Transformer area electric energy quality treatment method and system based on coordination control of multiple one-machine multi-purpose devices

Abstract

The invention relates to the technical field of electric energy management, and discloses a transformer area electric energy quality management method and system based on coordination control of multiple one-machine multi-purpose devices, and the method comprises the following steps: collecting the operation information of a power grid of a transformer area and the operation information of power equipment, the power equipment comprising an electric energy quality management device; generating graph structure data based on the topological graph of the power grid of the transformer area; respectively inputting the sequence data into a first hidden layer of the distributed model, and respectively outputting a first hidden state of each node; inputting the graph structure data and the first hidden state of the node into a second hidden layer of the distributed model, and outputting a second hidden state by the second hidden layer; the output layer inputs a second hidden state and outputs a control parameter of the electric energy quality control device; according to the invention, each treatment device is provided with an intelligent agent. The intelligent agents cooperate with one another through a communication network, and optimal control over the whole system is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric energy management, and more specifically, to a method and system for managing power quality in an electric power distribution area based on coordinated control of multiple multi-purpose devices. Background Art

[0002] With the rapid development of the economy and society, energy demand is increasing day by day, and power quality issues are becoming increasingly serious. Power quality issues not only affect the normal operation of industrial production and civil equipment, but also lead to inefficient use and waste of energy. The traditional single device management method can no longer meet the power quality control needs in complex environments. Summary of the invention

[0003] The present invention provides a method and system for managing power quality in an area based on coordinated control of multiple one-machine-multi-purpose devices, which solves the technical problem in related technologies that a single device management method can no longer meet the power quality control needs in complex environments.

[0004] The present invention provides a method for controlling power quality in a substation based on coordinated control of multiple one-machine-multi-purpose devices, comprising the following steps:

[0005] Step 11, collecting the operation information of the power grid and the operation information of the power equipment in the substation area, the power equipment includes the power quality management device;

[0006] Step 12, generating graph structure data based on the topological graph of the power grid in the substation, wherein the graph structure data includes nodes and edges, wherein the nodes represent the power equipment in the power grid or substation, and there are edges between the nodes corresponding to the power equipment directly connected on the topological graph of the power grid, and there are edges between the nodes representing the power grid and all the nodes representing the power equipment in the substation;

[0007] Encode the operation information of the power equipment in the substation represented by each node into sequence data, wherein the sequence data includes n sequence units, the nth sequence unit represents the operation information at the current moment, and the nsth sequence unit represents the operation information at the sth moment before the current moment;

[0008] Step 13, input the sequence data into the first hidden layer of the distributed model respectively, and output the first hidden state of each node respectively;

[0009] Inputting the graph structure data and the first hidden state of the node into the second hidden layer of the distributed model, and the second hidden layer outputs the second hidden state;

[0010] The output layer inputs the second hidden state and outputs the control parameters of the power quality management device;

[0011] Each power quality control device is controlled by a control parameter output by a distributed model.

[0012] Furthermore, the topological map of the power grid in the substation includes topological information between various power devices in the power grid.

[0013] Furthermore, the calculation formula of the first hidden layer is as follows:

[0014] u (t) =σ(W u X (t) +W u H (t-1) +b u )

[0015] r (t) =σ(W r X (t) +W r H (t-1) +b r )

[0016] C (t) =tanh(W c X (t) +W c r (t) ⊙H (t-1) +b c )

[0017] H (t) =(1-u (t) )⊙C (t) +u (t) ⊙H (t-1)

[0018] Among them, W u , W r , W c represents the weight parameter, b u , b r , b c represents the bias parameter, ⊙ represents the dot product, where X (t) Represents the tth sequence unit of the sequence data of the node, H (t) and H (t-1) They represent the t-th and t-1-th sequence hidden states respectively. The n-th sequence hidden state is the first hidden state of the node. n ≥ t ≥ 1, where n represents the total number of sequence units of the sequence data. When t = 1, H (t) =X (t) , tanh is the hyperbolic tangent function, and σ represents the S-type function.

