Power distribution network terminal power quality treatment cluster system and power quality treatment method

Abstract

The application relates to the technical field of power quality treatment, and discloses a power quality treatment cluster system at the end of a power distribution network and a power quality treatment method. The power quality treatment cluster system comprises a decision unit, an inverter and a control module. The decision unit is used for encoding operation data of an electric network to obtain electric network characteristics, encoding operation data of an inverter to obtain inverter characteristics, and constructing graph structure data based on an electric network topology graph. The graph structure data, the electric network characteristics and the inverter characteristics are input into a first model, and then the first model outputs the value of reactive power required to be emitted by the inverter. The control module is used for emitting a control instruction to the inverter. After receiving the control instruction, the inverter can adjust the reactive power to the reactive power required to be emitted output by the decision unit. Each self-decision control module takes the voltage at the inverter grid connection point as a treatment target, and realizes voltage treatment of the entire transformer area through a point-to-plane control method.

Description

Technical Field

[0001] This invention relates to the field of power quality management technology, and more specifically, to a power quality management cluster system and power quality management method at the end of a distribution network. Background Technology

[0002] The grid-connected operation of large-scale distributed photovoltaic (PV) power generation systems within a distribution area may cause power flow reversal issues, leading to voltage rise or overvoltage at the PV system's point of common coupling (PCC). Voltage rise not only affects the power quality for local users but also increases losses in transmission and distribution equipment such as lines and transformers, causing system overload. Increasing the reactive power of the PV inverters can increase voltage drop, thereby suppressing voltage rise and lowering the voltage. The PV regulation strategy reads the inverter status and real-time grid operation data in real time and determines whether the voltage exceeds the set voltage limit. When a sustained voltage exceedance is detected, the strategy autonomously analyzes and calculates the required reactive power output based on grid and inverter data, and sends a control command to the inverter. When the voltage returns to normal, the strategy controls the inverter to slowly reduce reactive power output, meeting voltage compliance requirements while minimizing impact on the power factor. Since the inverters are distributed across multiple grid connection points with varying parameters and operating states, unified closed-loop control based solely on grid voltage is not feasible. Summary of the Invention

[0003] This invention provides a power quality management cluster system and power quality management method for the end of a distribution network, solving technical problems in related technologies.

[0004] This invention provides a power quality management cluster system for the end of a distribution network, comprising:

[0005] A self-decision control device, used to control the reactive power output of the inverter, includes:

[0006] The power grid operation data reading unit is used to read the operation data of the power grid to which the inverter is connected;

[0007] Inverter operation data acquisition unit, which is used to collect the inverter's operation data;

[0008] The decision-making unit is used to encode the power grid's operational data to obtain power grid characteristics;

[0009] Inverter characteristics are obtained by encoding the inverter's operating data.

[0010] Based on the power grid topology, graph structure data is constructed. The graph structure data includes nodes and edges connecting nodes. A node represents an inverter or a power grid. There are edges between the nodes representing the power grid and all nodes representing the inverters. There are edges between the nodes corresponding to the inverters connected to the same grid connection point.

[0011] The graph structure data, grid characteristics, and inverter characteristics are input into the first model, and then the first model outputs the value of reactive power that the inverter needs to generate;

[0012] An online training control unit is used to control the online training of the first model. The online training control unit inputs the power grid characteristics of n time intervals between the current time into the second model. The second model outputs the start time of the online training of the first model and the number of experience points collected during the training in the time interval between the current time and the next time interval.

[0013] The first model is trained online based on the start time of online training output by the second model and the number of experience points collected during training.

[0014] The control module is used to send control commands to the inverter. After receiving the control commands, the inverter can adjust the reactive power to the reactive power required by the decision unit.

