Transformer substation automatic voltage control method based on voltage prediction

Abstract

The invention provides a transformer substation automatic voltage control method based on voltage prediction, and belongs to the field of automatic voltage control. The method comprises the steps that a preset voltage prediction model is utilized to obtain a bus voltage prediction value of each substation control unit, and the voltage prediction model is formed by combining a Conv1D one-dimensional convolutional neural network and an LSTM (Long Short Term Memory) network; determining the voltage trend of the substation control unit based on the bus voltage predicted value; and generating a pre-scheduling strategy of the substation control unit based on the voltage trend. According to the invention, the voltage is dynamically adjusted according to the voltage change trend, the stable voltage level can be maintained at the load peak and valley, and excessive adjustment and frequent control actions are reduced.

Description

Technical Field

[0001] The present invention belongs to the field of automatic voltage control, and particularly relates to a substation automatic voltage control method based on voltage prediction. Background Art

[0002] With the continuous expansion of the power grid scale, it has become increasingly difficult to regulate the voltage of the power system. In power grid operation scheduling, the Automatic Voltage Control (AVC) system is used to achieve a reasonable distribution of voltage and reactive power. So far, there are mainly three mainstream AVC control modes in China, namely, the secondary control mode represented by RWE in Germany and the three-level control mode based on zoning represented by EDF ( de France), and the three-level control mode based on soft zoning proposed by the Dispatching Automation Laboratory of the Department of Electrical Engineering of Tsinghua University. In the dispatching centers at all levels in China, the three-level control mode based on soft zoning proposed by Tsinghua University is mainly used for AVC.

[0003] The automatic reactive power voltage control system has become an important system for reactive power voltage operation management in today's power grid dispatching, ensuring the hierarchical and zonal balance of reactive power, local balance, meeting the voltage requirements, and realizing the loss reduction of the power grid. However, in the actual application process of AVC, there are still many problems and deficiencies. For example, due to the different load characteristics in each region, if the action of reactive power voltage regulating equipment cannot be reasonably planned during the peak-valley alternation period, it may cause the problem of frequent action of reactive power voltage regulating equipment. In the actual control system, each reactive power voltage regulating equipment has an upper limit on the number of action times. In the above situation, it will cause the control number upper limit locking of the reactive power voltage regulating equipment, which will directly affect the reactive power voltage control quality in this region. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the existing technologies and propose a substation automatic voltage control method based on voltage prediction. The present invention dynamically adjusts the voltage according to the voltage change trend, and can achieve a stable voltage level during peak and valley loads, reducing over-regulation and frequent control actions.

[0005] An embodiment of the present invention proposes a substation automatic voltage control method based on voltage prediction, including:

[0006] Using a preset voltage prediction model to obtain the predicted bus voltage values of each substation control unit, where the voltage prediction model is composed of a combination of a one-dimensional convolutional neural network (Conv1D) and a long short-term memory network (LSTM);

[0007] Based on the predicted bus voltage values, determining the voltage trend of the substation control unit;

[0008] Generate a pre-scheduling strategy for the substation control unit based on the voltage trend.

[0009] In a specific embodiment of the present invention, the voltage prediction model includes an input layer, a one-dimensional convolutional neural network layer, a pooling layer, a long short-term memory network layer, a fully connected layer, and an output layer connected in sequence; the input of the voltage prediction model is the historical data sequence of active power load, the historical data sequence of transformer tap positions, the historical state data sequence of capacitors and reactors in the past period corresponding to the substation control unit, and the output is the bus voltage data sequence of the substation control unit in the future period.

[0010] In a specific embodiment of the present invention, before obtaining the bus voltage prediction values of each substation control unit by using the preset voltage prediction model, the method further includes:

[0011] Training the voltage prediction model;

[0012] The training of the voltage prediction model includes:

[0013] 1) Establish a corresponding voltage prediction model for each substation control unit;

[0014] 2) Establish a training set for the voltage prediction model corresponding to each substation control unit. The specific steps are as follows:

[0015] 2-1) Obtain the historical data of the substation control unit, including: historical data of active power load, historical data of transformer tap positions, historical state data of capacitors and reactors, and historical data of bus voltage;

[0016] 2-2) Divide the data obtained in step 2-1) into input data sequences and output data sequences respectively according to a set sliding time window;

[0017] Among them, the length of the sliding time window of the input data in the training set is equal to the length of the past period of the model input, and the length of the sliding time window of the output data is equal to the length of the future period of the model output;

[0018] 2-3) Use the results of step 2-2) to form a training sample with each group of input data sequences and their corresponding output data sequences, and all training samples constitute the training set;

[0019] 3) Use the training set obtained in step 2) to train the voltage prediction model constructed in step 1) to obtain a trained model.

