Power distribution network adaptive control method and system based on deep neural network

Abstract

The invention relates to the technical field of smart power grids, in particular to a power distribution network adaptive control method and system based on a deep neural network, and the method comprises the steps: obtaining a data set; a deep neural network is trained through the data set to obtain a prediction model, the input of the prediction model is the influence parameter, and the output of the prediction model is the optimal control coefficient; and generating a real-time optimal control coefficient based on the prediction model, and performing real-time control on a system inverter. According to the technical scheme, the power supply uncertainty data of the renewable energy power supply system are added into the data set, joint optimization is carried out by simultaneously adopting the P-f droop curve and the Q-v droop curve, various uncertainties can be considered, and the method is suitable for practical application.

Description

Technical Field

[0001] The present disclosure relates to the field of smart grid technology, and in particular to a distribution network adaptive control method and system based on a deep neural network. Background Art

[0002] In recent years, the number of distributed energy resources and inverter energy resources deployed in distribution networks has been increasing due to the demand for renewable energy and the reduction in the cost of renewable energy resources such as smart inverters. However, distributed energy resources do not have any rotating parts and will not have an inherent impact on the inertia of the distribution network. If the inverter energy resources are not well controlled, the reduction in inertia will lead to frequency deviation and tripping events such as load imbalance. At the same time, the user-side inverter energy in the distribution system can be controlled through information and operation technology. Therefore, it is necessary to study effective voltage change optimization design and power flow injection optimization design. However, these optimization controls are for snapshots and do not consider the control of the inverter. In addition, this optimization process does not involve time series optimization and cannot effectively cope with dynamic changes in actual operation. Summary of the invention

[0003] In order to solve the problems in the related art, the embodiments of the present disclosure provide a distribution network adaptive control method and system based on deep neural network.

[0004] In a first aspect, the present disclosure provides a method for adaptively controlling a power distribution network based on a deep neural network, including: Get the dataset, including: Acquire power supply uncertainty data of the renewable energy power supply system as an influencing parameter; the influencing parameter includes one or more of the following data: penetration rate of renewable energy to load, load dynamics, irradiance, and temperature; The objective function is determined as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters; the constraint conditions are set as the bus voltage is within a first predetermined range and the parameter value of the inverter droop control is within a second predetermined range, wherein the droop control adopts both the Pf droop curve and the Qv droop curve; Under the constraints, solve the objective function and obtain the optimal control coefficient of droop control; Taking the influencing parameters and the corresponding optimal control coefficients as samples in a data set; Using the data set to train a deep neural network to obtain a prediction model, wherein the input of the prediction model is the influencing parameter and the output is the optimal control coefficient; A real-time optimal control coefficient is generated based on the prediction model, and the system inverter is controlled in real time.

[0005] In one implementation of the present disclosure, the objective function is expressed as: in, It is The energy system in Active power per hour, The entire energy system is hours of power loss, is the weight coefficient, which can be 0 or 1.

[0006] In one implementation of the present disclosure, the constraint condition is expressed as: in, It is a busbar The bus voltage at calculate, satisfy: , in, It is Hourly branch The voltage amplitude, and It is a branch road The resistance and reactance, and It is Hourly branch The active and reactive power, It is The square of the hourly node current; VVC1 and VVC2 are the parameters of the Qv droop curve; VWC1 and VWC2 are the parameters of the Pf droop curve; the symbol “~” indicates satisfaction.

[0007] In an implementation of the present disclosure, after acquiring the data set, the method further includes: Performing anomaly detection on the data set, removing abnormal samples in the data set or removing abnormal data in abnormal samples in the data set; The method of using the data set to train a deep neural network to obtain a prediction model includes: The prediction model is obtained by training a deep neural network using the data set after removing abnormal samples or the data set after removing abnormal data.

[0008] In an implementation of the present disclosure, the anomaly detection adopts a density-based local outlier factor algorithm.

[0009] In one implementation of the present disclosure, the prediction model is an LSTM-GRU model, which is composed of LSTM units and GRU units; the last three layers of the prediction model are a fully connected layer, a Dropout layer and an output layer.

[0010] In one implementation of the present disclosure, the prediction model also includes a normalization layer connected to each layer of the LSTM unit and the GRU unit.

