There is no feasible solution for the dispatching and operation of isolated microgrids. A method for dynamic capacity increase of short-term lines
By generating daily load curve samples and building economic scheduling models, combining neural network prediction and constraint correction, dynamic capacity increase of island microgrid lines is achieved, solving the problem of unfeasible solutions for scheduling operation, and improving the efficiency and feasibility of scheduling operation. Suitable for peak load periods and new energy access.
Patent Information
- Application Number
- CN202411118261.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-08-15
AI Technical Summary
There is no feasible solution when scheduling in the island microgrid. The existing methods rely on manual experience, are inefficient and difficult to guarantee. There is a lack of fast and effective boundary adjustment methods to cope with peak load periods and high proportion of new energy access.
By generating daily load curve samples of island microgrids, an economic scheduling model is constructed, and the dynamic capacity increase of line current carrying capacity is solved, and neural networks are used to predict the lines that need to be increased and their capacity increase, combining equation constraint completion and inequality constraint correction. Implement the feasibility of dynamically increasing the line to restore scheduling operation.
Provide a fast and effective method. By dynamically increasing the capacity of the transmission line, the problem of unfeasible solutions for the scheduling operation of the island microgrid is solved, and the efficiency and feasibility of the scheduling operation are improved. It is suitable for peak load periods and high proportion new Energy access situation.
Smart Images

Figure CN118646005B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power grid dispatching, and in particular to a method for dynamically increasing the capacity of short-term lines with no feasible solution for dispatching and operating isolated island microgrids. Background Art
[0002] like Figure 4 The figure shows the topological structure of an isolated microgrid. When the isolated microgrid is dispatched and operated, for safety and stability reasons, the actual limit value of the line is usually set to operate at a lower proportion of the rated value, allowing the flow to exceed the limit value in a short period of time. Therefore, when there is no feasible solution for the short-term peak load dispatch operation, the critical transmission line limit between the power source and the load is increased in a short period of time to meet the sudden short-term system load peak demand. To this end, the economic dispatch model of the isolated microgrid can be solved to obtain the output of each unit in the microgrid, and the flow value of each transmission line can be calculated based on the power transfer distribution factor. The capacity of some over-limit transmission lines is dynamically increased in a short time, and then this adjustment strategy is learned based on artificial intelligence methods, so as to achieve rapid feasibility recovery of the dispatch operation problem of the isolated microgrid.
[0003] At present, faced with the situation that there is no feasible solution for the dispatching and operation of isolated microgrids, most dispatchers can only rely on past experience to judge the boundary conditions and obtain feasible solutions by repeatedly adjusting the unit output, operating time or load reduction. This traditional adjustment method based on manual experience is inefficient, labor-intensive, and due to its certain randomness, its effect is often difficult to guarantee.
[0004] Therefore, there is an urgent need for an effective and fast boundary adjustment method to solve the analysis and adjustment problems when the dispatching and operation of isolated microgrids do not converge. That is, by studying the short-term online adjustment method of the dynamic transmission capacity of the transmission line, an effective reference can be provided for the dispatching and operation of isolated microgrids during peak load periods and when a high proportion of new energy is connected. Summary of the invention
[0005] The purpose of the present invention is to overcome the shortcomings of the prior art, propose a short-term line dynamic capacity increase method for an isolated microgrid dispatching operation with no feasible solution, and realize the feasibility recovery of the isolated microgrid dispatching operation problem by dynamically increasing the capacity of the transmission line.
[0006] The present invention solves the technical problem by adopting the following technical solutions:
[0007] The method for dynamic capacity increase of short-term lines without feasible solution for dispatching and operating isolated microgrids includes the following steps:
[0008] Step 1, generating a daily load curve sample of an island microgrid;
[0009] Step 2: construct an economic dispatch model for an isolated microgrid, and solve the dynamic capacity increase of the line current carrying capacity according to the daily load curve sample of the isolated microgrid generated in step 1;
[0010] Step 3: Introduce the steps of equality constraint completion and inequality constraint correction, and use the neural network to predict the lines that need to be dynamically increased in capacity and their dynamic increase capacity.
