Power distribution network line power loss prediction method

Abstract

The invention relates to a power distribution network line power loss prediction method, in particular to a line power loss prediction method, and solves the problem of poor prediction precision of the line power loss under mass data in the existing prediction method. The method comprises the steps of obtaining historical line loss electric quantity data; preprocessing the obtained historical line loss electric quantity data to generate a training set; establishing a line loss electric quantity prediction model based on an improved artificial bee colony joint algorithm, and optimizing the line loss electric quantity prediction model by using the generated training set to obtain an optimized line loss electric quantity prediction model; and inputting to-be-predicted power distribution network line power loss data into the obtained optimized line power loss prediction model, outputting line power loss, and completing prediction of the power distribution network line power loss. The method has the beneficial effect of improving the prediction precision of the power loss of the power distribution network line.

Description

Technical Field

[0001] The invention belongs to the field of low-voltage distribution network analysis, and in particular relates to a line loss electricity prediction method. Background Art

[0002] With the rapid development of the national economy, the scale of traditional and emerging industries has expanded, and the process of urbanization and electrification has continued to accelerate, the demand for load and energy will continue to increase; electric energy is a clean energy and the most widely used secondary energy; electric energy will inevitably produce losses in the process of power generation, transmission, conversion and distribution in the power system. The increase in losses will not only reduce the efficiency of electricity use, but also have an adverse effect on the quality of electric energy and the operation of the power grid. The line loss rate is an important economic indicator of the power system. Accurate analysis and management of the line loss rate can effectively improve the technical level and management level of power supply enterprises and achieve energy saving and efficiency improvement.

[0003] The distribution network is a key component of the power system. The current continuous increase in electricity demand has gradually increased the structural complexity of the distribution network. In recent years, the automation level of the distribution network has been comprehensively improved, and various automatic data acquisition devices have been produced. While providing reliable data for the calculation of the line loss rate of the distribution network, it also brings huge challenges in huge data processing and efficient application. Although the accuracy of data acquisition and line loss rate calculation of the distribution network has been improved, it is still impossible to accurately predict the future development trend of the line loss rate. Therefore, researchers in related fields have carried out a lot of research on the prediction methods of line loss electricity and line loss rate. Due to the complex structure of the current distribution network and the many factors affecting line loss, the existing prediction methods are difficult to adapt to the increasingly high requirements for prediction accuracy under massive data. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that the existing prediction methods have poor prediction accuracy for line loss electricity under massive data, and propose a distribution network line loss electricity prediction method.

[0005] A method for predicting power loss in a power distribution network according to the present invention comprises the following steps:

[0006] Step 1: Obtain historical line loss power data;

[0007] Step 2: pre-process the historical line loss power data obtained in step 1 to generate a training set;

[0008] Step 3: Establish a line loss electricity prediction model based on the improved artificial bee colony joint algorithm, use the training set generated in step 2 to optimize the line loss electricity prediction model, and obtain the optimized line loss electricity prediction model;

[0009] Step 4: Input the distribution network line loss electricity data to be predicted into the optimized line loss electricity prediction model obtained in step 3, output the line loss electricity, and complete the prediction of the distribution network line loss electricity.

[0010] Furthermore, the specific method for obtaining the historical line loss electricity data in step 1 is: obtaining the data of the line loss characteristic factors in time series as the line loss characteristic data, and each of the line loss characteristic data constitutes the historical line loss electricity data.

[0011] Furthermore, the specific method for generating the training set in step 2 is:

[0012] Step 21: De-noise and smooth the historical line loss power data to generate a power data set;

[0013] Step 22: Divide the electricity data set generated in step 21 into: missing values, abnormal values ​​and normal values;

[0014] Step 23: For the missing values ​​divided in step 22, patch them vertically and horizontally to generate patch values;

[0015] Step 24: for the abnormal values ​​classified in step 22, perform abnormal value processing to generate replacement values;

[0016] Step 25: Perform data set normalization processing on the normal values ​​divided in step 22, the repair values ​​generated in step 23, and the replacement values ​​generated in step 24 to generate a training set.

