point, A QDeployed at the top Q of the transmission line tower as the suspension point, P and Q form the baseline.

[0089] Define the other end of Q as P', the horizontal plane containing the two adjacent fixed points PP' as H, the vertical plane as V, and the projection of Q onto H as Q'. H The projection of V is Q V , ∠Q H PP′ is the wind deflection angle α, ∠Q V PP′ is the deflection angle β;

[0090] Define Q function: A Q Receive BeiDou III satellite S i radio frequency carrier signal R i The carrier phase is obtained through analysis. Elevation angle θ i Azimuth Q will use Lora to Send to P; P function: A P Receive BeiDou III satellite S i radio frequency carrier signal R i The carrier phase is obtained through analysis. P receives Q sent via Lora. Integer ambiguity is solved by double difference of two carrier phases, and the solution is based on integer ambiguity. Attitude angles are used to calculate wind deflection angle and yaw angle, which are then transmitted to a remote network server via 4G signal.

[0091] A method for monitoring wind deflection and skew angle of transmission lines based on BeiDou III attitude measurement is used to monitor the galloping of transmission lines, and the following steps are taken:

[0092] definition The heading angle ω is the projection PQ H The angle ω between the north direction and true north ranges from 0° to 360°.

[0093] definition The pitch angle ψ is PQ and PQ H The angle between them The range is -90° to 90°;

[0094] Step 1: Install the BeiDou III attitude measurement device at the top of the guide tower;

[0095] Q is placed in the middle of the transmission line to ensure that the suspension point is in the maximum position when the line gallops. P is placed at the connection between the transmission line and the transmission tower to ensure that the measured wind deflection angle α and skew angle β are accurate.

[0096] Step 2: Analyze and transmit the radio frequency carrier signal of the BeiDou III satellite;

[0097] Analyze A using P and Q. P A Q Received R i Obtain the BeiDou III carrier phase The elevation angle θ of the BeiDou-3 satellite is calculated by interpolating the receiver location, reception time, and ephemeris data. i Beidou III satellite azimuth angle

[0098] Step 3: Construct a baseline on P Attitude angle model;

[0099] Step 3.1, Establish A P A Q Carrier phase model on;

[0100] based on Positioning accuracy is achieved at the millimeter level, and A is established respectively. P A Q Carrier phase model, connection With respect to physical state (pseudo-distance ρ + clock δt + integer cycle N + noise ε),

[0101]

[0102] It is A P A Q The carrier phase of the i-th Beidou III satellite is received; λ is the wavelength of the radio frequency carrier signal (the wavelength of Beidou III in B1 is 0.19029367m); It is A P A Q The pseudorange relative to the i-th Beidou III satellite is the "measured distance" from the satellite to the receiver. It is called pseudorange because of reception errors; c is the speed of light (299,792,458 m / s); δt P δt Q It is the P and Q clock error; δt i This is the clock error of the i-th Beidou III satellite; It is A P A Q The integer ambiguity of the i-th Beidou III satellite; ε is the random observation noise, which is the error caused by thermal noise, multipath, vibration, etc. during the reception process.

[0103] Based on the carrier phase double difference elimination of BeiDou III satellite clock error and receiver clock error, the following is obtained:

[0104]

[0105] in This is the integer coefficient.

[0106] Step 3.4: Replace ρ with ω, ψ, θ. Obtain the attitude angle model;

[0107] pseudo-range Represented as the projection of the baseline onto the cosine of the satellite direction:

[0108]

[0109] in, b is Length; e(ω,ψ) is Direction vector, according to The attitude angles (ω and ψ) give its three-dimensional direction vector; u i Let θ be the unit direction vector of the i-th Beidou III satellite, and let the direction of satellite i (θ) be... Convert θ to a three-dimensional unit vector. i The i-th satellite pair The altitude angle; The i-th satellite pair The azimuth angle.

[0110] The dot product of two direction vectors is obtained.

[0111]

[0112] Using attitude angles and satellite geometric information, the pseudorange difference can be directly represented. Substituting formula (10) into formula (6) yields...

[0113]

[0114] Substituting formula (11) into formula (5), we get:

[0115]

[0116] The attitude angle model can be obtained from the above formula, reflecting the relationship between N and ω and ψ.

[0117]

[0118] A nonlinear functional relationship was established between integer ambiguity N and baseline attitude angles (ω, ψ), forming the mathematical basis for attitude resolution. If a set of (ω, ψ) can be found that makes N close to an integer, a correct attitude solution can likely be obtained.