[0019] Furthermore, the sequence units are encoded into features through feature engineering before being input into the first hidden layer.

[0020] Furthermore, the calculation formula of the second hidden layer is as follows:

[0021]

[0022] where h i represents the hidden state of the i-th node, φ i represents the aggregation coefficient of the i-th node, K i represents the set of nodes that have edges with the i-th node, σ represents the activation function, and W h represents the state weight coefficient, x j Represents the node features of the jth node;

[0023]

[0024] Z i =W z x i

[0025] Z j =W z x j

[0026] where x i and x j Represent the first hidden state of the i-th and j-th nodes respectively, W1 represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transpose, exp represents the exponential function with natural constant as the base, LeakyRelu represents the modified linear unit function, K i Represents the set of nodes that have edges with the i-th node.

[0027] Furthermore, the calculation formula of the output layer is as follows:

[0028] y v =σ(W y h v +b y )

[0029] where y v represents the control parameter vector of the power quality management device represented by the vth node. A component of the control parameter vector represents the probability value of an action. The set of all actions is called the action space, which contains all possible actions that may be selected. v represents the hidden state of the vth node, v∈K v , K v represents the set of nodes corresponding to all power quality management devices, W y is the weight parameter, b y is the bias parameter and σ represents the S-type function.

[0030] Furthermore, an action is represented as a vector, and the j-th component of the vector represents the j-th control parameter of the power quality management device.

[0031] Furthermore, the method further comprises the following steps:

[0032] Step 21, generating a first model set, wherein the online distributed model in the first model set is of the same type as the power quality management device controlled by the offline distributed model;

[0033] If the first model set is not empty, randomly selecting a distributed model in the first model set as the model to be migrated;

[0034] If the first model set is empty, randomly select an online distributed model as the model to be migrated;

[0035] Step 22, randomly combine the actions in the action space of the offline distributed model and the model to be migrated to obtain a combined action space;

[0036] Step 23, combining the parameters of the offline distributed model and the parameters of the model to be migrated to obtain combined parameters;

[0037] Step 24, replace the parameters of the model to be migrated with the combined parameters;

[0038] Step 25, retrain the model to be migrated, and the trained model to be migrated outputs the control parameters of the offline distributed model and the power quality management device controlled by the model to be migrated.

[0039] Furthermore, the calculation formula for combining the parameters of the offline distributed model and the parameters of the model to be migrated is as follows:

[0040]

[0041] θ t represents the combination parameter, α∈[0,1] represents the combination coefficient, which controls the ratio of the parameter combination, and θ s Represents the parameters of the offline distributed model, Indicates the parameters of the model to be migrated before replacement.

[0042] Furthermore, the model to be migrated is still a distributed model, and can be trained using the distributed model training method. The distributed model is trained using Q learning.

[0043] Step 31, build the targetQ network, the targetQ network replicates the distributed model;

[0044] Step 32, randomly obtain an experience from the experience pool, the experience includes the state S at time t t, action A executed at time t t , the state S at time t+1 t+1 , perform action A t Rewards received R t+1 ;

[0045] Then calculate the state S t The output of the targetQ network corresponds to A t The probability value Q targett ;

[0046]

[0047] Indicates S t+1 The maximum probability value output by the targetQ network;

[0048] Step 33, according to and Calculate the loss value and update the distributed model;

[0049]

[0050] Indicates S t The output of the distributed model corresponds to A t The probability value of

[0051] γ is the discount factor, a value between 0 and 1;

[0052] Step 34, update the targetQ network every fixed number of network updates to make its parameters the same as those of the current distributed model;

[0053] Step 35, until the number of collected experiences reaches the set threshold or the first model converges, the step is terminated;

[0054] Reward R t+1 The calculation formula is as follows:

[0055]

[0056] THD t+1 Represents the average value of the total harmonic distortion rate of the power grid from time t to time t+1, power t+1 Represents the total value of the reactive power loss of the power grid from time t to time t+1.

