Self-adaptive portable power supply control method for power transformation hot-line work

Abstract

The invention relates to a power transformation hot-line work adaptive portable power supply control method, which comprises the steps of collecting power supply demand data under various working conditions in a hot-line work process through a data acquisition system, constructing a random forest model, training the random forest model by using the data, and generating an initial power supply control scheme. A self-adaptive ant lion optimization algorithm is adopted to finely adjust parameters of a random forest model, the optimized random forest model is applied to dynamic adjustment and optimization of a power supply control scheme, and by monitoring operation conditions in real time and adaptively adjusting power supply output, it is ensured that optimal power supply support can be provided under various working conditions; according to the method, historical data and real-time monitoring data are combined, and the hyper-parameters of the random forest model are dynamically adjusted by using the adaptive ant lion optimization algorithm, so that the power supply control scheme in the hot-line work is optimized, and the safety and efficiency of the hot-line work are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic power management, and particularly to an adaptive portable power control method for live working on power transformation equipment. Background Art

[0002] In live working, the lack of an effective optimization scheme for portable power control is a common problem. At present, the commonly used power control methods mainly rely on the experience of operators for manual adjustment. This method is not only inefficient but also difficult to ensure the stability and safety of power supply. In addition, many live working scenarios require the use of a fixed power supply scheme throughout the process. This single power management method cannot adapt to various dynamic changes that may occur during the operation, such as power demand fluctuations and environmental condition changes. This results in low power usage efficiency, and sometimes even over-consumption or insufficient supply, increasing the operation risk. Therefore, it is necessary to develop a more intelligent and adaptable power control optimization scheme. Summary of the Invention

[0003] The present invention provides an adaptive portable power control method for live working on power transformation equipment. By combining historical data and real-time monitoring data, the hyperparameters of the random forest model are dynamically adjusted using the adaptive ant lion optimization algorithm, thereby optimizing the power control scheme in live working and improving the safety and efficiency of live working.

[0004] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0005] An adaptive portable power control method for live working on power transformation equipment includes the following steps:

[0006] S1. Collect power demand data under various working conditions during live working through a data acquisition system, and construct a random forest model;

[0007] S2. Use this data to train the random forest model to generate an initial power control scheme;

[0008] S3. Use the adaptive ant lion optimization algorithm to finely adjust the parameters of the random forest model;

[0009] S4. Apply the optimized random forest model to the dynamic adjustment and optimization of the power control scheme. By real-time monitoring the operation situation, adaptively adjust the power output to ensure that the best power support can be provided under various working conditions.

[0010] Further, the power demand data includes the power, usage frequency, and power temperature data under various working conditions during historical live working.

[0011] Further, the construction of the random forest model includes the following steps:

[0012] S1.1. Divide the standardized dataset D' into a training set D train and a validation set D val ;

[0013] S1.2. Build a random forest model to predict future power demands: Let the random forest model be RF(X; Θ), where X is the input feature and the initial hyperparameter set Θ = {M, d, s}. The specific parameter settings are as follows:

[0014] M: The number of decision trees;

[0015] d: The maximum number of features, which refers to the maximum number of features considered when looking for the best split;

[0016] s: The minimum number of samples in each leaf node;

[0017] S1.3. Randomly generate an initial ant colony X: X = {x1, x2, …, x N}), where each x i is a combination of hyperparameters of a random forest;

[0018] S1.4. Initialize the initial pits T: T = {t1, t2, …, t N}), where t i is the position of the i-th pit, and the initial value of each t i is the same as the corresponding x i ;

[0019] S1.5. Calculate the fitness value f(x i ) of each ant to evaluate the performance of each ant. The fitness value can be measured by the prediction effect on the validation set. For example, use the mean squared error MSE:

[0020]

[0021] where m is the number of samples in the validation set,

[0022] X j is the j-th sample in the validation set and is a subset of the input feature set in the validation set;

[0023] Y j is the actual power demand.