[0020] In a specific embodiment of the present invention, the method further includes:

[0021] Perform data smoothing and missing data value processing on the bus voltage prediction values.

[0022] In a specific embodiment of the present invention, the data smoothing includes:

[0023] For any substation control unit, if the voltage prediction value at the t-th moment of the nd-th bus is a distorted data point, it is smoothed according to the following formula:

[0024]

[0025] where V unit (nd,t) represents the voltage prediction value at the t-th moment of the nd-th bus of this substation control unit, and V′ unit (nd,t) represents the voltage prediction value at the t-th moment of the nd-th bus of this substation control unit after smoothing.

[0026] In a specific embodiment of the present invention, the processing of data missing values includes:

[0027] For each V unit (nd,t) data, if data is missing at t = t 0 moment, find the two moments before and after the t 0 moment without data missing, and record them as t -1 and t 1 moments respectively, and then supplement the data missing value according to the following formula:

[0028]

[0029] In a specific embodiment of the present invention, determining the voltage trend of the substation control unit based on the bus voltage prediction value includes:

[0030] Using the regression analysis method to perform linear fitting on the bus voltage prediction value, where k is recorded as the slope obtained by linear fitting, representing the degree of voltage change trend;

[0031] Set Kup and Kdown as the preset voltage rise slope threshold and voltage fall slope threshold respectively, where Kup is a positive number and Kdown is a negative number, and then determine:

[0032] If k > Kup, it is determined that the voltage trend is rising;

[0033] If k < Kdown, it is determined that the voltage trend is falling;

[0034] Otherwise, it is determined that the voltage trend is stable.

[0035] In a specific embodiment of the present invention, generating a pre-scheduling strategy for the substation control unit based on the voltage trend includes:

[0036] For any substation control unit, if the voltage trend is rising or falling, reactive power resource pre-scheduling is performed; when performing reactive power resource pre-scheduling, first form a list of controllable reactive power regulation devices according to the electrical control times and locking states of reactive power compensation devices; traverse the list of controllable reactive power regulation devices of this substation control unit, and sort the devices from high to low according to priority; then, based on the reactive power-voltage sensitivity, predict whether the bus voltage of this station will exceed the limit after the reactive power regulation device is put into operation: if the limit is exceeded, no pre-scheduling strategy is generated; if the limit is not exceeded, a pre-scheduling strategy is generated, that is, the devices in the list of controllable reactive power regulation devices are put into or withdrawn.

[0037] Features and beneficial effects of the present invention:

[0038] 1) Through historical data and real-time monitoring data, combined with machine learning or data-driven models, the present invention can obtain relatively accurate prediction results of future voltage change trends.

[0039] 2) Based on the prediction results, the present invention adjusts the voltage control strategy in advance to avoid voltage fluctuations or overlimits.

[0040] 3) The present invention is applicable to multiple voltage levels (such as 500 kV, 220 kV, 110 kV, etc.) in the substation to ensure global optimal regulation. Description of the Drawings

[0041] Figure 1 It is the overall flowchart of an automatic voltage control method for a substation based on voltage prediction according to an embodiment of the present invention.

[0042] Figure 2 It is the structural schematic diagram of a voltage prediction model in a specific embodiment of the present invention. Detailed Embodiments

[0043] The present invention proposes an automatic voltage control method for a substation based on voltage prediction, which is further described in detail below with reference to the drawings and specific embodiments.

[0044] An embodiment of the present invention proposes an automatic voltage control method for a substation based on voltage prediction, including:

[0045] Using a preset voltage prediction model, obtain the bus voltage prediction values of each substation control unit, and the voltage prediction model is composed of a combination of a Conv1D one-dimensional convolutional neural network and an LSTM long short-term memory network;

[0046] Based on the bus voltage prediction values, determine the voltage trend of the substation control unit;

[0047] Based on the voltage trend, generate a pre-scheduling strategy for the substation control unit.

[0048] In a specific embodiment of the present invention, the automatic voltage control method for a substation based on voltage prediction has an overall process as follows Figure 1 shown, and includes the following steps:

[0049] 1) Establish a corresponding voltage prediction model for each substation control unit.

