Optimized scheduling method and system for power system aggregator

Abstract

The invention discloses an optimal scheduling method for a power system aggregator. The method comprises the following steps: acquiring load data information of a target power system; classifying the load of the target power system; respectively establishing a user load side demand response optimization model for an aggregator, a power distribution network operator and a residential user in the target power system; and solving the constructed model and completing the optimal scheduling of the target power system aggregator. The invention also discloses a system for realizing the optimal scheduling method of the power system aggregator. According to the method, the user loads in the power system are classified, modeling and solving are carried out on the aggregator, the power distribution network operator and the resident user, optimal scheduling of the power system aggregator is achieved, multiple factors of the aggregator, the power distribution network operator and the resident in the power system are considered, and the optimal scheduling of the power system aggregator is achieved. The operation reliability and safety of the power system can be effectively improved, the reliability of the scheme is higher, and the accuracy is better.

Description

Technical Field

[0001] The present invention belongs to the field of electrical automation, and in particular relates to an optimization dispatching method and system for a power system aggregator. Background Art

[0002] With the development of economy and technology and the improvement of people's living standards, electricity has become an indispensable secondary energy source in people's production and life, bringing endless convenience to people's production and life. Therefore, ensuring the stable and reliable supply of electricity has become one of the most important tasks of the power system.

[0003] At present, with the rapid development of the power market, load aggregators (LAs) have become demand-side load regulation service agencies. As an entity that gathers huge adjustable resources, load aggregators have great ability to participate in demand response, which can greatly balance power supply and demand, reduce peak loads, and relieve grid pressure, thereby improving the stability and reliability of the power system.

[0004] However, in the current optimization and dispatching process of the power system, the power system still lacks a targeted solution to use load aggregators as demand-side load regulation service agencies and participate in the power system demand response. Therefore, the current power system aggregators have not participated in the power system demand response process on a large scale, which has greatly reduced the safety and reliability of the power system. Summary of the invention

[0005] One of the purposes of the present invention is to provide an optimization dispatching method for power system aggregators with high reliability and good accuracy.

[0006] A second object of the present invention is to provide a system for implementing the optimization dispatching method of the power system aggregator.

[0007] The optimization dispatching method of the power system aggregator provided by the present invention comprises the following steps:

[0008] S1. Obtain load data information of the target power system;

[0009] S2. Classify the loads of the target power system based on clustering algorithm and similarity index;

[0010] S3. According to the classification results obtained in step S2, a user load-side demand response optimization model is established for the aggregators, distribution network operators and residential users in the target power system;

[0011] S4. Solve the model constructed in step S3 to complete the optimal dispatch of the target power system aggregator.

[0012] The step S1 of acquiring the load data information of the target power system specifically includes the following steps:

[0013] Obtain load data information of a target power system; the load data information includes household load data information of the target power system.

[0014] The load classification of the target power system based on the clustering algorithm and the similarity index described in step S2 includes the following steps:

[0015] The K-means clustering algorithm is used to cluster the residential users of the target power system, taking the Euclidean distance as the indicator for evaluating similarity and combining it with the feature attention mechanism.

[0016] The step S2 specifically includes the following steps:

[0017] A. Extract load usage habit features and construct feature vectors:

[0018] The load usage habit characteristics include power consumption characteristics, load usage frequency characteristics, load usage time and duration characteristics, and load usage correlation characteristics;

[0019] Among them, the power consumption feature is the daily power consumption of each load; the load usage frequency feature is the daily usage frequency of each load; the load usage time and duration feature is the start time and working duration of each load, which is used to measure the user's fixed power usage time preference; the load usage correlation feature is used to describe the correlation usage of different loads;

[0020] For user u, the constructed feature vector is expressed as

[0021]

[0022] In the formula is the electricity consumption feature vector; is the daily power consumption of load i of user u; is the daily power consumption of load I of user u; Use frequency eigenvectors for loads; is the daily usage frequency of load i of user u; is the daily usage frequency of load I of user u; is the load usage time and duration characteristic vector; is the start time of load i of user u; is the start time of load I of user u; is the usage time of load i of user u; is the usage time of load I of user u; Use relevance eigenvectors for loads; is the correlation coefficient between load i and load j of user u; is the correlation coefficient between load I-1 and load I of user u;

[0023] Constructing the total feature vector for

[0024] Perform standardization to obtain the total feature vector after standardization for Where μ is the mean of the eigenvector and σ is the standard deviation of the eigenvector;

[0025] B. Introduce feature attention mechanism to adjust feature weights:

[0026] The total feature vector after normalization As input, the attention weight vector is calculated using a multi-layer perceptron

[0027]

[0028] In the formula is the attention weight coefficient corresponding to the input feature v of user u, V is the total number of features; σ() is the sigma function; W e is the training weight matrix of the multilayer perceptron; b e is the bias vector of the multilayer perceptron;

[0029] After training is completed, a weight value is generated for each feature of each user. Then calculate the standardized weight for

[0030] After the feature attention layer, the processed feature data Expressed as in is the feature data of the input feature v of user u before processing;

[0031] According to the external separation of clusters, the following formula is used as the network training loss function:

[0032]

[0033] Where L se The loss function value for network training; is the kth cluster center vector; is the jth cluster center vector; K is the number of cluster centers;

[0034] C. Initialize cluster centers:

[0035] Randomly select K cluster centers and assign weighted data points to the cluster center closest to them. The distance is calculated using the following formula:

[0036]

[0037] In the formula For data points and cluster centers The distance between is the cluster center The value on the feature dimension v;

[0038] D. Update cluster centers:

[0039] Calculate the new center point of each cluster, use the following formula to iteratively calculate and update the mean vector of all points in each class:

[0040]

[0041] In the formula is the mean vector of all points in each class; S k is the aggregation of all points in the current cluster k; |S k | is the number of all points in the current cluster k;

[0042] The evaluation index of cluster number is calculated using the following formula:

[0043]

[0044] Where WCSS(k) is the evaluation index of the number of clusters;

[0045] Iterative calculation and updating are performed until the change value of the evaluation index of the number of clusters obtained in two adjacent iterative processes is less than the set threshold value, and the iterative calculation and updating process is completed to obtain the optimal number of clusters.

