Load optimization regulation and control method and device

Abstract

The invention relates to the technical field of intelligent power utilization, and particularly provides a load optimization regulation and control method and device, and the method comprises the steps: taking user load data of a current time period as the input of a pre-trained impact load prediction model, and obtaining a user impact load of a prediction time period outputted by the pre-trained impact load prediction model; based on the user impact load of the prediction time period, determining the demand response amount required by the impact load of the prediction time period and the demand response amount reserved in the day to deal with the short-time impact load; substituting the demand response quantity required by the impact load in the prediction time period and the demand response quantity reserved in the day to cope with the short-time impact load into a pre-constructed load optimization scheduling model, and solving to obtain an optimization result; according to the technical scheme provided by the invention, through load identification and screening, demand response capacity prediction and day-ahead and intra-day optimization scheduling, the problem of influence on power utilization of users caused by a load rigid control mode is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent electricity utilization technology, and in particular to a load optimization control method and device. Background Art

[0002] With the in-depth implementation of the energy revolution strategy, the proportion of renewable energy installed capacity and power generation has continued to increase, and the proportion of terminal electrification has continued to increase. However, the rapid development of renewable energy is not matched with the planning and development of the power grid, resulting in increasingly serious problems of wind and solar power abandonment. The high volatility and randomness of both the source and the load have put forward higher requirements for the balance and regulation of the power grid, and the safe and reliable operation of the power grid and power supply guarantee are facing huge challenges.

[0003] With the rapid development of industrialization and urbanization, people's demand for electricity is growing day by day. However, in many areas, the power grid was built earlier, and the technology and equipment are relatively backward, which cannot meet the peak load power demand. It is necessary to expand the power grid to improve the power supply capacity of the power grid. At the same time, the rapid development of distributed power sources (such as solar energy) has brought about changes in the power grid trend, making the problem of power grid expansion more complicated.

[0004] There is a type of load in the power grid that imposes a sudden and large amount of power demand on the power grid system in a very short period of time, but the duration is limited. These loads are called impact loads. The excessive instantaneous power of the impact load occupies a large amount of power grid capacity, forcing the power grid to set up redundant systems or adopt higher-standard equipment to ensure sufficient capacity and flexibility to ensure safe operation. Considering that there is a problem of high cost to directly optimize power grid equipment to increase capacity, especially many devices are located in urban areas, optimization is difficult and costly. At present, in areas where capacity expansion is difficult, the main method used is load rigidity control, which requires non-critical loads to downgrade or shut down, and resume normal operation after the impact subsides. However, load rigidity control requires other users to reduce their loads, which affects users' electricity consumption and production operations.

[0005] Demand Response (DR) is an electricity management strategy that uses economic means to adjust electricity consumption behavior on the demand side in response to changes in grid supply and demand. It is seen as one of the effective means to achieve grid expansion and improve grid flexibility and reliability. Demand response can reduce pressure during peak load periods and improve grid operation efficiency, thereby indirectly achieving "virtual expansion" of the grid. Through demand response measures, electricity demand can be adjusted dynamically. When a large-scale load increase is predicted, the electricity demand of some non-critical loads can be reduced through economic incentives or direct control of agreements, thereby reducing the impact of shock loads on the grid. Demand response provides economic compensation for loads and delays the replacement or addition of new equipment in the grid, which can achieve a win-win situation for grid companies and users.

[0006] Currently, there are patents for research on demand response, such as:

[0007] The invention discloses a distribution network optimization dispatching method considering the uncertainty of comprehensive demand response (CN117543582A), comprising: constructing a price-type demand response model and an interruptible load model to obtain a comprehensive load demand response uncertainty model; determining a first objective function, establishing operation constraints, and obtaining a distribution network robust optimization dispatching model; constructing an uncertain opportunity constraint programming min model based on the comprehensive load constraints, the first objective function and the operation constraints; converting the opportunity constraint programming min model into an obstacle term of the objective function by using an interior point method, obtaining a second objective function, and reconstructing a clear class equivalent distribution network optimization dispatching model; using YALMIP modeling and solving it with the help of CPLEX commercial software to obtain a dispatching result of the distribution network, which can overcome the influence of demand response uncertainty on the dispatching of the distribution network and improve the economy and stability.

[0008] However, this dispatching method dispatches the power grid based on economic optimization, and fails to solve problems such as insufficient power supply capacity of the power grid as the load increases, and shortage of line capacity caused by excessive instantaneous load changes when there are impact loads in the distribution network.

[0009] For the identification of impact load, a method for identifying impact load attributes on the distribution transformer outgoing line side (CN114757232A) includes: obtaining the instantaneous power signal and average power signal of the equipment that generates impact load at each time node within a preset time period; generating multiple load attributes that may constitute the instantaneous power signal of the corresponding time node at each time node; evaluating the respective occurrence probabilities of the multiple load attributes according to the average power at each time node; determining the optimal solution of the load attribute according to the occurrence probability, and outputting it as the load attribute of the equipment. The present invention is a non-invasive, high-precision method that can be set up on the outgoing line side of the distribution transformer, and is suitable for impact load behavior.

[0010] This method mainly considers the instantaneous power signal and average power signal of the load, and does not consider indicators that describe the load power consumption characteristics such as load kurtosis, dispersion coefficient, waveform factor and pulse factor, and cannot fully and accurately characterize the impact load characteristics. Summary of the invention

[0011] In order to overcome the above-mentioned defects, the present invention proposes a load optimization control method and device.

[0012] In a first aspect, a load optimization control method is provided, the load optimization control method comprising:

[0013] Using the user load data of the current period as the input of the pre-trained impact load prediction model, and obtaining the user impact load of the prediction period output by the pre-trained impact load prediction model;

[0014] Determine the required demand response amount for the impact load in the forecast period and the demand response amount reserved in the day ahead to cope with the short-term impact load based on the user impact load in the forecast period;

[0015] Substituting the demand response amount required for the impact load in the forecast period and the demand response amount reserved for the short-term impact load in the day ahead into the pre-built load optimization scheduling model and solving it to obtain the optimization result;

[0016] Based on the optimization results, a load optimization control scheme is obtained;

[0017] Among them, the optimization result includes at least one of the following: photovoltaic power generation power, gas turbine output power, energy storage charging and discharging power, and demand response output power of load-side resources.

[0018] Preferably, the training process of the pre-trained impact load prediction model includes:

[0019] Obtain historical user load data, and determine impact load data in the historical user load data;

[0020] Using historical user load data and impact load data in the historical user load data to construct training data;

[0021] Using the training data to train an initial impact load prediction model to obtain the pre-trained impact load prediction model;

[0022] Wherein, the initial impact load prediction model is based on a convolutional neural network model of a long short-term memory network.

[0023] Further, the determining of the impact load data in the historical user load data includes:

[0024] Obtain the index characteristics of historical user load data, and use the principal component analysis algorithm to obtain the dimensionality reduction characteristics corresponding to the index characteristics of the historical user load data;

[0025] An outlier factor detection algorithm is used to select historical user load data whose outlier factors exceed a threshold value from the dimensionality reduction features, and the selected historical user load data is used as impact load data in the historical user load data.