[0015] Furthermore, the calculation formula for the first model is as follows:

[0016]

[0017]

[0018]

[0019]

[0020]

[0021] This represents the output vector, where one component of the output vector corresponds to a probability value of a discrete value of reactive power. This represents the hidden feature of the i-th node, which is the node corresponding to the inverter controlled by the decision-making unit. and Let i and j represent the node features of the i-th and j-th nodes, respectively. This represents the weight parameters of the first hidden layer. This represents the weight vector of the first hidden layer. Let represent the set of nodes that are connected to the i-th node by an edge. This represents the activation function. Indicates attention weights. express Activation function Represents the sigmoid function;

[0022] The node characteristics of a node in a power grid are equal to the characteristics of the power grid, and the node characteristics of a node in an inverter are equal to the characteristics of the inverter.

[0023] Furthermore, the calculation formula for the second model is as follows:

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030] in, and Let n and L represent the 1st and Lth power grid features, respectively, where n ≥ L ≥ 1. When L = n, the power grid feature at the current moment is input; when L = 1, the power grid feature at the (n-1)th moment before the current moment is input. , , and All represent trainable weight parameters. , , and All of these represent trainable bias parameters. Represents the dot product. Indicates the first The first intermediate feature, Indicates the first A second intermediate feature, Indicates the first The third intermediate feature, and They represent the first -1 and the first Each output feature Let represent the second output vector, where the value of the i-th component represents the probability value of the i-th second action, and tanh represents the tanh function. This represents the Sigmoid function.

[0031] Furthermore, the decision model is trained using Q-learning, and the state of Q-learning is defined as the set of environmental information from time tn to time t.

[0032] The reward function for Q-learning is:

[0033]

[0034] in, It is the reward obtained by taking the second action chosen between time t and t+1. It is the resource utilization rate of the processor of the self-decision control device from time t to t+1.

[0035] Furthermore, the formula for calculating processor resource utilization is as follows:

[0036]

[0037] CPU time refers to the time that the processor spends executing programs and operating system code within a certain period of time;

[0038] Total CPU time refers to the total running time of the processor within this time period.

[0039] Furthermore, the online training steps for the first model include:

[0040] Step 101, Build network, Network replication first model;

[0041] Step 102: Randomly obtain an experience from the experience pool. The experience includes the state at time t. The first action at time t The state at time t+1 Execute the first action Rewards ;

[0042] Then calculate the state. Down The network output corresponds to probability value ;

[0043]

[0044] express Down The highest probability value output by the network;

[0045] Step 103, according to and Calculate the loss value and update the first model;

[0046]

[0047] express The output of the first model corresponds to The probability value;

[0048] It is the discount factor, a value between 0 and 1;

[0049] The first action originates from the first action space, and a first action in the first action space represents a discrete value of reactive power.

[0050] Step 104: Update at fixed network update intervals. The network is configured so that its parameters are the same as those of the current first model;

[0051] Step 105 continues until the number of experiences collected equals the number of experiences collected in this training session, at which point the process terminates.

[0052] Furthermore, at each moment, the first action with the highest probability value is selected.

[0053] Furthermore, rewards The calculation formula is as follows:

[0054]

[0055] in This represents the maximum value of the grid voltage deviation from time t to t+1. Indicates the standard voltage value. This represents the maximum value of the power grid frequency deviation from time t to t+1. This indicates the standard value of the frequency.

[0056] Furthermore, the state refers to the input of the first model.

[0057] This invention provides a power quality management method, comprising the following steps:

[0058] Step 100: Read the operating data of the power grid to which the inverter is connected;

[0059] Step 200, Inverter Operation Data Acquisition Unit, is used to collect the inverter's operation data.

[0060] Step 300: Encode the power grid operation data to obtain power grid characteristics.

[0061] Step 400: Encode the inverter's operating data to obtain inverter characteristics;

[0062] Step 500: Construct graph structure data based on the power grid topology. The graph structure data includes nodes and edges connecting nodes. A node represents an inverter or a power grid. There are edges between the nodes representing the power grid and all nodes representing the inverters. There are edges between the nodes corresponding to the inverters connected to the same grid connection point.

[0063] Step 600: Input the graph structure data, grid characteristics and inverter characteristics into the first model, and then the first model outputs the value of reactive power that the inverter needs to generate.