[0057] The beneficial effects of the present invention are:

[0058] The present invention configures an intelligent agent for each treatment device (active filter, passive filter, dynamic reactive power compensation device, etc.), and the intelligent agents cooperate with each other through a communication network to achieve optimal control of the overall system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0060] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0061] At least one embodiment of the present invention discloses a method for controlling power quality in a substation based on coordinated control of multiple multi-purpose devices. Figure 1 As shown, the following steps are included:

[0062] Step 11, collecting the operation information of the power grid and the operation information of the power equipment in the substation area, the power equipment includes the power quality management device;

[0063] Step 12, generating graph structure data based on the topological graph of the power grid in the substation, wherein the graph structure data includes nodes and edges, wherein the nodes represent the power equipment in the power grid or substation, and there are edges between the nodes corresponding to the power equipment directly connected on the topological graph of the power grid, and there are edges between the nodes representing the power grid and all the nodes representing the power equipment in the substation;

[0064] Encode the operation information of the power equipment in the substation represented by each node into sequence data, wherein the sequence data includes n sequence units, the nth sequence unit represents the operation information at the current moment, and the nsth sequence unit represents the operation information at the sth moment before the current moment;

[0065] Step 13, input the sequence data into the first hidden layer of the distributed model respectively, and output the first hidden state of each node respectively;

[0066] Inputting the graph structure data and the first hidden state of the node into the second hidden layer of the distributed model, and the second hidden layer outputs the second hidden state;

[0067] The output layer inputs the second hidden state and outputs the control parameters of the power quality management device;

[0068] Each power quality control device is controlled by a control parameter output by a distributed model.

[0069] In one embodiment of the present invention, the topological map of the power grid of the substation includes topological information between various power devices in the power grid.

[0070] In one embodiment of the present invention, the calculation formula of the first hidden layer is as follows:

[0071] u (t) =σ(W u X (t) +W u H (t-1) +b u )

[0072] r (t) =σ(W r X (t) +W r H (t-1) +b r )

[0073] C (t) =tanh(W c X (t) +W c r (t) ⊙H (t-1) +b c )

[0074] H (t) =(1-u (t) )⊙C (t) +u (t) ⊙H (t-1)

[0075] Among them, W u , W r , W c represents the weight parameter, b u , b r , b c represents the bias parameter, ⊙ represents the dot product, where X (t) Represents the tth sequence unit of the sequence data of the node, H (t) and H (t-1) They represent the t-th and t-1-th sequence hidden states respectively. The n-th sequence hidden state is the first hidden state of the node. n ≥ t ≥ 1, where n represents the total number of sequence units of the sequence data. When t = 1, H (t) =X (t) , tanh is the hyperbolic tangent function, and σ represents the S-type function.

[0076] In one embodiment of the present invention, the sequence units are encoded into features through feature engineering before being input into the first hidden layer.

[0077] In one embodiment of the present invention, the calculation formula of the second hidden layer is as follows:

[0078]

[0079] where h i represents the hidden state of the i-th node, φ i represents the aggregation coefficient of the i-th node, K i represents the set of nodes that have edges with the i-th node, σ represents the activation function, and W h represents the state weight coefficient, x j Represents the node features of the jth node;

[0080]

[0081] Z i =W z x i

[0082] Z j =W z x j

[0083] where x i and x j Represent the first hidden state of the i-th and j-th nodes respectively, W1 represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transpose, exp represents the exponential function with natural constant as the base, LeakyRelu represents the modified linear unit function, K i Represents the set of nodes that have edges with the i-th node;

[0084] In one embodiment of the present invention, the calculation formula of the output layer is as follows:

[0085] y v =σ(W y h v +b y )

[0086] where y v represents the control parameter vector of the power quality management device represented by the vth node. A component of the control parameter vector represents the probability value of an action. The set of all actions is called the action space, which contains all possible actions that may be selected. v represents the hidden state of the vth node, v∈K v , K v represents the set of nodes corresponding to all power quality management devices, W y is the weight parameter, b yis the bias parameter and σ represents the S-type function.