[0026] Furthermore, the indicator characteristics of the historical user load data are as follows:

[0027]

[0028] In the above formula, σ is the standard deviation of historical user load data, x i is the load of the ith sampling point, is the mean of historical user load data, n is the number of sampling points, Ck is the load kurtosis, C d is the load dispersion coefficient, C s is the load waveform factor, C if is the load pulse factor.

[0029] Furthermore, before using the training data to train the initial impact load prediction model, the method includes:

[0030] The crown porcupine optimization algorithm is used to optimize the convolution kernel step size of the convolutional neural network in the initial impact load prediction model. During the optimization process, the fitness used is the prediction accuracy of the impact load prediction model.

[0031] Preferably, the demand response amount required for the impact load in the forecast period is as follows:

[0032]

[0033] In the above formula, DS is the downstream node set of node i i The required demand response amount for the impact load in period t, D im,i,t is the user impact load of node i in period t, D k,t is the load of node k in period t, S l,i is the line capacity of node i.

[0034] Furthermore, the demand response quantity reserved for short-term impact load is as follows:

[0035]

[0036] In the above formula, The demand response amount reserved for node i to deal with short-term impact loads before time period t, D impact,i,in is the maximum value of user impact load of node i in period t.

[0037] Preferably, the pre-built load optimization scheduling model includes: an objective function with the goal of minimizing the total scheduling cost and its corresponding constraints.

[0038] Furthermore, the objective function is as follows:

[0039] C Total =min(C Oper +C DR )

[0040] In the above formula, C Total is the objective function value, C Oper is the total operating cost of resources on the load side, C DR Compensation for interrupted load.

[0041] Furthermore, the total operating cost of the load-side resources is as follows:

[0042]

[0043] The interruption load compensation fee is as follows:

[0044]

[0045] In the above formula, P Bat (t) is the charging and discharging power of the battery during period t, K Bat is the life loss cost coefficient caused by the battery charge and discharge, △t is the time interval, T is the total number of optimization periods, γ GT is the fuel cost of the microturbine, is the startup cost of the gas turbine during period t, is the shutdown cost of the gas turbine during period t, is the operating coefficient of the gas turbine during period t, is the shutdown coefficient of the gas turbine during period t, λ e is the demand response unit compensation coefficient, △L e,cut_t is the demand response dispatch quantity in period t, λ re is the day-ahead reserved demand response coefficient, △L re,t Reserve demand response capacity for time period t.

[0046] Furthermore, the constraints are as follows:

[0047] P Load (t)+P Bat_Cha (t)+P Sell (t)≤P PV (t)+P Bat_Dis (t)+P GT (t)+P Buy (t)+P DR (t)

[0048]

[0049] U DR P DR_min，t <P DR (t) DR P DR_max,t

[0050] |P DR (t)-P DR (t-1)| <P DR_ramp

[0051]

[0052] ​In the above formula, P Load (t) is the total load demand power of the resource alliance during period t, P Bat_Cha (t) is the charging power of the battery pack during period t, P Sell (t) is the power sold by the load-side resources to the large power grid during period t, P PV (t) is the photovoltaic power generation power in period t, P Bat_Dis (t) is the discharge power of the battery pack during period t, P GT (t) is the gas turbine output power during period t, P Buy (t) is the power purchased from the large power grid by the load-side resources during period t, P DR (t) is the demand response output power of the load-side resources during period t, is the demand response reactive power of node j in period t, is the reactive power demanded by the load at node j during period t, is the active power required by the load of node j in period t, is the active power of node j in period t, B DR is the set of demand response nodes, U DR is a 0-1 variable indicating whether the demand response is invoked, P DR_min，t is the minimum output power of the load-side resource demand response during period t, P DR_max,t is the maximum output power of the load-side resource demand response during period t, P DR (t-1) is the demand response output power of the load-side resources in period t-1, P DR_ramp is the rate of change of demand response output power per unit time, is the day-ahead demand response of node j in period t, P j,Bat_Cha,t is the charging power of the battery pack at node j during period t, P j,Bat_Dis,t is the discharge power of the battery pack at node j during period t, DS is the downstream node set of node i i The demand response required for the impact load during period t, DS i is the set of downstream nodes of node i, is the demand response of node j during period t, P Bat_Cha,max is the maximum charging power of the battery pack, The demand response capacity reserved for node i in time period t to cope with short-term impact load.

[0053] In a second aspect, a load optimization control device is provided, the load optimization control device comprising:

[0054] A first analysis module is used to use the user load data of the current period as the input of the pre-trained impact load prediction model to obtain the user impact load of the prediction period output by the pre-trained impact load prediction model;

[0055] A determination module, configured to determine, based on the user impact load in the forecast period, the required demand response amount for the impact load in the forecast period and the demand response amount reserved in the day ahead to cope with the short-term impact load;

[0056] The second analysis module is used to substitute the demand response amount required for the impact load in the forecast period and the demand response amount reserved for the short-term impact load in the day before into the pre-built load optimization scheduling model and solve it to obtain the optimization result;

[0057] A third analysis module is used to obtain a load optimization control scheme based on the optimization result;

[0058] Among them, the optimization result includes at least one of the following: photovoltaic power generation power, gas turbine output power, energy storage charging and discharging power, and demand response output power of load-side resources.

[0059] Preferably, the training process of the pre-trained impact load prediction model includes:

[0060] Obtain historical user load data, and determine impact load data in the historical user load data;

[0061] Using historical user load data and impact load data in the historical user load data to construct training data;

[0062] Using the training data to train an initial impact load prediction model to obtain the pre-trained impact load prediction model;

[0063] Wherein, the initial impact load prediction model is based on a convolutional neural network model of a long short-term memory network.

[0064] Further, the determining of the impact load data in the historical user load data includes:

[0065] Obtain the index characteristics of historical user load data, and use the principal component analysis algorithm to obtain the dimensionality reduction characteristics corresponding to the index characteristics of the historical user load data;

[0066] An outlier factor detection algorithm is used to select historical user load data whose outlier factors exceed a threshold value from the dimensionality reduction features, and the selected historical user load data is used as impact load data in the historical user load data.

[0067] Furthermore, the indicator characteristics of the historical user load data are as follows:

[0068]

[0069] In the above formula, σ is the standard deviation of historical user load data, x i is the load of the ith sampling point, is the mean of historical user load data, n is the number of sampling points, C k is the load kurtosis, C d is the load dispersion coefficient, C s is the load waveform factor, C if is the load pulse factor.

[0070] Furthermore, before using the training data to train the initial impact load prediction model, the method includes:

[0071] The crown porcupine optimization algorithm is used to optimize the convolution kernel step size of the convolutional neural network in the initial impact load prediction model. During the optimization process, the fitness used is the prediction accuracy of the impact load prediction model.