[0046] The step S3 of establishing a user load side demand response optimization model for the aggregator specifically includes the following steps:

[0047] The following formula is used as the objective function of the aggregator's user load side demand response optimization model:

[0048]

[0049] Where C total,t is the total electricity sales revenue of the aggregator to the demand response users during period t, and p t (k) is the time-of-use electricity price issued by the aggregator to the k-th user, P t (k) is the power consumption of the kth user in time period t, Δt is the length of the time interval, and K is the total number of user categories; is the electricity purchase cost of the aggregator during period t, and c t is the price of electricity purchased by the aggregator from the distribution network operator during period t; is the demand response subsidy during period t, and is the total power consumption during the peak period t before demand response, is the total power consumption during the peak period t after demand response, is the incentive price for demand response subsidies, is the penalty price of demand response subsidy; T h For peak hours;

[0050] The following formula is used as the constraint condition of the aggregator's user load side demand response optimization model:

[0051] Power balance constraints:

[0052]

[0053] Where P t (k) is P t (k) is the power consumption of the kth category of users in period t;

[0054] Power purchase price constraints:

[0055]

[0056] Where c fmin is the lowest electricity price during peak hours; c fmax is the highest electricity price during peak hours; c f is the peak-time electricity price; c pmin The lowest electricity price during normal hours; c pmax The highest electricity price during normal hours; c p is the electricity price during normal hours; c gmin is the lowest electricity price during off-peak hours; c gmax is the highest electricity price during the off-peak period; c g The electricity price during off-peak hours;

[0057] Electricity sales price constraints:

[0058] p min ≤p t (k)≤p max

[0059] Where p min is the minimum electricity price; p max is the maximum electricity price;

[0060] Revenue constraints:

[0061] L>0

[0062] Where L is the objective function value of the aggregator’s user load side demand response optimization model.

[0063] The step S3 of establishing a user load side demand response optimization model for the distribution network operator specifically includes the following steps:

[0064] The following formula is used as the objective function of the user load side demand response optimization model of the distribution network operator:

[0065]

[0066] In the formula is the electric power sold by the distribution network operator to the aggregator during period t;

[0067] The following formula is used as the constraint condition of the user load side demand response optimization model of the distribution network operator:

[0068] Power Constraints:

[0069]

[0070] In the formula is the electric power sold directly to users in the t period before demand response; T is the total number of periods;

[0071] Peak-to-valley difference constraint:

[0072]

[0073] The peak-valley difference constraint indicates that the peak-valley difference of the total user load curve after demand response is not greater than the peak-valley difference of the total user load curve before demand response.

[0074] The step S3 of establishing a user load side demand response optimization model for residential users specifically includes the following steps:

[0075] The household loads of residential users are set to include shiftable loads and other loads; among which, shiftable loads are loads whose operating time can be shifted within a period acceptable to users, which cannot be interrupted once in operation, and whose working time can be adjusted without changing the working state;

[0076] For one duty cycle of the load, it is defined as

[0077]

[0078] In the formula P is the electric power consumed by load i in period t; i is the rated power of load i when working; is the working state of load i in period t; is the end period of load i; is the starting period of load i; s i is the number of continuous working periods of load i;

[0079] After clustering residential users, the aggregator evenly distributes the load on-time to make the number of loads on in each period as close as possible to reduce the load peak in any single period;

[0080] The actual number of working hours of the k-th user in period t Expressed as

[0081]

[0082] In the formula is the number of users who start load i for the mth time in the kth user group; is the duration of the mth working cycle of load i in the kth user group, and is the mth working start period of load i in the kth user group; is the total number of periods with the same number of workloads within the period, and is the mth working termination period of load i in the kth user group; is the floor rounding function;

[0083] Electricity power P of the kth user in period t after demand response t (k) is expressed as

[0084]

[0085] In the formula is the power consumed by the k-th user's shiftable load i in period t; is the total power consumed by other loads of the k-th user in period t;

[0086] The total electricity cost C(k) of the k-th user on that day is expressed as

[0087]

[0088] The electricity cost C of the kth user before demand response 0 (k) is expressed as

[0089]

[0090] In the formula is the number of working loads of the kth user i in period t before demand response; is the working power of the kth user load i in period t before demand response; c t The peak, off-peak and normal electricity prices set by distribution network operators to aggregators, and T fis the peak period, T p For normal period, T g It is the valley period;

[0091] The following formula is used as the first indicator of the user:

[0092]

[0093] The following formula is used as the user's second indicator:

[0094]

[0095] Where M is the first index value; S is the second index value; is the number of working loads of the kth user i in period t before demand response; is the working power value of the k-th user load i in period t before demand response; is the rated power of the kth user group when load i is working, The value is equal to the average power of all users of the kth category working with the same load rating; The working status of the load after demand response; is the working state of the k-th user load i in period t before demand response; is the load regulation potential coefficient of the load i of the kth user group.

