Multi-type user resource scheduling dynamic optimization method and system based on load prediction

Abstract

The invention belongs to the technical field of power system optimization scheduling, and provides a multi-type user resource scheduling dynamic optimization method and system based on load prediction. The optimization method comprises the following steps: acquiring multiple types of user resource parameters, and predicting a user load; adjusting the scheduling cost of the user resources through a preset day-ahead scheduling model according to a user load prediction result to obtain a day-ahead optimal scheduling scheme, and constructing and solving an intra-day dynamic scheduling optimization model of the user resources in combination with model prediction control to obtain an intra-day optimal scheduling plan; if the obtained intra-day optimal scheduling plan meets a preset control target, completing scheduling dynamic optimization of multiple types of user resources according to the obtained intra-day optimal scheduling plan; otherwise, the scheduling cost of the user resources is readjusted.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system optimization and dispatching, and in particular relates to a method and system for dynamic optimization of multi-type user resource dispatching based on load forecasting. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] As various types of power equipment, represented by new energy, are connected to the power grid, maintaining the safety and stability of the entire network requires more and more resource allocation and technical support. Therefore, optimizing the resource allocation of the power system through the interaction of "source-grid-load-storage" has become an important solution.

[0004] With the gradual improvement of the construction of the power market, a multi-dimensional, multi-level and multi-functional power market is taking shape, providing conditions for various resources with regulatory capabilities to enter the market and participate in the dispatch of the power system. From a narrow perspective, there are many types of flexible loads on the load side that are regarded as user-side resources that can participate in dispatch. From the perspective of "source-grid-load-storage", the dispatchable resources include not only the load side, but also distributed energy storage, micro gas turbines and other equipment. These resources can participate in market transactions such as demand response, frequency regulation and peak regulation, and standby services if they meet the market access conditions.

[0005] The current electricity spot market structure is a multi-level market with the electricity energy market as the main body and the ancillary service market as the auxiliary market, which is actually determined by the coupling of the power system. This coupling is transmitted to the market, resulting in the market equilibrium condition of multiple markets jointly clearing, which is constrained by the safe and stable operation of the entire network. In other words, the joint clearing of multiple markets must also meet the security constrained economic dispatch (SCED) condition constraints of the power grid dispatch.

[0006] Market participants need to face market risks caused by market competition and changes in supply and demand, as well as output deviation risks caused by the uncertainty of output of units such as new energy. In the day-ahead stage, market participants need to analyze all information to determine their own decisions; among the analyzed information, load forecast is a key information that represents the market's estimate of the scale of demand. Since electricity has the characteristics of instantaneousness and difficulty in storage, the supply and demand mechanism of electricity is "supply determined by demand". Load forecasts are generally released by dispatchers based on load curves in the day-ahead stage, and are a reference for market participants' bidding strategies in the day-ahead stage; existing load forecasts only include the disclosure of information such as load levels and key load characteristics in each time period, focusing on the balance of supply and demand in the electric energy market, and insufficient information support for the ancillary service market, resulting in reduced efficiency in the joint clearing of multiple markets.

[0007] The safe and stable operation of the power grid in the intraday stage is the constraint condition for the joint clearing of multiple markets, that is, the market clearing condition and the power grid dispatching target are the same. Therefore, when the dispatcher makes load forecasts, it is necessary to dynamically combine the SCED constraints of the power grid with the forecasting process, that is, to use the constraints of the previous period as the starting conditions for the forecast results of the next period, and finally achieve market equilibrium.

[0008] From the perspective of the interaction of "source-grid-load-storage", we should not narrowly examine only the flexible load resources on the user side, but should incorporate resources such as distributed energy storage into the dispatching process, and incorporate new goals such as improving the consumption of new energy into the dispatching goals to improve the overall resource utilization efficiency. How to combine load forecasting with resource scheduling to improve the reliability of load forecasting results is a difficult problem that needs to be solved urgently. Summary of the invention

[0009] In order to solve the above problems, the present invention proposes a dynamic optimization method and system for multi-type user resource scheduling based on load forecasting, which constructs a load synergy prediction model through graph comparison learning, combines model predictive control MPC, and constructs a dynamic optimization model for the entire network scheduling according to the current power market transaction cycle to obtain dynamic rolling scheduling of allocation resources.