[0072] Preferably, the demand response amount required for the impact load in the forecast period is as follows:

[0073]

[0074] In the above formula, DS is the downstream node set of node i i The required demand response amount for the impact load in period t, D im,i,t is the user impact load of node i in period t, D k,t is the load of node k in period t, S l,i is the line capacity of node i.

[0075] Furthermore, the demand response quantity reserved for short-term impact load is as follows:

[0076]

[0077] In the above formula, The demand response amount reserved for node i to deal with short-term impact loads before time period t, D impact,i,in is the maximum value of user impact load of node i in period t.

[0078] Preferably, the pre-built load optimization scheduling model includes: an objective function with the goal of minimizing the total scheduling cost and its corresponding constraints.

[0079] Furthermore, the objective function is as follows:

[0080] C Total =min(C Oper +C DR )

[0081] In the above formula, C Total is the objective function value, C Oper is the total operating cost of resources on the load side, C DR Compensation for interrupted load.

[0082] Furthermore, the total operating cost of the load-side resources is as follows:

[0083]

[0084] The interruption load compensation fee is as follows:

[0085]

[0086] In the above formula, P Bat (t) is the charging and discharging power of the battery during period t, K Bat is the life loss cost coefficient caused by the battery charge and discharge, △t is the time interval, T is the total number of optimization periods, γ GT is the fuel cost of the microturbine, is the startup cost of the gas turbine during period t, is the shutdown cost of the gas turbine during period t, is the operating coefficient of the gas turbine during period t, is the shutdown coefficient of the gas turbine during period t, λ e is the demand response unit compensation coefficient, △L e,cut_t is the demand response dispatch quantity in period t, λ re is the day-ahead reserved demand response coefficient, △L re,t Reserve demand response capacity for time period t.

[0087] Furthermore, the constraints are as follows:

[0088] P Load (t)+P Bat_Cha (t)+P Sell (t)≤P PV (t)+P Bat_Dis (t)+P GT (t)+P Buy (t)+P DR (t)

[0089]

[0090] U DR P DR_min，t <P DR (t) DR P DR_max,t

[0091] |P​DR (t)-P DR (t-1)| <P DR_ramp

[0092]

[0093] In the above formula, P Load (t) is the total load demand power of the resource alliance during period t, P Bat_Cha (t) is the charging power of the battery pack during period t, P Sell (t) is the power sold by the load-side resources to the large power grid during period t, P PV (t) is the photovoltaic power generation power in period t, P Bat_Dis (t) is the discharge power of the battery pack during period t, P GT (t) is the gas turbine output power during period t, P Buy (t) is the power purchased from the large power grid by the load-side resources during period t, P DR (t) is the demand response output power of the load-side resources during period t, is the demand response reactive power of node j in period t, is the reactive power demanded by the load at node j during period t, is the active power required by the load of node j in period t, is the active power of node j in period t, B DR is the set of demand response nodes, U DR is a 0-1 variable indicating whether the demand response is invoked, P DR_min，t is the minimum output power of the load-side resource demand response during period t, P DR_max,t is the maximum output power of the load-side resource demand response during period t, P DR (t-1) is the demand response output power of the load-side resources in period t-1, P DR_ramp is the rate of change of demand response output power per unit time, is the day-ahead demand response of node j in period t, P j,Bat_Cha,t is the charging power of the battery pack at node j during period t, P j,Bat_Dis,t is the discharge power of the battery pack at node j during period t, DS is the downstream node set of node i i The demand response required for the impact load during period t, DS i is the set of downstream nodes of node i, is the demand response of node j during period t, P Bat_Cha,max is the maximum charging power of the battery pack, The demand response capacity reserved for node i in time period t to cope with short-term impact load.

[0094] In a third aspect, a computer device is provided, comprising: one or more processors;

[0095] The processor is configured to execute one or more programs;

[0096] When the one or more programs are executed by the one or more processors, the load optimization and control method is implemented.

[0097] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed, the load optimization and control method is implemented.

[0098] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0099] The present invention provides a load optimization control method, the method comprising: using the user load data of the current period as the input of a pre-trained impact load prediction model to obtain the user impact load of the prediction period output by the pre-trained impact load prediction model; determining the required demand response amount of the impact load of the prediction period and the demand response amount reserved for short-term impact loads based on the user impact load of the prediction period; substituting the required demand response amount of the impact load of the prediction period and the demand response amount reserved for short-term impact loads into a pre-built load optimization scheduling model and solving them to obtain an optimization result; obtaining a load optimization control scheme based on the optimization result; wherein the optimization result includes at least one of the following: photovoltaic power generation power, gas turbine output power, energy storage charging and discharging power, and demand response output power of load-side resources. The technical scheme provided by the present invention improves the problem of user power consumption caused by the load rigid control method through load identification and screening, demand response capacity prediction and day-ahead intra-day optimization scheduling, improves the user power consumption experience, and can realize the coordinated operation of multiple resources to meet the power grid's safe power supply needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] Figure 1 It is a schematic flow chart of the main steps of the load optimization control method according to an embodiment of the present invention;

[0101] Figure 2 This is a local abnormal factor distribution diagram of an embodiment of the present invention.

[0102] Figure 3 This is a 33-node distribution network structure diagram of an embodiment of the present invention.

[0103] Figure 4 is a schematic diagram of the day-ahead demand response scheduling amount according to an embodiment of the present invention;

[0104] Figure 5 Schematic diagram of daily demand response volume in an embodiment of the present invention. DETAILED DESCRIPTION

[0105] The specific implementation modes of the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0106] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0107] As disclosed in the background technology, with the in-depth implementation of the energy revolution strategy, the proportion of renewable energy installed capacity and power generation has continued to increase, and the proportion of terminal electrification has continued to increase. However, the rapid development of renewable energy is not matched with the planning and development of the power grid, resulting in increasingly serious problems of wind and solar power abandonment. The high volatility and randomness of both the source and the load have put forward higher requirements for the balance regulation of the power grid, and the safe and reliable operation of the power grid and power supply guarantee are facing huge challenges.

[0108] With the rapid development of industrialization and urbanization, people's demand for electricity is growing day by day. However, in many areas, the power grid was built earlier, and the technology and equipment are relatively backward, which cannot meet the peak load power demand. It is necessary to expand the power grid to improve the power supply capacity of the power grid. At the same time, the rapid development of distributed power sources (such as solar energy) has brought about changes in the power grid trend, making the problem of power grid expansion more complicated.

[0109] There is a type of load in the power grid that imposes a sudden and large amount of power demand on the power grid system in a very short period of time, but the duration is limited. These loads are called impact loads. The excessive instantaneous power of the impact load occupies a large amount of power grid capacity, forcing the power grid to set up redundant systems or adopt higher-standard equipment to ensure sufficient capacity and flexibility to ensure safe operation. Considering that there is a problem of high cost to directly optimize power grid equipment to increase capacity, especially many devices are located in urban areas, optimization is difficult and costly. At present, in areas where capacity expansion is difficult, the main method used is load rigidity control, which requires non-critical loads to downgrade or shut down, and resume normal operation after the impact subsides. However, load rigidity control requires other users to reduce their loads, which affects users' electricity consumption and production operations.