[0011] Moreover, the specific implementation method of step 1 is: selecting the actual load data of the isolated island microgrid to form daily load curve data, clustering the daily load curve data using the k-means algorithm, determining the number of clusters using the elbow method and the silhouette coefficient method, and classifying according to the number of clusters to obtain the cluster center curve;
[0012] set up Factor, which distributes the load on each bus according to The factors are linearly scaled and normalized:
[0013]
[0014] in, is the percentage of each bus in the island microgrid to the total system load, is the linear scaling factor;
[0015] Calculate the hourly change of total load of the microgrid :
[0016]
[0017] In the formula, is the total load of the microgrid in time period t, and the clustered daily load curve samples are calculated for each category. The mean With variance , get a random number , , ,in is the load value of the cluster center curve of each type of sample in the first hour of the microgrid, Multiplying a random number Get the total daily load curve sample of the microgrid, and multiply the total daily load curve sample of the microgrid by the randomly generated load distribution coefficient on each bus , generate daily load curve samples on each bus of the microgrid.
[0018] Moreover, the objective function of the economic dispatch model of the isolated island microgrid in step 2 is to minimize the daily operating cost:
[0019]
[0020] In the formula, , Respectively represent the time period and the serial number of the unit; , Respectively represent the collection of time periods and units; For the crew In time period contribution; , , For each unit The constant term, linear term, and quadratic term coefficients of the running cost function are approximated by a piecewise linear function for the quadratic running cost function.
[0021] Moreover, the constraints of the island microgrid economic dispatch model include: power balance constraint, upper and lower limit constraints of unit output and unit climbing constraint, among which,
[0022] Power balance constraints:
[0023]
[0024] In the formula, Indicates the node's sequence number; Represents a collection of nodes; For Node In time period Load;
[0025] Unit output upper and lower limit constraints:
[0026]
[0027] In the formula, and For each unit The minimum and maximum output of
[0028] Unit climbing constraints:
[0029]
[0030] In the formula, and For each unit Limitation of the rate of change of output between adjacent time periods.
[0031] Moreover, the specific implementation method of solving the dynamic increase capacity of the line current carrying capacity in step 2 is:
[0032]
[0033]
[0034]
[0035] In the formula, For transmission lines In time period The meritorious trend; is the power transfer distribution factor; For transmission lines In time period Increased capacity; For Line The maximum power flow limit value is calculated based on the unit output value obtained by the economic dispatch model of the isolated island microgrid, and the transmission line In time period The dynamic capacity that needs to be increased within the network is used to generate the training data set.
[0036] Moreover, the specific implementation method of step 3 is: taking the economic dispatch model of the isolated microgrid and the dynamic capacity increase of the transmission line as a constrained optimization problem:
[0037]
[0038] In the formula, represents the parameters of the optimization problem, Represents the solution to the optimization problem, training the neural network through supervised learning Output the complete solution to the optimization problem Partial set variables ; Complete solution to the optimization problem By partial set variables After completing the equality constraints, it is calculated; after the inequality constraint correction step, Corrected to , making it closer to satisfying the inequality constraint, the neural network Training is performed using the error back propagation algorithm on the SmoothL1Loss loss function:
[0039] .
[0040] Furthermore, the neural network A multi-head neural network is used to complete the prediction task, and each head is specially trained to learn the mapping relationship between the microgrid load and the output of a specific unit. It includes three main structures: shared layer, parallel head and integrated output. The shared layer is used to extract the common features of the input data; the parallel head is responsible for learning these features and generating independent prediction values; the final output is obtained by integrating the prediction values output by each head;
[0041] At the same time, Sigmoid activation function is used between layers to realize nonlinear mapping to enhance the expressive power of the network. In addition, batch normalization layer is used to accelerate the training process and alleviate the gradient vanishing problem, thereby improving the stability and training efficiency of the model. Random inactivation layer is used to randomly discard the output of neurons to prevent overfitting and improve the generalization ability of the model.