[0010] According to some embodiments, a first solution of the present invention provides a method for dynamic optimization of resource scheduling of multiple types of users based on load forecasting, which adopts the following technical solution:

[0011] A dynamic optimization method for resource scheduling of multiple types of users based on load forecasting, comprising:

[0012] Get multiple types of user resource parameters;

[0013] Predict user load based on the acquired user resource parameters;

[0014] According to the user load forecast results, the scheduling cost of user resources is adjusted through the preset day-ahead scheduling model to obtain the day-ahead optimal scheduling plan;

[0015] Based on the obtained optimal scheduling plan for the day ahead, combined with model predictive control, a user resource intraday dynamic scheduling optimization model is constructed;

[0016] Solve the constructed intraday dynamic scheduling optimization model of user resources to obtain the optimal intraday scheduling plan;

[0017] If the obtained intraday optimal scheduling plan meets the preset control target, the scheduling dynamic optimization of multi-type user resources is completed according to the obtained intraday optimal scheduling plan.

[0018] As a further technical limitation, the specific process of predicting user load is: using the Informer model, a deep learning model for time series prediction, to predict user load and obtain short-term load characteristic variables and deployable resource state variables.

[0019] Furthermore, the process of predicting user load by the Informer model is as follows: short-term load characteristic variables and deployable resource state variables are input into a preset Informer encoder for linear transformation of feature vectors to obtain a characterization query matrix, a characterization key matrix, and a characterization value matrix of equal column lengths, and the obtained three matrices are processed by a multi-head probabilistic sparse self-attention mechanism to output the user load prediction result, i.e., the short-term load prediction value and the active / reactive output that can be provided by the deployable resources are obtained.

[0020] As a further technical limitation, the preset day-ahead scheduling model takes the minimum total scheduling cost of user resources as the objective function, and the objective function F DA for Where, T represents the total decision cycle, which is 24 hours; t represents the time period; μ loss,t represents the unit line loss cost in period t; P loss,t represents the power line loss of the entire network in period t; C ess,t represents the cost of energy storage participating in dispatch during period t; C mt,t represents the cost of micro gas turbine participating in the dispatch in period t; C oltc,t It represents the reactive power regulation cost of the transformer on-load tap changer (i.e. OLTC equipment) during period t; It represents the cost of type A demand response (i.e., demand response in which users can reduce their loads) decomposed into period t.

[0021] As a further technical limitation, the constructed user resource intraday dynamic scheduling optimization model constructs an objective function with the goal of minimizing the comprehensive cost within a single decision cycle within the day, the node voltage deviation and the resources called within the decision cycle. The constraints of the constructed objective function include at least flow constraints and power constraints.

[0022] As a further technical limitation, if the obtained intraday optimal scheduling plan does not meet the preset control target, the scheduling cost of the user resources is readjusted. According to some embodiments, the second solution of the present invention provides a dynamic optimization system for scheduling multi-type user resources based on load forecasting, which adopts the following technical solution:

[0023] A dynamic optimization system for resource scheduling of multiple types of users based on load forecasting, comprising:

[0024] An acquisition module, which is configured to acquire multiple types of user resource parameters;

[0025] A prediction module, configured to predict user load according to the acquired user resource parameters;

[0026] A day-ahead scheduling module is configured to adjust the scheduling cost of user resources through a preset day-ahead scheduling model according to the user load forecast result to obtain the day-ahead optimal scheduling plan;

[0027] The intraday scheduling module is configured to build a user resource intraday dynamic scheduling optimization model based on the obtained optimal scheduling plan for the day ahead and in combination with model predictive control;

[0028] A solution module, which is configured to solve the constructed intraday dynamic scheduling optimization model of user resources to obtain an intraday optimal scheduling plan;

[0029] The dynamic optimization module is configured to complete the dynamic optimization of scheduling of multi-type user resources according to the obtained intra-day optimal scheduling plan if the obtained intra-day optimal scheduling plan meets the preset control target.

[0030] As a further technical limitation, in the prediction module, the specific process of predicting user load is: using the Informer model, a deep learning model for time series prediction, to predict user load and obtain short-term load characteristic variables and deployable resource state variables.

[0031] Furthermore, the process of predicting user load by the Informer model is as follows: short-term load characteristic variables and deployable resource state variables are input into a preset Informer encoder for linear transformation of feature vectors to obtain a characterization query matrix, a characterization key matrix, and a characterization value matrix of equal column lengths, and the obtained three matrices are processed by a multi-head probabilistic sparse self-attention mechanism to output the user load prediction result, i.e., the short-term load prediction value and the active / reactive output that can be provided by the deployable resources are obtained.