[0050] In this embodiment, the structure of the voltage prediction model is as follows Figure 2 shown, and includes an input layer, a Conv1D one-dimensional convolutional neural network layer, a pooling layer, an LSTM long short-term memory network layer, a fully connected layer, and an output layer that are connected in sequence.

[0051] Among them, each substation control unit corresponds to a voltage prediction model. In a specific embodiment of the present invention, the number of convolutional kernels in the one-dimensional convolutional neural network layer of each voltage prediction model is 64, and the size of the convolutional kernel is 3; the pooling size of the pooling layer is 2; the number of units in the long short-term memory network layer is 50. Each voltage prediction model is trained separately according to the historical data of the corresponding substation control unit.

[0052] In this embodiment, for any substation control unit, the constructed voltage prediction model is denoted as Model unit , unit = 1,......., N. One substation control unit corresponds to one model, and N is the total number of substation control units. The input of this voltage prediction model is the historical data sequence of active power load, the historical data sequence of transformer tap positions, and the historical state data sequence of capacitors and reactors in the past period of the corresponding substation control unit, and the output is the bus voltage data sequence in the future period of this substation control unit. Among them, the lengths of the past period and the future period may be inconsistent. In a specific embodiment of the present invention, the input of this voltage prediction model is the historical data sequence of active power load, the historical data sequence of transformer tap positions, and the historical state data sequence of capacitors and reactors in the past 1 hour, and the output is the bus voltage data sequence in the future 30 minutes.

[0053] 2) Establish a training set for the voltage prediction model corresponding to each substation control unit.

[0054] In this embodiment, for any substation control unit, the specific steps are as follows:

[0055] 2-1) Obtain the historical data of the substation control unit.

[0056] In this embodiment, historical data of the substation control unit for a set historical period is obtained from the historical database of the EMS (Energy Management System) platform (in this embodiment, the duration of the historical period is at least one year). The historical data includes: historical active load data, historical transformer tap data, historical status data of capacitors and reactors, and historical bus voltage data.

[0057] 2-2) The data obtained in step 2-1) is divided into input data sequences and output data sequences respectively according to a set sliding time window.

[0058] In this embodiment, the length of the sliding time window of the input data in the training set is equal to the length of the past period input to the model (in a specific embodiment of the present invention, it is 1 hour); the length of the sliding time window of the output data is equal to the length of the future period output by the model (in a specific embodiment of the present invention, it is 30 minutes).

[0059] Specifically, for the input data, denote the historical active load data of any substation control unit as P_INPUT unit , unit = 1,...., N; denote the historical transformer tap data of any substation control unit as Oltc_INPUT unit , unit = 1,...., N; denote the historical reactive power compensation capacity data of any substation control unit as Qnorm_INPUT unit , unit = 1,...., N.

[0060] In this embodiment, the historical input data with a length of one year is divided by 5-minute sampling points, and there are a total of P_SUM = 105120 data points. The time stride of an input data sequence TIME_STEPS = 12, then the number of input data sequences obtained after division SAMPLES = (P_SUM - TIME_STEPS + 1) = 105109.

[0061] For the output data, denote the bus voltage data of any substation control unit as V_INPUT unit , unit = 1,...., N, where the bus in this embodiment refers to the 10kV bus. In this embodiment, the historical output data with a length of one year is divided by 5-minute sampling points, and there are a total of P_SUM = 105120 data points. The time stride of an output data sequence TIME_STEPS = 6, then the number of output data sequences obtained after division SAMPLES = (P_SUM - TIME_STEPS + 1) = 10515.

[0062] It should be noted that in this embodiment, for different substation control units, the selected historical period and its length, as well as the sampling frequency, can be different; however, the time window lengths of the obtained input data sequence and output data sequence must be the same.

[0063] 2-3) Construct a training set using the data sequences obtained by partitioning in step 2-2)

[0064] In this embodiment, using the input data sequence and output data sequence obtained by partitioning in step 2-2), an input data sequence for the past one hour (including: the historical data sequence of power load in the past 1 hour, the historical data sequence of transformer tap positions, and the historical state data sequence of capacitors and reactors) and its corresponding output data sequence for the next 30 minutes form a training sample, and all training samples constitute a training set.

[0065] 3) Train the voltage prediction model constructed in step 1) using the training set obtained in step 2) to obtain a trained model.

[0066] In this embodiment, use the training set of each substation control unit to train the voltage prediction model corresponding to this substation control unit. When the set training end condition is reached, obtain the final voltage prediction model of this substation control unit.