[0024] Furthermore, step S3 specifically includes the following steps:

[0025] S3.1. Build pits:

[0026] Attract the better ants to the pit positions according to the fitness value f(x i ); If f(ti ) > f(x i ), then update the pothole position t i , where f(t i ) is the fitness value corresponding to the i-th pothole position:

[0027] t i = x i (2)

[0028] Use a trap to capture ants; for each ant x i with the included position vector, update its position to a random point A(trap) near the pothole t i :

[0029] x i (trap) = t i + rand(-1, 1) × (A(trap) - x i ) (3)

[0030] where A(trap) is a randomly generated point within the foraging range;

[0031] S3.2. Ant movement:

[0032] Randomly move the ants; for each ant x i , update its position to a new random point x i (move):

[0033] x i (move) = x i + rand(-1, 1) × (sign(rand() - 0.5) × (Δ(t) × (A(move) - x i ) (4)

[0034] where A(move) and A(trap) are randomly generated points related to the current ant position, rand() represents a random number generated in the range [0, 1), the initial step size Δ(0) and the adaptive adjustment formula are:

[0035]

[0036] where exp is the exponential function;

[0037] t is the current iteration number;

[0038] Δ(T) is the adaptive adjustment step size, representing the step size of the ant movement at time step t;

[0039] T max is the maximum number of iterations;

[0040] S3.3. Fitness Update and Probability Adaptation:

[0041] Calculate the updated fitness value f(x i (move));

[0042] Introduce the adaptive exploration probability P a (t):

[0043]

[0044] Determine whether the ant falls into the pit according to the fitness value and the adaptive exploration probability:

[0045]

[0046] random() represents generating a random number between [0, 1);

[0047] S3.4. Update the population

[0048] Update the pit position t according to the newly found ant position i :

[0049] t i = x i (next) (8)

[0050] Reconstruct the ant colony: If random() > P a (t), then the ant position x i is updated to x i (move):

[0051] x i = x i (move) (9)

[0052] S3.5. Random Forest Model Training and Evaluation:

[0053] Use the hyperparameter combination Θ corresponding to each ant (solution) i to train the random forest model RF(X; Θ i );

[0054] Evaluate the performance of each random forest model on the validation set D val and calculate its mean squared error MSE;

[0055] S3.6. Iterative Optimization: Repeat steps S3.2 to S3.6 until the predetermined number of iterations T max or the fitness value reaches a satisfactory accuracy;

[0056] S3.7. Select the Best Solution: Among the fitness values f(x of all ants i)Select the hyperparameter combination Θ with the minimum MSE best as the optimal solution.

[0057] A method for adaptive portable power control in live electrical substation work according to claim 1, characterized in that the input of the optimized random forest model is real-time monitoring data, the output is the best future power supply plan, and corresponding power regulation is carried out.

[0058] Compared with the prior art, the beneficial effects of the present invention are:

[0059] 1) The parameters of the random forest model are finely adjusted by using the adaptive ant lion optimization algorithm to improve the prediction accuracy and stability of the model;

[0060] 2) The optimized random forest model is applied to the dynamic adjustment and optimization of the power control scheme. By real-time monitoring the operation situation, the power output is adaptively adjusted to ensure that the best power support can be provided under various working conditions;

[0061] 3) Combining historical data and real-time monitoring data, the hyperparameters of the random forest model are dynamically adjusted by using the adaptive ant lion optimization algorithm, thereby optimizing the power control scheme in live electrical work and improving the safety and efficiency of live electrical work. Description of the Drawings

[0062] Figure 1 is the flowchart of the random forest of the present invention.

[0063] Figure 2 is the flowchart of the power control method for optimizing the random forest based on the adaptive ant lion optimization algorithm of the present invention. Detailed Embodiments

[0064] The following further describes the specific embodiments of the present invention with reference to the drawings:

[0065] An adaptive portable power control method for live working in a substation. First, a data acquisition system is used to collect power demand data under various working conditions during live working, including factors such as power magnitude, power usage frequency, and power temperature. Each set of data corresponds to an optimal power supply scheme. For example, a 12V power supply is composed of four 6V batteries. Two groups of batteries are connected in parallel and then in series to form a 12V power supply. Then, there are power supply schemes such as: all four batteries operating simultaneously, two 6V batteries connected in series operating, and two in parallel plus one in series operating simultaneously. Then, these data are used to train a random forest model to generate an initial power control scheme. Next, an adaptive ant lion optimization algorithm is used to finely adjust the parameters of the random forest model to improve the prediction accuracy and stability of the model. Subsequently, the optimized random forest model is applied to the dynamic adjustment and optimization of the power control scheme. By real-time monitoring the working conditions, the power output is adaptively adjusted to ensure the best power support under various working conditions.