[0032] As a further technical limitation, in the day-ahead scheduling module, the preset day-ahead scheduling model takes the minimum total scheduling cost of user resources as the objective function, and the objective function F DA for Where, T represents the total decision cycle, which is 24 hours; t represents the time period; μ loss,t represents the unit line loss cost in period t; P loss,t represents the power line loss of the entire network in period t; C ess,t represents the cost of energy storage participating in dispatch during period t; C mt,t represents the cost of micro gas turbine participating in the dispatch in period t; C oltc,t represents the reactive power regulation cost of the transformer on-load tap changer during period t; It represents the cost of class A demand response decomposed into period t.

[0033] As a further technical limitation, in the intraday scheduling module, the constructed user resource intraday dynamic scheduling optimization model constructs an objective function with the goal of minimizing the comprehensive cost within a single decision cycle within the day, the node voltage deviation and the resources called within the decision cycle. The constraints of the constructed objective function include at least flow constraints and power constraints.

[0034] As a further technical limitation, in the dynamic optimization module, if the obtained optimal scheduling plan within the day does not meet the preset control target, it returns to the day-ahead scheduling module to readjust the scheduling cost of user resources.

[0035] According to some embodiments, a third solution of the present invention provides a computer-readable storage medium, which adopts the following technical solution:

[0036] A computer-readable storage medium stores a program, which, when executed by a processor, implements the steps of the dynamic optimization method for multi-type user resource scheduling based on load prediction as described in the first solution of the present invention.

[0037] According to some embodiments, a fourth solution of the present invention provides an electronic device, which adopts the following technical solution:

[0038] An electronic device comprises a memory, a processor and a program stored in the memory and running on the processor, wherein when the processor executes the program, the steps in the dynamic optimization method for resource scheduling of multiple types of users based on load prediction as described in the first scheme of the present invention are implemented.

[0039] According to some embodiments, a fifth solution of the present invention provides a computer program product, which adopts the following technical solution:

[0040] A computer program product includes software codes, wherein the program in the software codes executes the steps in the dynamic optimization method for resource scheduling of multiple types of users based on load prediction as described in the first solution of the present invention.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention combines load forecasting with the SCED target and dispatching process of the power grid, which conforms to the inherent mechanism of the power market, namely, the joint clearing of the electric energy market and the auxiliary market. It takes the safe and economic dispatch of the power grid as a constraint condition, and determines the optimal configuration of the deployable resources in each time period, thereby providing intuitive and feasible data support for the specific trading strategies of various trading varieties and various trading scenarios in the power market.

[0043] The present invention constructs a load synergy prediction model through graph comparison learning, combines model predictive control MPC, and constructs a dynamic optimization model for the whole network scheduling according to the current transaction cycle of the power market to obtain dynamic rolling scheduling of allocation resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings in the specification that constitute a part of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments of this embodiment and their descriptions are used to explain this embodiment and do not constitute improper limitations on this embodiment.

[0045] Figure 1 This is a flow chart of a method for dynamic optimization of resource scheduling for multiple types of users based on load forecasting in Embodiment 1 of the present invention;

[0046] Figure 2 This is a structural block diagram of a dynamic optimization system for resource scheduling of multiple types of users based on load forecasting in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0047] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0048] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0049] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0050] In the present invention, terms such as "upper", "lower", "left", "right", "front", "back", "vertical", "horizontal", "side", "bottom", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings. They are relational words determined only for the convenience of describing the structural relationships of the various parts or elements of the present invention, and do not specifically refer to any part or element in the present invention and should not be understood as limitations on the present invention.

[0051] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0052] Embodiment 1

[0053] Embodiment 1 of the present invention introduces a dynamic optimization method for resource scheduling of multiple types of users based on load prediction.

[0054] like Figure 1 The method for dynamic optimization of multi-type user resource scheduling based on load forecasting comprises the following steps:

[0055] Get multiple types of user resource parameters;

[0056] Predict user load based on the acquired user resource parameters;

[0057] According to the user load forecast results, the scheduling cost of user resources is adjusted through the preset day-ahead scheduling model to obtain the day-ahead optimal scheduling plan;

[0058] Based on the obtained optimal scheduling plan for the day ahead, combined with model predictive control, a user resource intraday dynamic scheduling optimization model is constructed;

[0059] Solve the constructed intraday dynamic scheduling optimization model of user resources to obtain the optimal intraday scheduling plan;

[0060] If the obtained intraday optimal scheduling plan meets the preset control target, the scheduling dynamic optimization of multi-type user resources is completed according to the obtained intraday optimal scheduling plan; otherwise, the scheduling cost of user resources is readjusted.