[0017] Furthermore, the specific formula for normalizing the data set in step 25 is:

[0018]

[0019] where x * is the normalized data; x i is the original data; x max is the maximum value in the data; x min is the minimum value in the data.

[0020] Furthermore, the specific steps of obtaining the optimized line loss power prediction model in step 3 are as follows:

[0021] Step 31: Build the LSSVM training model, determine the size of the population, the number of nectar source mining times, the ratio of employed bees, observer bees, and scout bees, and generate an initial solution within the search range;

[0022] Step 32: Set the fitness function and calculate the fitness value;

[0023] Step 33: Set the loop conditions and start the loop;

[0024] Step 34: Hired Bee Utilization Formula Make a new interpretation ij Search and calculate the fitness value; where v ij A new solution for hired bees; ij is the current solution of the employed bees; x jk is the random solution of hired bees; is the kernel parameter of employed bees;

[0025] Step 35: Hire bees to make greedy choices. That is, if v ij The fitness value is better than x ij , then use v ij Replace x ij , v ij As the current best solution, otherwise keep x ij constant;

[0026] Step 36: According to the fitness value of the employed bees, select the target through roulette, that is, calculate the probability p of the food source i , where f i is the fitness of the ith solution; N is the total number of solutions;

[0027] Step 37: Follow the bee according to probability p i Select a solution or food source, according to the formula Continue to new solution i The search produces a new solution v i , and calculate its fitness; where v i A new solution for following bees; x i is the current solution of the follower bee; x k is the random solution of following bees; is the kernel parameter of the follower bee;

[0028] Step 38: Follow the bee to make greedy choices; That is, if v i The fitness is better than x i , then use v i Replace x i , v i As the current best solution, otherwise keep x i constant;

[0029] Step 39: After the follower bee completes the search, if the solution is not further updated after a limited number of cycles, the follower bee corresponding to the food source is converted into a scout bee, and the scout bee randomly generates a new solution, x ij =x min +rand(0,1)·(x max -x min ), where xmax is the upper bound of the solution; x min is the lower bound of the solution, that is, start searching for neighborhood solutions in the new area;

[0030] Step 30: mutate the current population, calculate the fitness values ​​of the individuals in the mutated bee colony, and retain the individuals with better fitness values;

[0031] Step 31: record the optimal solution of the current cycle and determine whether to stop; if the stop condition is not met, return to execute steps 3 and 4 until the judgment condition is met, obtain the optimal parameter value, and obtain the optimized line loss power prediction model.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention adopts an improved artificial bee colony algorithm that introduces Cauchy mutation operator and Gaussian mutation operator interference to obtain the initial parameters of the least squares support vector machine (LSSVM) to construct a line loss electricity prediction model to predict the line loss electricity of the distribution network. The line loss electricity prediction model is suitable for electricity prediction under massive data, can achieve accurate prediction of the line loss electricity of the distribution network, meets engineering requirements, has fast calculation speed, and has low training cost of the line loss electricity prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flow chart of a method for predicting power loss in a distribution network according to a specific implementation mode 1;

[0035] Figure 2 This is a flow chart of a specific method for obtaining an optimized line loss power prediction model in specific implementation mode five. DETAILED DESCRIPTION

[0036] Specific implementation method 1. Combination Figure 1 This embodiment describes a method for predicting power loss in a distribution network, and the method comprises the following steps:

[0037] Step 1: Obtain historical line loss power data;

[0038] Step 2: pre-process the historical line loss power data obtained in step 1 to generate a training set;

[0039] Step 3: Establish a line loss electricity prediction model based on the improved artificial bee colony joint algorithm, use the training set generated in step 2 to optimize the line loss electricity prediction model, and obtain the optimized line loss electricity prediction model;

[0040] Step 4: Input the distribution network line loss electricity data to be predicted into the optimized line loss electricity prediction model obtained in step 3, output the line loss electricity, and complete the prediction of the distribution network line loss electricity.

