Power transmission line dynamic current-carrying capacity prediction and capacity increase method based on double-model adaptive fusion
By using a dual-model adaptive fusion method, combining meteorological-driven and data-driven models, the problem of insufficient accuracy in traditional current carrying capacity prediction is solved, and dynamic capacity expansion of transmission lines with higher accuracy and reliability is achieved.
Patent Information
- Application Number
- CN202511941415.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-22
- Publication Date
- 2026-05-01
AI Technical Summary
Traditional current carrying capacity prediction methods rely on static heat balance equations or single data models, which fail to fully consider the temporal characteristics of meteorological changes and historical load data, resulting in limited prediction accuracy and difficulty in meeting the real-time and reliability requirements of power grid dispatch.
A dual-model adaptive fusion approach is adopted, combining meteorological and data-driven models. The kernel extreme learning machine is optimized through empirical mode decomposition and artificial heron optimization algorithm, and the LSTM model is used for load prediction. Dynamic capacity expansion is achieved through adaptive weighted fusion and confidence interval evaluation.
It significantly improves the accuracy of current carrying capacity prediction and the model's adaptability to complex weather and load changes, thereby enhancing the safety and reliability of dynamic capacity expansion of transmission lines.
Smart Images

Figure CN121959410A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of transmission line monitoring and dynamic capacity expansion technology, specifically involving a method for predicting and expanding the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion. Background Technology
[0002] As power systems increasingly demand higher efficiency and safety from transmission lines, dynamic capacity expansion technology has become a crucial means to enhance transmission capacity. Traditional current carrying capacity prediction methods often rely on static heat balance equations or single data models, failing to fully consider the dynamic impact of meteorological changes and the temporal characteristics of historical load data. This results in limited prediction accuracy, making it difficult to meet the real-time and reliability requirements of power grid dispatching. Summary of the Invention
[0003] The purpose of this invention is to provide a method for dynamic current carrying capacity prediction and capacity expansion of transmission lines based on dual-model adaptive fusion. By fusing two types of prediction models, namely meteorological-driven and data-driven, the method aims to improve the accuracy and reliability of current carrying capacity prediction and realize dynamic capacity expansion and safe operation of transmission lines.
[0004] To achieve the above objectives, the technical solution adopted by this invention is a method for predicting and increasing the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion, characterized by the following steps: S1, Data Acquisition and Preprocessing Historical meteorological data and historical dynamic current carrying capacity data of the transmission line corridor were collected. Outliers were identified and removed using the Raida criterion. The valid data were then Z-Score standardized to obtain meteorological data including ambient temperature, wind direction, solar radiation intensity and wind speed, as well as current carrying capacity data. S2, Weather-Driven Forecasting Meteorological parameter decomposition: The Empirical Mode Decomposition (EMD) algorithm is used to decompose the dataset of various meteorological parameters. X The time series was decomposed into multiple intrinsic mode function components and one residual component; the meteorological parameters included: ambient temperature ( t a ),wind direction( f ), solar radiation intensity ( J ) and wind speed ( v w The corresponding initial dataset can be represented as: The specific steps of the Empirical Mode Decomposition (EMD) algorithm are as follows: First, find the original signal sequence. x ( tThe system identifies the maxima and minima of the original data; then, using cubic spline interpolation, it connects all the maxima and minima of the original data to form an upper and lower envelope. Finally, it calculates the mean of the upper and lower envelopes. m 1 ( t Finally, calculate the original signal sequence. x ( t ) and the mean of the envelope m 1 ( t If there are negative local maxima and positive local minima between the two values, it means that this is not an eigenmode function and we need to continue "filtering" until the difference between the two is a constant.
[0005] Model parameter optimization: The artificial heron optimization algorithm is adopted to adaptively adjust the penalty factor and kernel parameters of the kernel extreme learning machine, using the inverse of the mean squared error as the fitness function to obtain the optimal meteorological forecast model; the optimized kernel extreme learning machine model is then used to optimize the intrinsic mode function components. and residual components The future values are predicted; the prediction results of each component are superimposed and reconstructed to obtain a prediction dataset of future meteorological parameters. ; This is the predicted ambient temperature value. This is the predicted wind direction value. This is the predicted value of solar radiation intensity. This is the predicted wind speed value; Thermal balance calculation: This involves analyzing a dataset of predicted future meteorological parameters. Substituting into the conductor's thermal balance equation, the first predicted current-carrying capacity is calculated; S3, Data-Driven Prediction Carrying capacity data decomposition: Variational mode decomposition is used to decompose the carrying capacity data into multiple modal components. Dynamic sensing and precise capture optimization algorithms are used to optimize the decomposition parameters. The fitness function is to minimize the sum of the envelope entropy of the decomposed components. The number of decomposition layers and the penalty factor of variational mode decomposition are iteratively optimized to obtain the optimal combination of hyperparameters. Component prediction and reconstruction: The historical dynamic carrying capacity data is decomposed into components using a variational mode decomposition algorithm with the optimal hyperparameter combination. n Each modal component is input into an LSTM model for prediction, and a separate Long Short-Term Memory (LSTM) network model is built for prediction of each modal component. n The prediction results of each modal component are superimposed to reconstruct the second predicted value of the current carrying capacity. S4, Adaptive Weighted Fusion Based on the root mean square error of the meteorological driving model and the data-driven model on the recent validation set, the weight coefficients are dynamically calculated and normalized, and the two predicted values are weighted and fused to obtain the final predicted carrying capacity value. S5, Confidence Interval and Dynamic Capacity Increase Based on the historical prediction error data, the error probability distribution is fitted, and the confidence interval of the future current carrying capacity is calculated by combining the predicted value and the error distribution. The point prediction value is used as the reference setting value for dynamic capacity increase, and the upper limit of the confidence interval is used as the safety limit to realize the dynamic adjustment of the current carrying capacity of the transmission line.