[0087] In one embodiment of the present invention, an action is represented by a vector, and the jth component of the vector represents the jth control parameter of the power quality management device;

[0088] The actions in the action space are obtained by random combinations of all possible values ​​of the control parameters.

[0089] For control parameters with continuous values, the mean is discretized into point values, and then combined with optional values ​​of other control parameters to obtain the action.

[0090] In one embodiment of the present invention, the power quality control device includes:

[0091] Active filters for harmonic suppression;

[0092] Reactive power and reactive power compensation devices;

[0093] Dynamic reactive power compensation device;

[0094] Servo-controlled voltage regulator, etc.

[0095] The operating information of the active filter includes: filter output voltage / current, compensation harmonic current content, and controller operating parameters;

[0096] The operation information of the reactive power compensation device includes: compensated reactive power value, actual value of power factor, capacitor / reactor switching status;

[0097] The operation information of the dynamic reactive power compensation device includes: real-time reactive power demand, response time of the compensation device;

[0098] The operating information of the servo-controlled voltage regulator includes: actual output voltage value, voltage fluctuation suppression, and servo controller parameters;

[0099] The operation information of the energy storage system includes: battery SOC (remaining power), charge and discharge power, and energy storage system working mode;

[0100] In one embodiment of the present invention, a power quality control device is provided, which has the following functions:

[0101] Terminal low voltage control: When the voltage at the user end of the distribution station area is lower than the normal voltage and is unqualified, it will be discharged according to the set coefficient through automatic detection, and output reactive power support to raise the power factor of the line, ensure reliable electricity use for residents, and improve power supply reliability.

[0102] Severe overload control in substations: When the transformer capacity in the substation is insufficient or the transformer is severely overloaded, the power supply capacity of the substation is improved through discharge to support the normal operation of the substation, save the investment cost of capacity increase and site deployment, shorten the transformation cycle and improve power supply reliability.

[0103] New energy access: Photovoltaic energy is first stored in the battery, and the surplus power is supplied to the load. When the photovoltaic energy is insufficient, the load is powered by the energy storage, and the energy storage system smoothes the power generation and power consumption. It eliminates the instability and random fluctuations of photovoltaic power generation, improves the quality of power, and provides continuous power supply.

[0104] In practical applications, some carriers of the distributed model may fail or go offline. In this case, the present invention provides a method for managing power quality in a substation based on coordinated control of multiple one-machine-multi-purpose devices, which also includes the following steps:

[0105] Step 21, generating a first model set, wherein the online distributed model in the first model set is of the same type as the power quality management device controlled by the offline distributed model;

[0106] If the first model set is not empty, randomly selecting a distributed model in the first model set as the model to be migrated;

[0107] If the first model set is empty, randomly select an online distributed model as the model to be migrated;

[0108] Step 22, randomly combine the actions in the action space of the offline distributed model and the model to be migrated to obtain a combined action space;

[0109] Step 23, combining the parameters of the offline distributed model and the parameters of the model to be migrated to obtain combined parameters;

[0110] Step 24, replace the parameters of the model to be migrated with the combined parameters;

[0111] Step 25, retrain the model to be migrated, and the trained model to be migrated outputs the control parameters of the offline distributed model and the power quality management device controlled by the model to be migrated.

[0112] In one embodiment of the present invention, the calculation formula for combining the parameters of the offline distributed model and the parameters of the model to be migrated is as follows:

[0113]

[0114] θ t represents the combination parameter, α∈[0,1] represents the combination coefficient, which controls the ratio of the parameter combination, and θ s Represents the parameters of the offline distributed model, Represents the parameters of the model to be migrated before replacement. The greater the ratio of the importance of the offline distributed model to the power quality management device controlled by the model to be migrated, the greater the value of α.

[0115] In one embodiment of the present invention, the model to be migrated is still a distributed model, and can be trained by a distributed model training method. The distributed model is trained by Q learning.