[0042] The advantages and positive effects of the present invention are:
[0043] The present invention generates daily load curve samples of isolated microgrids; constructs an economic dispatch model of isolated microgrids, solves the dynamic increase capacity of line current carrying capacity based on the generated daily load curve samples of isolated microgrids, adds equation constraint completion and inequality constraint correction steps, and uses neural networks to predict the transmission lines and their increased capacity that need to be dynamically increased. The present invention firstly solves the boundary adjustment problem when there is no feasible solution for the dispatch operation of isolated microgrids by dynamically increasing the capacity of transmission lines; secondly, a method for generating daily load curve samples that considers the uncertainty of load distribution coefficient and system load change is proposed, which provides rich load scenarios for deep learning model training; finally, the steps of equality constraint completion and inequality constraint correction are introduced, and the quadratic programming problem is solved by neural networks, which can quickly obtain the transmission lines that need to be adjusted, and calculate the dynamic increase capacity of line current carrying capacity, providing an effective reference for dispatching personnel to deal with the dispatch operation of isolated microgrids in peak load periods and high proportion of new energy access. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 is a flow chart of the present invention;
[0045] Figure 2 A schematic diagram of a model for introducing the steps of equality constraint completion and inequality constraint correction in the present invention;
[0046] Figure 3 is a structural diagram of the neural network of the present invention;
[0047] Figure 4 This is the topological structure diagram of the isolated island microgrid. DETAILED DESCRIPTION
[0048] The present invention is further described in detail below with reference to the accompanying drawings.
[0049] There is no feasible solution for the dispatching and operation of isolated microgrids. There is no method for dynamic capacity increase of short-term lines, such as Figure 1 As shown, the following steps are included:
[0050] Step 1: Generate a sample of daily load curve of the island microgrid.
[0051] The specific implementation method of step 1 is: select the actual load data of an isolated microgrid for 365 days, with a time resolution of 1 hour, that is, 24 points per day. Cluster the 365 daily load curve data based on the k-means algorithm, and use the elbow method and silhouette coefficient method to determine the optimal number of clusters K=4, and divide them into 4 categories;
[0052] Sample generation is performed based on 4 cluster center curves. Factor, which distributes the load on each bus according to The factors are linearly scaled and normalized:
[0053]
[0054] in, is the percentage of each bus in the island microgrid to the total system load, is the linear scaling factor, ;
[0055] Calculate the hourly change of total load of the microgrid :
[0056]
[0057] In the formula, is the total load of the microgrid in time period t, and the clustered daily load curve samples are calculated for each category. The mean With variance , get a random number , , ,in is the load value of the cluster center curve of each type of sample in the first hour of the microgrid, Multiplying a random number Get the total daily load curve sample of the microgrid, and multiply the total daily load curve sample of the microgrid by the randomly generated load distribution coefficient on each bus , generate daily load curve samples on each bus of the microgrid.
[0058] Each type of sample is multiplied by 5 sets of randomly generated load distribution coefficients on each bus, generating a total of 400 daily load curve samples, involving 9,600 data. These data constitute a rich load scenario and lay the foundation for subsequent model training.
[0059] Step 2: Build an economic dispatch model for an isolated microgrid, and solve the dynamic capacity increase of the line current carrying capacity based on the daily load curve sample of the isolated microgrid generated in step 1.
[0060] The objective function of the economic dispatch model of the isolated microgrid is to minimize the daily operating cost:
[0061]
[0062] In the formula, , Respectively represent the time period and the serial number of the unit; , Respectively represent the collection of time periods and units; For the crew In time period contribution; , , For each unit The constant term, linear term, and quadratic term coefficients of the running cost function are approximated by a piecewise linear function for the quadratic running cost function.
[0063] The constraints of the economic dispatch model of the isolated microgrid include: power balance constraint, upper and lower limit constraints of unit output, and unit climbing constraint.
[0064] Power balance constraints:
[0065]
[0066] In the formula, Indicates the node's sequence number; Represents a collection of nodes; For Node In time period Load;
[0067] Unit output upper and lower limit constraints:
[0068]
[0069] In the formula, and For each unit The minimum and maximum output of
[0070] Unit climbing constraints:
[0071]
[0072] In the formula, and For each unit Limitation of the rate of change of output between adjacent time periods.
[0073] The generated 400 daily load curve samples correspond to the economic dispatch models of 400 isolated island microgrids. The Python-based IBM CPLEX solver is used to solve these 400 models to obtain the unit output values in each time period.
[0074] The specific implementation method for solving the dynamic increase capacity of line current carrying capacity is:
[0075]
[0076]
[0077]
[0078] In the formula, For transmission lines In time period The meritorious trend; is the power transfer distribution factor; For transmission lines In time period Increased capacity; For Line The maximum power flow limit value is calculated based on the unit output value obtained by the economic dispatch model of the isolated island microgrid, and the transmission line In time period The dynamic capacity that needs to be increased within the network is used to generate the training data set.
[0079] Step 3: introduce the steps of equality constraint completion and inequality constraint correction, and use the neural network to predict the transmission lines that need to be dynamically increased in capacity and their dynamic increased capacity.