[0110] Demand Response (DR) is an electricity management strategy that uses economic means to adjust electricity consumption behavior on the demand side in response to changes in grid supply and demand. It is seen as one of the effective means to achieve grid expansion and improve grid flexibility and reliability. Demand response can reduce pressure during peak load periods and improve grid operation efficiency, thereby indirectly achieving "virtual expansion" of the grid. Through demand response measures, electricity demand can be adjusted dynamically. When a large-scale load increase is predicted, the electricity demand of some non-critical loads can be reduced through economic incentives or direct control of agreements, thereby reducing the impact of shock loads on the grid. Demand response provides economic compensation for loads and delays the replacement or addition of new equipment in the grid, which can achieve a win-win situation for grid companies and users.

[0111] Currently, there are patents for research on demand response, such as:

[0112] The invention discloses a distribution network optimization dispatching method considering the uncertainty of comprehensive demand response (CN117543582A), comprising: constructing a price-type demand response model and an interruptible load model to obtain a comprehensive load demand response uncertainty model; determining a first objective function, establishing operation constraints, and obtaining a distribution network robust optimization dispatching model; constructing an uncertain opportunity constraint programming min model based on the comprehensive load constraints, the first objective function and the operation constraints; converting the opportunity constraint programming min model into an obstacle term of the objective function by using an interior point method, obtaining a second objective function, and reconstructing a clear class equivalent distribution network optimization dispatching model; using YALMIP modeling and solving it with the help of CPLEX commercial software to obtain a dispatching result of the distribution network, which can overcome the influence of demand response uncertainty on the dispatching of the distribution network and improve the economy and stability.

[0113] However, this dispatching method dispatches the power grid based on economic optimization, and fails to solve problems such as insufficient power supply capacity of the power grid as the load increases, and shortage of line capacity caused by excessive instantaneous load changes when there are impact loads in the distribution network.

[0114] For the identification of impact load, a method for identifying impact load attributes on the distribution transformer outgoing line side (CN114757232A) includes: obtaining the instantaneous power signal and average power signal of the equipment that generates impact load at each time node within a preset time period; generating multiple load attributes that may constitute the instantaneous power signal of the corresponding time node at each time node; evaluating the respective occurrence probabilities of the multiple load attributes according to the average power at each time node; determining the optimal solution of the load attribute according to the occurrence probability, and outputting it as the load attribute of the equipment. The present invention is a non-invasive, high-precision method that can be set up on the outgoing line side of the distribution transformer, and is suitable for impact load behavior.

[0115] This method mainly considers the instantaneous power signal and average power signal of the load, and does not consider indicators that describe the load power consumption characteristics such as load kurtosis, dispersion coefficient, waveform factor and pulse factor, and cannot fully and accurately characterize the impact load characteristics.

[0116] In order to improve the above problems, the present invention provides a load optimization control method, the method comprising: taking the user load data of the current period as the input of a pre-trained impact load prediction model, obtaining the user impact load of the prediction period output by the pre-trained impact load prediction model; determining the required demand response amount of the impact load of the prediction period and the demand response amount reserved for short-term impact loads based on the user impact load of the prediction period; substituting the required demand response amount of the impact load of the prediction period and the demand response amount reserved for short-term impact loads into the pre-built load optimization scheduling model and solving them to obtain the optimization result; obtaining the load optimization control scheme based on the optimization result; wherein the optimization result includes at least one of the following: photovoltaic power generation power, gas turbine output power, energy storage charging and discharging power, and demand response output power of load-side resources. The technical scheme provided by the present invention improves the problem of user power consumption caused by the load rigid control mode through load identification and screening, demand response capacity prediction and day-ahead intra-day optimization scheduling, improves the user power consumption experience, and can realize the coordinated operation of multiple resources to meet the power grid's safe power supply needs.

[0117] The above scheme is described in detail below.

[0118] Example 1

[0119] See attached Figure 1 , Figure 1 It is a flow chart of the main steps of a load optimization control method according to an embodiment of the present invention.

[0120] like Figure 1 As shown, the load optimization control method in the embodiment of the present invention mainly includes the following steps:

[0121] Step S101: using the user load data of the current period as the input of a pre-trained impact load prediction model to obtain the user impact load of the prediction period output by the pre-trained impact load prediction model;

[0122] Determine the required demand response amount for the impact load in the forecast period and the demand response amount reserved in the day ahead to cope with the short-term impact load based on the user impact load in the forecast period;

[0123] Substituting the demand response amount required for the impact load in the forecast period and the demand response amount reserved for the short-term impact load in the day ahead into the pre-built load optimization scheduling model and solving it to obtain the optimization result;

[0124] Based on the optimization results, a load optimization control scheme is obtained;

[0125] Among them, the optimization result includes at least one of the following: photovoltaic power generation power, gas turbine output power, energy storage charging and discharging power, and demand response output power of load-side resources.

[0126] In this embodiment, the training process of the pre-trained impact load prediction model includes:

[0127] Obtain historical user load data, and determine impact load data in the historical user load data;

[0128] Using historical user load data and impact load data in the historical user load data to construct training data;

[0129] Using the training data to train an initial impact load prediction model to obtain the pre-trained impact load prediction model;

[0130] Wherein, the initial impact load prediction model is based on a convolutional neural network model of a long short-term memory network.

[0131] In one embodiment, determining the impact load data in the historical user load data includes:

[0132] Obtain the index characteristics of historical user load data, and use the principal component analysis algorithm to obtain the dimensionality reduction characteristics corresponding to the index characteristics of the historical user load data;

[0133] An outlier factor detection algorithm is used to select historical user load data whose outlier factors exceed a threshold value from the dimensionality reduction features, and the selected historical user load data is used as impact load data in the historical user load data.

[0134] Specifically, the principal component analysis algorithm uses an orthogonal transformation to transform a series of possibly linearly correlated variables into a set of linearly unrelated new variables, also called principal components, so as to use the new variables to display the characteristics of the data in a smaller dimension. The main steps are as follows:

[0135] 1) Organize the feature data into a matrix X with N rows and M columns.

[0136] 2) Zero-mean each row of X, that is, subtract the mean of each row.

[0137] 3) Find the covariance matrix C of X.

[0138] 4) Find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix C. The eigenvalues ​​of C are the variances of the elements of each dimension of Y, which are also the diagonal elements of D. They are arranged along the diagonal from large to small to form D.

[0139] 5) Arrange the eigenvectors into a matrix by row from top to bottom according to the corresponding eigenvalue size. According to the actual business scenario, take the first R rows with a contribution rate of 99% to form the matrix P.

[0140] 6)Y=PX is the target matrix after being reduced to R dimensions.