[0066] (1) Ant Lion Optimizer (ALO)

[0067] The core idea of the ALO algorithm is to simulate the hunting mechanism of ant lions preying on ants to achieve global optimization. Before hunting, ant lions dig a funnel-shaped trap in sandy soil using their large jaws and hide at the bottom of the trap waiting for prey. Once a randomly wandering ant falls into the trap, the ant lion quickly preys on it and then repairs the trap to wait for the next hunt. The ALO algorithm optimizes the problem by numerically simulating the interaction between ants and ant lions: introducing the random wandering of ants to achieve global search, ensuring the diversity of the population and the optimization performance of the algorithm through the roulette wheel strategy and the elite strategy. The ant lion is equivalent to the solution of the optimization problem, and the approximate optimal solution is updated and saved by hunting high-fitness ants. The main components include ant lions (optimizers), ants (solutions), and pits (the positions where ant lions prey). The working principle is as follows:

[0068] 1. Initialization

[0069] 1.1 Randomly generate an initial ant colony X: X = {x1, x2, …, x N}

[0070] 1.2 Initial pits T: T = {t1, t2, …, t N}

[0071] 1.3 Fitness value f(x): Evaluate the performance of each ant.

[0072] 2. Construct pits

[0073] 2.1 Attract the better ants to the pit positions according to the fitness values.

[0074] 2.2 Update the location of the pothole: t i =x i iff(t i )>f(x i )

[0075] 2.3 Use traps to capture ants: x i (trap)=t i +rand()×(A(trap)-x i )

[0076] Among them, A(trap) is a randomly generated point in the exploration area.

[0077] 3. Ant Movement

[0078] 3.1 Randomly moving ants: x i (move) = x i +rand(-1,1)×(|A(move)-x i |+|A(trap)-x i |)

[0079] Among them, A(move) and A(trap) are randomly generated points related to the current ant position.

[0080] 3.2 Determine whether the ant falls into the pit: x i (next) = x i (trap)if random() <P a

[0081] Among them, P a is the probability controlled by the fitness function f(x).

[0082] 4. Update population

[0083] 4.1 Update the hole position according to the newly found ant position: t i =x i (next)

[0084] 4.2 Reconstructing the ant colony: x i =x i (move)if random()>P a

[0085] 5. Termination Conditions

[0086] Reach the predetermined number of iterations TmaxTmax or the fitness meets the requirements.

[0087] (2) Adaptive Ant Lion Optimizer (AALO): Based on the standard ant lion optimizer, an adaptive mechanism is introduced to dynamically adjust the key parameters in the algorithm to improve the search efficiency and accuracy. The working steps are as follows:

[0088] 1. Adaptive exploration probability P a :

[0089] 1.1 Initial exploration probability: P a (0) = 0.6

[0090] 1.2 Adjust according to the number of iterations:

[0091] where t is the current number of iterations and T max is the maximum number of iterations.

[0092] 2. Adaptive step size adjustment:

[0093] 2.1 Initial step size Δ: Δ(0) = max(range) - min(range)

[0094] 2.2 Adjust according to the number of iterations:

[0095] where exp is the exponential function.

[0096] (3) Random forest algorithm: An ensemble learning algorithm that improves the prediction ability and generalization ability of the model by constructing multiple decision trees and combining their prediction results. See Figure 1 , and the working principle is as follows:

[0097] 1. Generate multiple subsamples:

[0098] 1.1 Sample with replacement from the original dataset to generate subsamples for multiple decision trees;

[0099] 1.2 Assume there are M decision trees, and each subsample has n samples;

[0100] 2. Construct multiple decision trees:

[0101] 2.1 Construct a decision tree for each subsample;

[0102] 2.2 When selecting each node, randomly select a subset of features for splitting decisions instead of all features;

[0103] 2.3 The depth d of the decision tree and the minimum number of samples s in the leaf nodes can be adjusted as hyperparameters;

[0104] 3. Prediction: Make predictions using weighted voting (for classification problems) or the average value (for regression problems).

[0105] 4. Hyperparameters:

[0106] M: The number of decision trees;

[0107] s: The minimum number of samples in each leaf node.

[0108] See Figure 2 , a power control method based on an adaptive ant lion optimization algorithm to optimize the random forest, specifically including the following steps:

[0109] S1. Data collection and preprocessing:

[0110] S1.1. Collect data on power consumption, usage frequency, and power temperature under various working conditions during historical live working processes;

[0111] S1.2. Perform data cleaning to handle missing values and abnormal data;

[0112] Standardize or normalize the data for subsequent analysis and modeling. Assume the dataset is D, and the standardized dataset is D'.