[0006] Furthermore, in the adaptive weighted fusion of step S4: the method for determining the weight coefficients is as follows: calculate the prediction error index of the first prediction model and the second prediction model on the recent validation set respectively; dynamically calculate the weight coefficients based on the prediction error index, and the prediction model with a smaller prediction error index is assigned a higher weight coefficient.
[0007] Furthermore, in step S5, regarding the confidence interval and dynamic scaling: the probability modeling fits a normal or Weiber distribution for the prediction error distribution.
[0008] Furthermore, when the error distribution is a normal distribution, the confidence interval is calculated as follows: using the final predicted value of the carrying capacity point as the mean, combined with the standard deviation, the confidence interval at the preset confidence level is calculated, where is the standard normal distribution quantile at the corresponding confidence level.
[0009] Furthermore, in step S5, the confidence interval and dynamic capacity expansion also include a dynamic capacity expansion application step: the final predicted current carrying capacity point value is used as the reference setting value for dynamic capacity expansion; the upper limit of the confidence interval is used as the safety limit value for dynamic capacity expansion; and the power grid dispatching system dynamically adjusts the current carrying capacity of the transmission line under the constraint of the safety limit value.
[0010] The beneficial effects of this invention are: by integrating meteorological and data-driven models, the accuracy of current carrying capacity prediction is significantly improved; an adaptive weighting mechanism is introduced to enhance the model's adaptability to complex weather and load changes; and confidence interval assessment is combined to improve the safety and reliability of dynamic capacity expansion of transmission lines. Attached Figure Description
[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0012] Figure 1This is a flowchart illustrating the method for predicting and increasing the dynamic current carrying capacity of transmission lines provided in an embodiment of the present invention.
[0013] Figure 2 This diagram illustrates the results of EMD decomposition of a temperature signal in an embodiment of the present invention.
[0014] Figure 3 A comparison chart of the optimization capabilities of the dynamic sensing and precise capture optimization algorithm with other algorithms is provided for embodiments of the present invention.
[0015] Figure 4 This invention provides a comparison chart of the prior error and correlation coefficient of the artificial heron optimization algorithm with three other algorithms for temperature prediction in an embodiment of the invention. Detailed Implementation
[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0017] Figure 1 This is one embodiment provided by the present invention. For example... Figure 1 As shown, a method for predicting and increasing the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion is presented. This method specifically includes: S1, Data Acquisition and Preprocessing.
[0018] Historical meteorological data and historical dynamic current carrying capacity data near the transmission line were collected. First, outliers were initially identified based on the Raida criterion. Combining real-time data trends with historical data from the same period, it was determined whether extreme values in the meteorological and current carrying capacity data were true extreme values. If they were true extreme values, they were retained; if they were false readings, the corresponding outlier data were removed. Then, the data after removing outliers was standardized using the Z-Score standardization method. Finally, the processed meteorological data was obtained.
[0019] The Z-Score standardization formula is as follows: In the formula, x This represents the initial data for any type of meteorological data or carrying capacity. m This represents the mean of any meteorological or carrying capacity data. s The standard deviation of any meteorological or carrying capacity data; x norm Standardized data representing any type of meteorological data or flow rate data.
[0020] The meteorological data obtained above includes: ambient temperature ( t a ),wind direction( f ), solar radiation intensity ( J ) and wind speed ( v w The carrying capacity data obtained above Y .
[0021] S2, weather-driven forecast.
[0022] Empirical mode decomposition is used to decompose the time series of various meteorological parameters into multiple intrinsic mode function components and a residual component.
[0023] The empirical mode decomposition (EMD) process for temperature signals is as follows: First, find the original signal sequence. x ( t The maximum and minimum points of the original signal are identified; then, cubic spline interpolation is used to connect all the maximum and minimum points of the original signal to form the upper and lower envelopes, and finally, the mean of the upper and lower envelopes is calculated. m 1 ( t Finally, the original signal sequence is calculated. x ( t ) and the mean of the envelope m 1 ( t If the difference between the two (i.e., the difference between the two local maxima and local minima) still exists, it indicates that this is not yet an eigenmode function and further "filtering" is needed until the difference between the two is a constant. Original signal sequence x ( t This can be expressed as the sum of the IMF and the remainder, i.e. In the formula: q This represents the T-distribution degrees of freedom parameter. i Indicates the number of algorithm iterations. C i ( t ) represents the modal components of the original signal sequence decomposed. r n ( t The residual components are shown; EMD decomposition is performed on the original ambient temperature data sequence, wind speed data sequence, wind direction data sequence and solar radiation intensity data sequence, and different IMF components are decomposed in turn. The data is decomposed into 5 components with small fluctuations and 1 residual component.