[0096] The following formula is used as the objective function of the user load side demand response optimization model for residents:

[0097] max M β -ζ·S γ

[0098] Where β is the elasticity coefficient of the first indicator; γ is the elasticity coefficient of the second indicator; ζ is the set weight value;

[0099] The following formula is used as the constraint condition of the user load side demand response optimization model for residents:

[0100] Load working time constraints:

[0101]

[0102] In the formula is the earliest allowed start time of the k-th user load i, and r ij is the load-related utilization factor; The time when the k-th user load i is turned on for the mth time; is the latest allowed start time of the k-th user load i, and is the set of working cycles of the k-th user load i; is the time of m-th shutdown of load i of the k-th user;

[0103] Load working state constraints:

[0104]

[0105] Power Constraints:

[0106]

[0107] In the formula is the rated power of the kth type user load i;

[0108] Electricity cost constraints:

[0109] C(k)≤C 0 (k)

[0110] Power cap constraints:

[0111]

[0112] Where L max The upper limit of power that a household power supply line can carry.

[0113] Step S4 of solving the model constructed in step S3 specifically includes the following steps:

[0114] The user load side demand response optimization model of the distribution network operator is used as the upper model, the user load side demand response optimization model of the aggregator is used as the middle model, and the user load side demand response optimization model of the residents is used as the lower model;

[0115] The KKT method is used to transform the user load side demand response optimization model of residents into the additional constraints of the user load side demand response optimization model of the aggregator, thereby transforming the three-layer optimization model into a two-layer optimization model. The upper model is the user load side demand response optimization model of the distribution network operator, and the lower model is the user load side demand response optimization model of the aggregator after the conversion.

[0116] According to the obtained two-layer optimization model, based on the nested genetic algorithm, the following steps are used to solve it:

[0117] a. Set parameters: set the number of user groups to K, the number of time periods to T, the maximum number of iterations of the converted upper model to GT, and the maximum number of iterations of the converted lower model to GE;

[0118] b. Initialization of the upper model population: Generate a random initial population of the decision variables of the upper model, encode them, and use them as the current population of the upper model;

[0119] c. Determine whether the current population of the upper model meets the constraints: If it meets the constraints, the current population of the upper model is passed to the genetic algorithm of the lower model; otherwise, the fitness value of the current population of the upper model is set to 0, and jump to step g;

[0120] d. Initialization of the lower model population: Generate a random initial population of the lower model based on the current population transmitted by the upper model, perform encoding processing, and use it as the current population of the lower model;

[0121] e. Evaluate whether the current population of the lower model meets the constraints: If it meets the constraints, the objective function of the lower model is used as the fitness function, and the fitness value of each individual in the current population of the lower model is evaluated; if it does not meet the constraints, the fitness value of the current population of the lower model is set to 0;

[0122] f. Determine the termination condition of the lower model: Determine whether the number of iterations of the lower model reaches the set number GE: If it reaches, stop the calculation and record the optimal solution; if it does not reach, perform selection, crossover and mutation operations on the current population of the lower model, generate a new population and use it as the current population of the lower model, and return to step e;

[0123] g. Determine the termination condition of the upper model: Determine whether the number of iterations of the upper model reaches the set number GT: If it reaches it, stop the calculation and record the optimal solution; if it does not reach it, perform selection, crossover and mutation operations on the current population of the upper model, generate a new population and use it as the current population of the upper model, and return to step c;

[0124] h. Based on the optimal solution finally obtained, complete the solution of the model constructed in step S3.

[0125] The present invention also provides a system for implementing the optimization scheduling method of the power system aggregator, comprising a data acquisition module, a load classification module, a model building module and an optimization scheduling module; the data acquisition module, the load classification module, the model building module and the optimization scheduling module are connected in series in sequence; the data acquisition module is used to acquire the load data information of the target power system, and upload the data information to the load classification module; the load classification module is used to classify the load of the target power system based on the received data information, based on the clustering algorithm and the similarity index, and upload the data information to the model building module; the model building module is used to establish user load side demand response optimization models for the aggregators, distribution network operators and residential users in the target power system according to the received data information and the obtained classification results, and upload the data information to the optimization scheduling module; the optimization scheduling module is used to solve the constructed model according to the received data information to complete the optimization scheduling of the target power system aggregator.

[0126] The method and system for optimizing the scheduling of power system aggregators provided by the present invention classifies the user loads in the power system and models and solves the aggregators, distribution network operators and residential users of the power system, thereby not only realizing the optimized scheduling of the power system aggregators, but also taking into account multiple factors of the aggregators, distribution network operators and residents in the power system, and being able to effectively improve the operational reliability and safety of the power system, and the present invention has higher reliability and better accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0127] Figure 1 The figure is a schematic diagram of the method flow of the present invention.