[0119] Step 4: Solve the attitude angle mathematical model using a reinforcement learning framework driven by a multiphase liquid neural network to obtain the... The attitude angle;

[0120] Traditional search algorithms are prone to getting trapped in local optima in complex environments and are computationally intensive. To address this, this paper proposes a reinforcement learning framework based on an improved Liquid Neural Network (LNN) for efficiently solving integer ambiguities and retrieving attitude angles. This step maps the integer ambiguity problem into a multi-dimensional reinforcement learning problem, enabling efficient search for ψ and ω.

[0121] Step 4.1, Strategy Definition:

[0122] Liquid neural networks are a novel type of neural network model inspired by biological nervous systems. Their core idea is to simulate the continuous, dynamic, and differentiable interactions between neurons. Unlike traditional artificial neural networks (such as fully connected networks and LSTMs), which perform computation discretely and in time-step increments, liquid neural networks treat the network's internal state (hidden state) h(t) as a continuously evolving physical system over time. The dynamic behavior of liquid neural networks is as follows:

[0123]

[0124] t represents the time step; h(t) represents the hidden state vector at time t; γ is the time leakage coefficient; σ is the nonlinear activation function; W x Input weight matrix; W h The hidden strongly coupled weight matrix; the bias vector d; and x(t) represent the input features at time t. This equation describes the evolution of the hidden state h(t) over time, exhibiting memory and nonlinear response capabilities, making it suitable for capturing dynamic trends during the pose search process.

[0125] Although liquid neural networks perform well in handling dynamic time-series problems, a single-phase, unified LNN will expose the following key problems when facing complex, high-dimensional optimization problems such as attitude angle solving: Attitude angle solving (especially integer ambiguity search) is a typical "non-convex" optimization problem, with a large number of local optima in its solution space. For a single-phase LNN, the state h(t) of the entire network is a unified vector, and its search path in the solution space is singular. The attitude angle contains two key components—heading angle θ and pitch angle ψ. Although these two components are related, their physical meanings, variation laws, and search characteristics are different. A unified hidden state h(t) attempts to encode and process these two complex information of different dimensions simultaneously using the same dynamic mechanism. To solve the above problems, we designed a "multi-phase fusion mechanism". Its core idea is to decompose a complex, single search task into multiple parallel and relatively independent "phases". Based on (14), we model the multi-phase information field.

[0126]

[0127] G(t) represents the stealth combination state of the entire network at time t; K represents the number of phase fields; h [k] Represents the state vector of the k-th phase; This represents the flow component (output weight) of the k-th phase. The network is divided into K "phases," each evolving independently, and then weighted... By integrating outputs, the expressive power of the model is enhanced.

[0128] In this process, the liquid dynamics of each phase are modeled.

[0129]

[0130] h [k] (t), h [l] (t) represents the stealth state vector of phase k and l at time t; γ [k] σ represents the time leakage coefficient of the k-th phase (describing the rate of information decay); [k] Represents the nonlinear activation function of the k-th phase; This represents the input weight matrix for the k-th phase. d represents the hidden strong-linked weight matrix of the k-th phase; [k] x represents the bias vector of the k-th phase; [k] Let η represent the input characteristics of the k-th phase at time t. Coupling coefficient η lk This enables inter-phase information transfer. For example, once the pitch angle subspace has been initially locked into a reasonable range, prior information can be transferred from the pitch angle subspace to the yaw angle subspace via the η direction to achieve coordinated convergence.

[0131] The overall state fusion multiphase liquid neural network modeling strategy function π(a|s) is used.

[0132]

[0133] The policy function determines the next action 'a', and the value function evaluates the quality of the current state; together, they guide the search direction. Modeling the value function v(s) t+1 ) = MLP(s t ), s t+1 s t Let be the states at time t+1 and t, and MLP(s) be a multilayer vector machine. By comparing the value of the next state that can be reached after different actions, the algorithm can make more informed decisions. If a new action, although its immediate reward is not high, can lead the system to a new state s with a high value, then... t+1 (v(s t+1If a state has a high value, then that action is worth exploring. Conversely, if all possible actions lead to a low-value state, then the algorithm should tend to utilize the best known policy. The policy function determines "what to do" (i.e., which action to choose), while the value function evaluates "how good the current state is." They work together to guide the search process towards a fast and stable convergence towards the optimal solution.

[0134] After constructing the "multiphase liquid neural network," we developed a powerful strategy for intelligently solving attitude angle problems. Next, to enable this "agent" to learn and evolve autonomously, we updated the objective to maximize the expected return:

[0135]

[0136] Where a t It is the estimate of the dominance function:

[0137] a t =r t +γv(s t+1 )-v(s t (14)

[0138] clip(s) is the Clip coefficient, a t+1 a t For the actions at times t+1 and t, r t Let γ be the reward at time t, and γ be the weighting coefficient. By maximizing the expected return, the network is guided to achieve a balance between exploration and exploitation, avoiding blind search.