[0024] The penalty factor of the kernel extreme learning machine is optimized using the artificial heron optimization algorithm.K and kernel parameters α Optimization is performed, where the initial value of K is 1.8, and K∈[0.01,10]. α The initial value is 1.2. α ∈[0.01,10], these two parameters are used as inputs to the Artificial Heron algorithm, and the root mean square error of the predictions output by KELM is used as the fitness for cross-validation to obtain the optimal penalty factor. K’ and parameters α’ The specific process is as follows: In the artificial heron optimization algorithm, guided foraging is a global search strategy, while territorial foraging is a local search strategy. In each iteration, the probability of choosing either guided foraging or territorial foraging is 50%. However, this proportional global exploration and local search can easily lead the algorithm into local optima. To address this issue, this invention proposes an adaptive probability threshold. To improve the heron's choice between two search strategies, the probability of selecting each strategy is adjusted by using different individual fitness values. The dynamic threshold adjustment value is as follows: In the formula: In the first t In the nth iteration i The adaptive threshold value for a single heron; 𝑃(𝑋 𝑖,𝑡 ) for the first t The fitness value of the heron at the nth value in the nth iteration; , , Let represent the worst, median, and best fitness values in the t-th iteration, respectively.
[0025] The formula for migratory foraging behavior is as follows: In the formula: Indicates the first t The new position of the heron with the worst fitness in +1 iteration. The weight perturbation factor is a weight value that changes over time (number of iterations). This represents the upper bound of the search space in the t-th iteration. This represents the lower bound of the search space in the t-th iteration. Indicates the current iteration number. This indicates the maximum number of iterations.
[0026] A T-distribution mutation strategy is introduced to perturb individuals in the heron's location update process. The number of iterations is used as the degree of freedom parameter of the T-distribution. The algorithm can enhance global search capabilities and expand the solution space in the early stages using Cauchy mutation; and strengthen local exploitation capabilities in the later stages using Gaussian mutation, achieving a refined search for the neighborhood optimum. The T-distribution mechanism formula is: In the formula: This represents the current position of the i-th heron in the t-th iteration. This indicates that the degree of freedom is n Random numbers distributed according to the T-distribution; In the first t The position of the best heron individual after mutation in the next iteration; q These are the T-distribution degrees of freedom parameters.
[0027] Define a fitness function to evaluate the predictive performance of the model under the current parameters.
[0028] Preferably, the fitness function is the reciprocal of the mean squared error (MSE), as shown in the following formula: In the formula, Indicates fitness. Indicates mean square error; These are actual meteorological values. These are the model's predicted values. n This represents the number of samples in the test set. A higher fitness value indicates higher model prediction accuracy.
[0029] Finally, an optimization iteration is performed, using the initialized penalty factor. K and kernel parameters α Substitute the model into the temporal convolutional network model, train the model using the training set, and then calculate the initial fitness using the test set. For each newly generated candidate solution of parameters ( K’ , α’ Substitute the values into the model to calculate the fitness. Fit .like Fit If the fitness is greater than the current optimal fitness, then update the optimal parameters. K , α ) and corresponding fitness Fit Substitute the optimal parameters ( K , α The optimal weather forecast model is obtained by incorporating the weather forecast model into the weather forecast model.
[0030] The processed raw meteorological data includes ambient temperature. t a ),wind direction( f ), solar radiation intensity ( J) and wind speed ( v w The components are fed into the optimal weather forecast model, and the forecast results of each component are output. These components are then reconstructed to obtain the final weather forecast result.
[0031] Substituting the predicted meteorological parameters into the heat balance equation, the first predicted carrying capacity value is calculated.
[0032] When a conductor is in thermal equilibrium, its heat absorption power is equal to its heat release power, as shown in the following equation: In the formula: W S This refers to the solar heat absorption power. W R This refers to the power of radiative heat dissipation. W F This refers to the convective heat dissipation power; T C For the temperature of the conductor; R ( T C ) represents temperature T C AC resistance of the conductor; I This refers to the current carrying capacity of the conductor.
[0033] 1) Radiative heat dissipation formula W R for: In the formula: D The diameter of the wire; e The radiation heat dissipation coefficient of the conductor is taken in the range of [0.9, 0.95]. , These refer to the ambient temperature and the maximum allowable temperature of the conductor, respectively.
[0034] 2) Convection heat dissipation power W F for: In the formula: v w Wind speed; r air density; m The dynamic viscosity of air; k f The thermal conductivity of air; k a This is the wind direction factor.
[0035] 3) Solar heat absorption power W S for: In the formula: α The heat absorption coefficient of the conductor surface is taken as 0.35 to 0.46 for bright new wire. J The intensity of sunlight on the conductor.
[0036] 4) The AC resistance value is: In the formula: R ( T C () indicates the AC resistance value. T C For the temperature of the conductor, T low As a low temperature reference point, R ( T high ) is the temperature T high AC resistance of the conductor at that time; R ( T low ) is the temperature T low The AC resistance of the conductor at that time. From empirical values, when selecting... T high It is 75℃. T low The error between the resistance value of the conductor at 25℃ and the actual resistance value is negligible.