[0061] This embodiment uses the Informer model to predict user load.

[0062] In this embodiment, the deployable resources include not only the flexible load at the load end, but also other deployable resources; that is, the deployable resources in this embodiment specifically include distributed energy storage, EV charging stations, micro gas turbines, and OLTC equipment.

[0063] This embodiment classifies distributed energy storage and EV charging stations into energy storage for research, which can provide both active and reactive power scheduling; the micro gas turbine MT mainly provides active power scheduling; and the OLTC equipment can only provide reactive power scheduling.

[0064] The short-term forecasting model mainly predicts short-term loads and deployable resources in a given time period. In the short-term load forecasting, the flexible resources that can participate in the dispatch can be calculated through the total load level according to historical data. Therefore, the short-term load forecasting of the model actually includes the estimation of the flexible load level.

[0065] The independent variables of the prediction model include the characteristic variables of short-term load and the state variables of deployable resources (such as the charge state of energy storage, the active power and climbing state of micro gas turbines, etc.), and the dependent variables are the short-term load forecast value and the active / reactive output that can be provided by deployable resources.

[0066] Let the set of independent variables be X={X1,X2,…X n}; where element X i represents the time series set of the ith independent variable, which can be expressed as X i ={x i,1 ,x i,2 ,…x i,t}, the subscript t represents the length of the time series.

[0067] This embodiment uses the Informer model to predict the independent variable. After X is input into the Informer encoder for linear transformation of the feature vector, the matrix Ω representing the query with equal column length can be obtained. q , the matrix Ω representing the bond k And the matrix Ω representing the value v ; The three matrices obtained are processed by the multi-head probabilistic sparse self-attention mechanism in the encoder, that is, the sequence processed by the multi-head probabilistic sparse self-attention mechanism for in, Represents the matrix Ω q The matrix after probability thinning; Represents Ω k is the transpose of ; d represents the column vector length of the above three matrices.

[0068] In order to compress the dimension of feature sequence, this embodiment extracts the advantage attention mechanism. Informer adds the distillation process of self-attention in addition to the multi-head probabilistic sparse self-attention mechanism. The update process of the distillation process can be expressed as in, represents the input of the j-th layer attention module; MaxPool represents the maximum pooling function; elu represents the activation function; convld represents the one-dimensional convolution function; express The output after transformation by the self-attention module.

[0069] When the above distillation process reaches the maximum number of steps, the encoder outputs the feature map to the decoder. After the multi-head self-attention mechanism operation of several layers of parallel decoder, the fully connected layer outputs the final prediction result at one time, which is recorded as I = {I1, I2, ... I n}, element I in I i Yes X i prediction results.

[0070] In a long time series, apart from the trend and periodicity in the series, the series itself contains the inertia of historical information, which is called the Historical Inertia theory. iIndicates X i prediction, we can intercept X i The subsequence closest to the predicted target value is directly used as H i The estimated value of Among them, L H Indicates the length of the subsequence closest to the predicted target value. By integrating the Informer and HI mechanisms, the composite prediction model can be obtained as follows: in, and They represent weight coefficients respectively; I and H represent the prediction results of Informer and HI for X respectively; Represents the prediction result of Informer and HI composite model for X. and The value of can be determined by the least squares method OLS, that is, Among them, y i represents the true value of the i-th dependent variable; represents the predicted value of the i-th dependent variable; I i and H i Represent the predicted values ​​of the Informer and HI models respectively. The extreme value conditions of the above formula can be obtained to obtain the parameters and The best value of .

[0071] The above model is used to predict short-term load and deployable resources for a given time period, and the prediction results will serve as the initial state set of the system in a single decision cycle.