[0011] In a second aspect, the present disclosure provides a distribution network adaptive control system based on a deep neural network, including: The acquisition module is configured to acquire a data set, including: Acquire power supply uncertainty data of the renewable energy power supply system as an influencing parameter; the influencing parameter includes one or more of the following data: penetration rate of renewable energy to load, load dynamics, irradiance, and temperature; The objective function is determined as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters; the constraint conditions are set as the bus voltage is within a first predetermined range and the parameter value of the inverter droop control is within a second predetermined range, wherein the droop control adopts both the Pf droop curve and the Qv droop curve; Under the constraints, solve the objective function and obtain the optimal control coefficient of droop control; Taking the influencing parameters and the corresponding optimal control coefficients as samples in a data set; A training module is configured to train a deep neural network using the data set to obtain a prediction model, wherein the input of the prediction model is the influencing parameter and the output is the optimal control coefficient; The control module is configured to generate a real-time optimal control coefficient based on the prediction model and perform real-time control on the system inverter.

[0012] In one implementation of the present disclosure, the objective function is expressed as: in, It is The energy system in Active power per hour, The entire energy system is hours of power loss, is the weight coefficient, which can be 0 or 1.

[0013] In one implementation of the present disclosure, the constraint condition is expressed as: in, It is a busbar The bus voltage at calculate, satisfy: , in, It is Hourly branch The voltage amplitude, and It is a branch road The resistance and reactance, and It is Hourly branch The active and reactive power, It is The square of the hourly node current; VVC1 and VVC2 are the parameters of the Qv droop curve; VWC1 and VWC2 are the parameters of the Pf droop curve; the symbol “~” indicates satisfaction.

[0014] In one implementation of the present disclosure, it also includes: An anomaly detection module is configured to perform anomaly detection on the data set, remove abnormal samples in the data set or remove abnormal data in abnormal samples in the data set; The training module is configured as follows: The prediction model is obtained by training a deep neural network using the data set after removing abnormal samples or the data set after removing abnormal data.

[0015] In an implementation of the present disclosure, the anomaly detection adopts a density-based local outlier factor algorithm.

[0016] In one implementation of the present disclosure, the prediction model is an LSTM-GRU model, which is composed of LSTM units and GRU units; the last three layers of the prediction model are a fully connected layer, a Dropout layer and an output layer.

[0017] In one implementation of the present disclosure, the prediction model also includes a normalization layer connected to each layer of the LSTM unit and the GRU unit.

[0018] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, and wherein the one or more computer instructions are executed by the processor to implement a method as described in any one of the first aspects.

[0019] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed by a processor, the method described in any one of the first aspects is implemented.

[0020] In a fifth aspect, an embodiment of the present disclosure provides a chip, the chip including a processor, the processor being used to call a computer program in a memory to execute the above-mentioned distribution network adaptive control method based on deep neural network.

[0021] The technical effects provided by the embodiments of the present disclosure may include the following beneficial effects: According to the technical solution provided by the embodiment of the present disclosure, the adaptive control method of the distribution network based on the deep neural network includes obtaining a data set; using the data set to train the deep neural network to obtain a prediction model, wherein the input of the prediction model is the influencing parameter and the output is the optimal control coefficient; generating a real-time optimal control coefficient based on the prediction model, and performing real-time control on the system inverter. In the above technical solution, the power supply uncertainty data of the renewable energy power supply system is added to the data set, and by simultaneously using the Pf droop curve and the Qv droop curve for joint optimization, multiple uncertainties can be considered, which is suitable for practical applications.

[0022] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flow chart of a method for adaptive control of a power distribution network based on a deep neural network according to an embodiment of the present disclosure is shown.

[0024] Figure 2 A structural block diagram of a power distribution network adaptive control system based on a deep neural network according to an embodiment of the present disclosure is shown.

[0025] Figure 3 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0026] Figure 4 A schematic diagram showing the structure of a computer system suitable for implementing the method according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0027] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.

[0028] In the present disclosure, it should be understood that terms such as "include" or "have" are intended to indicate the presence of features, numbers, steps, behaviors, components, parts, or a combination thereof disclosed in the present specification, and are not intended to exclude the possibility that one or more other features, numbers, steps, behaviors, components, parts, or a combination thereof exist or are added.

[0029] It should also be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0030] In recent years, the number of distributed energy resources and inverter energy resources deployed in distribution networks has been increasing due to the demand for renewable energy and the reduction in the cost of renewable energy resources such as smart inverters. However, distributed energy resources do not have any rotating parts and will not have an inherent impact on the inertia of the distribution network. If the inverter energy resources are not well controlled, the reduction in inertia will lead to frequency deviation and tripping events such as load imbalance. At the same time, the user-side inverter energy in the distribution system can be controlled through information and operation technology. Therefore, it is necessary to study effective voltage change optimization design and power flow injection optimization design. However, these optimization controls are for snapshots and do not consider the control of the inverter. In addition, this optimization process does not involve time series optimization and cannot effectively cope with dynamic changes in actual operation.