[0041] In this embodiment, support vector machine (SVM) is the main representative of machine learning and is a very common classifier in various fields. SVM adopts the principle of structural risk minimization to reduce the amount of calculation by setting the kernel function, and can be better applied to nonlinear problems. It has a simple structure and strong generalization ability. It requires a small amount of sample data and is suitable for complex and multi-dimensional small sample data. This structure will not be restricted by constraints in practical applications. Each point in the training sample can obtain the corresponding linear model by fitting, so as to control the calculation error within a reasonable range and achieve high-precision data prediction.

[0042] The Least Squares Support Vector Machine (LSSVM) is an improvement on SVM by Suykens and Vandewalle. The inequality constraints in the original algorithm are changed to equality constraints, and the sum of squared errors is used as the empirical loss. By constructing a loss function, the quadratic optimization is transformed into solving a simpler linear equation. LSSVM has greatly improved in terms of noise resistance and calculation speed; LSSVM has good performance in terms of noise resistance and calculation speed, but its calculation speed and prediction accuracy are greatly affected by the initial parameters; the artificial bee colony algorithm (ABC) is a group optimization method proposed by Karaboga that imitates the behavior of bees. Its main feature is that it does not require special information about the problem, but only needs to compare the pros and cons of the problem. Through the local optimization behavior of each artificial bee individual, the global optimal value is finally revealed in the group. It has a fast convergence speed and can accurately solve multivariable function optimization problems; this method uses an improved artificial bee colony algorithm to optimize the initial parameters of LSSVM to construct a line loss power prediction model to predict the line loss power of the distribution network. The line loss power prediction model is suitable for power prediction under massive data, and can achieve accurate prediction of the line loss power of the distribution network. It meets engineering needs while having a fast calculation speed and low training cost of the line loss power prediction model.

[0043] Specific implementation method 2. This implementation method further limits the distribution network line loss electricity prediction method described in specific implementation method 1. In this implementation method, the specific method for obtaining historical line loss electricity data in step 1 is: obtaining the data of the line loss characteristic factors in time series as line loss characteristic data, and each of the line loss characteristic data constitutes the historical line loss electricity data.

[0044] In this embodiment, the historical line loss electricity data needs to have a sufficient time span, such as a forecast representative of 30 days or more before the day.

[0045] Specific implementation method 3: This implementation method further limits the method for predicting power loss in a distribution network described in specific implementation method 1. In this implementation method, the specific method for generating a training set in step 2 is:

[0046] Step 21: De-noise and smooth the historical line loss power data to generate a power data set;

[0047] Step 22: Divide the electricity data set generated in step 21 into: missing values, abnormal values ​​and normal values;

[0048] Step 23: For the missing values ​​divided in step 22, patch them vertically and horizontally to generate patch values;

[0049] Step 24: for the abnormal values ​​classified in step 22, perform abnormal value processing to generate replacement values;

[0050] Step 25: Perform data set normalization processing on the normal values ​​divided in step 22, the repair values ​​generated in step 23, and the replacement values ​​generated in step 24 to generate a training set.

[0051] In this embodiment, in step 21, the historical line loss power data is denoised and smoothed to generate a power data set as follows:

[0052] Decompose the historical data sequence x(t) into n intrinsic mode functions (IMFs), u k(t) , k = 1, 2, 3...n, to ensure that each decomposition sequence is a modal component with a limited bandwidth and a center frequency, and the sum of the estimated bandwidths of each mode is minimized, the constraint condition is that the sum of all modes is equal to the original signal, then the VMD constrained variational model is as follows:

[0053]

[0054] Among them, u k(t) is each mode function, ω k is the center frequency of each mode function;

[0055] The Lagrangian method is used to find the optimal solution, and the quadratic penalty factor α and the Lagrangian multiplier λ(t) are introduced to establish the augmented Lagrangian function to transform the above constrained variational problem into an unconstrained variational problem.