[0141] The core of the outlier factor detection algorithm is to pick out the points with abnormal outliers in the characteristic distribution. Since the outliers in the load curve data only account for a small part, compared with the unsupervised clustering algorithm, the outlier factor detection algorithm is more targeted to cluster the points with similar abnormal distribution. At the same time, it considers more information than the traditional statistical method and has high flexibility. When the unknown type of abnormal data is scattered for various reasons and far away from the normal data set, the outlier factor detection algorithm has good performance. The load curve has a basic distinction after extracting features and reducing the dimension, which is very suitable for outlier detection algorithm. For a given point p, LOF k (p) represents the neighborhood point N of object p k The LOF of object p is the average of the ratio of the local reachable density of point p to the local reachable density of point p (p). The closer its value is to 1, the smaller the difference between the density of point p and its neighboring points is, and they belong to the same cluster. If the ratio is greater than 1, it means that the density of point p is less than that of its neighboring points, and point p may be an outlier. To detect outliers, it is necessary to calculate the LOF of each sample and set a threshold. The threshold is compared with the outlier factor to determine whether the sample point is an outlier.

[0142] The main detection steps are as follows:

[0143] 1) Process the collected load data, fill in the missing points (interpolate), and remove the data with 96 load points being 0.

[0144] 2) Extract the time domain and waveform characteristics of each user's T-1 load curve.

[0145] 3) Use PCA to reduce the dimension of feature data and extract the main features.

[0146] 4) Take the two new features after dimensionality reduction as input and set the neighborhood K and outlier threshold λ in the LOF algorithm.

[0147] 5) Calculate the outlier factor LOFk(p) of sample p.

[0148] 6) Compare the outlier factor LOFk(p) with the outlier threshold. When LOFk(p)>λ, the sample is an abnormal load, otherwise it is a normal load, thereby distinguishing data with mutations from data without mutations. The local factor distribution diagram is shown in Figure 2.

[0149] In one embodiment, the indicator characteristics of the historical user load data are as follows:

[0150]

[0151]

[0152] In the above formula, σ is the standard deviation of historical user load data, x i is the load of the ith sampling point, is the mean of historical user load data, n is the number of sampling points, C k is the load kurtosis, C d is the load dispersion coefficient, C s is the load waveform factor, C if is the load pulse factor.

[0153] In one embodiment, before using the training data to train the initial impact load prediction model, the method includes:

[0154] The Crested Porcupine Optimizer (CPO) algorithm is used to optimize the convolution kernel step size of the convolutional neural network in the initial impact load prediction model. During the optimization process, the fitness used is the prediction accuracy of the impact load prediction model. In a specific implementation, it specifically includes:

[0155] (1) Initialize the CPO population: In the CPO algorithm, we first need to initialize a population that represents possible solutions, i.e., different convolution kernel step size configurations.

[0156] (2) Iterative optimization of convolution step size: Use the number of iterations and the generated random numbers to simulate the path search in the chemical reaction and find the optimal convolution kernel step size. In each iteration, the population is updated according to the fitness of the current population (i.e., the evaluation index) and a new candidate solution is generated.

[0157] (3) Fitness function solution: The fitness function is used to evaluate the quality of each convolution kernel step size, which is usually related to the performance of CNN on the training data set, such as accuracy, loss function value, etc. By calculating the fitness function, the optimal convolution kernel step size is selected as the parameter of the CNN convolution layer.

[0158] (4) Training the optimized CNN: Use the determined optimal convolution kernel step size to train the CNN model. Use the training data set to train the CNN and adjust the network weights so that it can learn the characteristics of the data.

[0159] (5) Test the generalization ability of CNN: Use the test dataset to evaluate the trained CNN model to determine its generalization and detection capabilities. The test dataset should be different from the training dataset to ensure that the model can handle unseen samples.

[0160] (6) Use LSTM for load forecasting: After completing the optimization and testing of CNN, the LSTM algorithm is used to extract features and forecast the time series data of the load. The LSTM network is particularly suitable for processing time series data and can capture long-term dependencies.

[0161] (7) Load forecasting model training: Train the LSTM model, use historical load data as input, and predict future load changes. Adjust the parameters of the LSTM network to optimize the prediction accuracy.

[0162] (8) Model evaluation and application: Evaluate the LSTM load forecasting model and use actual load data to verify the accuracy of the forecast results. Based on the evaluation results, decide whether to apply the model to the actual load forecasting task.

[0163] The core of the entire process is to optimize the convolution step size of CNN through the CPO algorithm, improve the network's learning ability and prediction accuracy, and realize effective prediction of load time series through the LSTM model.

[0164] In this embodiment, the required demand response amount of the impact load in the forecast period is as follows:

[0165]

[0166] In the above formula, DS is the downstream node set of node i i The required demand response amount for the impact load in period t, D im,i,t is the user impact load of node i in period t, D k,t is the load of node k in period t, S l,i is the line capacity of node i.

[0167] In one embodiment, the demand response amount reserved for dealing with short-term impact loads is as follows:

[0168]

[0169] In the above formula, The demand response amount reserved for node i to deal with short-term impact loads before time period t, D impact,i,in is the maximum value of user impact load of node i in period t.

[0170] In this embodiment, the pre-constructed load optimization scheduling model includes: an objective function with the goal of minimizing the total scheduling cost and its corresponding constraints.

[0171] In one embodiment, the objective function is as follows:

[0172] CTotal =min(C Oper +C DR )

[0173] In the above formula, C Total is the objective function value, C Oper is the total operating cost of resources on the load side, C DR Compensation for interrupted load.

[0174] In one embodiment, the total operating cost of the load-side resources is as follows:

[0175]

[0176] The interruption load compensation fee is as follows:

[0177]

[0178] In the above formula, P Bat (t) is the charging and discharging power of the battery during period t, K Bat is the life loss cost coefficient caused by the battery charge and discharge, △t is the time interval, T is the total number of optimization periods, γ GT is the fuel cost of the microturbine, is the startup cost of the gas turbine during period t, is the shutdown cost of the gas turbine during period t, is the operating coefficient of the gas turbine during period t, is the shutdown coefficient of the gas turbine during period t, λ e is the demand response unit compensation coefficient, △L e,cut_t is the demand response dispatch quantity in period t, λ re is the day-ahead reserved demand response coefficient, △L re,t Reserve demand response capacity for time period t.