[0064] The beneficial effects of this invention are as follows: each self-decision control module takes the inverter grid connection point voltage as the control target and achieves voltage control of the entire distribution area through a point-to-area control method. Attached Figure Description

[0065] Figure 1 This is a flowchart of the inverter adjustment process of the present invention. Detailed Implementation

[0066] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0067] This invention provides a local self-decision distributed photovoltaic control architecture, in which each inverter controls the output of reactive power through a corresponding self-decision control device.

[0068] At least one embodiment of the present invention discloses a power quality management cluster system for the end of a distribution network, comprising:

[0069] A self-decision control device, used to control the reactive power output of the inverter, includes:

[0070] The power grid operation data reading unit is used to read the operation data of the power grid to which the inverter is connected;

[0071] Inverter operation data acquisition unit, which is used to collect the inverter's operation data;

[0072] The power grid's operational data includes:

[0073] I. Voltage:

[0074] The effective values ​​of line voltage and phase voltage;

[0075] Phase and frequency of voltage waveform;

[0076] Voltage imbalance (voltage difference between different phases);

[0077] Voltage fluctuations and flicker;

[0078] II. Current:

[0079] Effective value of current per phase;

[0080] The phase of the current waveform;

[0081] Zero-sequence current;

[0082] Current harmonic content;

[0083] III. Power:

[0084] Active power;

[0085] Reactive power;

[0086] Apparent power;

[0087] Power factor;

[0088] IV. Electrical Energy:

[0089] Active electrical energy;

[0090] Reactive power;

[0091] V. Frequency deviation.

[0092] The inverter's operating data includes:

[0093] Voltage – the voltage output by the inverter and the grid voltage. This includes phase voltage and line voltage.

[0094] Current – ​​the current output from the inverter to the grid. Typically, the current in each phase is monitored.

[0095] Frequency – The inverter's output frequency must match the grid frequency. In most countries, the grid frequency is 50Hz or 60Hz.

[0096] Power factor (PF) represents the ratio of actual power (active power) to apparent power. An ideal power factor is 1, meaning that all electricity is used efficiently.

[0097] Active power – the power actually consumed by the load, measured in kilowatts (kW).

[0098] Reactive power is the energy exchanged between inductor and capacitor components in an AC circuit, measured in kilovars (kVar).

[0099] Apparent power is the vector sum of active and reactive power, measured in kilovolt-amperes (kVA).

[0100] Harmonic Distortion (THD) - Harmonics are the non-ideal components of a sinusoidal waveform that cause waveform distortion. Total Harmonic Distortion (THD) measures the degree of influence of harmonics on the fundamental frequency.

[0101] Operating temperature of key components inside the inverter, such as IGBTs.

[0102] The total electricity generated by an inverter is usually measured in kilowatt-hours (kWh).

[0103] The decision-making unit is used to encode the power grid's operational data to obtain power grid characteristics;

[0104] Inverter characteristics are obtained by encoding the inverter's operating data.

[0105] Based on the power grid topology, graph structure data is constructed. The graph structure data includes nodes and edges connecting nodes. A node represents an inverter or a power grid. There are edges between the nodes representing the power grid and all nodes representing the inverters. There are edges between the nodes corresponding to the inverters connected to the same grid connection point.

[0106] The graph structure data, grid characteristics, and inverter characteristics are input into the first model, and then the first model outputs the value of reactive power that the inverter needs to generate;

[0107] The calculation formula for the first model is as follows:

[0108]

[0109]

[0110]

[0111]

[0112]

[0113] This represents the output vector, where one component of the output vector corresponds to a probability value of a discrete value of reactive power. This represents the hidden feature of the i-th node, which is the node corresponding to the inverter controlled by the decision-making unit. and Let i and j represent the node features of the i-th and j-th nodes, respectively. This represents the weight parameters of the first hidden layer. This represents the weight vector of the first hidden layer. Let represent the set of nodes that are connected to the i-th node by an edge. This represents the activation function. Indicates attention weights. express Activation function Represents the sigmoid function;

[0114] The node characteristics of a node in a power grid are equal to the characteristics of the power grid, and the node characteristics of a node in an inverter are equal to the characteristics of the inverter.