[0056]

[0057] The optimal solution of the Lagrangian augmented function is obtained by using the alternating direction multiplier algorithm (fixing the other two variables and updating one of the variables), thereby decomposing the input original historical data into n IMF components;

[0058]

[0059] According to the equivalence of Fourier transform, the above equation is converted to the frequency domain and solved in the frequency domain using Parseval's theorem;

[0060]

[0061] Taking the derivative of the above formula, we can get:

[0062]

[0063] Similarly, using the center frequency expression, the center of the current modal power spectrum can be obtained as:

[0064]

[0065] After decomposing the input historical data sequence, the degree of correlation between each mode and the source data is determined, and the irrelevant modes are removed as noise;

[0066] In step 2 and 3: a combination of horizontal and vertical methods is used to process missing data points in historical line loss power data due to abnormal line maintenance or sensor abnormalities;

[0067] For non-continuous missing points, the horizontal processing method is used, and the average value of the two line loss points before and after the missing point is used to replace the data of the non-continuous missing point. The specific formula is:

[0068]

[0069] Among them, y t is the desired alternative data, y t-1 is the data before the missing point; y t+1 is the data after the missing point; y (D-1)t is the data at the same time point one day before the missing point; y (D+1)t It is the data at the same time point one day after the missing point;

[0070] For consecutive missing points, a vertical processing method is used, and the weighted average of line loss electricity at the same time a few days before and after the missing point is used to replace the missing point data. The specific formula is:

[0071]

[0072] Among them, y t is the desired alternative data, y (D-1)t is the data at the same time point one day before the missing point; y (D+1)t is the data at the same time point one day after the missing point; y (D-2)tis the data at the same time point two days before the missing point; y (D+2)t It is the data at the same time point two days after the missing point;

[0073] Step 24: for the abnormal values ​​classified in step 22, perform abnormal value processing to generate replacement values;

[0074] Set an empirical threshold to filter suspected abnormal data in historical data, and determine that data exceeding 1.2 times the average line loss power at the same time in the previous and next four days is suspected abnormal data. The specific formula is:

[0075]

[0076] Among them, y is the data of each point to be determined in the historical data, y (D-1)t is the data at the same time point one day before the missing point; y (D+1)t is the data at the same time point one day after the missing point; y (D-2)t is the data at the same time point two days before the missing point; y (D+2)t It is the data at the same time point two days after the missing point;

[0077] Calculate the difference between the loads of two adjacent sampling points other than the suspected abnormal data point on the day of the suspected abnormal data, and record it as the difference sequence x1, x2, …, x t If the load difference between the suspected abnormal data point and the two points before and after it is greater than any value in the difference sequence at the same time, the data at this point is judged to be abnormal data;

[0078] Eliminate the data that are judged to be abnormal, and use the methods in steps two and three to repair them.

[0079] Specific implementation method 4: This implementation method further limits the method for predicting power loss in a distribution network described in specific implementation method 1. In this implementation method, the specific formula for normalizing the data set in step 25 is:

[0080]

[0081] where x * is the normalized data; x i is the original data; x max is the maximum value in the data; x min is the minimum value in the data.

[0082] In this embodiment, the purpose of data set normalization is to unify the order of magnitude of different types of data and improve the prediction accuracy of model output results.

[0083] Specific implementation method V. Combination Figure 2This embodiment is described. This embodiment further limits the method for predicting power line loss in a distribution network described in the first embodiment. In this embodiment, the specific steps for obtaining the optimized power line loss prediction model in step 3 are as follows:

[0084] Step 31: Build the LSSVM training model, determine the size of the population, the number of nectar source mining times, the ratio of employed bees, observer bees, and scout bees, and generate an initial solution within the search range;

[0085] Step 32: Set the fitness function and calculate the fitness value;

[0086] Step 33: Set the loop conditions and start the loop;

[0087] Step 34: Hired Bee Utilization Formula Make a new interpretation ij Search and calculate the fitness value; where v ij A new solution for hired bees; ij is the current solution of the employed bees; x jk is the random solution of hired bees; is the kernel parameter of employed bees;