[0072] To keep consistent with the day-ahead market, in the day-ahead scheduling stage, the model takes 1 hour as the decision cycle and examines the total cycle of 24 hours within the day. In the day-ahead stage, the system power deviation is not considered, only the economic efficiency of the system decision is considered, so the decision-making goal is to minimize the scheduling cost, and the objective function F DA for Where, T represents the total decision cycle, which is 24 hours; t represents the time period; μ loss,t represents the unit line loss cost in period t; P loss,t represents the power line loss of the entire network in period t; C ess,t represents the cost of energy storage participating in dispatch during period t; C mt,t represents the cost of micro gas turbine participating in the dispatch in period t; C oltc,t represents the reactive power regulation cost of OLTC equipment during period t; It represents the cost of class A demand response decomposed into period t.

[0073] The power line loss of the entire network during period t is P loss,t It can be expressed as Where N represents the total number of branches in the entire network; Pij,t represents the active power between branches i and j during period t; P` ij,t represents the reactive power between branches i and j during period t; V i,t represents the voltage amplitude at node i during period t; r ij Represents the resistance between branches i and j.

[0074] The dispatch cost of energy storage in period t is C ess,t It can be expressed as Among them, Ω ess Represents the set of energy storage nodes connected to the distribution network; μ ess,t represents the unit cost of energy storage dispatch in period t; represents the active power of the i-th energy storage in period t; It represents the reactive power of the i-th energy storage in period t.

[0075] The dispatch cost C of the micro gas turbine in period t mt,t It can be expressed as Where, Ωess represents all micro gas turbine nodes connected to the distribution network; μmt,t represents the unit cost of calling the micro gas turbine power; It indicates the absolute value of the micro gas turbine climbing power called in this period.

[0076] The cost C of reactive power regulation of OLTC equipment during period t oltc,t It can be expressed as C oltc,t =μ oltc,t λ oltc,t ; where μ oltc,t represents the cost of a single switch of OLTC equipment in period t; oltc,t Indicates the number of switching actions of OLTC equipment in period t.

[0077] Pay the cost of effective DR in period t It can be expressed as Among them, μ dr,t represents the unit price of demand response during period t; It represents the load reduction level of the class A demand response during period t decomposed into period t.

[0078] In the intraday stage, the model predictive control MPC mechanism is introduced to achieve dynamic optimization of the model in a given decision cycle. In this embodiment, the decision cycle of the intraday stage is set to 15 minutes.

[0079] There is a deviation in settlement in the intraday market, so the scheduling target of the decision cycle must take into account both economic efficiency and risk level assessment. The decision cycle remains unchanged, the specific cost items are different, and the objective function adds the risk value assessment. Therefore, the first objective function of the intraday stage can be expressed as Among them, F RT,1 represents the first objective function in the intraday stage; Δt represents a single decision cycle in the day, which is 15 minutes; C dr,t represents the payment cost of other demand responses except Class A in period t; θ represents the unit penalty coefficient of the risk indicator; Γ R represents the global risk index, Γ R =ε pv Γ pv +ε wt Γ wt +ε ess Γ ess , Γ pv , Γ wt , Γ ess Respectively represent the risk indicators of photovoltaic, wind power and energy storage; ε pv , ε wt , ε ess Represent the weights of various risk indicators respectively.

[0080] The single-category risk index is characterized by the probability distribution of the voltage over-limit of its nodes, namely:

[0081]

[0082] Among them, V i,min Indicates the lower limit of the over-limit voltage; V i,max Indicates the upper limit of the over-limit voltage; f(V i,t ) represents the probability density function of the node voltage. The above formula represents the risk index of a single node, and the overall risk index of each type of unit can be obtained by summing up the set, which will not be repeated here.

[0083] After considering the overall economy and risk, due to the strictness of the power system flow safety constraints, the node must not have an over-limit accident. Therefore, the second objective function of the intraday stage is to minimize the node voltage deviation, that is, Where N represents the set of all nodes; E(V i,t ) represents the expected value of the voltage at node i.

[0084] The goal of system scheduling also includes minimizing the resources called within the decision cycle, so the third objective function can be expressed as Among them, χ j is a dummy variable. When the previous priority resource of the j-th class of deployable resources cannot meet the total demand of the scheduling period, it takes 1, and when the j-th class of resources does not need to be called, it takes 0; it contains |Φ j,t+Δt -Φ j.t | represents the value of the jth type of deployable resource deployed in the scheduling period; represents the maximum adjustable value of the j-th resource; Ψ represents the priority and category set of the deployable resources, The order of elements indicates the scheduling priority of various resources; SOC t Indicates the state of charge of all energy storage during period t; represents the available power of all energy storage in period t; represents the available power of all micro gas turbines during period t; Represents all flexible loads that can be reduced or transferred during period t.