[0115] An online training control unit is used to control the online training of the first model. The online training control unit inputs the power grid characteristics of n time intervals between the current time into the second model. The second model outputs the start time of the online training of the first model and the number of experience points collected during the training in the time interval between the current time and the next time interval.

[0116] The first model is trained online based on the start time of online training output by the second model and the number of experience points collected during training.

[0117] The calculation formula for the second model is as follows:

[0118]

[0119]

[0120]

[0121]

[0122]

[0123]

[0124] in, and Let n and L represent the 1st and Lth power grid features, respectively, where n ≥ L ≥ 1. When L = n, the power grid feature at the current moment is input; when L = 1, the power grid feature at the (n-1)th moment before the current moment is input. , , and All represent trainable weight parameters. , , and All of these represent trainable bias parameters. Represents the dot product. Indicates the first The first intermediate feature, Indicates the first A second intermediate feature, Indicates the first The third intermediate feature, and They represent the first -1 and the first Each output feature Let represent the second output vector, where the value of the i-th component represents the probability value of the i-th second action, and tanh represents the tanh function. This represents the Sigmoid function.

[0125] In one embodiment of the present invention, the second action belongs to a second action space. The method for generating the second action in the second action space is as follows:

[0126] Discretize 0% to 100% into multiple point values, and then combine them with empirical numbers acquired through optional training.

[0127] For example, a second action can be represented as: ;

[0128] in, % represents the ratio of the time difference between the start of training and time t to the time difference between time t and time t+1. %; 12 represents the number of experience points collected during training.

[0129] It should be noted that, % indicates that training is not started.

[0130] In one embodiment of the present invention, the decision model is trained using the Q-learning method, and the state of Q-learning is defined as the set of environmental information from time tn to time t.

[0131] The reward function for Q-learning is:

[0132]

[0133] in, It is the reward obtained by taking the second action chosen between time t and t+1. It is the resource utilization rate of the processor of the self-decision control device from time t to t+1.

[0134] In one embodiment of the present invention, the formula for calculating the processor's resource utilization is as follows:

[0135]

[0136] CPU time refers to the time that the processor spends executing programs and operating system code within a certain period of time;

[0137] Total CPU time refers to the total running time of the processor within this time period.

[0138] It should be noted that the training of the second model is carried out during the intervals of the self-decision control device's operation.

[0139] The steps for online training of the first model include:

[0140] Step 101, Build network, Network replication first model;

[0141] Step 102: Randomly obtain an experience from the experience pool. The experience includes the state at time t. The first action at time t The state at time t+1 Execute the first action Rewards ;

[0142] Then calculate the state. Down The network output corresponds to probability value ;

[0143]

[0144] express Down The highest probability value output by the network;

[0145] Step 103, according to and Calculate the loss value and update the first model;

[0146]

[0147] express The output of the first model corresponds to The probability value;

[0148] It is a discount factor, a value between 0 and 1, used to balance the weight of immediate rewards and future rewards, with a default value of 0.6;

[0149] The first action originates from the first action space, and a first action in the first action space represents a discrete value of reactive power.

[0150] Step 104: Update at fixed network update intervals. The network is configured so that its parameters are the same as those of the current first model;

[0151] The default value for the fixed interval of network updates is 10.

[0152] Step 105 continues until the number of experiences collected equals the number of experiences collected in this training session, at which point the process terminates.

[0153] In one embodiment of the present invention, the first action with the highest probability value is selected at each time step.

[0154]

[0155] in This represents the maximum value of the grid voltage deviation from time t to t+1. Indicates the standard voltage value. This represents the maximum value of the power grid frequency deviation from time t to t+1. This indicates the standard value of the frequency.