[0031] In view of the above-mentioned defects, the adaptive control method of the distribution network based on deep neural network provided by the present disclosure includes obtaining a data set; using the data set to train the deep neural network to obtain a prediction model, wherein the input of the prediction model is the influencing parameter and the output is the optimal control coefficient; generating a real-time optimal control coefficient based on the prediction model, and performing real-time control on the system inverter. In the above-mentioned technical solution, the power supply uncertainty data of the renewable energy power supply system is added to the data set, and by simultaneously using the Pf droop curve and the Qv droop curve for joint optimization, multiple uncertainties can be considered, which is suitable for practical applications.

[0032] Figure 1 A flow chart of a method for adaptive control of a power distribution network based on a deep neural network according to an embodiment of the present disclosure is shown.

[0033] like Figure 1 As shown, the distribution network adaptive control method based on deep neural network includes the following steps S110-S130: In step S110, a data set is obtained; In step S120, a deep neural network is trained using the data set to obtain a prediction model, wherein the input of the prediction model is an influencing parameter and the output is an optimal control coefficient; In step S130, a real-time optimal control coefficient is generated based on the prediction model, and the system inverter is controlled in real time.

[0034] In one implementation of the present disclosure, obtaining a data set in step S110 includes: Acquire power supply uncertainty data of the renewable energy power supply system as an influencing parameter; the influencing parameter includes one or more of the following data: penetration rate of renewable energy to load, load dynamics, irradiance, and temperature; The objective function is determined as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters; the constraint conditions are set as the bus voltage is within a first predetermined range and the parameter value of the inverter droop control is within a second predetermined range, wherein the droop control adopts both the Pf droop curve and the Qv droop curve; Under the constraints, solve the objective function and obtain the optimal control coefficient of droop control; The influencing parameters and the corresponding optimal control coefficients are used as samples in a data set.

[0035] The adaptive control method for distribution networks based on deep neural networks in the disclosed embodiments adds power supply uncertainty data of renewable energy power supply systems to a data set. By simultaneously using the Pf droop curve and the Qv droop curve for joint optimization, multiple uncertainties can be considered, which is suitable for practical applications.

[0036] In the disclosed method, the constructed renewable energy power supply system meets the following conditions: System inverter: The energy control mode of the inverter adopts voltage-ford (VCC) control and voltage-watt (VWC) control; Load model: The constant active power and reactive power load model is replaced by a resident dynamic load, where the average active and reactive power load values ​​are constant load values ​​and the dynamic part of the load is used as the uncertain factor of the system.

[0037] In the present disclosure, the objective function is expressed as: in, It is The energy system in Active power per hour, The entire energy system is hours of power loss, is the weight coefficient, which takes a value of 0 or 1, thus converting the objective function into a single-objective optimization. When is 0, the objective function is to minimize the sum of the active powers of N energy systems; when When is 1, the objective function is to minimize the sum of the power losses of N energy systems.

[0038] In the disclosed method, two independent Pf droop curves and Qv droop curves are used to determine the injection of active and reactive power according to the monitored bus voltage, wherein the control parameters of the Qv droop curve are symmetrical, counted as VVC1 and VVC2, and the control parameters of the Pf droop curve are counted as VWC1 and VWC2.

[0039] The slope of the Qv droop curve is m i The calculation formula is: ; The slope of the Pf droop curve is m p The calculation formula is: ; Reactive power According to the Qv droop curve, the formula is: ; Active Power According to the Pf droop curve, the formula is: ; in, Indicates the maximum value of the inverter reactive power; Indicates the maximum value of the inverter active power; Indicates The energy system in +1 hour reactive power flow; Indicates The energy system in +1 hour of active power flow; Indicates voltage amplitude; Indicates the dead zone voltage; in, Indicates the middle point of the dead zone. When the voltage is higher than When the reactive power The above calculation formula is replaced by: ; In the present disclosure, and This can be calculated from the active power circuit loss. By using the openDSS software, the active power loss can be directly derived from the communication port.

[0040] In one implementation of the present disclosure, the constraint condition is expressed as: in, It is a busbar The bus voltage at calculate, satisfy: , in, It is Hourly branch The voltage amplitude, and It is a branch road The resistance and reactance, and It is Hourly branch The active and reactive power, It is The square of the node current in an hour; the symbol “~” indicates satisfaction.

[0041] In the disclosed method, the variation range of the bus voltage is controlled within 0.05 pu. When the bus voltage exceeds the range, a maximum value is artificially set for the objective function value. In the disclosed method, a particle swarm-based optimization algorithm is used to solve the objective function. The solution process can refer to the prior art and is not the focus of the disclosed method.