[0179] In one embodiment, the constraints are as follows:

[0180] P Load (t)+P Bat_Cha (t)+P Sell (t)≤P PV (t)+P Bat_Dis (t)+P GT (t)+P Buy (t)+P DR (t)

[0181]

[0182] U DR P DR_min，t <P DR (t) <UDR P DR_max,t

[0183] |P DR (t)-P DR (t-1)| <P DR_ramp

[0184]

[0185] In the above formula, P Load (t) is the total load demand power of the resource alliance during period t, P Bat_Cha (t) is the charging power of the battery pack during period t, P Sell (t) is the power sold by the load-side resources to the large power grid during period t, P PV (t) is the photovoltaic power generation power in period t, P Bat_Dis (t) is the discharge power of the battery pack during period t, P GT (t) is the gas turbine output power during period t, P Buy (t) is the power purchased from the large power grid by the load-side resources during period t, P DR (t) is the demand response output power of the load-side resources during period t, is the demand response reactive power of node j in period t, is the reactive power demanded by the load at node j during period t, is the active power required by the load of node j in period t, is the active power of node j in period t, B DR is the set of demand response nodes, U DR is a 0-1 variable indicating whether the demand response is invoked, P DR_min，t is the minimum output power of the load-side resource demand response during period t, P DR_max,t is the maximum output power of the load-side resource demand response during period t, P DR (t-1) is the demand response output power of the load-side resources in period t-1, P DR_ramp is the rate of change of demand response output power per unit time, is the day-ahead demand response of node j in period t, P j,Bat_Cha,t is the charging power of the battery pack at node j during period t, P j,Bat_Dis,t is the discharge power of the battery pack at node j during period t, DR need,t,DSi DS is the downstream node set of node i i The demand response required for the impact load during period t, DS i is the set of downstream nodes of node i, is the demand response of node j during period t, P Bat_Cha,max is the maximum charging power of the battery pack, DR re,t,DSiThe demand response capacity reserved for node i in time period t to cope with short-term impact load.

[0186] In a specific embodiment, Figure 3 This is a 33-node distribution network structure diagram. The case considers resources such as photovoltaic PV, energy storage ESS, reactive power compensation device SVC, on-load tap changer OLTC, circuit breaker CB, and power demand response DR.

[0187] First, the characteristic data is collected, as shown in Table 1, and the local abnormal factor of the measured load is obtained, as shown in Table 2.

[0188] Table 1

[0189]

[0190] Table 2

[0191]

[0192] According to the local abnormal factor values ​​of the user load data calculated in Table 2, the LOF values ​​of 19, 20, 26, and 35 are larger than the LOF values ​​of other user loads, so these four loads are used as the abnormal loads screened by the present invention and also as the required impact loads. Take one of the load data for one day and perform step 2 for 96 nodes, and compare the predicted value and the actual value, as shown in Table 3. Determine the demand response resource scheduling result, such as Figure 4 , Figure 5 shown.

[0193] Table 3

[0194]

[0195]

[0196] Example 2

[0197] Based on the same inventive concept, the present invention also provides a load optimization control device, the load optimization control device comprising:

[0198] A first analysis module is used to use the user load data of the current period as the input of the pre-trained impact load prediction model to obtain the user impact load of the prediction period output by the pre-trained impact load prediction model;

[0199] A determination module, configured to determine, based on the user impact load in the forecast period, the required demand response amount for the impact load in the forecast period and the demand response amount reserved in the day ahead to cope with the short-term impact load;

[0200] The second analysis module is used to substitute the demand response amount required for the impact load in the forecast period and the demand response amount reserved for the short-term impact load in the day before into the pre-built load optimization scheduling model and solve it to obtain the optimization result;

[0201] A third analysis module is used to obtain a load optimization control scheme based on the optimization result;

[0202] Among them, the optimization result includes at least one of the following: photovoltaic power generation power, gas turbine output power, energy storage charging and discharging power, and demand response output power of load-side resources.

[0203] Preferably, the training process of the pre-trained impact load prediction model includes:

[0204] Obtain historical user load data, and determine impact load data in the historical user load data;

[0205] Using historical user load data and impact load data in the historical user load data to construct training data;

[0206] Using the training data to train an initial impact load prediction model to obtain the pre-trained impact load prediction model;

[0207] Wherein, the initial impact load prediction model is based on a convolutional neural network model of a long short-term memory network.

[0208] Further, the determining of the impact load data in the historical user load data includes:

[0209] Obtain the index characteristics of historical user load data, and use the principal component analysis algorithm to obtain the dimensionality reduction characteristics corresponding to the index characteristics of the historical user load data;

[0210] An outlier factor detection algorithm is used to select historical user load data whose outlier factors exceed a threshold value from the dimensionality reduction features, and the selected historical user load data is used as impact load data in the historical user load data.

[0211] Furthermore, the indicator characteristics of the historical user load data are as follows:

[0212]

[0213] In the above formula, σ is the standard deviation of historical user load data, x i is the load of the ith sampling point, is the mean of historical user load data, n is the number of sampling points, C k is the load kurtosis, C d is the load dispersion coefficient, C s is the load waveform factor, Cif is the load pulse factor.

[0214] Furthermore, before using the training data to train the initial impact load prediction model, the method includes:

[0215] The crown porcupine optimization algorithm is used to optimize the convolution kernel step size of the convolutional neural network in the initial impact load prediction model. During the optimization process, the fitness used is the prediction accuracy of the impact load prediction model.

[0216] Preferably, the demand response amount required for the impact load in the forecast period is as follows:

[0217]

[0218] In the above formula, DS is the downstream node set of node i i The required demand response amount for the impact load in period t, D im,i,t is the user impact load of node i in period t, D k,t is the load of node k in period t, S l,i is the line capacity of node i.

[0219] Furthermore, the demand response quantity reserved for short-term impact load is as follows:

[0220]

[0221] In the above formula, The demand response amount reserved for node i to deal with short-term impact loads before time period t, D impact,i,in is the maximum value of user impact load of node i in period t.

[0222] Preferably, the pre-built load optimization scheduling model includes: an objective function with the goal of minimizing the total scheduling cost and its corresponding constraints.

[0223] Furthermore, the objective function is as follows:

[0224] C Total =min(C Oper +C DR )

[0225] In the above formula, C Total is the objective function value, C Oper is the total operating cost of resources on the load side, C DR Compensation for interrupted load.

[0226] Furthermore, the total operating cost of the load-side resources is as follows:

[0227]

[0228] The interruption load compensation fee is as follows:

[0229]

[0230] In the above formula, P Bat (t) is the charging and discharging power of the battery during period t, K Bat is the life loss cost coefficient caused by the battery charge and discharge, △t is the time interval, T is the total number of optimization periods, γ GT is the fuel cost of the microturbine, is the startup cost of the gas turbine during period t, is the shutdown cost of the gas turbine during period t, is the operating coefficient of the gas turbine during period t, is the shutdown coefficient of the gas turbine during period t, λ e is the demand response unit compensation coefficient, △L e,cut_t is the demand response dispatch quantity in period t, λ re is the day-ahead reserved demand response coefficient, △L re,t Reserve demand response capacity for time period t.