[0085] The intraday objective function includes three functions, which have different dimensions and scales. However, since the three functions are all seeking the minimum value, the three functions can be normalized and linearly combined to obtain the final objective function. in, represents the weight coefficient; Represent the initial values ​​of the three objective functions respectively.

[0086] In this embodiment, the constraints of the intraday dynamic scheduling optimization model of user resources include power flow constraints and power constraints; specifically:

[0087] The power flow constraint can be expressed as Among them, P ij,t and P i ` j,t represents the active power and reactive power between nodes i and j respectively; k:j→k represents other nodes that generate power flow with j except node i; I ij represents the current between nodes i and j; P j,t and P j ` ,t They represent the net active power and reactive power injected into node j during period t respectively; L j,t and L` j,t They represent the active power and reactive power of the load at node j respectively; and They represent the energy storage active power and reactive power of node j respectively; and They represent the photovoltaic active power and reactive power of node j respectively; and They represent the wind power active power and reactive power of node j respectively; represents the active power of the micro gas turbine at node j.

[0088] The system needs to meet voltage safety conditions, which has been reflected in the objective function and will not be repeated.

[0089] The output constraint of renewable energy is in, and Represent the rated power of photovoltaic and wind power respectively.

[0090] The power constraint of the micro gas turbine is: in, and They represent the minimum and maximum output of the micro gas turbine respectively; represents the ramp rate threshold of the micro gas turbine; Represents the ramp rate threshold of the micro gas turbine.

[0091] The power and state of charge constraints of energy storage are: That is, they respectively indicate that the capacity constraint, charge and discharge power constraint, charge and discharge state constraint, initial state and final state of charge of ESS are the same; among them, SOC min and SOC max Respectively represent the lower and upper limits of the state of charge; κ ch (t) and k dis (t) is a dummy variable, which takes the value of 1 when the ESS is in the charging / discharging state.

[0092] After completing the prediction model and the day-ahead-day scheduling model, this embodiment determines the optimal scheduling plan for the day and the optimal call amount of the deployable resources; specifically

[0093] Step 1: In the day-ahead phase, input the dispatch plan for the next 24 hours and the initial values ​​of the respective variables, including load forecast, deployable resources, and renewable energy;

[0094] Step 2: Informer outputs the short-term forecast results and inputs them into the model to generate the scheduling plan for the current decision cycle within the day;

[0095] Step 3: Take the true value of the independent variable in this decision cycle as the independent variable, input it into the prediction model, and output the scheduling plan for the next stage;

[0096] Step 4: Solve the dynamic scheduling optimization model and output the optimal decision result according to the objective function requirements;

[0097] It should be noted that the decision variables in the scheduling plan given in step 3 are input into the model as parameters, and the objective function is solved under the constraints, so that the optimal scheduling plan for the next stage can be obtained, completing the correction and optimization of the scheduling plan obtained in step 3.

[0098] Solving the dynamic scheduling optimization model in this embodiment is actually a multi-objective linear programming problem. The big M method is used. For each objective function, the corresponding number of artificial variables is set according to the number of constraints, and the initial M value is set, which is generally 10000. The predicted value given in step 3 is used as the boundary value of the constraint condition, and the decision variables in the scheduling plan generated in step 3 are used as the coefficient term of the objective function. The Simplemax function in matlab is called to give the optimal solution under the given big M value.

[0099] Step 5: Set the control target and determine whether the result of step 4 meets the target. If yes, go to step 6. If not, go back to step 2.

[0100] It should be noted that in the process of setting control targets, a threshold interval is first set to determine whether the output result of step 4 is within the threshold, such as whether the deviation between the predicted value and the true value is less than the threshold interval, and whether the proportion of the penalty item to the operating cost is less than the threshold interval; the basis for setting includes industry norms and the control objectives of the operating entity, which can be set according to actual needs, such as requiring the prediction error to be less than 5%.

[0101] Step 6: Based on the decision results, output the final dynamic rolling scheduling plan.

[0102] This embodiment combines load forecasting with the SCED target and dispatching process of the power grid, which is in line with the inherent mechanism of the power market, namely, the joint clearing of the electric energy market and the auxiliary market. It takes the safe and economic dispatching of the power grid as a constraint condition to determine the optimal configuration of the deployable resources in each time period, thereby providing intuitive and feasible data support for the specific trading strategies of various trading varieties and various trading scenarios in the power market. Through graph comparison learning, a load synergy prediction model is constructed, combined with model predictive control MPC, and a dynamic optimization model for the whole network dispatch is constructed according to the current trading cycle of the power market, so as to obtain dynamic rolling dispatching of dispatched resources.