[0067] 4) Use the model trained in step 3) to perform bus voltage prediction.

[0068] In this embodiment, for any substation control unit, denote the current moment as Tn, and read the historical data of this substation control unit for the past one hour from the EMS platform historical database; then input this historical data into the voltage prediction model trained in step 3), and this model outputs the bus voltage prediction value of this substation control unit for the next 30 minutes.

[0069] Furthermore, in this embodiment, in order to avoid abnormal values and missing values in the bus voltage prediction value, which may affect the accuracy of subsequent substation strategy calculation, it is necessary to preprocess the bus voltage prediction value, which may specifically include data smoothing and data missing value processing.

[0070] In this embodiment, if the distorted data is not smoothed, it will affect the accuracy of substation calculation. However, fuzzy inference has a certain robustness in dealing with fluctuations and noises in the data. Therefore, without increasing the system complexity, this embodiment designs a simple smoothing method to effectively reduce the influence of noise and optimize the prediction effect. Specifically, if the voltage prediction value V(nd,t) at the t-th moment of the nd-th bus is a distorted data point, it is smoothed according to the following formula:

[0071]

[0072] Among them, V unit (nd, t) represents the predicted voltage value of the nd-th bus at the t-th moment of the substation control unit, and V′ unit (nd, t) represents the predicted voltage value of the nd-th bus at the t-th moment after smoothing processing by the substation control unit.

[0073] Furthermore, missing values will affect the calculation of statistical indicators, resulting in deviations in the analysis results; by filling or processing missing data, data deviations can be reduced to ensure the accuracy of the analysis results. In this embodiment, the linear interpolation method is used for data missing processing. For example, in V unit (nd, t) data, if data is missing at t = t 0 moment, at this time, find the two moments before and after the t 0 moment without missing data and record them as t -1 and t 1 moments respectively, then the missing data can be supplemented by the following calculation:

[0074]

[0075] 5) Determine the voltage trend using the predicted bus voltage value obtained in step 4).

[0076] Specifically, in this embodiment, the regression analysis method is used to perform linear fitting on the predicted bus voltage value obtained in step 4), and the voltage trend is judged according to the slope obtained from the fitting.

[0077] Among them, let k be the slope obtained from the linear fitting, representing the degree of voltage change trend.

[0078] Set Kup and Kdown as the preset voltage rise slope threshold and voltage drop slope threshold respectively, where Kup is a positive number and Kdown is a negative number, and then judge:

[0079] If k > Kup, it is determined that the voltage trend is rising. In this embodiment, the voltage trend flag trend = 1 is returned to indicate that the voltage trend is rising;

[0080] If k < Kdown, it is determined that the voltage trend is falling. In this embodiment, the voltage trend flag trend = 0 is returned to indicate that the voltage trend is falling;

[0081] Otherwise, it is considered that the voltage trend change is not obvious and the voltage is stable. In this embodiment, the voltage trend flag trend = 2 is returned to indicate that the voltage trend is stable.

[0082] In this embodiment, k up and k down can be configured according to the actual requirements of each application site.

[0083] 6) Perform reactive power resource pre-scheduling based on the result of step 5).

[0084] In this embodiment, after the voltage trend of the substation control unit is judged, if the voltage trend is rising or falling at this time, the pre-scheduling strategy calculation of reactive power resources is performed. By pre-adjusting the input of reactive power compensation devices (such as capacitor banks, SVCs, etc.), in this way, when the load fluctuates rapidly, the AVC system can respond earlier without waiting for the voltage deviation to exceed the threshold before making adjustments. This pre-scheduling can help the system maintain a stable voltage level during peak and trough loads, reducing over-regulation and frequent control actions. Specifically:

[0085] For any substation control unit, according to the voltage trend obtained in step 5), if the voltage trend is rising or falling, reactive power resource pre-scheduling is performed.

[0086] When performing reactive power resource pre-scheduling, first form a list of controllable reactive power regulation devices according to the electrical control times and blocking states of reactive power voltage compensation devices. Traverse the list of controllable reactive power regulation devices of this substation control unit, and sort the devices from high to low priority (where devices that are not currently running but can be put into operation are given priority, and devices that are already close to the maximum / minimum limit are ranked behind). Then, according to the sensitivity (generally referring to the reactive power voltage sensitivity, this data can be read from the on-site AVC automatic voltage control system), judge whether the bus voltage of this station will exceed the limit after the reactive power regulation device is put into operation: if the limit is exceeded, no pre-scheduling strategy is generated; if the limit is not exceeded, a pre-scheduling strategy is generated, that is, the devices in the list of controllable reactive power regulation devices are put into or withdrawn.