[0156] The control module is used to send control commands to the inverter. After receiving the control commands, the inverter can adjust the reactive power to the reactive power required by the decision unit.

[0157] Harmonic currents generated by some new energy equipment and nonlinear power loads enter the power system during use, causing abnormal changes in current and voltage. Changes in grid load and the addition or removal of inverters will alter the overall state of the grid. Therefore, the first model needs to be trained online to adapt to the dynamic changes in the grid.

[0158] When the grid load is high, voltage may change frequently. In this case, the data sampling frequency and the inverter control frequency may be dynamically adjusted. The processor of the self-decision control device will be under heavy load. If the first model is still learned online at a fixed frequency, it may cause the processor to be overloaded and the control to fail. Therefore, a second model is introduced to supervise the training start and training iteration of the first model to avoid the control failure and to update the first model as much as possible.

[0159] The advantages of this system are:

[0160] (1) Seamless access with communication management function. The device will not affect the original communication and function of the equipment when it is connected.

[0161] (2) It has strong adaptability and can automatically identify and convert to the standard protocol according to the protocol of different manufacturers' equipment, so as to achieve protocol unification and communication without obstacles. Especially for photovoltaic inverters with only one communication interface, no additional interface expansion is required for direct communication.

[0162] (3) The device is easy to install and the equipment is simple and convenient to install. There is no need to shut down the power for a long time for installation. Some devices can be installed while the power is on.

[0163] At least one embodiment of the present invention discloses a power quality management method, comprising the following steps:

[0164] Step 100: Read the operating data of the power grid to which the inverter is connected;

[0165] Step 200, Inverter Operation Data Acquisition Unit, is used to collect the inverter's operation data.

[0166] Step 300: Encode the power grid operation data to obtain power grid characteristics.

[0167] Step 400: Encode the inverter's operating data to obtain inverter characteristics;

[0168] Step 500: Construct graph structure data based on the power grid topology. The graph structure data includes nodes and edges connecting nodes. A node represents an inverter or a power grid. There are edges between the nodes representing the power grid and all nodes representing the inverters. There are edges between the nodes corresponding to the inverters connected to the same grid connection point.

[0169] Step 600: Input the graph structure data, grid characteristics and inverter characteristics into the first model, and then the first model outputs the value of reactive power that the inverter needs to generate.

[0170] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A power quality management cluster system for the end of a distribution network, characterized in that, include: A self-decision control device, used to control the reactive power output of the inverter, includes: The power grid operation data reading unit is used to read the operation data of the power grid to which the inverter is connected; Inverter operation data acquisition unit, which is used to collect the inverter's operation data; The decision-making unit is used to encode the power grid's operational data to obtain power grid characteristics; Inverter characteristics are obtained by encoding the inverter's operating data. Based on the power grid topology, graph structure data is constructed. The graph structure data includes nodes and edges connecting nodes. A node represents an inverter or a power grid. There are edges between the nodes representing the power grid and all nodes representing the inverters. There are edges between the nodes corresponding to the inverters connected to the same grid connection point. The graph structure data, grid characteristics, and inverter characteristics are input into the first model, and then the first model outputs the value of reactive power that the inverter needs to generate; An online training control unit is used to control the online training of the first model. The online training control unit inputs the power grid characteristics of n time intervals between the current time into the second model. The second model outputs the start time of the online training of the first model and the number of experience points collected during the training in the time interval between the current time and the next time interval. The first model is trained online based on the start time of online training output by the second model and the number of experience points collected during training. The calculation formula for the first model is as follows: ； ； ； ； ； This represents the output vector, where one component of the output vector corresponds to a probability value of a discrete value of reactive power. Let represent the hidden feature of the i-th node, where the i-th node is the node corresponding to the inverter controlled by the decision unit. and Let i and j represent the node features of the i-th and j-th nodes, respectively. This represents the weight parameters of the first hidden layer. This represents the weight vector of the first hidden layer. Let represent the set of nodes that are connected to the i-th node by an edge. This represents the activation function. Indicates attention weights. express Activation function; The node characteristics of a node in a power grid are equal to the characteristics of the power grid; the node characteristics of a node in an inverter are equal to the characteristics of the inverter. The decision model is trained using the Q-learning method, and the state of Q-learning is defined as the set of environmental information from time tn to time t. The reward function for Q-learning is: ； in, It is the reward obtained by taking the second action chosen between time t and t+1. It is the resource utilization rate of the processor of the self-decision control device from time t to t+1; The control module is used to send control commands to the inverter. After receiving the control commands, the inverter can adjust the reactive power to the reactive power required by the decision unit.