[0103] Embodiment 2

[0104] Embodiment 2 of the present invention introduces a dynamic optimization system for resource scheduling of multiple types of users based on load forecasting.

[0105] like Figure 2 A multi-type user resource scheduling dynamic optimization system based on load forecasting is shown, comprising:

[0106] An acquisition module, which is configured to acquire multiple types of user resource parameters;

[0107] A prediction module, configured to predict user load according to the acquired user resource parameters;

[0108] A day-ahead scheduling module is configured to adjust the scheduling cost of user resources through a preset day-ahead scheduling model according to the user load forecast result to obtain the day-ahead optimal scheduling plan;

[0109] The intraday scheduling module is configured to build a user resource intraday dynamic scheduling optimization model based on the obtained optimal scheduling plan for the day ahead and in combination with model predictive control;

[0110] A solution module, which is configured to solve the constructed intraday dynamic scheduling optimization model of user resources to obtain an intraday optimal scheduling plan;

[0111] The dynamic optimization module is configured to complete the dynamic optimization of scheduling of multiple types of user resources according to the obtained intraday optimal scheduling plan if the obtained intraday optimal scheduling plan meets the preset control target; otherwise, the scheduling cost of user resources is readjusted.

[0112] As one or more implementation modes, in the prediction module, the specific process of predicting user load is: using the Informer model to predict user load, that is, predicting short-term load characteristic variables and deployable resource state variables; specifically, inputting the short-term load characteristic variables and deployable resource state variables into a preset Informer encoder for linear transformation of feature vectors to obtain a representation query matrix, a representation key matrix, and a representation value matrix with equal column lengths, performing multi-head probabilistic sparse self-attention mechanism processing on the obtained three matrices, and outputting the user load prediction result, that is, obtaining the short-term load prediction value and the active / reactive output that can be provided by the deployable resources.

[0113] As one or more implementation modes, in the day-ahead scheduling module, the preset day-ahead scheduling model takes the minimum total scheduling cost of user resources as the objective function, that is, the objective function F DA for Where, T represents the total decision cycle, which is 24 hours; t represents the time period; μ loss,t represents the unit line loss cost in period t; P loss,t represents the power line loss of the entire network in period t; C ess,t represents the cost of energy storage participating in dispatch during period t; C mt,t represents the cost of micro gas turbine participating in the dispatch in period t; C oltc,t represents the reactive power regulation cost of OLTC equipment during period t; It represents the cost of class A demand response decomposed into period t.

[0114] As one or more implementation modes, in the intraday scheduling module, the constructed user resource intraday dynamic scheduling optimization model constructs an objective function with the goal of minimizing the comprehensive cost within a single decision cycle within the day, the node voltage deviation and the resources called within the decision cycle. The constraints of the constructed objective function include at least flow constraints and power constraints.

[0115] The detailed steps are the same as the dynamic optimization method for multi-type user resource scheduling based on load forecasting provided in Example 1, and will not be repeated here.

[0116] Embodiment 3

[0117] Embodiment 3 of the present invention provides a computer-readable storage medium.

[0118] A computer-readable storage medium stores a program, which, when executed by a processor, implements the steps of the dynamic optimization method for resource scheduling of multiple types of users based on load prediction as described in Embodiment 1 of the present invention.

[0119] The detailed steps are the same as the dynamic optimization method for multi-type user resource scheduling based on load forecasting provided in Example 1, and will not be repeated here.

[0120] Embodiment 4

[0121] A fourth embodiment of the present invention provides an electronic device.

[0122] An electronic device comprises a memory, a processor and a program stored in the memory and running on the processor, wherein when the processor executes the program, the steps in the dynamic optimization method for resource scheduling of multiple types of users based on load prediction as described in the first embodiment of the present invention are implemented.

[0123] The detailed steps are the same as the dynamic optimization method for multi-type user resource scheduling based on load forecasting provided in Example 1, and will not be repeated here.

[0124] Embodiment 5

[0125] Embodiment 5 of the present invention provides a computer program product.