[0113] S2. Random forest model construction:

[0114] S1.1. Divide the standardized dataset D' into a training set D train and a validation set D val ;

[0115] S1.2. Construct a random forest model to predict future power demands: Let the random forest model be RF(X; Θ), where X is the input feature, and the initial hyperparameter set Θ = {M, d, s}. The specific parameter settings are as follows:

[0116] M: The number of decision trees;

[0117] d: The maximum number of features, which refers to the maximum number of features considered when finding the best split;

[0118] s: The minimum number of samples in each leaf node;

[0119] S1.3. Randomly generate an initial ant colony X: X = {x1, x2, …, x N}], where each x i is a combination of hyperparameters of a random forest;

[0120] S1.4. Initialize the initial pits T: T = {t1, t2, …, t N}], where t i is the position of the i-th pit, and each t iInitial value and corresponding x i are the same;

[0121] S1.5. Calculate the fitness value f(x i ), which is used to evaluate the performance of each ant. The fitness value can be measured by the prediction effect of the validation set. For example, the mean square error MSE is used:

[0122]

[0123] where m is the number of samples in the validation set,

[0124] X j is the j-th sample in the validation set and is a subset of the input feature set in the validation set;

[0125] Y j is the actual power demand.

[0126] S3. Construct potholes:

[0127] Attract the better ants to the pothole positions according to the fitness value f(x i ); if f(t i ) > f(x i ), then update the pothole position t i , where f(t i ) is the fitness value corresponding to the i-th pothole position:

[0128] t i = x i (11)

[0129] Use traps to capture ants; for the position vector included in each ant x i , update its position to a random point A(trap) near the pothole t i :

[0130] x i (trap) = t i + rand(-1, 1)×(A(trap) - x i ) (12)

[0131] where A(trap) is a point randomly generated within the foraging range.

[0132] S4. Ant movement:

[0133] Randomly move the ants; for each ant x i , update its position to a new random point x i (move):

[0134] x i(move) = x i + rand(-1, 1) × (sign(rand() - 0.5) × (Δ(t) × (A(move) - x i ) (13)

[0135] where A(move) and A(trap) are randomly generated points related to the current ant position, rand() represents generating a random number between [0, 1), the initial step size Δ(0) and the adaptive adjustment formula are:

[0136]

[0137] where exp is the exponential function, t is the current iteration number, Δ(t) is the adaptive adjustment step size, and T max is the maximum iteration number.

[0138] S5. Fitness update and probability adaptation:

[0139] Calculate the updated fitness value f(x i (move));

[0140] Introduce the adaptive exploration probability P a (t):

[0141]

[0142] Determine whether the ant falls into the pit according to the fitness value and the adaptive exploration probability:

[0143]

[0144] random() represents generating a random number between [0, 1).

[0145] S6. Update the population

[0146] Update the pit position t according to the newly found ant position i :

[0147] t i = x i (next) (17)

[0148] Reconstruct the ant colony: If random() > P a (t), then the ant position x i is updated to x i (move):

[0149] x i = x i (move) (18)

[0150] S7. Random Forest Model Training and Evaluation:

[0151] Use the hyperparameter combination Θ corresponding to each ant (solution). i Train the random forest model RF(X; Θ i ).

[0152] Evaluate the performance of each random forest model on the validation set D val and calculate its mean squared error MSE.

[0153] S8. Iterative Optimization: Repeat steps S4 to S8 until the predetermined number of iterations T max or the fitness value reaches a satisfactory accuracy.

[0154] S9. Select the Best Solution: Select the hyperparameter combination Θ with the minimum MSE among the fitness values f(x i ) of all ants best as the best solution.

[0155] S10. Dynamic Power Control:

[0156] During the actual live working process, real-time monitor information such as power, usage frequency, and temperature, and input the real-time monitoring data into the optimized random forest model RF(X; Θ best ), predict the best future power supply plan, and perform corresponding power regulation.

[0157] S11. Effect Evaluation and Iteration: Evaluate the effectiveness of the optimized power control method through actual operation data, adjust the parameters of the adaptive ant lion optimization algorithm according to the evaluation results (such as the adaptive step size Δ(t) and the adaptive exploration probability P a (t)), and retrain the random forest model, and perform multiple iterative optimizations to improve the accuracy and stability of the power control plan.

[0158] Through the above steps, combining historical data and real-time monitoring data, use the adaptive ant lion optimization algorithm to dynamically adjust the hyperparameters of the random forest model, thereby optimizing the power control plan in live working.

[0159] The above embodiments are implemented on the premise of the technical solution of the present invention, and detailed implementation manners and specific operation processes are given, but the protection scope of the present invention is not limited to the above embodiments. The methods used in the above embodiments are all conventional methods unless otherwise specified.