[0042] In the disclosed method, the penetration rate of renewable energy to load The calculation formula is as follows: ; in, is the number of buses in the circuit, It is a busbar The average active power on.

[0043] In the disclosed method, the penetration rate of renewable energy to load, load dynamics, irradiance, and temperature in the power supply uncertainty data are generated according to the preset ranges of different uncertainties.

[0044] In the disclosed method, power supply uncertainty data (influencing parameters) and optimal control coefficients are written into the data center as a data set. A sample in the data set includes the corresponding influencing parameters and the optimal control coefficient obtained by solving the objective function under the influencing parameters.

[0045] In an implementation of the present disclosure, after acquiring the data set, the method further includes: Performing anomaly detection on the data set, removing abnormal samples in the data set or removing abnormal data in abnormal samples in the data set; The method of using the data set to train a deep neural network to obtain a prediction model includes: The prediction model is obtained by training a deep neural network using the data set after removing abnormal samples or the data set after removing abnormal data.

[0046] In the disclosed method, clean data is obtained by deleting abnormal samples or abnormal data of the data set obtained in step S110, so as to reduce the influence of model errors on the generated data and improve the prediction accuracy of the prediction model.

[0047] In an implementation of the present disclosure, the anomaly detection adopts a density-based local outlier factor algorithm.

[0048] Specifically, the local outlier factor algorithm calculates the density between different control variables and finds low-density data that are identified as abnormal. The local outlier factor score of is expressed as: in, Represents the local reachability density of the minimum number of objects selected from the dataset. The set of distances representing objects identifies anomalies as:

[0049] in, The LOF score is calculated percentage, Indicates the proportion of abnormal points. If the LOF score is higher than the threshold, the point will be deleted.

[0050] In one implementation of the present disclosure, the prediction model is an LSTM-GRU model, which is composed of LSTM units and GRU units; the last three layers of the prediction model are a fully connected layer, a Dropout layer, and an output layer. The fully connected layer is used to reshape the output into one-dimensional data, and the Dropout layer is used to improve the robustness of the prediction model.

[0051] Since the optimization of the control coefficients of the Pf droop curve and the Qv droop curve are both nonlinear optimization problems, the use of deep neural networks, especially the LSTM-GRU model, can capture complex nonlinear relationships through large-scale data training, improving the accuracy and robustness of the model. In addition, the system inverter can be controlled in real time based on the real-time optimal control coefficient, large-scale data can be processed in real time, and the learning ability of the LSTM-GRU model can be used to quickly generate adaptive optimization strategies to improve the system response speed and optimization effect.

[0052] In the disclosed method, by stacking 3 layers of LSTM and GRU units, the data features in the data set are learned and extracted, the output is evaluated using the mean square error (MSE), and the optimal control coefficients VVC1, ​​VVC2, VWC1 and VWC2 of the droop control are predicted.

[0053] In the LSTM unit: (1) the forget gate is used to filter redundant information, (2) a new candidate feature is created and the state of the last feature is updated, and (3) the unit output is calculated according to the tanh and sigmoid functions. In the GRU unit: the reset gate and update gate are used, and the reset gate is responsible for saving renewable energy information.

[0054] In one implementation of the present disclosure, the prediction model also includes a normalization layer connected to each layer of the LSTM unit and the GRU unit.

[0055] In the disclosed method, considering that the data set is nonlinearly distributed, in order to avoid overfitting of the model, a regularization scheme is adopted, and a normalization layer is connected to each layer of the LSTM unit and the GRU unit to form an LSTM-GN model with GRU and normalization layers. The normalization layer fits the features of each layer to a normal distribution, and its constraints are ,in and denote the expected value and variance respectively.

[0056] In one implementation of the present disclosure, the prediction model output is expressed as: , which is simplified to: in, Indicates that it has parameters A nonlinear function of the input (such as the number of layers and the number of epochs during training) that can learn and filter features from the input. and are weight and bias respectively.

[0057] The distribution network adaptive control method based on deep neural network of the embodiment of the present disclosure determines the objective function as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters, while taking into account the variation range of the bus voltage in the smart distribution network, determining the optimal control coefficient, and then training the LSTM-GRU model and using the model for real-time prediction. It can minimize the computational complexity of the distribution network voltage control optimization while maximizing the collaborative operation performance, effectively solve the inverter optimization control problem in the distribution network, realize the real-time prediction and dynamic adjustment of the inverter control coefficient, and effectively improve the adaptive optimization capability of the distribution network.