[0080] The economic dispatch model of the isolated microgrid and the dynamic capacity increase of the transmission line are considered as a constrained optimization problem:
[0081]
[0082] In the formula, represents the parameters of the optimization problem, Represents the solution to the optimization problem, training the neural network through supervised learning Output the complete solution to the optimization problem Partial set variables ; Complete solution to the optimization problem By partial set variables After completing the equation constraint, it is calculated to satisfy the DC power flow model; after the inequality constraint correction step, Corrected to , making it closer to satisfying the inequality constraint, the neural network Training is performed using the error back propagation algorithm on the SmoothL1Loss loss function:
[0083] .
[0084] like Figure 3 As shown, the neural network A multi-head neural network is used to complete the prediction task, and each head is specially trained to learn the mapping relationship between the microgrid load and the output of a specific unit. It includes three main structures: shared layer, parallel head and integrated output. The shared layer is used to extract the common features of the input data; the parallel head is responsible for learning these features and generating independent prediction values; the final output is obtained by integrating the prediction values output by each head;
[0085] At the same time, Sigmoid activation function is used between layers to realize nonlinear mapping to enhance the expressive power of the network. In addition, batch normalization layer is used to accelerate the training process and alleviate the gradient vanishing problem, thereby improving the stability and training efficiency of the model. Random inactivation layer is used to randomly discard the output of neurons to prevent overfitting and improve the generalization ability of the model.
[0086] The specific implementation method of equality constraint completion is:
[0087] First, assume In elements have been identified, the others elements can be solved explicitly based on the equality constraints. Formally, assuming that access to a function Make , here yes Determined in The partial set of solutions formed by the neural network After training, the output is: Depend on After completion with Formed in series, Then, for the short-term dynamic capacity increase problem of the isolated microgrid, assuming that the number of units in the system is , the number of lines is , the load quantity is , then , , feasible point . Considering the system input load values, and the feasible points are predicted by the neural network Before elements, corresponding to the system The output value of each unit is then obtained by the DC power flow model Dynamic capacity increase of transmission lines to achieve feasible points The completion of .
[0088] Inequality Constraint Correction
[0089] In order to further ensure the feasibility of the inequality constraints, a gradient-based inequality constraint correction step is proposed. This method calculates the gradient of the points satisfying the equality constraints in the variables. Take a gradient step towards the feasible region in the direction of , and map the output of the equality constraint completion step to the feasible region. Assume Represents the input point The operation of Let the partial set variable of the solution Move one step in the opposite direction of the gradient that violates the inequality constraint to make it closer to satisfying the inequality constraint. That is, for the learning rate , then:
[0090]
[0091]
[0092]
[0093] It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation manner. Any other implementation manners derived by those skilled in the art based on the technical solution of the present invention also fall within the scope of protection of the present invention.
Claims
1. A method for dynamic capacity increase of short-term lines with no feasible solution for dispatching and operating isolated microgrids, characterized in that: The following steps are involved: Step 1, generating a daily load curve sample of an island microgrid; Step 2: construct an economic dispatch model for an isolated microgrid, and solve the dynamic capacity increase of the line current carrying capacity according to the daily load curve sample of the isolated microgrid generated in step 1; The specific implementation method for solving the dynamic increase capacity of the line current carrying capacity in step 2 is: Where PL l,t is the active power flow of transmission line l in time period t; PIDF is the power transfer distribution factor; s l,t is the capacity added to the transmission line l in time period t; is the maximum power flow limit value of line l. The objective function of the economic dispatch model of the isolated microgrid is to minimize the daily operating cost. Based on the unit output value obtained by the economic dispatch model of the isolated microgrid, the dynamic capacity that needs to be increased for each transmission line l in time period t is calculated to generate a training data set, P g,t is the output of unit g in time period t, D b,t is the load of node b in time period t, t and g represent the sequence number of time period and unit respectively; T and G represent the set of time period and unit respectively; Step 3: introduce the steps of equality constraint completion and inequality constraint correction, and use the neural network to predict the lines that need to be dynamically increased in capacity and their dynamic increase capacity; The economic dispatch model of the isolated microgrid and the dynamic capacity increase of the transmission line are considered as a constrained optimization problem: Where x∈R d Represents the parameters of the optimization problem, y∈R n Represents the solution to the optimization problem, and trains the neural network N through supervised learning θ (x) outputting the partial set variable z of the complete solution y of the optimization problem; the complete solution y of the optimization problem is obtained by calculating the partial set variable z through equation constraint completion; After the inequality constraint correction step, y is corrected to To make it closer to satisfying the inequality constraint, the neural network N θ (x) Training using the error back propagation algorithm on the SmoothL1Loss loss function: Equation constraint completion: For the short-term dynamic capacity increase problem of isolated microgrid, the number of units in the system is ng, the number of lines is nbr, and the number of loads is m. x=[D1,D2,...,D m ], feasible point y = [P1, ..., P ng , s1, ..., s nbr ]; Considering the m load values input by the system, the neural network predicts the first ng elements of the feasible point y, which corresponds to the output values of the ng units in the system, and then the dynamic capacity increase of the nbr transmission lines is obtained by the DC power flow model to complete the feasible point y; Inequality constraint correction: ρ x (y) represents the input point The operation of x (y) Move the partial variable z of the solution one step in the opposite direction of the gradient that violates the inequality constraint, so that it is closer to satisfying the inequality constraint.