[0139] Step 4.2, State Space Definition:

[0140] The state space s is the set of all information about the entire problem that a multiphase liquid neural network can "perceive" at any decision time t. The current network state is defined as the combination of each variable and the current candidate solution (ψ). t ,ω t ), It provides a comprehensive description of the network's current search state, which is then used by reinforcement learning algorithms to make decisions.

[0141] Step 4.3, Define the action space:

[0142] The action space is the set of all possible actions that a multiphase liquid neural network can choose to execute at any decision time t. A state vector is constructed using relevant physical quantities in the attitude angle model, and a policy network is built to predict the attitude angle. Specifically: ψ is the search subspace, limited to [0, 90°]; ω is the search subspace, limited to [0, 360°].

[0143] The action space is defined based on the ranges defined by ψ and ω. a ω ∈[0,2π], the action space is limited to a physically reasonable range to prevent invalid jumps.

[0144] Step 4.4, Definition of reward function:

[0145] The reward function serves as the "survival rule" and "behavioral guide" for multiphase liquid neural networks. It is used to immediately provide the agent with a numerical feedback (reward r) after the multiphase liquid neural network performs an action and enters a new state. t Or punish r t The reward function is designed based on the error between the current predicted ambiguity and the integer ambiguity.

[0146]

[0147] |N p -N| 2 Current prediction ambiguity N p The smaller the difference between the integer ambiguity N and the target value, the better; τ1 and τ2 are the weights of the reward functions; the magnitude of changes in |Δψ| and |Δω| actions penalizes large changes to stabilize the search process; D(ψ,ω) is an auxiliary penalty or reward term for action selection, guiding action stability. Once a phase (e.g., pitch angle) has initially locked onto a candidate region with a small ambiguity error, the coupling coefficient η drives the interaction mechanism. The stable solution of the pitch angle subspace (phase 1) transmits information to the heading angle subspace (phase 2) through the coupling coefficient η; the heading angle subspace quickly receives feedback from the region where the pitch angle has been determined, narrowing the search range and converging rapidly; the two phases drive and feedback each other, accelerating the search speed of the overall network and quickly achieving fine focusing. The reward mechanism embodies the design principles of "accuracy priority, stable search, and collaborative optimization".

[0148] Step 4.5, Multi-path parallel exploration mechanism:

[0149] To address the efficiency and stability challenges faced by traditional search algorithms and basic reinforcement learning algorithms in the specific scenario of high-precision GNSS attitude calculation, we propose a multi-path parallel exploration mechanism. Based on the current state s... t And the policy function π(a|s) generates multiple candidate inference paths a={a1,a2,…,a… k},a i ∈[-0.1,0.1] 2 Let the current state be s. t This includes GNSS observation data, attitude angle estimates, etc. Policy model π θ (a|s) Output an action distribution a t ~π(a|s).

[0150] Each action corresponds to an attitude angle adjustment [Δθ, Δφ], thus generating k different inference paths:

[0151] trajector i =f(s) t ,a i ), i = 1, ..., k (16)

[0152] Where f is the environmental dynamics function, representing the impact of the action on the state. By performing parallel simulation and evaluation of multiple candidates and multiple paths at each step, it transforms "single-threaded blind exploration" into "multi-threaded intelligent deduction," thereby achieving a double leap in search speed and solution accuracy.

[0153] Step 4.6: The environment executes all actions and generates a trajectory, based on the candidate action a for each step. i and current state s t Perform the operation and update the state through a physical model or simulation mechanism to form a complete trajectory s t+1 =f(s) t ,a t ), S={s1,s2,…,s k},s i =f(s) t ,a i For each a) i Calculate the new attitude angle:

[0154]

[0155] The ambiguity solution N is calculated based on the new attitude angle. This information is then encapsulated into structured text for subsequent scoring. The PRM scores the trajectory as scores = PRM(S). The trajectory with the highest score is selected as the current Step action.

[0156]

[0157] The relevant parameters of the policy function π(a|s) are updated based on the current action. This mechanism is similar to "thought chain" reasoning, allowing for parallel exploration of multiple paths and improving global search capabilities.

[0158] Step 4.8, Termination Condition

[0159] The search process terminates when any of the following conditions are met, and the final determined attitude angle and integer ambiguity are output as the final solution for calculating the wind deflection angle and skew angle of the transmission line:

[0160] (1) The ambiguity prediction error meets the predetermined accuracy requirement, i.e., |N p -N| 2The predetermined convergence threshold has been reached;

[0161] (2) Reach the predetermined maximum number of iterations.