[0058] Figure 2 The structural block diagram of the power distribution network adaptive control system based on deep neural network according to the embodiment of the present disclosure is shown. The device can be implemented as part or all of the electronic device through software, hardware or a combination of both.

[0059] like Figure 2 As shown, a distribution network adaptive control system 200 based on a deep neural network includes: The acquisition module 210 is configured to acquire a data set, including: Acquire power supply uncertainty data of the renewable energy power supply system as an influencing parameter; the influencing parameter includes one or more of the following data: penetration rate of renewable energy to load, load dynamics, irradiance, and temperature; The objective function is determined as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters; the constraint conditions are set as the bus voltage is within a first predetermined range and the parameter value of the inverter droop control is within a second predetermined range, wherein the droop control adopts both the Pf droop curve and the Qv droop curve; Under the constraints, solve the objective function and obtain the optimal control coefficient of droop control; Taking the influencing parameters and the corresponding optimal control coefficients as samples in a data set; A training module 220 is configured to train a deep neural network using the data set to obtain a prediction model, wherein the input of the prediction model is the influencing parameter and the output is the optimal control coefficient; The control module 230 is configured to generate a real-time optimal control coefficient based on the prediction model and perform real-time control on the system inverter.

[0060] The adaptive control method for distribution networks based on deep neural networks in the disclosed embodiments adds power supply uncertainty data of renewable energy power supply systems to a data set. By simultaneously using the Pf droop curve and the Qv droop curve for joint optimization, multiple uncertainties can be considered, which is suitable for practical applications.

[0061] In the disclosed method, the constructed renewable energy power supply system meets the following conditions: System inverter: The energy control mode of the inverter adopts voltage-ford (VCC) control and voltage-watt (VWC) control; Load model: The constant active power and reactive power load model is replaced by a resident dynamic load, where the average active and reactive power load values ​​are constant load values ​​and the dynamic part of the load is used as the uncertain factor of the system.

[0062] In one implementation of the present disclosure, the objective function is expressed as: in, It is The energy system in Active power per hour, The entire energy system is hours of power loss, is the weight coefficient, which takes a value of 0 or 1, thus converting the objective function into a single-objective optimization; that is, when When is 0, the objective function is to minimize the sum of the active powers of N energy systems; when When is 1, the objective function is to minimize the sum of the power losses of N energy systems.

[0063] In the disclosed method, two independent Pf droop curves and Qv droop curves are used to determine the injection of active and reactive power according to the monitored bus voltage, wherein the control parameters of the Qv droop curve are symmetrical, counted as VVC1 and VVC2, and the control parameters of the Pf droop curve are counted as VWC1 and VWC2.

[0064] The slope of the Qv droop curve is m i The calculation formula is: ; The slope of the Pf droop curve is m p The calculation formula is: ; Reactive power According to the Qv droop curve, the formula is: ; Active Power According to the Pf droop curve, the formula is: ; in, Indicates the maximum value of the inverter reactive power; Indicates the maximum value of the inverter active power; Indicates The energy system in +1 hour reactive power flow; Indicates The energy system in +1 hour of active power flow; Indicates voltage amplitude; Indicates the dead zone voltage; in, Indicates the middle point of the dead zone. When the voltage is higher than When the reactive power The above calculation formula is replaced by: ; In the present disclosure, and This can be calculated from the active power circuit loss. By using the openDSS software, the active power loss can be directly derived from the communication port.

[0065] In one implementation of the present disclosure, the constraint condition is expressed as: in, It is a busbar The bus voltage at calculate, satisfy: , in, It is Hourly branch The voltage amplitude, and It is a branch road The resistance and reactance, and It is Hourly branch The active and reactive power, It is The square of the node current in an hour; the symbol “~” indicates satisfaction.

[0066] In the disclosed method, the variation range of the bus voltage is controlled within 0.05 pu. When the bus voltage exceeds the range, a maximum value is artificially set for the objective function value. In the disclosed method, a particle swarm-based optimization algorithm is used to solve the objective function. The solution process can refer to the prior art and is not the focus of the disclosed method.

[0067] In the disclosed method, the penetration rate of renewable energy to load The calculation formula is as follows: ; in, is the number of buses in the circuit, It is a busbar The average active power on.

[0068] In the disclosed method, the penetration rate of renewable energy to load, load dynamics, irradiance, and temperature in the power supply uncertainty data are generated according to the preset ranges of different uncertainties.

[0069] In the disclosed method, power supply uncertainty data (influencing parameters) and optimal control coefficients are written into the data center as a data set. A sample in the data set includes the corresponding influencing parameters and the optimal control coefficient obtained by solving the objective function under the influencing parameters.