[0128] Figure 2 Schematic diagram of the functional modules of the system of the present invention. DETAILED DESCRIPTION

[0129] like Figure 1 The method flow chart of the method of the present invention is shown as follows: The optimization dispatching method of the power system aggregator disclosed in the present invention comprises the following steps:

[0130] S1. Obtaining load data information of the target power system; specifically comprising the following steps:

[0131] Obtaining load data information of a target power system; the load data information includes household load data information of the target power system;

[0132] S2. Classify the loads of the target power system based on the clustering algorithm and the similarity index; comprising the following steps:

[0133] The K-means clustering algorithm is used to cluster the residential users of the target power system, taking the Euclidean distance as the indicator for evaluating similarity and combining it with the feature attention mechanism;

[0134] The specific implementation includes the following steps:

[0135] A. Extract load usage habit features and construct feature vectors:

[0136] The load usage habit characteristics include power consumption characteristics, load usage frequency characteristics, load usage time and duration characteristics, and load usage correlation characteristics;

[0137] Among them, the power consumption feature is the daily power consumption of each load; the load usage frequency feature is the daily usage frequency of each load; the load usage time and duration feature is the start time and working duration of each load, which is used to measure the user's fixed power usage time preference; the load usage correlation feature is used to describe the correlation usage of different loads;

[0138] For user u, the constructed feature vector is expressed as

[0139]

[0140] In the formula is the electricity consumption feature vector; is the daily power consumption of load i of user u; is the daily power consumption of load I of user u; Use frequency eigenvectors for loads; is the daily usage frequency of load i of user u; is the daily usage frequency of load I of user u; is the load usage time and duration characteristic vector; is the start time of load i of user u; is the start time of load I of user u; is the usage time of load i of user u; is the usage time of load I of user u; Use relevance eigenvectors for loads; is the correlation coefficient between load i and load j of user u; is the correlation coefficient between load I-1 and load I of user u;

[0141] Constructing the total feature vector for

[0142] Perform standardization to obtain the total feature vector after standardization for Where μ is the mean of the eigenvector and σ is the standard deviation of the eigenvector;

[0143] B. Introduce feature attention mechanism to adjust feature weights:

[0144] The user's feature vector has many features, so the feature attention mechanism is introduced to calculate the importance score of each feature to reflect the importance of different features to clustering. The training phase of the attention mechanism network guides the attention network learning through the loss function to effectively assign weights to each feature.

[0145] The total feature vector after normalization As input, the attention weight vector is calculated using a multi-layer perceptron

[0146]

[0147] In the formula is the attention weight coefficient corresponding to the input feature v of user u, V is the total number of features; σ() is the sigma function; We is the training weight matrix of the multilayer perceptron; b e is the bias vector of the multilayer perceptron;

[0148] After training is completed, a weight value is generated for each feature of each user. Then calculate the standardized weight for

[0149] After the feature attention layer, the processed feature data Expressed as in is the feature data of the input feature v of user u before processing;

[0150] According to the external separation of clusters, the following formula is used as the network training loss function:

[0151]

[0152] Where L se The loss function value for network training; is the kth cluster center vector; is the jth cluster center vector; K is the number of cluster centers;

[0153] C. Initialize cluster centers:

[0154] Randomly select K cluster centers and assign weighted data points to the cluster center closest to them. The distance is calculated using the following formula:

[0155]

[0156] In the formula For data points and cluster centers The distance between is the cluster center The value on the feature dimension v;

[0157] D. Update cluster centers:

[0158] Calculate the new center point of each cluster, use the following formula to iteratively calculate and update the mean vector of all points in each class:

[0159]

[0160] In the formula is the mean vector of all points in each class; S k is the aggregation of all points in the current cluster k; |S k | is the number of all points in the current cluster k;

[0161] The evaluation index of cluster number is calculated using the following formula:

[0162]

[0163] Where WCSS(k) is the evaluation index of the number of clusters;

[0164] Iterative calculation and updating are performed until the change value of the evaluation index of the number of clusters obtained in two adjacent iterative processes is less than the set threshold value, and the iterative calculation and updating process is completed to obtain the optimal number of clusters;

[0165] S3. According to the classification results obtained in step S2, a user load-side demand response optimization model is established for the aggregators, distribution network operators and residential users in the target power system;

[0166] The scenario of this step is that the distribution network operator is responsible for managing the power supply of a region and entrusts an aggregator to gather and classify users in a certain community. The aggregator then formulates differentiated electricity prices for different residential user groups and provides a dispatching strategy that can shift loads. By signing an agreement with the user, the user load-side demand response is achieved.

[0167] In the specific implementation, a user load side demand response optimization model is established for the aggregator, which specifically includes the following steps:

[0168] The following formula is used as the objective function of the aggregator's user load side demand response optimization model:

[0169]

[0170] Where C total,t is the total electricity sales revenue of the aggregator to the demand response users during period t, and p t (k) is the time-of-use electricity price issued by the aggregator to the k-th user, P t (k) is the power consumption of the kth user in time period t, Δt is the length of the time interval, and K is the total number of user categories; is the electricity purchase cost of the aggregator during period t, and c t is the price of electricity purchased by the aggregator from the distribution network operator during period t; is the demand response subsidy during period t, and is the total power consumption during the peak period t before demand response, is the total power consumption during the peak period t after demand response, is the incentive price for demand response subsidies, is the penalty price of demand response subsidy; T h For peak hours;

[0171] The following formula is used as the constraint condition of the aggregator's user load side demand response optimization model:

[0172] Power balance constraint: The power purchased by the aggregator is equal to the sum of the power sold to various demand response users, expressed as

[0173]

[0174] Where P t (k) is P t (k) is the power consumption of the kth category of users in period t;

[0175] Power purchase price constraints:

[0176]