[0088] Step 35: Hire bees to make greedy choices. That is, if v ij The fitness value is better than x ij , then use v ij Replace x ij , v ij As the current best solution, otherwise keep x ij constant;

[0089] Step 36: According to the fitness value of the employed bees, select the target through roulette, that is, calculate the probability p of the food source i , where f i is the fitness of the ith solution; N is the total number of solutions;

[0090] Step 37: Follow the bee according to probability p i Select a solution or food source, according to the formula Continue to new solution i The search produces a new solution v i , and calculate its fitness; where v i A new solution for following bees; x i is the current solution of the follower bee; x k is the random solution of following bees; is the kernel parameter of the follower bee;

[0091] Step 38: Follow the bee to make greedy choices; That is, if vi The fitness is better than x i , then use v i Replace x i , v i As the current best solution, otherwise keep x i constant;

[0092] Step 39: After the follower bee completes the search, if the solution is not further updated after a limited number of cycles, the follower bee corresponding to the food source is converted into a scout bee, and the scout bee randomly generates a new solution, x ij =x min +rand(0,1)·(x max -x min ), where x max is the upper bound of the solution; x min is the lower bound of the solution, that is, start searching for neighborhood solutions in the new area;

[0093] Step 30: mutate the current population, calculate the fitness values ​​of the individuals in the mutated bee colony, and retain the individuals with better fitness values;

[0094] Step 31: record the optimal solution of the current cycle and determine whether to stop; if the stop condition is not met, return to execute steps 3 and 4 until the judgment condition is met, obtain the optimal parameter value, and obtain the optimized line loss power prediction model.

[0095] In this embodiment, a line loss power prediction model based on the least squares support vector machine (LSSVM) of the improved artificial bee colony is established; the classification prediction effect of the least squares support vector machine (LSSVM) is affected by the initial parameter penalty parameter (γ) and the kernel parameter The impact is large, and it is difficult to find accurate parameter values ​​by manual trial and error. In the present invention, the artificial bee colony algorithm (ABC) is used to find the optimal parameters of the model;

[0096] ABC is a group optimization algorithm with strong stability, but it may fall into the local optimum during the optimization process. In this method, the Cauchy mutation operator and Gaussian mutation operator are introduced to interfere, increase the diversity of the population, and jump out of the local optimum. new (t) = x best (t)(1+ω1causs(σ)+ω2cauchy(δ)), compare the fitness values ​​of the new individual after mutation with the original individual to determine whether replacement has occurred, and retain the individual with better fitness value;

[0097] In addition to generating a training set, the data set processed in step 2 can also be divided into a part of the data to form a validation set; the training set data is used to determine the parameters of the least squares support vector machine prediction model, the validation set is used to test the prediction effect of the least squares support vector machine prediction model, the model is calibrated, and the optimal prediction model is obtained, and the test set is used to predict line loss electricity;

[0098] Constructing optimal decision functions in high-dimensional space, in is a mapping function that transforms the original training set data (x i ,y i ) is mapped to a high-dimensional space, ω is the weight vector that determines the direction of the decision boundary in the high-dimensional space, and b is the bias term that determines the position of the decision boundary in the high-dimensional space;

[0099] The constraints are, Where γ is the penalty parameter;

[0100] The weights and biases are solved by the larrange multiplier method to obtain the nonlinear approximation function of the LSSVM of the training data set.

[0101] The kernel function To satisfy any symmetric function of Mercer, α i is the Lagrange multiplier;

[0102] Obtaining the optimized prediction model;

[0103] The test set data is input into the optimal prediction model to conduct error analysis and evaluation of the prediction results, and the model prediction performance evaluation index;

[0104] Relative error Mean absolute error Mean absolute percentage error Root mean square error Four evaluation indicators are used to evaluate the prediction effect of the model, where n is the number of prediction samples, y′ t ,y t is the predicted value at time t and the actual value of line loss electricity.