2. The power quality management cluster system for the end of a distribution network according to claim 1, characterized in that, The calculation formula for the second model is as follows: ； ； ； ； ； ； in, and Let n and L represent the 1st and Lth power grid features, respectively, where n ≥ L ≥ 1. When L = n, the power grid feature at the current moment is input; when L = 1, the power grid feature at the (n-1)th moment before the current moment is input. , , and All represent trainable weight parameters. , , and All of these represent trainable bias parameters. Represents the dot product. Indicates the first The first intermediate feature, Indicates the first A second intermediate feature, Indicates the first The third intermediate feature, and They represent the first -1 and the first Each output feature Let represent the second output vector, where the value of the i-th component represents the probability value of the i-th second action, and tanh represents the tanh function. This represents the Sigmoid function.

3. The power quality management cluster system for the end of a distribution network according to claim 1, characterized in that, The formula for calculating processor resource utilization is as follows: ； CPU time refers to the time that the processor spends executing programs and operating system code within a certain period of time; Total CPU time refers to the total running time of the processor within this time period.

4. The power quality management cluster system for the end of a distribution network according to claim 1, characterized in that, The steps for online training of the first model include: Step 101, Build network, Network replication first model; Step 102: Randomly obtain an experience from the experience pool. The experience includes the state at time t. The first action at time t The state at time t+1 Execute the first action Rewards ; Then calculate the state. Down The network output corresponds to probability value ; ； express Down The highest probability value output by the network; Step 103, according to and Calculate the loss value and update the first model; ； express The output of the first model corresponds to The probability value; It is the discount factor, a value between 0 and 1; The first action originates from the first action space, and a first action in the first action space represents a discrete value of reactive power. Step 104: Update at fixed network update intervals. The network is configured so that its parameters are the same as those of the current first model; Step 105 continues until the number of experiences collected equals the number of experiences collected in this training session, at which point the process terminates.

5. The power quality management cluster system for the end of a distribution network according to claim 4, characterized in that, At each moment, select the action with the highest probability value as the first action.

6. The power quality management cluster system for the end of a distribution network according to claim 4, characterized in that, award The calculation formula is as follows: ； in This represents the maximum value of the grid voltage deviation from time t to t+1. Indicates the standard voltage value. This represents the maximum value of the power grid frequency deviation from time t to t+1. This indicates the standard value of the frequency.

7. The power quality management cluster system for the end of a distribution network according to claim 4, characterized in that, The state refers to the input of the first model.

8. A method for improving power quality, characterized in that, Based on the power quality management cluster system for the power distribution network terminal as described in any one of claims 1-7, the following steps are performed: Step 100: Read the operating data of the power grid to which the inverter is connected; Step 200, Inverter Operation Data Acquisition Unit, is used to collect the inverter's operation data. Step 300: Encode the power grid operation data to obtain power grid characteristics. Step 400: Encode the inverter's operating data to obtain inverter characteristics; Step 500: Construct graph structure data based on the power grid topology. The graph structure data includes nodes and edges connecting nodes. A node represents an inverter or a power grid. There are edges between the nodes representing the power grid and all nodes representing the inverters. There are edges between the nodes corresponding to the inverters connected to the same grid connection point. Step 600: Input the graph structure data, grid characteristics and inverter characteristics into the first model, and then the first model outputs the value of reactive power that the inverter needs to generate.

Images