[0177] Where c fmin is the lowest electricity price during peak hours; c fmax is the highest electricity price during peak hours; c f is the peak-time electricity price; c pmin The lowest electricity price during normal hours; c pmax The highest electricity price during normal hours; c p is the electricity price during normal hours; c gmin is the lowest electricity price during off-peak hours; c gmax is the highest electricity price during the off-peak period; c g The electricity price during off-peak hours;

[0178] Electricity sales price constraints:

[0179] p min ≤p t (k)≤p max

[0180] Where p min is the minimum electricity price; p max is the maximum electricity price;

[0181] Revenue constraints:

[0182] L>0

[0183] Where L is the objective function value of the aggregator’s user load side demand response optimization model;

[0184] Establishing a user load side demand response optimization model for distribution network operators includes the following steps:

[0185] The goal of the distribution network operator is to minimize the peak-to-valley difference of the total load power curve of residential users within the jurisdiction. However, in reality, only the daily load curve of users participating in demand response has changed, and the load curves of other users have not changed. Therefore, the goal of the distribution network operator is to minimize the peak-to-valley difference of the power curve sold to the aggregator; therefore, the following formula is used as the objective function of the user load side demand response optimization model of the distribution network operator:

[0186]

[0187] In the formula is the electric power sold by the distribution network operator to the aggregator during period t;

[0188] The following formula is used as the constraint condition of the user load side demand response optimization model of the distribution network operator:

[0189] Electricity Constraint: After demand response, the total electricity sold by the distribution network operator to the aggregator should be equal to the total electricity sold directly to users before demand response, expressed as

[0190]

[0191] In the formula is the electric power sold directly to users in the t period before demand response; T is the total number of periods;

[0192] Peak-valley difference constraint: After demand response, the peak-valley difference of the user's total load curve should be smaller than the initial peak-valley difference, expressed as

[0193]

[0194]

[0195] The peak-valley difference constraint indicates that the peak-valley difference of the total user load curve after demand response is not greater than the peak-valley difference of the total user load curve before demand response;

[0196] A user load side demand response optimization model is established for residential users, which specifically includes the following steps:

[0197] The household loads of residential users are set to include shiftable loads and other loads; among which, shiftable loads are loads whose operating time can be shifted within a period acceptable to users, which cannot be interrupted once in operation, and whose working time can be adjusted without changing the working state;

[0198] For one duty cycle of the load, it is defined as

[0199]

[0200] In the formula P is the electric power consumed by load i in period t; i is the rated power of load i when working; is the working state of load i in period t; is the end period of load i; is the starting period of load i; s i is the number of continuous working periods of load i;

[0201] After clustering residential users, the aggregator optimizes the working time of the load that can be shifted for each type of user. In order to avoid power consumption peaks caused by the load of all users being shifted to the same time period during optimization, a balanced startup strategy is adopted, that is, the load startup time is evenly distributed, so that the number of loads started in each time period is as close as possible, so as to reduce the load peak in any single time period.

[0202] Therefore, the actual number of users in the kth category working at load i in period t is Expressed as

[0203]

[0204] In the formula is the number of users who start load i for the mth time in the kth user group; is the duration of the mth working cycle of load i in the kth user group, and is the mth working start period of load i in the kth user group; is the total number of periods with the same number of workloads within the period, and is the mth working termination period of load i in the kth user group; is the floor rounding function;

[0205] The electricity cost of residential users is composed of the product of the total electricity consumed by the two types of loads, namely, the shiftable load and other loads, and the corresponding electricity price. The operating time of other loads is not allowed to be adjusted.

[0206] Electricity power P of the kth user in period t after demand response t (k) is expressed as

[0207]

[0208] In the formula is the power consumed by the k-th user's shiftable load i in period t; is the total power consumed by other loads of the k-th user in period t;

[0209] The total electricity cost C(k) of the k-th user on that day is expressed as

[0210]

[0211] The electricity cost C of the kth user before demand response 0 (k) is expressed as

[0212]

[0213] In the formula is the number of working loads of the kth user i in period t before demand response; is the working power of the kth user load i in period t before demand response; c t The peak, off-peak and normal electricity prices set by distribution network operators to aggregators, and T f is the peak period, T p For normal period, T g It is the valley period;

[0214] The following formula is used as the first indicator of the user:

[0215]

[0216] The following formula is used as the user's second indicator, where the second indicator is used to reflect the user's comfort level in electricity use:

[0217]

[0218] Where M is the first index value; S is the second index value; is the number of working loads of the kth user i in period t before demand response; is the working power value of the k-th user load i in period t before demand response; is the rated power of the kth user group when load i is working, The value is equal to the average power of all users of the kth category working with the same load rating; is the working state of the load; is the working state of the k-th user load i in period t before demand response; is the load regulation potential coefficient of the load i of the kth user group.