[0231] Furthermore, the constraints are as follows:

[0232] P Load (t)+P Bat_Cha (t)+P Sell (t)≤P PV (t)+P Bat_Dis (t)+P GT (t)+P Buy (t)+P DR (t)

[0233]

[0234] U DR P DR_min，t <P DR (t) DR P DR_max,t

[0235] |P DR (t)-P DR (t-1)| <P DR_ramp

[0236]

[0237] In the above formula, P Load (t) is the total load demand power of the resource alliance during period t, P Bat_Cha (t) is the charging power of the battery pack during period t, P​Sell (t) is the power sold by the load-side resources to the large power grid during period t, P PV (t) is the photovoltaic power generation power in period t, P Bat_Dis (t) is the discharge power of the battery pack during period t, P GT (t) is the gas turbine output power during period t, P Buy (t) is the power purchased from the large power grid by the load-side resources during period t, P DR (t) is the demand response output power of the load-side resources during period t, is the demand response reactive power of node j in period t, is the reactive power demanded by the load at node j during period t, is the active power required by the load of node j in period t, is the active power of node j in period t, B DR is the set of demand response nodes, U DR is a 0-1 variable indicating whether the demand response is invoked, P DR_min，t is the minimum output power of the load-side resource demand response during period t, P DR_max,t is the maximum output power of the load-side resource demand response during period t, P DR (t-1) is the demand response output power of the load-side resources in period t-1, P DR_ramp is the rate of change of demand response output power per unit time, is the day-ahead demand response of node j in period t, P j,Bat_Cha,t is the charging power of the battery pack at node j during period t, P j,Bat_Dis,t is the discharge power of the battery pack at node j during period t, DS is the downstream node set of node i i The demand response required for the impact load during period t, DS i is the set of downstream nodes of node i, is the demand response of node j during period t, P Bat_Cha,max is the maximum charging power of the battery pack, The demand response capacity reserved for node i in time period t to cope with short-term impact load.

[0238] Example 3

[0239] Based on the same inventive concept, the present invention also provides a computer device, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of a load optimization control method in the above embodiment.

[0240] Example 4

[0241] Based on the same inventive concept, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It can be understood that the computer-readable storage medium here can include both built-in storage media in a computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space that stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of a load optimization control method in the above embodiment.

[0242] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0243] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0244] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0245] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0246] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A load optimization control method, characterized in that: The method comprises: Using the user load data of the current period as the input of the pre-trained impact load prediction model, and obtaining the user impact load of the prediction period output by the pre-trained impact load prediction model; Determine the required demand response amount for the impact load in the forecast period and the demand response amount reserved in the day ahead to cope with the short-term impact load based on the user impact load in the forecast period; Substituting the demand response amount required for the impact load in the forecast period and the demand response amount reserved for the short-term impact load in the day ahead into the pre-built load optimization scheduling model and solving it to obtain the optimization result; Based on the optimization results, a load optimization control scheme is obtained; Among them, the optimization result includes at least one of the following: photovoltaic power generation power, gas turbine output power, energy storage charging and discharging power, and demand response output power of load-side resources.

2. The method according to claim 1, characterized in that The training process of the pre-trained impact load prediction model includes: Obtain historical user load data, and determine impact load data in the historical user load data; Using historical user load data and impact load data in the historical user load data to construct training data; Using the training data to train an initial impact load prediction model to obtain the pre-trained impact load prediction model; Wherein, the initial impact load prediction model is based on a convolutional neural network model of a long short-term memory network.

3. The method according to claim 2, characterized in that The step of determining the impact load data in the historical user load data includes: Obtain the index characteristics of historical user load data, and use the principal component analysis algorithm to obtain the dimensionality reduction characteristics corresponding to the index characteristics of the historical user load data; An outlier factor detection algorithm is used to select historical user load data whose outlier factors exceed a threshold value from the dimensionality reduction features, and the selected historical user load data is used as impact load data in the historical user load data.

4. The method according to claim 3, characterized in that The indicator characteristics of the historical user load data are as follows: In the above formula, σ is the standard deviation of historical user load data, x i is the load of the ith sampling point, is the mean of historical user load data, n is the number of sampling points, C k is the load kurtosis, C d is the load dispersion coefficient, C s is the load waveform factor, C if is the load pulse factor.

5. The method according to claim 2, characterized in that Before using the training data to train the initial impact load prediction model, the method includes: The crown porcupine optimization algorithm is used to optimize the convolution kernel step size of the convolutional neural network in the initial impact load prediction model. During the optimization process, the fitness used is the prediction accuracy of the impact load prediction model.

6. The method according to claim 1, characterized in that The demand response required for the impact load during the forecast period is as follows: In the above formula, DS is the downstream node set of node i i The required demand response amount for the impact load in period t, D im,i,t is the user impact load of node i in period t, D k,t is the load of node k in period t, S l,i is the line capacity of node i.

7. The method according to claim 6, characterized in that The demand response quantity reserved for short-term impact loads is as follows: In the above formula, The demand response amount reserved for node i to deal with short-term impact loads before time period t, D impact,i,in is the maximum value of user impact load of node i in period t.

8. The method according to claim 1, characterized in that The pre-built load optimization scheduling model includes: an objective function with the goal of minimizing the total scheduling cost and its corresponding constraints.

9. The method according to claim 8, characterized in that The objective function is as follows: C Total =min(C Oper +C DR ) In the above formula, C Total is the objective function value, C Oper is the total operating cost of resources on the load side, C DR Compensation for interrupted load.

10. The method according to claim 9, characterized in that The total operating cost of the load-side resources is as follows: The interruption load compensation fee is as follows: In the above formula, P Bat (t) is the charging and discharging power of the battery during period t, K Bat is the life loss cost coefficient caused by the battery charge and discharge, △t is the time interval, T is the total number of optimization periods, γ GT is the fuel cost of the microturbine, is the startup cost of the gas turbine during period t, is the shutdown cost of the gas turbine during period t, is the operating coefficient of the gas turbine during period t, is the shutdown coefficient of the gas turbine during period t, λ e is the demand response unit compensation coefficient, △L e,cut_t is the demand response dispatch quantity in period t, λ re is the day-ahead reserved demand response coefficient, △L re,t Reserve demand response capacity for time period t.

11. The method according to claim 10, characterized in that The constraints are as follows: P Load (t)+P Bat_Cha (t)+P Sell (t)≤P PV (t)+P Bat_Dis (t)+P GT (t)+P Buy (t)+P DR (t) U DR P DR_min，t <P DR (t)<U DR P DR_max,t |P DR (t)-P DR (t-1)|<P DR_ramp In the above formula, P Load (t) is the total load demand power of the resource alliance during period t, P Bat_Cha (t) is the charging power of the battery pack during period t, P Sell (t) is the power sold by the load-side resources to the large power grid during period t, P PV (t) is the photovoltaic power generation power in period t, P Bat_Dis (t) is the discharge power of the battery pack during period t, P GT (t) is the gas turbine output power during period t, P Buy (t) is the power purchased from the large power grid by the load-side resources during period t, P DR (t) is the demand response output power of the load-side resources during period t, is the demand response reactive power of node j in period t, is the reactive power demanded by the load at node j during period t, is the active power required by the load of node j in period t, is the active power of node j in period t, B DR is the set of demand response nodes, U DR is a 0-1 variable indicating whether the demand response is invoked, P DR_min，t is the minimum output power of the load-side resource demand response during period t, P DR_max,t is the maximum output power of the load-side resource demand response during period t, P DR (t-1) is the demand response output power of the load-side resources in period t-1, P DR_ramp is the rate of change of demand response output power per unit time, is the day-ahead demand response of node j in period t, P j,Bat_Cha,t is the charging power of the battery pack at node j during period t, P j,Bat_Dis,t is the discharge power of the battery pack at node j during period t, DS is the downstream node set of node i i The demand response required for the impact load during period t, DS i is the set of downstream nodes of node i, is the demand response of node j during period t, P Bat_Cha,max is the maximum charging power of the battery pack, The demand response capacity reserved for node i in time period t to cope with short-term impact load.