[0070] In one implementation of the present disclosure, the distribution network adaptive control system based on a deep neural network further includes: An anomaly detection module is configured to perform anomaly detection on the data set, remove abnormal samples in the data set or remove abnormal data in abnormal samples in the data set; The training module is configured as follows: The prediction model is obtained by training a deep neural network using the data set after removing abnormal samples or the data set after removing abnormal data.

[0071] In the disclosed method, clean data is obtained by deleting abnormal samples or abnormal data in the acquired data set, so as to reduce the influence of model errors on the generated data and improve the prediction accuracy of the prediction model.

[0072] In an implementation of the present disclosure, the anomaly detection adopts a density-based local outlier factor algorithm.

[0073] Specifically, the local outlier factor algorithm calculates the density between different control variables and finds low-density data that are identified as abnormal. The local outlier factor score of is expressed as: in, Represents the local reachability density of the minimum number of objects selected from the dataset. The set of distances representing objects identifies anomalies as:

[0074] in, The LOF score is calculated percentage, Indicates the proportion of abnormal points. If the LOF score is higher than the threshold, the point will be deleted.

[0075] In one implementation of the present disclosure, the prediction model is an LSTM-GRU model, which is composed of LSTM units and GRU units; the last three layers of the prediction model are a fully connected layer, a Dropout layer, and an output layer. The fully connected layer is used to reshape the output into one-dimensional data, and the Dropout layer is used to improve the robustness of the prediction model.

[0076] Since the optimization of the control coefficients of the Pf droop curve and the Qv droop curve are both nonlinear optimization problems, the use of deep neural networks, especially the LSTM-GRU model, can capture complex nonlinear relationships through large-scale data training, improving the accuracy and robustness of the model. In addition, the system inverter can be controlled in real time based on the real-time optimal control coefficient, large-scale data can be processed in real time, and the learning ability of the LSTM-GRU model can be used to quickly generate adaptive optimization strategies to improve the system response speed and optimization effect.

[0077] In the disclosed method, by stacking 3 layers of LSTM and GRU units, the data features in the data set are learned and extracted, the output is evaluated using the mean square error (MSE), and the optimal control coefficients VVC1, ​​VVC2, VWC1 and VWC2 of the droop control are predicted.

[0078] In the LSTM unit: (1) the forget gate is used to filter redundant information, (2) a new candidate feature is created and the state of the last feature is updated, and (3) the unit output is calculated according to the tanh and sigmoid functions. In the GRU unit: the reset gate and update gate are used, and the reset gate is responsible for saving renewable energy information.

[0079] In one implementation of the present disclosure, the prediction model also includes a normalization layer connected to each layer of the LSTM unit and the GRU unit.

[0080] In the disclosed method, considering that the data set is nonlinearly distributed, in order to avoid overfitting of the model, a regularization scheme is adopted, and a normalization layer is connected to each layer of the LSTM unit and the GRU unit to form an LSTM-GN model with GRU and normalization layers. The normalization layer fits the features of each layer to a normal distribution, and its constraints are ,in and denote the expected value and variance respectively.

[0081] In one implementation of the present disclosure, the prediction model output is expressed as: , which is simplified to: in, Indicates that it has parameters A nonlinear function of the input (such as the number of layers and the number of epochs during training) that can learn and filter features from the input. and are weight and bias respectively.

[0082] The distribution network adaptive control method based on deep neural network of the embodiment of the present disclosure determines the objective function as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters, while taking into account the variation range of the bus voltage in the smart distribution network, determining the optimal control coefficient, and then training the LSTM-GRU model and using the model for real-time prediction. It can minimize the computational complexity of the distribution network voltage control optimization while maximizing the collaborative operation performance, effectively solve the inverter optimization control problem in the distribution network, realize the real-time prediction and dynamic adjustment of the inverter control coefficient, and effectively improve the adaptive optimization capability of the distribution network.

[0083] The present disclosure also discloses an electronic device, Figure 3 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0084] like Figure 3 As shown, the electronic device includes a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method according to an embodiment of the present disclosure.

[0085] The distribution network adaptive control method based on deep neural network includes: Get the dataset, including: Acquire power supply uncertainty data of the renewable energy power supply system as an influencing parameter; the influencing parameter includes one or more of the following data: penetration rate of renewable energy to load, load dynamics, irradiance, and temperature; The objective function is determined as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters; the constraint conditions are set as the bus voltage is within a first predetermined range and the parameter value of the inverter droop control is within a second predetermined range, wherein the droop control adopts both the Pf droop curve and the Qv droop curve; Under the constraints, solve the objective function and obtain the optimal control coefficient of droop control; Taking the influencing parameters and the corresponding optimal control coefficients as samples in a data set; Using the data set to train a deep neural network to obtain a prediction model, wherein the input of the prediction model is the influencing parameter and the output is the optimal control coefficient; A real-time optimal control coefficient is generated based on the prediction model, and the system inverter is controlled in real time.