[0219] The following formula is used as the objective function of the user load side demand response optimization model for residents:

[0220] max M β -ζ·S γ

[0221] Where β is the elasticity coefficient of the first indicator; γ is the elasticity coefficient of the second indicator; ζ is the set weight value;

[0222] The following formula is used as the constraint condition of the user load side demand response optimization model for residents:

[0223] Load working time constraint: When adjusting the shiftable period of the load to be optimized, consider the load association usage habits and use the association coefficient r ij Add users' electricity usage habits to time constraints to avoid disrupting the original load coordination relationship during optimization; the load-related usage habits of typical users in each category of users are regarded as the overall electrical appliance usage habits of the user group in this category;

[0224] The constraint formula is expressed as

[0225]

[0226] In the formula is the earliest allowed start time of the k-th user load i, and r ij is the load-related utilization factor; The time when the k-th user load i is turned on for the mth time; is the latest allowed start time of the k-th user load i, and is the set of working cycles of the k-th user load i; is the time of m-th shutdown of load i of the k-th user;

[0227] Load working state constraints:

[0228]

[0229] Power Constraints:

[0230]

[0231] In the formula is the rated power of the kth type user load i;

[0232] Electricity cost constraints:

[0233] C(k)≤C 0 (k)

[0234] Power cap constraints:

[0235]

[0236] Where L max The upper limit of the power that can be carried by the household power supply line;

[0237] S4. Solve the model constructed in step S3 to complete the optimal dispatch of the target power system aggregator; specifically, the steps include:

[0238] The user load side demand response optimization model of the distribution network operator is used as the upper model, the user load side demand response optimization model of the aggregator is used as the middle model, and the user load side demand response optimization model of the residents is used as the lower model;

[0239] The KKT method is used to transform the user load side demand response optimization model of residents into the additional constraints of the user load side demand response optimization model of the aggregator, thereby transforming the three-layer optimization model into a two-layer optimization model. The upper model is the user load side demand response optimization model of the distribution network operator, and the lower model is the user load side demand response optimization model of the aggregator after the conversion.

[0240] According to the obtained two-layer optimization model, based on the nested genetic algorithm, the following steps are used to solve it:

[0241] a. Set parameters: set the number of user groups to K, the number of time periods to T, the maximum number of iterations of the converted upper model to GT, and the maximum number of iterations of the converted lower model to GE;

[0242] b. Initialization of the upper model population: Generate a random initial population of the decision variables of the upper model, encode them, and use them as the current population of the upper model;

[0243] c. Determine whether the current population of the upper model meets the constraints: If it meets the constraints, the current population of the upper model is passed to the genetic algorithm of the lower model; otherwise, the fitness value of the current population of the upper model is set to 0, and jump to step g;

[0244] d. Initialization of the lower model population: Generate a random initial population of the lower model based on the current population transmitted by the upper model, perform encoding processing, and use it as the current population of the lower model;

[0245] e. Evaluate whether the current population of the lower model meets the constraints: If it meets the constraints, the objective function of the lower model is used as the fitness function, and the fitness value of each individual in the current population of the lower model is evaluated; if it does not meet the constraints, the fitness value of the current population of the lower model is set to 0;

[0246] f. Determine the termination condition of the lower model: Determine whether the number of iterations of the lower model reaches the set number GE: If it reaches, stop the calculation and record the optimal solution; if it does not reach, perform selection, crossover and mutation operations on the current population of the lower model, generate a new population and use it as the current population of the lower model, and return to step e;

[0247] g. Determine the termination condition of the upper model: Determine whether the number of iterations of the upper model reaches the set number GT: If it reaches it, stop the calculation and record the optimal solution; if it does not reach it, perform selection, crossover and mutation operations on the current population of the upper model, generate a new population and use it as the current population of the upper model, and return to step c;

[0248] h. Based on the optimal solution finally obtained, complete the solution of the model constructed in step S3.

[0249] like Figure 2 The figure shows a schematic diagram of the functional modules of the system of the present invention: the system for realizing the optimization scheduling method of the power system aggregator disclosed in the present invention comprises a data acquisition module, a load classification module, a model construction module and an optimization scheduling module; the data acquisition module, the load classification module, the model construction module and the optimization scheduling module are connected in series in sequence; the data acquisition module is used to acquire the load data information of the target power system, and upload the data information to the load classification module; the load classification module is used to classify the load of the target power system based on the clustering algorithm and the similarity index according to the received data information, and upload the data information to the model construction module; the model construction module is used to establish user load side demand response optimization models for the aggregators, distribution network operators and residential users in the target power system according to the received data information and the obtained classification results, and upload the data information to the optimization scheduling module; the optimization scheduling module is used to solve the constructed model according to the received data information to complete the optimization scheduling of the target power system aggregator.

Claims

1. An optimization dispatching method for a power system aggregator, comprising the following steps: S1. Obtain load data information of the target power system; S2. Classify the loads of the target power system based on clustering algorithm and similarity index; S3. According to the classification results obtained in step S2, a user load-side demand response optimization model is established for the aggregators, distribution network operators and residential users in the target power system; S4. Solve the model constructed in step S3 to complete the optimal dispatch of the target power system aggregator.

2. The optimization dispatching method of the power system aggregator according to claim 1 is characterized in that The step S1 of acquiring the load data information of the target power system specifically includes the following steps: Obtain load data information of a target power system; the load data information includes household load data information of the target power system.

3. The optimization dispatching method of the power system aggregator according to claim 1 is characterized in that The load classification of the target power system based on the clustering algorithm and the similarity index described in step S2 includes the following steps: The K-means clustering algorithm is used to cluster the residential users of the target power system, taking the Euclidean distance as the indicator for evaluating similarity and combining it with the feature attention mechanism.