Claims

1. A substation automatic voltage control method based on voltage prediction, characterized in that: Including: Using a preset voltage prediction model to obtain the predicted bus voltage values of each substation control unit, where the voltage prediction model is composed of a combination of a Conv1D one-dimensional convolutional neural network and an LSTM long short-term memory network; Based on the predicted bus voltage values, determining the voltage trend of the substation control unit; Based on the voltage trend, generating a pre-scheduling strategy for the substation control unit.

2. The method according to claim 1, characterized in that: The voltage prediction model includes an input layer, a one-dimensional convolutional neural network layer, a pooling layer, a long short-term memory network layer, a fully connected layer, and an output layer connected in sequence; the input of the voltage prediction model is the historical data sequence of active power load, the historical data sequence of transformer tap positions, the historical state data sequence of capacitors and reactors in the past period corresponding to the substation control unit, and the output is the future bus voltage data sequence of the substation control unit.

3. The method according to claim 2, characterized in that Before using the preset voltage prediction model to obtain the predicted bus voltage values of each substation control unit, the method further includes: Training the voltage prediction model; The training of the voltage prediction model includes: 1) Establishing a corresponding voltage prediction model for each substation control unit; 2) Establishing a training set for the voltage prediction model corresponding to each substation control unit, and the specific steps are as follows: 2-1) Obtaining the historical data of the substation control unit, including: historical data of active power load, historical data of transformer tap positions, historical state data of capacitors and reactors, and historical bus voltage data; 2-2) Dividing the data obtained in step 2-1) into input data sequences and output data sequences respectively according to a set sliding time window; Among them, the length of the sliding time window of the input data in the training set is equal to the length of the past period of the model input, and the length of the sliding time window of the output data is equal to the length of the future period of the model output; 2-3) Using the results of step 2-2), forming a training sample with each group of input data sequences and their corresponding output data sequences, and all training samples constitute the training set; 3) Training the voltage prediction model constructed in step 1) using the training set obtained in step 2) to obtain the trained model.

4. The method according to claim 3, characterized in that The method further includes: Performing data smoothing and data missing value processing on the predicted bus voltage values.

5. The method according to claim 4, characterized in that The data smoothing includes: For any substation control unit, if the predicted voltage value at the t-th moment of the nd-th bus is a distorted data point, then it is smoothed according to the following formula: Among them, V unit (nd,t) represents the voltage prediction value of the ndth busbar at the tth moment of the substation control unit, V′ unit (nd,t) represents the voltage prediction value of the ndth bus at the tth moment after smoothing by the substation control unit.

6. The method according to claim 4, characterized in that The data missing value processing includes: For each V unit (nd,t) data, if data is missing at t = t0, then find the two times before and after t0 where no missing data occurs and record them as t -1 and t1, and then fill in the missing values ​​according to the following formula:

7. The method according to claim 4, characterized in that The determining the voltage trend of the substation control unit based on the predicted bus voltage values includes: Using regression analysis to perform linear fitting on the predicted bus voltage values, where k is denoted as the slope obtained by linear fitting, indicating the degree of voltage change trend; Setting Kup and Kdown as the preset voltage rise slope threshold and voltage fall slope threshold respectively, where Kup is a positive number and Kdown is a negative number, and then determining: If k > Kup, then determine that the voltage trend is rising; If k < Kdown, then determine that the voltage trend is falling; Otherwise, determine that the voltage trend is stable.

8. The method according to claim 7, characterized in that The generating the pre-scheduling strategy for the substation control unit based on the voltage trend includes: For any substation control unit, if the voltage trend is rising or falling, reactive resource pre-dispatch is performed; When pre-dispatching reactive resources, first form a list of controllable reactive control devices based on the number of electrical controls and the locking status of the reactive voltage compensation equipment; traverse the list of controllable reactive control devices of the substation control unit and sort the devices from high to low priority; then predict whether the bus voltage of this station will be exceeded after the reactive control equipment is put into operation based on the reactive voltage sensitivity: if the limit is exceeded, no pre-dispatching strategy is generated; if the limit is not exceeded, a pre-dispatching strategy is generated, that is, the equipment in the list of controllable reactive control devices is put into operation or withdrawn.

Images