2. The method for dynamic capacity increase of short-term lines with no feasible solution for dispatching and operating an isolated island microgrid according to claim 1 is characterized by: The specific implementation method of step 1 is: selecting the actual load data of the isolated island microgrid to form daily load curve data, clustering the daily load curve data using the k-means algorithm, determining the number of clusters using the elbow method and the silhouette coefficient method, and classifying according to the number of clusters to obtain a cluster center curve; Setting up beta b factor, which distributes the load on each bus by the factor d b Press β b The factors are linearly scaled and normalized: Among them, d i is the percentage of each bus in the island microgrid to the total system load, β i is the linear scaling factor; Calculate the hourly change in total load of the microgrid γ t : Where D t is the total load of the microgrid in time period t, and for the clustered daily load curve samples, the γ t The mean μ t With variance Get a random number t=1,...,23,D1~U(0.95C,1.05C), where C is the load value of the cluster center curve of each type of sample in the first hour of the microgrid, and D1 is multiplied by the random number γ t The total daily load curve sample of the microgrid is obtained by multiplying the total daily load curve sample of the microgrid by the randomly generated load distribution coefficient d on each bus. i , generate daily load curve samples on each bus of the microgrid.
3. The method for dynamic capacity increase of short-term lines with no feasible solution for dispatching and operating an isolated island microgrid according to claim 1 is characterized by: The objective function of the economic dispatch model of the isolated island microgrid in step 2 is to minimize the daily operating cost: In the formula, t and g represent the time period and the sequence number of the unit respectively; T and G represent the time period and the set of units respectively; P g,t is the output of unit g in time period t; a g , b g 、c g are the constant term, linear term and quadratic term coefficients of the operating cost function of unit g respectively. The quadratic operating cost function is approximated by a piecewise linear function.
4. The method for dynamic capacity increase of short-term lines with no feasible solution for dispatching and operating an isolated island microgrid according to claim 3 is characterized by: The constraints of the island microgrid economic dispatch model include: power balance constraint, upper and lower limit constraints of unit output and unit climbing constraint, among which, Power balance constraints: In the formula, b represents the node number; B represents the set of nodes; D b,t is the load of node b in time period t; Unit output upper and lower limit constraints: In the formula, and are the minimum and maximum output of unit g respectively; Unit climbing constraints: Where, RD g and RU g They are respectively the limits on the output change rate of unit g in adjacent time periods.
5. The method for dynamically increasing the capacity of short-term lines when there is no feasible solution for dispatching and operating an isolated island microgrid according to claim 1 is characterized by: The neural network N θ (x) A multi-head neural network is used to complete the prediction task, and each head is specially trained to learn the mapping relationship between the microgrid load and the output of a specific unit. The neural network N θ (x) It includes three main structures: shared layer, parallel head and integrated output. The shared layer is used to extract the common features of the input data; The parallel heads are responsible for learning these features and generating independent predictions; the final output is obtained by integrating the predictions output by each head; At the same time, Sigmoid activation function is used between layers to achieve nonlinear mapping to enhance the network's expressiveness. In addition, batch normalization layer is used to accelerate the training process and alleviate the gradient vanishing problem, thereby improving the stability of the model and training efficiency. Use a random dropout layer to randomly discard the output of neurons to prevent overfitting and improve the generalization ability of the model.