4. The optimization dispatching method of the power system aggregator according to claim 3 is characterized in that The step S2 specifically includes the following steps: A. Extract load usage habit features and construct feature vectors: The load usage habit characteristics include power consumption characteristics, load usage frequency characteristics, load usage time and duration characteristics, and load usage correlation characteristics; Among them, the power consumption feature is the daily power consumption of each load; the load usage frequency feature is the daily usage frequency of each load; the load usage time and duration feature is the start time and working duration of each load, which is used to measure the user's fixed power usage time preference; the load usage correlation feature is used to describe the correlation usage of different loads; For user u, the constructed feature vector is expressed as In the formula is the electricity consumption feature vector; is the daily power consumption of load i of user u; is the daily power consumption of load I of user u; Use frequency eigenvectors for loads; is the daily usage frequency of load i of user u; is the daily usage frequency of load I of user u; is the load usage time and duration characteristic vector; is the start time of load i of user u; is the start time of load I of user u; is the usage time of load i of user u; is the usage time of load I of user u; Use relevance eigenvectors for loads; is the correlation coefficient between load i and load j of user u; is the correlation coefficient between load I-1 and load I of user u; Constructing the total feature vector for Perform standardization to obtain the total feature vector after standardization for Where μ is the mean of the eigenvector and σ is the standard deviation of the eigenvector; B. Introduce feature attention mechanism to adjust feature weights: The total feature vector after normalization As input, the attention weight vector is calculated using a multi-layer perceptron In the formula is the attention weight coefficient corresponding to the input feature v of user u, V is the total number of features; σ() is the sigma function; W e is the training weight matrix of the multilayer perceptron; b e is the bias vector of the multilayer perceptron; After training is completed, a weight value is generated for each feature of each user. Then calculate the standardized weight for After the feature attention layer, the processed feature data Expressed as in is the feature data of the input feature v of user u before processing; According to the external separation of clusters, the following formula is used as the network training loss function: Where L se The loss function value for network training; is the kth cluster center vector; is the jth cluster center vector; K is the number of cluster centers; C. Initialize cluster centers: Randomly select K cluster centers and assign weighted data points to the cluster center closest to them. The distance is calculated using the following formula: In the formula For data points and cluster centers The distance between is the cluster center The value on the feature dimension v; D. Update cluster centers: Calculate the new center point of each cluster, use the following formula to iteratively calculate and update the mean vector of all points in each class: In the formula is the mean vector of all points in each class; S k is the aggregation of all points in the current cluster k; |S k | is the number of all points in the current cluster k; The evaluation index of cluster number is calculated using the following formula: Where WCSS(k) is the evaluation index of the number of clusters; Iterative calculation and updating are performed until the change value of the evaluation index of the number of clusters obtained in two adjacent iterative processes is less than the set threshold value, and the iterative calculation and updating process is completed to obtain the optimal number of clusters.

5. The optimization dispatching method of the power system aggregator according to claim 4 is characterized in that The step S3 of establishing a user load side demand response optimization model for the aggregator specifically includes the following steps: The following formula is used as the objective function of the aggregator's user load side demand response optimization model: Where C total,t is the total electricity sales revenue of the aggregator to the demand response users during period t, and p t (k) is the time-of-use electricity price issued by the aggregator to the k-th user, P t (k) is the power consumption of the kth user in time period t, Δt is the length of the time interval, and K is the total number of user categories; is the electricity purchase cost of the aggregator during period t, and c t is the price of electricity purchased by the aggregator from the distribution network operator during period t; is the demand response subsidy during period t, and is the total power consumption during the peak period t before demand response, is the total power consumption during the peak period t after demand response, is the incentive price for demand response subsidies, is the penalty price of demand response subsidy; T h For peak hours; The following formula is used as the constraint condition of the aggregator's user load side demand response optimization model: Power balance constraints: Where P t (k) is P t (k) is the power consumption of the kth category of users in period t; Power purchase price constraints: Where c fmin is the lowest electricity price during peak hours; c fmax is the highest electricity price during peak hours; c f is the peak-time electricity price; c pmin The lowest electricity price during normal hours; c pmax The highest electricity price during normal hours; c p is the electricity price during normal hours; c gmin is the lowest electricity price during off-peak hours; c gmax is the highest electricity price during the off-peak period; c g The electricity price during off-peak hours; Electricity sales price constraints: p min ≤p t (k)≤p max Where p min is the minimum electricity price; p max is the maximum electricity price; Revenue constraints: L＞0 Where L is the objective function value of the aggregator’s user load side demand response optimization model.

6. The optimization dispatching method of the power system aggregator according to claim 5 is characterized in that The step S3 of establishing a user load side demand response optimization model for the distribution network operator specifically includes the following steps: The following formula is used as the objective function of the user load side demand response optimization model of the distribution network operator: In the formula is the electric power sold by the distribution network operator to the aggregator during period t; The following formula is used as the constraint condition of the user load side demand response optimization model of the distribution network operator: Power Constraints: In the formula is the electric power sold directly to users in the t period before demand response; T is the total number of periods; Peak-to-valley difference constraint: The peak-valley difference constraint indicates that the peak-valley difference of the total user load curve after demand response is not greater than the peak-valley difference of the total user load curve before demand response.