[0162] The algorithm is guaranteed to converge within a finite time, meeting real-time requirements.

[0163] Step 5: Calculate the wind deflection angle and yaw angle based on the attitude angle;

[0164] α and β are calculated based on the attitude angles, and the real-time monitored α and β are sent to a remote server.

[0165] α=arctan2(cosψ·sinω,cosψ·cosω) (19)

[0166] β=arcsin(sinψ) (20)

[0167] If α > 0: deviates from the right side (east side) of the main tower.

[0168] If α < 0: deviates from the left side (west side) of the main tower.

[0169] If β > 0: the wire curves upwards.

[0170] If β < 0: the conductor sags.

[0171] The final results are uploaded to a remote server via a 4G network for real-time monitoring and analysis by maintenance personnel.

[0172]

[0173]

[0174] This invention utilizes BeiDou III carrier phase double difference and liquid neural network optimization to reduce wind deflection / skew angle errors to within 0.15°, achieving millimeter-level (0.5cm) positioning accuracy. This represents a computational efficiency improvement of over 20 times compared to traditional video image methods. The key parameters used in the experiment and their values ​​are shown below:

[0175]

[0176] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0177] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0178] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for monitoring wind yaw angle and deflection angle of transmission conductor based on improved liquid neural network, characterized in that, Comprising the following steps: Step one: define the Beidou III satellite antenna A P , A Q , A P Deployed on the power transmission tower Beidou III attitude measurement device P, A P As a fixed point, A Q Deployed on the top of the power transmission tower Q as a hanging point, P, Q constitute a baseline Step two: define the other end of Q as P', the horizontal plane of two adjacent fixed points PP' as H, the vertical plane as V, Q in H projection as Q H , Q in V projection as Q V , ∠Q H PP' as wind deviation angle α, ∠Q V PP' as deflection angle β; Step 3: A Q Receive BeiDou III satellite S i radio frequency carrier signal R i The carrier phase is obtained through analysis. Elevation angle θ i Azimuth Q will use Lora to Send to P; P function: A P Receive BeiDou III satellite S i radio frequency carrier signal R i The carrier phase is obtained through analysis. P receives Q sent via Lora. Step four: double-difference solution of integer ambiguity by two carrier phases, based on integer ambiguity solution Attitude angles, from which a wind yaw angle and a drift angle are derived, are transmitted to the remote network server via the 4G signal.

2. The improved liquid neural network based method for monitoring wind yaw angle and drift angle of power transmission conductor as claimed in claim 1 wherein, definition The heading angle ω is the projection PQ H The angle ω between the north direction and true north ranges from 0° to 360°; defined The pitch angle ψ is PQ and PQ H The angle between them The range is -90° to 90°.

3. The improved liquid neural network based method for monitoring wind yaw angle and drift angle of power transmission conductor as claimed in claim 2 wherein, Arranging Q in the middle of the transmission line to ensure that the suspension point can be in the maximum position during galloping, and arranging P at the connection between the transmission line and the transmission tower to ensure that the measured wind yaw angle a and deflection angle b are accurate.