[0116] Step 31, build the targetQ network, the targetQ network replicates the distributed model;

[0117] Step 32, randomly obtain an experience from the experience pool, the experience includes the state S at time t t , action A executed at time t t , the state S at time t+1 t+1 , perform action A t Rewards received R t+1 ;

[0118] Then calculate the state S t The output of the targetQ network corresponds to A t The probability value

[0119]

[0120] Indicates S t+1 The maximum probability value output by the targetQ network;

[0121] Step 33, according to and Calculate the loss value and update the distributed model;

[0122]

[0123] Indicates S t The output of the distributed model corresponds to A t The probability value of

[0124] γ is the discount factor, a value between 0 and 1;

[0125] Step 34, update the targetQ network every fixed number of network updates to make its parameters the same as those of the current distributed model;

[0126] Step 35, until the number of collected experiences reaches the set threshold or the first model converges, the step is terminated.

[0127] Reward R t+1The calculation formula is as follows:

[0128]

[0129] THD t+1 Represents the average value of the total harmonic distortion rate of the power grid from time t to time t+1, power t+1 Represents the total value of the reactive power loss of the power grid from time t to time t+1.

[0130] In one embodiment of the present invention, the experience is derived from the record of historical action execution.

[0131] In one embodiment of the present invention, the parameters of the distributed model may be stored in a cloud server.

[0132] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A method for controlling power quality in a substation based on coordinated control of multiple multi-purpose devices, characterized in that: The following steps are involved: Step 11, collecting the operation information of the power grid and the operation information of the power equipment in the substation area, the power equipment includes the power quality management device; Step 12, generating graph structure data based on the topological graph of the power grid in the substation, wherein the graph structure data includes nodes and edges, wherein the nodes represent the power equipment in the power grid or substation, and there are edges between the nodes corresponding to the power equipment directly connected on the topological graph of the power grid, and there are edges between the nodes representing the power grid and all the nodes representing the power equipment in the substation; Encode the operation information of the power equipment in the substation represented by each node into sequence data, wherein the sequence data includes n sequence units, the nth sequence unit represents the operation information at the current moment, and the nsth sequence unit represents the operation information at the sth moment before the current moment; Step 13, input the sequence data into the first hidden layer of the distributed model respectively, and output the first hidden state of each node respectively; Inputting the graph structure data and the first hidden state of the node into the second hidden layer of the distributed model, and the second hidden layer outputs the second hidden state; The output layer inputs the second hidden state and outputs the control parameters of the power quality management device; Each power quality control device is controlled by a control parameter output by a distributed model.

2. According to claim 1, a method for controlling power quality in a substation based on coordinated control of multiple one-machine-multi-purpose devices, characterized in that: The topological diagram of the power grid in the substation area contains the topological information between various power equipment in the power grid.

3. According to claim 1, a method for controlling power quality in a substation based on coordinated control of multiple multi-purpose devices, characterized in that: The calculation formula for the first hidden layer is as follows: u (t) =σ(W u X (t) +W u H (t-1) +b u ) r (t) =σ(W r X (t) +W r H (t-1) +b r ) C (t) =tanh(W c X (t) +W c r (t) ⊙H (t-1) +b c ) H (t) =(1-u (t) )⊙C (t) +u (t) ⊙H (t-1) Among them, W u , W r , W c represents the weight parameter, b u 、b r 、b c represents the bias parameter, ⊙ represents the dot product, where X (t) Represents the tth sequence unit of the sequence data of the node, H (t) and H (t-1) They represent the t-th and t-1-th sequence hidden states respectively. The n-th sequence hidden state is the first hidden state of the node. n ≥ t ≥ 1, where n represents the total number of sequence units of the sequence data. When t = 1, H (t) =X (t) , tanh is the hyperbolic tangent function, and σ represents the S-type function.

4. According to claim 3, a method for controlling power quality in a substation based on coordinated control of multiple multi-purpose devices, characterized in that: The sequence units are encoded into features through feature engineering before being input into the first hidden layer.