[0086] In one implementation of the present disclosure, the objective function is expressed as: in, It is The energy system in Active power per hour, The entire energy system is hours of power loss, is the weight coefficient, which takes a value of 0 or 1, thus converting the objective function into a single-objective optimization; that is, when When is 0, the objective function is to minimize the sum of the active powers of N energy systems; when When is 1, the objective function is to minimize the sum of the power losses of N energy systems.

[0087] In one implementation of the present disclosure, the constraint condition is expressed as: in, It is a busbar The bus voltage at calculate, satisfy: , in, It is Hourly branch The voltage amplitude, and It is a branch road The resistance and reactance, and It is Hourly branch The active and reactive power, It is The square of the hourly node current; VVC1 and VVC2 are the parameters of the Qv droop curve; VWC1 and VWC2 are the parameters of the Pf droop curve; the symbol “~” indicates satisfaction.

[0088] In an implementation of the present disclosure, after acquiring the data set, the method further includes: Performing anomaly detection on the data set, removing abnormal samples in the data set or removing abnormal data in abnormal samples in the data set; The method of using the data set to train a deep neural network to obtain a prediction model includes: The prediction model is obtained by training a deep neural network using the data set after removing abnormal samples or the data set after removing abnormal data.

[0089] In an implementation of the present disclosure, the anomaly detection adopts a density-based local outlier factor algorithm.

[0090] In one implementation of the present disclosure, the prediction model is an LSTM-GRU model, which is composed of LSTM units and GRU units; the last three layers of the prediction model are a fully connected layer, a Dropout layer and an output layer.

[0091] In one implementation of the present disclosure, the prediction model also includes a normalization layer connected to each layer of the LSTM unit and the GRU unit.

[0092] Figure 4 A schematic diagram showing the structure of a computer system suitable for implementing the method according to an embodiment of the present disclosure is shown.

[0093] like Figure 4 As shown, the computer system includes a processing unit, which can execute various methods in the above-mentioned embodiments according to a program stored in a read-only memory (ROM) or a program loaded from a storage part into a random access memory (RAM). In the RAM, various programs and data required for the operation of the computer system are also stored. The processing unit, ROM and RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0094] The following components are connected to the I / O interface: an input part including a keyboard, a mouse, etc.; an output part including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage part including a hard disk, etc.; and a communication part including a network interface card such as a LAN card, a modem, etc. The communication part performs a communication process via a network such as the Internet. The drive is also connected to the I / O interface as needed. Removable media, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on the drive as needed so that the computer program read therefrom can be installed into the storage part as needed. Among them, the processing unit can be implemented as a processing unit such as a CPU, a GPU, a TPU, an FPGA, an NPU, etc.

[0095] In particular, according to an embodiment of the present disclosure, the method described above can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the above method. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium.

[0096] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0097] The units or modules involved in the embodiments described in the present disclosure may be implemented by software or programmable hardware. The units or modules described may also be set in a processor, and the names of these units or modules do not constitute limitations on the units or modules themselves in some cases.

[0098] As another aspect, the present disclosure further provides a chip, which includes at least one processor and can be used to implement the method involved in the above system embodiment.

[0099] In one possible design, the chip also includes a memory, which is used to store program instructions and data, and the memory is located inside or outside the processor.

[0100] As another aspect, the present disclosure further provides a computer-readable storage medium, which may be a computer-readable storage medium included in the electronic device or computer system in the above embodiment; or a computer-readable storage medium that exists independently and is not assembled into a device. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the method described in the present disclosure.

[0101] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present disclosure is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other.

Claims

1. A distribution network adaptive control method based on deep neural network, characterized in that: include: Get the dataset, including: Obtaining power supply uncertainty data of renewable energy power supply systems as influencing parameters; The influencing parameters include one or more of the following data: penetration rate of renewable energy to load, load dynamics, irradiance, temperature; The objective function is determined as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters; the constraint conditions are set as the bus voltage is within a first predetermined range and the parameter value of the inverter droop control is within a second predetermined range, wherein the droop control adopts both the Pf droop curve and the Qv droop curve; Under the constraints, solve the objective function and obtain the optimal control coefficient of droop control; Taking the influencing parameters and the corresponding optimal control coefficients as samples in a data set; Using the data set to train a deep neural network to obtain a prediction model, wherein the input of the prediction model is the influencing parameter and the output is the optimal control coefficient; A real-time optimal control coefficient is generated based on the prediction model, and the system inverter is controlled in real time.