7. The optimization dispatching method of the power system aggregator according to claim 6 is characterized in that The step S3 of establishing a user load side demand response optimization model for residential users specifically includes the following steps: The household loads of residential users are set to include shiftable loads and other loads; among which, shiftable loads are loads whose operating time can be shifted within a period acceptable to users, which cannot be interrupted once in operation, and whose working time can be adjusted without changing the working state; For one duty cycle of the load, it is defined as In the formula P is the electric power consumed by load i in period t; i is the rated power of load i when working; is the working state of load i in period t; is the end period of load i; is the starting period of load i; s i is the number of continuous working periods of load i; After clustering residential users, the aggregator evenly distributes the load on-time to make the number of loads on in each period as close as possible to reduce the load peak in any single period; The actual number of working hours of the k-th user in period t Expressed as In the formula is the number of users who start load i for the mth time in the kth user group; is the duration of the mth working cycle of load i in the kth user group, and is the mth working start period of load i in the kth user group; is the total number of periods with the same number of workloads within the period, and is the mth working termination period of load i in the kth user group; is the floor rounding function; Electricity power P of the kth user in period t after demand response t (k) is expressed as In the formula is the power consumed by the k-th user's shiftable load i in period t; is the total power consumed by other loads of the k-th user in period t; The total electricity cost C(k) of the k-th user on that day is expressed as The electricity cost C of the kth user before demand response 0 (k) is expressed as In the formula is the number of working loads of the kth user i in period t before demand response; is the working power of the kth user load i in period t before demand response; c t The peak, off-peak and normal electricity prices set by distribution network operators to aggregators, and T f is the peak period, T p For normal period, T g It is the valley period; The following formula is used as the first indicator of the user: The following formula is used as the second user indicator: Where M is the first index value; S is the second index value; is the number of working loads of the kth user i in period t before demand response; is the working power value of the k-th user load i in period t before demand response; is the rated power of the kth user group when load i is working, The value is equal to the average power of all users of the kth category working with the same load rating; is the working state of the load; is the working state of the k-th user load i in period t before demand response; is the load regulation potential coefficient of the load i of the k-th user group. ; The following formula is used as the objective function of the user load side demand response optimization model for residents: max M β -ζ·S γ Where β is the elasticity coefficient of the first indicator; γ is the elasticity coefficient of the second indicator; ζ is the set weight value; The following formula is used as the constraint condition of the user load side demand response optimization model for residents: Load working time constraints: In the formula is the earliest allowed start time of the k-th user load i, and r ij is the load-related utilization factor; The time when the k-th user load i is turned on for the mth time; is the latest allowed start time of the k-th user load i, and is the set of working cycles of the k-th user load i; is the time of m-th shutdown of load i of the k-th user; Load working state constraints: Power Constraints: In the formula is the rated power of the kth type user load i; Electricity cost constraints: C(k)≤C 0 (k) Power cap constraints: Where L max The upper limit of power that a household power supply line can carry.

8. The optimization dispatching method of the power system aggregator according to claim 7 is characterized in that Step S4 of solving the model constructed in step S3 specifically includes the following steps: The user load side demand response optimization model of the distribution network operator is used as the upper model, the user load side demand response optimization model of the aggregator is used as the middle model, and the user load side demand response optimization model of the residents is used as the lower model; The KKT method is used to transform the user load side demand response optimization model of residents into the additional constraints of the user load side demand response optimization model of the aggregator, thereby transforming the three-layer optimization model into a two-layer optimization model. The upper model is the user load side demand response optimization model of the distribution network operator, and the lower model is the user load side demand response optimization model of the aggregator after the conversion. According to the obtained two-layer optimization model, based on the nested genetic algorithm, the following steps are used to solve it: a. Set parameters: set the number of user groups to K, the number of time periods to T, the maximum number of iterations of the converted upper model to GT, and the maximum number of iterations of the converted lower model to GE; b. Initialization of the upper model population: Generate a random initial population of the decision variables of the upper model, encode them, and use them as the current population of the upper model; c. Determine whether the current population of the upper model meets the constraints: If it meets the constraints, the current population of the upper model is passed to the genetic algorithm of the lower model; otherwise, the fitness value of the current population of the upper model is set to 0, and jump to step g; d. Initialization of the lower model population: Generate a random initial population of the lower model based on the current population transmitted by the upper model, perform encoding processing, and use it as the current population of the lower model; e. Evaluate whether the current population of the lower model meets the constraints: If it meets the constraints, the objective function of the lower model is used as the fitness function, and the fitness value of each individual in the current population of the lower model is evaluated; if it does not meet the constraints, the fitness value of the current population of the lower model is set to 0; f. Determine the termination condition of the lower model: Determine whether the number of iterations of the lower model reaches the set number GE: If it reaches, stop the calculation and record the optimal solution; if it does not reach, perform selection, crossover and mutation operations on the current population of the lower model, generate a new population and use it as the current population of the lower model, and return to step e; g. Determine the termination condition of the upper model: Determine whether the number of iterations of the upper model reaches the set number GT: If it reaches it, stop the calculation and record the optimal solution; if it does not reach it, perform selection, crossover and mutation operations on the current population of the upper model, generate a new population and use it as the current population of the upper model, and return to step c; h. Based on the optimal solution finally obtained, complete the solution of the model constructed in step S3.

9. A system for implementing the optimization dispatching method of the power system aggregator according to any one of claims 1 to 8, characterized in that It includes a data acquisition module, a load classification module, a model building module and an optimization scheduling module; the data acquisition module, the load classification module, the model building module and the optimization scheduling module are connected in series in sequence; the data acquisition module is used to obtain the load data information of the target power system and upload the data information to the load classification module; The load classification module is used to classify the loads of the target power system according to the received data information based on the clustering algorithm and similarity index, and upload the data information to the model construction module; The model building module is used to establish user load side demand response optimization models for aggregators, distribution network operators and residential users in the target power system according to the received data information and the obtained classification results, and upload the data information to the optimization scheduling module; The optimization scheduling module is used to solve the constructed model according to the received data information and complete the optimization scheduling of the target power system aggregator.

Images