5. A method for controlling power quality in a substation based on coordinated control of multiple one-machine-multi-purpose devices according to claim 4, characterized in that: The calculation formula for the second hidden layer is as follows: where h i represents the hidden state of the i-th node, φ i represents the aggregation coefficient of the i-th node, K i represents the set of nodes that have edges with the i-th node, σ represents the activation function, and W h represents the state weight coefficient, x j Represents the node features of the jth node; Z i =W z x i Z j =W z x j where x i and x j Represent the first hidden state of the i-th and j-th nodes respectively, W1 represents the aggregation weight coefficient, represents the splicing weight coefficient, T represents transpose, exp represents the exponential function with natural constant as the base, LeakyRelu represents the modified linear unit function, K i Represents the set of nodes that have edges with the i-th node.

6. A method for controlling power quality in a substation area based on coordinated control of multiple one-machine-multi-purpose devices according to claim 5, characterized in that: The calculation formula of the output layer is as follows: y v =σ(W y h v +b y ) where y v represents the control parameter vector of the power quality management device represented by the vth node. A component of the control parameter vector represents the probability value of an action. The set of all actions is called the action space, which contains all possible actions that may be selected. v represents the hidden state of the vth node, v∈K v , K v represents the set of nodes corresponding to all power quality management devices, W y is the weight parameter, b y is the bias parameter and σ represents the S-type function.

7. The method for controlling power quality in a substation area based on coordinated control of multiple one-machine-multi-purpose devices according to claim 1 is characterized in that: An action is represented by a vector, and the j-th component of the vector represents the j-th control parameter of the power quality management device.

8. The method for controlling power quality in a substation area based on coordinated control of multiple one-machine-multi-purpose devices according to claim 1 is characterized in that: The following steps are also included: Step 21, generating a first model set, wherein the online distributed model in the first model set is of the same type as the power quality management device controlled by the offline distributed model; If the first model set is not empty, randomly selecting a distributed model in the first model set as the model to be migrated; If the first model set is empty, randomly select an online distributed model as the model to be migrated; Step 22, randomly combine the actions in the action space of the offline distributed model and the model to be migrated to obtain a combined action space; Step 23, combining the parameters of the offline distributed model and the parameters of the model to be migrated to obtain combined parameters; Step 24, replace the parameters of the model to be migrated with the combined parameters; Step 25, retrain the model to be migrated, and the trained model to be migrated outputs the control parameters of the offline distributed model and the power quality management device controlled by the model to be migrated.

9. A method for controlling power quality in a substation area based on coordinated control of multiple one-machine-multi-purpose devices according to claim 8, characterized in that: The calculation formula for combining the parameters of the offline distributed model and the parameters of the model to be migrated is as follows: θ t represents the combination parameter, α∈[0,1] represents the combination coefficient, which controls the ratio of the parameter combination, and θ s Represents the parameters of the offline distributed model, Indicates the parameters of the model to be migrated before replacement.

10. A method for controlling power quality in a substation area based on coordinated control of multiple one-machine-multi-purpose devices according to claim 9, characterized in that: The model to be migrated is still a distributed model and can be trained using the distributed model training method. The distributed model is trained using Q learning. Step 31, build the targetQ network, the targetQ network replicates the distributed model; Step 32, randomly obtain an experience from the experience pool, the experience includes the state S at time t t , action A executed at time t t , the state S at time t+1 t+1 , perform action A t Rewards received R t+1 ; Then calculate the state S t The output of the targetQ network corresponds to A t The probability value Q targett ; Indicates S t+1 The maximum probability value output by the targetQ network; Step 33, according to and Calculate the loss value and update the distributed model; Indicates S t The output of the distributed model corresponds to A t The probability value of γ is the discount factor, a value between 0 and 1; Step 34, update the targetQ network every fixed number of network updates to make its parameters the same as those of the current distributed model; Step 35, until the number of collected experiences reaches the set threshold or the first model converges, the step is terminated; Reward R t+1 The calculation formula is as follows: THD t+1 Represents the average value of the total harmonic distortion rate of the power grid from time t to time t+1, power t+1 Represents the total value of the reactive power loss of the power grid from time t to time t+1.

Images