[0037] By substituting the predicted meteorological parameters into the heat balance equation, the predicted value of the first carrying capacity is calculated. I 1 .
[0038] S3, data-driven prediction.
[0039] Traffic data Y The training set and the test set are divided in an 8:2 ratio. The training set is used to build the model, and the test set is used to verify the accuracy of the model.
[0040] An optimization algorithm combining dynamic sensing and precise capture is adopted, and the fitness function is selected as the sum of the envelope entropies of the components after VMD decomposition, thereby optimizing the number of decomposition layers and the penalty factor.
[0041] First, a certain number of VMD hyperparameters are randomly initialized. P and β And calculate the fitness value. The mathematical expression for envelope entropy is as follows: In the formula: E p Envelope entropy representing a DLR data component; Pj represent a j The normalized form; a j The envelope sequence obtained by Hilbert mediation of DLR data components; m The signal length; x ( j ) represents the original signal at point j The value of H[] represents the Hilbert transform.
[0042] Next, we introduce an optimized algorithm for dynamic sensing and precise capture, as shown below: Initialize the number of individuals in the algorithm: In the formula, x ij It is the first i The candidate solution specified by the first j The values of the variables, N It is the total number of members. m It is the signal length. rand It is a random number in the interval [0,1]. l j It is the first j A lower bound, u j It is the first of the problem variables j The upper limit.
[0043] After initialization, the algorithm performs an extensive search within the solution space to ensure that no potentially good regions are overlooked. Its expression is as follows: in Based on the first phase i The new state of an individual in the x-th dimension I It is a random number equal to 1 or 2. p j The facilitator is the individual in the first j The position in the middle, F p It is the value of its objective function. For the first i The individual in the first j Current position on the dimension Represents the fitness value of the i-th individual, parameter I It is randomly selected for each iteration and for each individual, when I When the value is 2, it will result in more displacement for the individual, which may guide the individual to a new region of the search space. Therefore, the parameter... IThe value of will affect the algorithm's exploration ability.
[0044] The individual's position is updated according to dynamically perceived rules. This typically involves considering the individual's current position, its historical best position, and information about other individuals in the group. It is crucial to ensure that the updated position remains within the defined search space. If an individual's position exceeds the search space boundary, it needs to be moved back within the boundary to prevent the algorithm from moving to a suboptimal region. The expression is as follows: In the formula: For the updated version i The location of each individual For the generated candidate new positions, For the candidate position The corresponding fitness value, For the candidate position The corresponding fitness value, For the current number i After the dynamic perception phase, the algorithm enters the precise capture phase, where it focuses on refining and improving the discovered potential solutions. The algorithm performs a more concentrated and detailed search around the known good solutions, aiming to find the optimal solution within the known good regions. As the algorithm progresses, the random component in the search is reduced to more accurately approximate the optimal solution. Its expression is as follows: In the formula: rand is a random number uniformly distributed in the range [0,1]. For the first i Individuals in the second phase j A new state on the dimension; R It is a constant. R The value is 0.2. R ∙(1− t / T ) for x i,j Neighborhood radius, t This represents the current iteration number. T This represents the maximum number of iterations.
[0045] If an individual's fitness at a new location is better than its historical best fitness, then its historical best location is updated according to the following formula. In the formula: For the first i Individuals in the second phase j The new state on the dimension, In order to be in The fitness value of the location; output the optimal hyperparameters after the iteration termination condition is met. P’ and β’ .
[0046] Next, the VMD algorithm is used to decompose the DLR data into... n These are components with different center frequencies but local stationary states. To avoid aliasing among the DLR components, a constraint equation is constructed with the objective of minimizing the sum of the bandwidths of each component, as shown below: In the formula: It is a set of modal functions; Let A be the set of center frequencies; min{A} and stB are the conditions that minimize A when B is satisfied. h t It is a partial derivative operator; (t) Here, j is the Dirac function; j is the imaginary unit. Pi; * represents the convolution operator; u k ( t ) is the first generation of DLR data decomposition k One component; w k ( t ) is the first k The frequency corresponding to each component; f ( t () represents undecomposed DLR data.
[0047] Subsequently, the alternating direction multiplier method is used for iterative updates. u k , w k , w k The optimal solution to the constraint variational equation is found, thereby completing the DLR data decomposition and obtaining the staged solution. n Each component.
[0048] Next, in sequence n Each component is fed into the LSTM model for prediction, yielding the component prediction value. The internal update process of the LSTM model can be represented as: In the formula: This is the hidden state vector from the previous time step; This is the cell state vector for the next time step; s Use the Sigmoid activation function; W f , W I ,W C , W O and b f , b I , b C , b O Forgotten Gate f t Input gate I t Memory cells C t and output gate O t The weight matrix and bias terms; X t Enter the current time; h t-1 The state was hidden in the previous moment; s Here, is the Sigmoi activation function; tanh is the activation function. Candidate state for memory cells; For Hada code multiplication.
[0049] Next, regarding n The predicted values of each component are superimposed and reconstructed to obtain the second predicted value of the carrying capacity. I 2 .