4. The improved liquid neural network based method for monitoring wind yaw angle and drift angle of power transmission conductor as claimed in claim 3 wherein, Resolving A by P, Q P , A Q Received R i , get compass III carrier phase Through the receiver position, receive time and ephemeris interpolation compass III satellite height angle θ i , compass III satellite azimuth 5. The improved liquid neural network based method for monitoring wind yaw angle and drift angle of power transmission conductor as claimed in claim 1 wherein, The step three comprises: A P , A Q carrier phase model, wherein, is A P , A Q is the carrier phase received from the i-th Beidou III satellite; and λ is the wavelength of the radio frequency carrier signal; is A P , A Q is the pseudo-range with respect to the i-th Beidou III satellite; c is the speed of light; and δt P , δt Q is the clock error of the P, Q receiver; δt i is the clock error of the i-th Beidou III satellite; is A P , A Q is the integer ambiguity with respect to the i-th Beidou III satellite; and ε is the random observation noise; Based on the carrier phase double difference, wherein is the integer coefficient; replace p with ω, ψ, θ, get the attitude angle model; The pseudo-range difference is expressed as a projection of the baseline-to-satellite direction cosine: wherein, b is length; e(ω,ψ) is direction vector; u i is the unit direction vector of the i-th Beidou III satellite; θ i is the elevation angle of the i-th satellite pair is the azimuth angle of the i-th satellite pair is the azimuth angle of the i-th satellite pair​ Two directional vector dot products are obtained, Substituting the formula gives Substituting the formula gives From the above formula, the attitude angle model can be obtained, 6. The improved liquid neural network based method for monitoring wind yaw angle and drift angle of power transmission conductor as claimed in claim 1 wherein, The step four comprises: solving the mathematical model of the attitude angle using a reinforcement learning framework driven by a multiphase liquid neural network, obtaining the attitude angle of the ​ The dynamic behavior of the liquid neural network is as follows, where t denotes time step; h(t) denotes hidden state vector at time t; γ time leaky coefficient; σ nonlinear activation function; W x input weight matrix; W h hidden strong connection weight matrix; d bias vector; x(t) denotes input feature at time t; Multi-phase information field modeling is performed: wherein G(t) represents the state of the hidden combination of the network at time t; K represents the number of phase fields; h [k] represents the state vector of the kth phase; represents the flow component of the kth phase; Wherein, the liquid dynamics modeling of each phase is as follows: where h [k] (t) represents the kth, lth hidden state vector at time t; γ [l] (t) represents the kth, lth hidden state vector at time t; γ [k] represents the kth time leakage coefficient; σ [k] represents the kth nonlinear activation function; represents the kth input weight matrix; represents the kth hidden-to-strong weight matrix; d [k ] represents the kth bias vector; x [k] represents the kth input feature at time t; The overall state fusion multi-phase liquid neural network modeling strategy function π(a|s) is: Modeling the value function v(s t+1 ) = MLP(s t ), s t+1 , s t are the states at time t+1, t, and MLP(s) is a multi-layer vector machine; The update target is to maximize the expected return: where a t is the advantage function estimate: a t = r t + γv(s t+1 ) - v(s t ); The clip(s) is a Clip coefficient, a t+1 , a t is the action at time t + 1, t t is the reward at time t, and γ is a weight coefficient. The current network state is defined as the combination of variables and the current candidate solution (ψ t ,ω t ) at the current iteration, The state vector is constructed using the relevant physical quantities in the attitude angle model, and the strategy network is constructed to predict the attitude angle; wherein: ψ searches the subspace, and the range is limited to [0, 90°]; ω searches the subspace, and the range is limited to [0, 360°]; The action space is defined by the range according to ψ and ω, a ω ∈ [0, 2π] The reward function is designed based on the error between the current predicted ambiguity and the whole-week ambiguity: where |N p -N 2 the current prediction ambiguity N p the difference between the integer ambiguity N and the fractional ambiguity N, τ1, τ2 reward function weight; |Δψ|, |Δω| change amplitude of action; D(ψ, ω) auxiliary penalty or reward item of action selection; According to the current state s t And the policy function π(a|s) generates a plurality of candidate reasoning paths a k}= {a1, a2, …, a i ∈ [-0.1, 0.1] 2}; Set the current state as s t , which contains GNSS observation data, attitude angle estimation value; The policy model π θ (a|s) outputs an action distribution a t ~ π(a|s); Each action corresponds to an attitude angle adjustment amount [Δθ, Δφ], thereby generating k different reasoning paths: trajector i = f(s t , a i ), i = 1,..., k; Wherein, f is the environmental dynamics function, which represents the influence on the state after executing the action; The environment executes all actions and generates trajectories, the environment according to the candidate action a of each step i and the current state s t Perform operations and update the state through a physical model or simulation mechanism to form a complete trajectory s t+1 = f(s t , a t ), S = {s1, s2, …, s k}, s i = f(s t , a i ); for each a i , calculate the new attitude angle: According to the new attitude angle, the ambiguity solution N is calculated, and the above information is packaged as structured text for subsequent scoring; the PRM scores the trajectory scores=PRM(S); select the highest score trajectory as the current Step action: According to the current action, update the related parameters of the strategy function π(a|s). The search process is terminated when any of the following conditions is met, and the final determined attitude angle and integer ambiguity are output as the final solution for calculating the windage and sag of the power transmission conductor: the ambiguity prediction error meets the predetermined accuracy requirement, i.e., |N p -N| 2 The predetermined convergence threshold is reached; the predetermined maximum number of iterations is reached.

7. The improved liquid neural network based method for monitoring windage and yaw angle of power transmission conductor according to claim 1, wherein, The wind yaw angle calculation formula is: a=arctan2(cosψ*sinω,cosψ*cosω); the deflection angle calculation formula is: b=arcsin(sinψ).

8. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the steps of the method of any one of claims 1 to 7.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method of any one of claims 1 to 7.

Images