12. A load optimization control device, characterized in that: The device comprises: A first analysis module is used to use the user load data of the current period as the input of the pre-trained impact load prediction model to obtain the user impact load of the prediction period output by the pre-trained impact load prediction model; A determination module, configured to determine, based on the user impact load in the forecast period, the required demand response amount for the impact load in the forecast period and the demand response amount reserved in the day ahead to cope with the short-term impact load; The second analysis module is used to substitute the demand response amount required for the impact load in the forecast period and the demand response amount reserved for the short-term impact load in the day before into the pre-built load optimization scheduling model and solve it to obtain the optimization result; A third analysis module is used to obtain a load optimization control scheme based on the optimization result; Among them, the optimization result includes at least one of the following: photovoltaic power generation power, gas turbine output power, energy storage charging and discharging power, and demand response output power of load-side resources.

13. The device according to claim 12, characterized in that The training process of the pre-trained impact load prediction model includes: Obtain historical user load data, and determine impact load data in the historical user load data; Using historical user load data and impact load data in the historical user load data to construct training data; Using the training data to train an initial impact load prediction model to obtain the pre-trained impact load prediction model; Wherein, the initial impact load prediction model is based on a convolutional neural network model of a long short-term memory network.

14. The device according to claim 13, characterized in that The step of determining the impact load data in the historical user load data includes: Obtain the index characteristics of historical user load data, and use the principal component analysis algorithm to obtain the dimensionality reduction characteristics corresponding to the index characteristics of the historical user load data; An outlier factor detection algorithm is used to select historical user load data whose outlier factors exceed a threshold value from the dimensionality reduction features, and the selected historical user load data is used as impact load data in the historical user load data.

15. The device according to claim 14, characterized in that The indicator characteristics of the historical user load data are as follows: In the above formula, σ is the standard deviation of historical user load data, x i is the load of the ith sampling point, is the mean of historical user load data, n is the number of sampling points, C k is the load kurtosis, C d is the load dispersion coefficient, C s is the load waveform factor, C if is the load pulse factor.

16. The device according to claim 13, characterized in that Before using the training data to train the initial impact load prediction model, the method includes: The crown porcupine optimization algorithm is used to optimize the convolution kernel step size of the convolutional neural network in the initial impact load prediction model. During the optimization process, the fitness used is the prediction accuracy of the impact load prediction model.

17. The device according to claim 12, characterized in that The demand response required for the impact load during the forecast period is as follows: In the above formula, DS is the downstream node set of node i i The required demand response amount for the impact load in period t, D im,i,t is the user impact load of node i in period t, D k,t is the load of node k in period t, S l,i is the line capacity of node i.

18. The device according to claim 17, characterized in that The demand response quantity reserved for short-term impact loads is as follows: In the above formula, DR re,t,DSi The demand response amount reserved for node i to deal with short-term impact loads before time period t, D impact,i,in is the maximum value of user impact load of node i in period t.

19. The device according to claim 12, characterized in that The pre-built load optimization scheduling model includes: an objective function with the goal of minimizing the total scheduling cost and its corresponding constraints.

20. The device according to claim 19, characterized in that The objective function is as follows: C Total =min(C Oper +C DR ) In the above formula, C Total is the objective function value, C Oper is the total operating cost of resources on the load side, C DR Compensation for interrupted load.

21. The device according to claim 20, characterized in that The total operating cost of the load-side resources is as follows: The interruption load compensation fee is as follows: In the above formula, P Bat (t) is the charging and discharging power of the battery during period t, K Bat is the life loss cost coefficient caused by the battery charge and discharge, △t is the time interval, T is the total number of optimization periods, γ GT is the fuel cost of the microturbine, is the startup cost of the gas turbine during period t, is the shutdown cost of the gas turbine during period t, is the operating coefficient of the gas turbine during period t, is the shutdown coefficient of the gas turbine during period t, λ e is the demand response unit compensation coefficient, △L e,cut_t is the demand response dispatch quantity in period t, λ re is the day-ahead reserved demand response coefficient, △L re,t Reserve demand response capacity for time period t.

22. The device according to claim 21, characterized in that The constraints are as follows: P Load (t)+P Bat_Cha (t)+P Sell (t)≤P PV (t)+P Bat_Dis (t)+P GT (t)+P Buy (t)+P DR (t) U DR P DR_min，t <P DR (t)<U DR P DR_max,t |P DR (t)-P DR (t-1)|<P DR_ramp In the above formula, P Load (t) is the total load demand power of the resource alliance during period t, P Bat_Cha (t) is the charging power of the battery pack during period t, P Sell (t) is the power sold by the load-side resources to the large power grid during period t, P PV (t) is the photovoltaic power generation power in period t, P Bat_Dis (t) is the discharge power of the battery pack during period t, P GT (t) is the gas turbine output power during period t, P Buy (t) is the power purchased from the large power grid by the load-side resources during period t, P DR (t) is the demand response output power of the load-side resources during period t, is the demand response reactive power of node j in period t, is the reactive power demanded by the load at node j during period t, is the active power required by the load of node j in period t, is the active power of node j in period t, B DR is the set of demand response nodes, U DR is a 0-1 variable indicating whether the demand response is invoked, P DR_min，t is the minimum output power of the load-side resource demand response during period t, P DR_max,t is the maximum output power of the load-side resource demand response during period t, P DR (t-1) is the demand response output power of the load-side resources in period t-1, P DR_ramp is the rate of change of demand response output power per unit time, is the day-ahead demand response of node j in period t, P j,Bat_Cha,t is the charging power of the battery pack at node j during period t, P j,Bat_Dis,t is the discharge power of the battery pack at node j during period t, DS is the downstream node set of node i i The demand response required for the impact load during period t, DS i is the set of downstream nodes of node i, is the demand response of node j during period t, P Bat_Cha,max is the maximum charging power of the battery pack, The demand response capacity reserved for node i in time period t to cope with short-term impact load.

23. A computer device, characterized in that: include: one or more processors; The processor is configured to execute one or more programs; When the one or more programs are executed by the one or more processors, the load optimization and control method according to any one of claims 1 to 12 is implemented.

24. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, the load optimization and control method as described in any one of claims 1 to 12 is implemented.

Images