[0050] S4. Based on the prediction error index of the first prediction model and the second prediction model on the recent validation set, the weight coefficients are dynamically calculated and adaptive weighted fusion is performed to obtain the final predicted value of the carrying capacity point.
[0051] The root mean square error is used as the prediction error index, as shown in the following formula: In the formula: MSE is the root mean square. This represents the actual carrying capacity. This is the predicted carrying capacity value. n This represents the number of samples in the test set.
[0052] Input the first predicted carrying capacity value I 1 Second Carrier Flow Prediction I 2 The output yields the root mean square error of both. m 1 , m 2 .
[0053] This serves as the base weight value for each model: In the formula: As the base value for weights, The root mean square error is used to obtain the basic weight values. w 1 , w 2 This is then normalized so that the sum of the weights equals 1, yielding the final weights. w’ 1 , w’ 2 The final predicted carrying capacity is calculated based on the final weights. I e .
[0054] S5, confidence interval and dynamic capacity expansion.
[0055] Based on historical prediction error data, a probability distribution is fitted. Using the predicted point value as the center, and combining the standard deviation and confidence level of the error distribution, the confidence interval for future air carrying capacity is calculated. Among the meteorological elements, temperature and solar radiation intensity follow a normal distribution, while wind speed follows a Weibull distribution, as shown in the following formula: In the formula: To indicate when temperature x The probability density at time , The solar radiation intensity is x The probability density at time , For wind speed x The probability density at time t, the parameters can be obtained by fitting. m , s , c , d , a , b The value of .
[0056] Finally, after determining the conductor type, and combining the heat balance equation, a current-carrying probability model based on the probability density function can be established under the known ambient temperature, solar radiation intensity, and wind speed distribution functions. This allows for the calculation of the probability density distribution of the current-carrying capacity over a 24-hour period, i.e., the confidence interval. The final predicted value is used as a reference setpoint for dynamic capacity expansion, and the upper limit of the confidence interval is used as a safety limit, enabling dynamic adjustment of the transmission line's current-carrying capacity.
[0057] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for predicting and increasing the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion, characterized in that... Includes the following steps: S1, Data Acquisition and Preprocessing Historical meteorological data and historical dynamic current carrying capacity data of the transmission line corridor were collected. Outliers were identified and removed using the Raida criterion. The valid data were then Z-Score standardized to obtain meteorological data including ambient temperature, wind direction, solar radiation intensity and wind speed, as well as current carrying capacity data. S2, Weather-Driven Forecasting Meteorological parameter decomposition: The empirical mode decomposition algorithm is used to decompose the dataset of various meteorological parameters. X The time series was decomposed into multiple intrinsic mode function components and one residual component; the meteorological parameters included: ambient temperature ( t a ),wind direction( φ ), solar radiation intensity ( J ) and wind speed ( v w The corresponding initial dataset can be represented as: The specific steps of the Empirical Mode Decomposition (EMD) algorithm are as follows: First, find the original signal sequence. x ( t The system identifies the maxima and minima of the original data; then, using cubic spline interpolation, it connects all the maxima and minima of the original data to form an upper and lower envelope. Finally, it calculates the mean of the upper and lower envelopes. m 1 ( t Finally, calculate the original signal sequence. x ( t ) and the mean of the envelope m 1 ( t If the difference between the two functions still has negative local maxima and positive local minima, it means that this is not an eigenmode function and we need to continue "filtering" until the difference between the two functions is a constant. Model parameter optimization: The artificial heron optimization algorithm is adopted to adaptively adjust the penalty factor and kernel parameters of the kernel extreme learning machine, using the inverse of the mean squared error as the fitness function to obtain the optimal meteorological forecast model; the optimized kernel extreme learning machine model is then used to optimize the intrinsic mode function components. and residual components The future values are predicted; the prediction results of each component are superimposed and reconstructed to obtain a prediction dataset of future meteorological parameters. ; This is the predicted ambient temperature value. This is the predicted wind direction value. This is the predicted value of solar radiation intensity. This is the predicted wind speed value; Thermal balance calculation: This involves analyzing a dataset of predicted future meteorological parameters. Substituting into the conductor's thermal balance equation, the first predicted current-carrying capacity is calculated; S3, Data-Driven Prediction Carrying capacity data decomposition: Variational mode decomposition is used to decompose the carrying capacity data into multiple modal components. Dynamic sensing and precise capture optimization algorithms are used to optimize the decomposition parameters. The fitness function is to minimize the sum of the envelope entropy of the decomposed components. The number of decomposition layers and the penalty factor of variational mode decomposition are iteratively optimized to obtain the optimal combination of hyperparameters. Component prediction and reconstruction: The historical dynamic carrying capacity data is decomposed into components using a variational mode decomposition algorithm with the optimal hyperparameter combination. n Each modal component is input into an LSTM model for prediction, and a separate Long Short-Term Memory (LSTM) network model is built for prediction of each modal component. n The prediction results of each modal component are superimposed to reconstruct the second predicted value of the current carrying capacity. S4, Adaptive Weighted Fusion Based on the root mean square error of the meteorological driving model and the data-driven model on the recent validation set, the weight coefficients are dynamically calculated and normalized, and the two predicted values are weighted and fused to obtain the final predicted carrying capacity value. S5, Confidence Interval and Dynamic Capacity Increase Based on the historical prediction error data, the error probability distribution is fitted, and the confidence interval of the future current carrying capacity is calculated by combining the predicted value and the error distribution. The point prediction value is used as the reference setting value for dynamic capacity increase, and the upper limit of the confidence interval is used as the safety limit to realize the dynamic adjustment of the current carrying capacity of the transmission line.