Claims

1. An adaptive portable power control method for live working on a substation, characterized in that, It includes the following steps: S1. Collect the power demand data under various working conditions during live working through a data acquisition system, and construct a random forest model; S2. Use this data to train the random forest model to generate an initial power control scheme; S3. Adopt an adaptive ant lion optimization algorithm to finely adjust the parameters of the random forest model; S4. Apply the optimized random forest model to the dynamic adjustment and optimization of the power control scheme. By real-time monitoring the working conditions, adaptively adjust the power output to ensure that the best power support can be provided under various working conditions.

2. The adaptive portable power control method for live substation operation according to claim 1, characterized in that, The power demand data includes the power, usage frequency, and power temperature data under various working conditions during historical live working.

3. A method for controlling an adaptive portable power supply for live working on a power transformation substation according to claim 1, characterized in that, The construction of the random forest model includes the following steps: S1.

1. Divide the standardized dataset D' into a training set D train and a validation set D val ; S1.

2. Construct a random forest model to predict future power demands: Let the random forest model be RF(X; Θ), where X is the input feature, and the initial hyperparameter set Θ = {M, d, s}, and the specific parameter settings are as follows: M: The number of decision trees; d: The maximum number of features, which refers to the maximum number of features considered when finding the best split; s: The minimum number of samples in each leaf node; S1.

3. Randomly generate the initial ant colony X: X = {x1, x2, …, x N}, where each x i is a hyperparameter combination of a random forest; S1.

4. Initialize the initial pits T: T = {t1, t2, …, t N}, where t i is the position of the i-th pit, and the initial value of each t i is the same as the corresponding x i . S1.

5. Calculate the fitness value f(x i ), which is used to evaluate the performance of each ant. The fitness value can be measured by the prediction effect of the validation set, for example, using the mean squared error MSE: where m is the number of samples in the validation set, X j The j-th sample in the validation set, which is a subset of the input feature set in the validation set; Y j is the actual power demand.

4. A self - adaptive portable power control method for live working on a power transformation substation according to claim 1, characterized in that, The specific steps of step S3 include the following steps: S3.

1. Construct pits; According to the fitness value f(x i ), attract the better ants to the pothole position; if f(t i ) > f(x i ), then update the pothole position t i , where f(t i ) is the fitness value corresponding to the i-th pothole position: t i = x i (2) Use traps to catch ants; for each ant x i For the included position vector, update its position to the pit t i A random point A(trap) nearby: x i (trap) = t i + rand(-1, 1) × (A(trap) - x i ) (3) where A(trap) is a point randomly generated within the foraging range; S3.

2. Ant movement; Randomly move the ants; for each ant x i , update its position to a new random point x i (move): x i (move) = x i + rand(-1, 1) × (sign(rand() - 0.5) × (Δ(t) × (A(move) - x i ) (4) where A(move) and A(trap) are randomly generated points related to the current ant position, rand() represents generating a random number between [0, 1), and the initial step size Δ(0) and the adaptive adjustment formula are: where exp is the exponential function; t is the current iteration number; Δ(t) is the adaptive adjustment step size, which represents the step size of the ant movement at time step t; T max is the maximum number of iterations; S3.

3. Fitness update and probability adaptation; Calculate the updated fitness value f(x i (move)); Introduce the adaptive exploration probability P a (t): Decide whether the ant falls into the pit according to the fitness value and the adaptive exploration probability: random() represents generating a random number between [0, 1); S3.

4. Update the population Update the position t of the pothole according to the newly found position of the ant i : t i = x i (next) (8) Reconstruct the ant colony: If random() > P a (t), then the ant position x i is updated to x i (move): x i = x i (move) (9) S3.

5. Random forest model training and evaluation; Use the hyperparameter combination Θ corresponding to each ant (solution) i Train the random forest model RF(X; Θ i ) Evaluate the performance of each random forest model on the validation set D val and calculate its mean squared error MSE; S3.

6. Iterative Optimization: Repeat steps S3.2 to S3.6 until the predetermined number of iterations T is reached max or the fitness value reaches a satisfactory accuracy; S3.

7. Select the best solution: Select the hyperparameter combination Θ with the smallest MSE in the fitness values f(x i ) among all ants as the best solution. best ​ 5. The adaptive portable power supply control method for live working on a power transformation substation according to claim 1, wherein, The input of the optimized random forest model is the real-time monitoring data, and the output is the best future power scheme, and corresponding power adjustment is carried out.

Images