[0105] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for predicting power loss in a distribution network, characterized in that: The line loss power prediction method comprises the following steps: Step 1: Obtain historical line loss power data; Step 2: pre-process the historical line loss power data obtained in step 1 to generate a training set; Step 3: Establish a line loss electricity prediction model based on the improved artificial bee colony joint algorithm, use the training set generated in step 2 to optimize the line loss electricity prediction model, and obtain the optimized line loss electricity prediction model; Step 4: Input the distribution network line loss electricity data to be predicted into the optimized line loss electricity prediction model obtained in step 3, output the line loss electricity, and complete the prediction of the distribution network line loss electricity.

2. A method for predicting power loss in a distribution network according to claim 1, characterized in that: The specific method for obtaining the historical line loss electricity data in step 1 is: obtaining the data of the line loss characteristic factors in time series as the line loss characteristic data, and each of the line loss characteristic data constitutes the historical line loss electricity data.

3. A method for predicting power loss in a distribution network according to claim 1, characterized in that: The specific method for generating the training set in step 2 is: Step 21: De-noise and smooth the historical line loss power data to generate a power data set; Step 22: Divide the electricity data set generated in step 21 into: missing values, abnormal values ​​and normal values; Step 23: For the missing values ​​divided in step 22, patch them vertically and horizontally to generate patch values; Step 24: for the abnormal values ​​classified in step 22, perform abnormal value processing to generate replacement values; Step 25: Perform data set normalization processing on the normal values ​​divided in step 22, the repair values ​​generated in step 23, and the replacement values ​​generated in step 24 to generate a training set.

4. A method for predicting power loss in a distribution network according to claim 1, characterized in that: The specific formula for normalizing the data set in step 25 is: where x * is the normalized data; x i is the original data; x max is the maximum value in the data; x min is the minimum value in the data.

5. A method for predicting power loss in a distribution network according to claim 1, characterized in that: The specific steps to obtain the optimized line loss power prediction model in step 3 are: Step 31: Build the LSSVM training model, determine the size of the population, the number of nectar source mining times, the ratio of employed bees, observer bees, and scout bees, and generate an initial solution within the search range; Step 32: Set the fitness function and calculate the fitness value; Step 33: Set the loop conditions and start the loop; Step 34: Hired Bee Utilization Formula Make a new interpretation ij Search and calculate the fitness value; Among them, v ij A new solution for hired bees; ij is the current solution of the employed bees; x jk is the random solution of hired bees; is the kernel parameter of employed bees; Step 35: Hire bees to make greedy choices. That is, if v ij The fitness value is better than x ij , then use v ij Replace x ij , v ij As the current best solution, otherwise keep x ij constant; Step 36: According to the fitness value of the employed bees, select the target through roulette, that is, calculate the probability p of the food source i , where f i is the fitness of the ith solution; N is the total number of solutions; Step 37: Follow the bee according to probability p i Select a solution or food source, according to the formula Continue to new solution i The search produces a new solution v i , and calculate its fitness; where v i A new solution for following bees; x i is the current solution of the follower bee; x k is the random solution of following bees; is the kernel parameter of the follower bee; Step 38: Follow the bee to make greedy choices; That is, if v i The fitness is better than x i , then use v i Replace x i , v i As the current best solution, otherwise keep x i constant; Step 39: After the follower bee completes the search, if the solution is not further updated after a limited number of cycles, the follower bee corresponding to the food source is converted into a scout bee, and the scout bee randomly generates a new solution, x ij =x min +rand(0,1)·(x max -x min ), where x max is the upper bound of the solution; x min is the lower bound of the solution, that is, start searching for neighborhood solutions in the new area; Step 30: mutate the current population, calculate the fitness values ​​of the individuals in the mutated bee colony, and retain the individuals with better fitness values; Step 31: record the optimal solution of the current cycle and determine whether to stop; if the stop condition is not met, return to execute steps 3 and 4 until the judgment condition is met, obtain the optimal parameter value, and obtain the optimized line loss power prediction model.

Images