2. The method for predicting and increasing the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion according to claim 1, characterized in that, In the adaptive weighted fusion of step S4: the weight coefficients are determined by calculating the prediction error indices of the first prediction model and the second prediction model on the recent validation set; and dynamically calculating the weight coefficients based on the prediction error indices. The prediction model with the smaller prediction error index is assigned a higher weight coefficient.
3. The method for predicting and increasing the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion according to claim 1, characterized in that, In step S5, during the confidence interval and dynamic scaling: the probability modeling fits a normal or Weiber distribution for the prediction error distribution.
4. The method for predicting and increasing the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion according to claim 3, characterized in that, When the error distribution is a normal distribution, the confidence interval is calculated as follows: using the final predicted value of the carrying capacity point as the mean, combined with the standard deviation, the confidence interval at the preset confidence level is calculated, where is the standard normal distribution quantile at the corresponding confidence level.
5. The method for predicting and increasing the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion according to claim 1, characterized in that, In step S5, the confidence interval and dynamic capacity expansion also include a dynamic capacity expansion application step: using the final predicted value of the carrying capacity point as the reference setting value for dynamic capacity expansion; and using the upper limit of the confidence interval as the safety limit value for dynamic capacity expansion. The power grid dispatching system dynamically adjusts the current-carrying capacity of transmission lines under the constraints of the aforementioned safety limits.
6. The method for predicting and increasing the dynamic current carrying capacity of transmission lines based on dual-model adaptive fusion according to claim 1, characterized in that, Specifically, it includes: S1, Data Acquisition and Preprocessing; Historical meteorological data and historical dynamic current carrying capacity data near the transmission line are collected. First, outliers are preliminarily identified based on the Raida criterion. Combining real-time data trends with historical data from the same period, it is determined whether the extreme values in the meteorological data and current carrying capacity data are true extreme values. If they are true extreme values, they are retained. If the data is a false positive, the corresponding outlier data is removed; then, the data after removing the outlier data is standardized using the Z-Score standardization method; finally, the processed meteorological data is obtained. The Z-Score standardization formula is as follows: In the formula, x This represents the initial data for any type of meteorological data or carrying capacity. μ This represents the mean of any meteorological or carrying capacity data. σ The standard deviation of any meteorological or carrying capacity data; x norm Standardized data representing any type of meteorological data or flow rate data; The meteorological data obtained above includes: ambient temperature ( t a ),wind direction( φ ), solar radiation intensity ( J ) and wind speed ( v w ); the carrying capacity data obtained above Y ; S2, weather-driven forecast; Empirical mode decomposition is used to decompose the time series of various meteorological parameters into multiple intrinsic mode function components and a residual component. The empirical mode decomposition (EMD) process for temperature signals is as follows: First, find the original signal sequence. x ( t The maximum and minimum points of the original signal are identified; then, cubic spline interpolation is used to connect all the maximum and minimum points of the original signal to form the upper and lower envelopes, and finally, the mean of the upper and lower envelopes is calculated. m 1 ( t Finally, the original signal sequence is calculated. x ( t ) and the mean of the envelope m 1 ( t If the difference between the two (i.e., the difference between the two) still has negative local maxima and positive local minima, it means that this is not yet an eigenmode function and needs to be further "filtered" until the difference between the two is a constant; original signal sequence x ( t This can be expressed as the sum of the IMF and the remainder, i.e. In the formula: q This represents the T-distribution degrees of freedom parameter. i Indicates the number of algorithm iterations. C i ( t ) represents the modal components of the original signal sequence decomposed. r n ( t The residual components are shown; EMD decomposition is performed on the original ambient temperature data sequence, wind speed data sequence, wind direction data sequence and solar radiation intensity data sequence, and different IMF components are decomposed in turn. The data is decomposed into 5 components with small fluctuations and 1 residual component. The penalty factor of the kernel extreme learning machine is optimized using the artificial heron optimization algorithm. K and kernel parameters α Optimization is performed, where the initial value of K is 1.8, and K∈[0.01,10]. α The initial value is 1.