2. The power distribution network adaptive control method according to claim 1, characterized in that: The objective function is expressed as: in, It is The energy system in Active power per hour, The entire energy system is hours of power loss, is the weight coefficient, which can be 0 or 1.

3. The adaptive control method for power distribution network according to claim 1, characterized in that: The constraints are expressed as: in, It is a busbar The bus voltage at calculate, satisfy: ， in, It is Hourly branch The voltage amplitude, and It is a branch road The resistance and reactance, and It is Hourly branch The active and reactive power, It is The square of the hourly node current; VVC1 and VVC2 are the parameters of the Qv droop curve; VWC1 and VWC2 are the parameters of the Pf droop curve; the symbol "~" indicates satisfaction.

4. The power distribution network adaptive control method according to claim 1, characterized in that: After acquiring the data set, the method further includes: Performing anomaly detection on the data set, removing abnormal samples in the data set or removing abnormal data in abnormal samples in the data set; The method of using the data set to train a deep neural network to obtain a prediction model includes: The prediction model is obtained by training a deep neural network using the data set after removing abnormal samples or the data set after removing abnormal data.

5. The power distribution network adaptive control method according to claim 4, characterized in that: The anomaly detection adopts a density-based local outlier factor algorithm.

6. The power distribution network adaptive control method according to claim 1, characterized in that: The prediction model is an LSTM-GRU model, which is composed of LSTM units and GRU units; the last three layers of the prediction model are a fully connected layer, a Dropout layer and an output layer.

7. The power distribution network adaptive control method according to claim 6, characterized in that: The prediction model also includes a normalization layer connected to each layer of LSTM units and GRU units.

8. A distribution network adaptive control system based on deep neural network, characterized in that: include: The acquisition module is configured to acquire a data set, including: Acquire power supply uncertainty data of the renewable energy power supply system as an influencing parameter; the influencing parameter includes one or more of the following data: penetration rate of renewable energy to load, load dynamics, irradiance, and temperature; The objective function is determined as maximizing the active power of the system inverter and minimizing the power loss under given influencing parameters; the constraint conditions are set as the bus voltage is within a first predetermined range and the parameter value of the inverter droop control is within a second predetermined range, wherein the droop control adopts both the Pf droop curve and the Qv droop curve; Under the constraints, solve the objective function and obtain the optimal control coefficient of droop control; Taking the influencing parameters and the corresponding optimal control coefficients as samples in a data set; A training module is configured to train a deep neural network using the data set to obtain a prediction model, wherein the input of the prediction model is the influencing parameter and the output is the optimal control coefficient; The control module is configured to generate a real-time optimal control coefficient based on the prediction model and perform real-time control on the system inverter.

9. The power distribution network adaptive control system according to claim 8, characterized in that: The objective function is expressed as: in, It is The energy system in Active power per hour, The entire energy system is hours of power loss, is the weight coefficient, which can be 0 or 1.

10. The adaptive control system for power distribution network according to claim 8, characterized in that: The constraints are expressed as: in, It is a busbar The bus voltage at calculate, satisfy: ， in, It is Hourly branch The voltage amplitude, and It is a branch road The resistance and reactance, and It is Hourly branch The active and reactive power, It is The square of the hourly node current; VVC1 and VVC2 are the parameters of the Qv droop curve; VWC1 and VWC2 are the parameters of the Pf droop curve; the symbol "~" indicates satisfaction.

11. The adaptive control system for power distribution network according to claim 8, characterized in that: Also includes: An anomaly detection module is configured to perform anomaly detection on the data set, remove abnormal samples in the data set or remove abnormal data in abnormal samples in the data set; The training module is configured as follows: The prediction model is obtained by training a deep neural network using the data set after removing abnormal samples or the data set after removing abnormal data.

12. The adaptive control system for power distribution network according to claim 11, characterized in that: The anomaly detection adopts a density-based local outlier factor algorithm.

13. The power distribution network adaptive control system according to claim 8, characterized in that: The prediction model is an LSTM-GRU model, which is composed of LSTM units and GRU units; the last three layers of the prediction model are a fully connected layer, a Dropout layer and an output layer.

14. The adaptive control system for power distribution network according to claim 13, characterized in that: The prediction model also includes a normalization layer connected to each layer of LSTM units and GRU units.

15. An electronic device, characterized in that: The method comprises a memory and a processor; wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method according to any one of claims 1 to 7.

16. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

17. A chip, characterized in that: include: At least one processor, configured to implement the method according to any one of claims 1 to 7.

Images