2. α ∈[0.01,10], these two parameters are used as inputs to the Artificial Heron algorithm, and the root mean square error of the predictions output by KELM is used as the fitness for cross-validation to obtain the optimal penalty factor. K’ and parameters α’ The specific process is as follows: In the artificial heron optimization algorithm, guided foraging is a global search strategy, while territorial foraging is a local search strategy. In each iteration, the probability of choosing either guided foraging or territorial foraging is 50%. However, the proportional global exploration and local search can easily lead the algorithm into local optima. To address this issue, an adaptive probability threshold is proposed. To improve the heron's choice between two search strategies, the selection probability of the two search strategies is adjusted by using different individual fitness values; the dynamic threshold adjustment value is as follows: In the formula: In the first t In the nth iteration i The adaptive threshold value for a single heron; 𝑃(𝑋 𝑖,𝑡 ) for the first t The fitness value of the heron at the nth value in the nth iteration; , , Let represent the worst, median, and best fitness values in the t-th iteration, respectively; The formula for migratory foraging behavior is as follows: In the formula: Indicates the first t The new position of the heron with the worst fitness in +1 iteration. The weight perturbation factor is a weight value that changes over time (number of iterations). This represents the upper bound of the search space in the t-th iteration. This represents the lower bound of the search space in the t-th iteration. Indicates the current iteration number. Indicates the maximum number of iterations; A T-distribution mutation strategy is introduced to perturb individuals in the heron's location update process. The number of iterations is used as the degree of freedom parameter of the T-distribution. The algorithm can enhance global search capabilities and expand the solution space in the early stages using Cauchy mutation; and strengthen local exploitation capabilities in the later stages using Gaussian mutation, achieving a refined search for the neighborhood optimum. The T-distribution mechanism formula is: In the formula: This represents the current position of the i-th heron in the t-th iteration. This indicates that the degree of freedom is n Random numbers distributed according to the T-distribution; In the first t The position of the best heron individual after mutation in the next iteration; q These are the T-distribution degrees of freedom parameters; Define a fitness function to evaluate the predictive performance of the model under the current parameters; Preferably, the fitness function is the reciprocal of the mean squared error (MSE), as shown in the following formula: In the formula, Indicates fitness. Indicates mean square error; These are actual meteorological values. These are the model's predicted values. n This represents the number of samples in the test set; a higher fitness value indicates higher model prediction accuracy. Finally, an optimization iteration is performed, using the initialized penalty factor. K and kernel parameters α Substitute the model into the temporal convolutional network model, train the model using the training set, and then calculate the initial fitness using the test set; for each newly generated candidate solution of parameters ( K’ , α’ Substitute the values into the model to calculate the fitness. Fit ;like Fit If the fitness is greater than the current optimal fitness, then update the optimal parameters. K , α ) and corresponding fitness Fit Substitute the optimal parameters ( K , α The optimal weather forecast model is obtained by incorporating the data into the weather forecast model. The processed raw meteorological data includes ambient temperature. t a ),wind direction( φ ), solar radiation intensity ( J ) and wind speed ( v w The components are fed into the optimal weather forecast model, and the forecast results of each component are output. These components are then reconstructed to obtain the final weather forecast result. Substituting the predicted meteorological parameters into the heat balance equation, the first predicted value of the carrying capacity is calculated. When a conductor is in thermal equilibrium, its heat absorption power is equal to its heat release power, as shown in the following equation: In the formula: W S This refers to the solar heat absorption power. W R This refers to the power of radiative heat dissipation. W F This refers to the convective heat dissipation power; T C For the temperature of the conductor; R ( T C ) represents temperature T C AC resistance of the conductor; I Current carrying capacity of the conductor; 1) Radiative heat dissipation formula W R for: In the formula: D The diameter of the wire; ε The radiation heat dissipation coefficient of the conductor is taken in the range of [0.9, 0.95]. , These are the ambient temperature and the maximum allowable temperature of the conductor, respectively. 2) Convection heat dissipation power W F for: In the formula: v w Wind speed; ρ air density; μ The dynamic viscosity of air; k f The thermal conductivity of air; k a Wind direction factor; 3) Solar heat absorption power W S for: In the formula: α The heat absorption coefficient of the conductor surface is taken as 0.35 to 0.46 for bright new wire. J The intensity of sunlight on the conductor; 4) The AC resistance value is: In the formula: R ( T C () indicates the AC resistance value. T C For the temperature of the conductor, T low As a low temperature reference point, R ( T high ) is the temperature T high AC resistance of the conductor at that time; R ( T low ) is the temperature T low The AC resistance of the conductor at that time; from empirical values, when selecting T high It is 75℃. T low The error between the resistance value of the conductor at 25℃ and the actual resistance value is negligible; By substituting the predicted meteorological parameters into the heat balance equation, the predicted value of the first carrying capacity is calculated. I 1 ; S3, data-driven forecasting; Traffic data Y The training set and the test set are divided in an 8:2 ratio. The training set is used to build the model, and the test set is used to verify the accuracy of the model. An optimization algorithm combining dynamic sensing and precise capture is adopted, and the fitness function is selected as the sum of the envelope entropy of the components after VMD decomposition, thereby optimizing the number of decomposition layers and the penalty factor. First, a certain number of VMD hyperparameters are randomly initialized. P and β And calculate the fitness value; the mathematical expression for the envelope entropy is as follows: In the formula: E p Envelope entropy representing a DLR data component; P j represent a j The normalized form; a j The envelope sequence obtained by Hilbert mediation of DLR data components; m The signal length; x ( j ) represents the original signal at point j The value of H[] represents the Hilbert transform; Next, we introduce an optimized algorithm for dynamic sensing and precise capture, as shown below: Initialize the number of individuals in the algorithm: In the formula, x ij It is the first i The candidate solution specified by the first j The values of the variables, N It is the total number of members. m It is the signal length. rand It is a random number in the interval [0,1]. l j It is the first j A lower bound, u j It is the first of the problem variables j An upper bound; After initialization, the algorithm performs an extensive search within the solution space to ensure that no potentially good regions are overlooked; its expression is as follows: in Based on the first phase i The new state of an individual in the x-th dimension I It is a random number equal to 1 or 2. p j The facilitator is the individual in the first j The position in the middle, F p It is its objective function value; For the first i The individual in the first j Current position on the dimension Represents the fitness value of the i-th individual, parameter I It is randomly selected for each iteration and for each individual, when I When the value is 2, it will result in more displacement for the individual, which may guide the individual to a new region of the search space; therefore, the parameter I The value of will affect the algorithm's exploration ability; The individual's position is updated according to dynamically perceived rules; this typically involves considering the individual's current position, its historical best position, and information about other individuals in the group; ensuring that the updated position remains within the defined search space; if an individual's position exceeds the boundary of the search space, it needs to be placed back within the boundary to prevent the algorithm from moving to a non-optimal region; its expression is as follows: In the formula: For the updated version i The location of each individual For the generated candidate new positions, For the candidate position The corresponding fitness value, For the candidate position The corresponding fitness value, For the current number i The algorithm locates the individual solutions; after the dynamic perception phase, it enters the precise capture phase. In this phase, the algorithm focuses more on refining and improving the discovered potential solutions. The algorithm performs a more concentrated and detailed search around the discovered good solutions, focusing on finding the optimal solution within the known good regions. As the algorithm progresses, the random component in the search is reduced to more accurately approximate the optimal solution; its expression is as follows: In the formula: rand is a random number uniformly distributed in the range [0,1]. For the first i Individuals in the second phase j A new state on the dimension; R It is a constant. R The value is 0.
2. R ∙(1− t / T ) for x i,j Neighborhood radius, t This represents the current iteration number. T This represents the maximum number of iterations. If an individual's fitness at a new location is better than its historical best fitness, then its historical best location is updated according to the following formula; In the formula: For the first i Individuals in the second phase j The new state on the dimension, In order to be in The fitness value of the location; output the optimal hyperparameters after the iteration termination condition is met. P’ and β’ ; Next, the VMD algorithm is used to decompose the DLR data into... n There are several components with different center frequencies but local stability. To avoid aliasing among the DLR components, a constraint equation is constructed with the goal of minimizing the sum of the bandwidths of each component, as shown below: In the formula: It is a set of modal functions; Let A be the set of center frequencies; min{A} and stB are the conditions that minimize A when B is satisfied. h t It is a partial derivative operator; (t) Here, j is the Dirac function; j is the imaginary unit. Pi; * represents the convolution operator; u k ( t ) is the first generation of DLR data decomposition k One component; w k ( t ) is the first k The frequency corresponding to each component; f ( t () represents undecomposed DLR data; Subsequently, the alternating direction multiplier method is used for iterative updates. u k , w k , w k The optimal solution to the constraint variational equation is found, thereby completing the DLR data decomposition and obtaining the staged solution. n One component; Next, in sequence n Each component is fed into the LSTM model for prediction, yielding the component prediction values; the internal update process of the LSTM model can be represented as: In the formula: This is the hidden state vector from the previous time step; This is the cell state vector for the next time step; σ Use the Sigmoid activation function; W f , W I , W C , W O and b f , b I , b C , b O Forgotten Gate f t Input gate I t Memory cells C t and output gate O t The weight matrix and bias terms; X t Enter the current time; h t-1 The state was hidden in the previous moment; σ Here, is the Sigmoi activation function; tanh is the activation function. Candidate state for memory cells; For Hada code multiplication; Next, regarding n The predicted values of each component are superimposed and reconstructed to obtain the second predicted value of the carrying capacity. I 2 ; S4. Based on the prediction error index of the first prediction model and the second prediction model on the recent validation set, dynamically calculate the weight coefficients, perform adaptive weighted fusion, and obtain the final predicted value of the carrying capacity point. The root mean square error is used as the prediction error index, as shown in the following formula: In the formula: MSE is the root mean square. This represents the actual carrying capacity. This is the predicted carrying capacity value. n This represents the number of samples in the test set. Input the first predicted carrying capacity value I 1 Second Carrier Flow Prediction I 2 The output yields the root mean square error of both. m 1 , m 2 ; This serves as the base weight value for each model: In the formula: As the base value for weights, The root mean square error is used to obtain the basic weight values. w 1 , w 2 This is then normalized so that the sum of the weights equals 1, yielding the final weights. w’ 1 , w’ 2 The final predicted carrying capacity is calculated based on the final weights. I e ; S5, confidence interval and dynamic capacity expansion; Based on historical prediction error data, its probability distribution is fitted; centered on the predicted point value, and combining the standard deviation and confidence level of the error distribution, the confidence interval for future carrying capacity is calculated; among meteorological elements, temperature and solar radiation intensity follow a normal distribution, and wind speed follows a Weibull distribution, as shown in the following formula: In the formula: To indicate when temperature x The probability density at time , The solar radiation intensity is x The probability density at time , For wind speed x The probability density at time t, the parameters can be obtained by fitting. μ , σ , γ , δ , a , b The value of .
Citation Information
Cited By
An intelligent exhibition hall temperature central control system based on an internet of things
CN122346212A