[0126] A computer program product includes software codes, wherein the program in the software codes executes the steps in the method for dynamic optimization of resource scheduling of multiple types of users based on load forecasting as described in the first embodiment of the present invention.

[0127] The detailed steps are the same as the dynamic optimization method for multi-type user resource scheduling based on load forecasting provided in Example 1, and will not be repeated here.

[0128] The above description is only a preferred embodiment of the present embodiment and is not intended to limit the present embodiment. For those skilled in the art, the present embodiment may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present embodiment shall be included in the protection scope of the present embodiment.

Claims

1. A dynamic optimization method for resource scheduling of multiple types of users based on load forecasting, characterized in that: include: Get multiple types of user resource parameters; Predict user load based on the acquired user resource parameters; According to the user load forecast results, the scheduling cost of user resources is adjusted through the preset day-ahead scheduling model to obtain the day-ahead optimal scheduling plan; Based on the obtained optimal scheduling plan for the day ahead, combined with model predictive control, a user resource intraday dynamic scheduling optimization model is constructed; Solve the constructed intraday dynamic scheduling optimization model of user resources to obtain the optimal intraday scheduling plan; If the obtained intraday optimal scheduling plan meets the preset control target, the scheduling dynamic optimization of multi-type user resources is completed according to the obtained intraday optimal scheduling plan.

2. A method for dynamic optimization of multi-type user resource scheduling based on load forecasting as claimed in claim 1, characterized in that: The specific process of predicting user load is: using the Informer model, a deep learning model for time series prediction, to predict user load and obtain short-term load characteristic variables and deployable resource state variables.

3. A method for dynamic optimization of resource scheduling for multiple types of users based on load forecasting as described in claim 2, characterized in that: The process of predicting user load by the Informer model is as follows: short-term load characteristic variables and deployable resource state variables are input into a preset Informer encoder for linear transformation of feature vectors to obtain a characterization query matrix, a characterization key matrix, and a characterization value matrix with equal column lengths, and the three matrices are processed by a multi-head probabilistic sparse self-attention mechanism to output the user load prediction result, that is, the short-term load prediction value and the active / reactive output that can be provided by the deployable resources are obtained.

4. A method for dynamic optimization of resource scheduling for multiple types of users based on load forecasting as claimed in claim 1, characterized in that: The preset day-ahead scheduling model takes the minimum total scheduling cost of user resources as the objective function. The objective function FDA is Wherein, T represents the total decision cycle, which is 24 hours; t represents the time period; μloss,t represents the unit line loss cost in time period t; Ploss,t represents the power line loss of the entire network in time period t; Cess,t represents the cost of energy storage participating in the dispatch in time period t; Cmt,t represents the cost of micro gas turbine participating in the dispatch in time period t; Coltc,t represents the reactive power regulation cost of the transformer on-load tap changer in time period t; It represents the cost of class A demand response decomposed into period t.

5. A method for dynamic optimization of resource scheduling for multiple types of users based on load forecasting as claimed in claim 1, characterized in that: The constructed user resource intraday dynamic scheduling optimization model constructs an objective function with the goal of minimizing the comprehensive cost within a single decision cycle within the day, the node voltage deviation and the resources called within the decision cycle. The constraints of the constructed objective function include at least flow constraints and power constraints.

6. A method for dynamic optimization of resource scheduling for multiple types of users based on load forecasting as claimed in claim 1, characterized in that: If the obtained optimal scheduling plan for the day does not meet the preset control objectives, the scheduling cost of user resources will be readjusted.

7. A dynamic optimization system for resource scheduling of multiple types of users based on load forecasting, characterized in that: include: An acquisition module, which is configured to acquire multiple types of user resource parameters; A prediction module, configured to predict user load according to the acquired user resource parameters; A day-ahead scheduling module is configured to adjust the scheduling cost of user resources through a preset day-ahead scheduling model according to the user load forecast result to obtain the day-ahead optimal scheduling plan; The intraday scheduling module is configured to build a user resource intraday dynamic scheduling optimization model based on the obtained optimal scheduling plan for the day ahead and in combination with model predictive control; A solution module, which is configured to solve the constructed intraday dynamic scheduling optimization model of user resources to obtain an intraday optimal scheduling plan; The dynamic optimization module is configured to complete the dynamic optimization of scheduling of multi-type user resources according to the obtained intra-day optimal scheduling plan if the obtained intra-day optimal scheduling plan meets the preset control target.

8. A multi-type user resource scheduling dynamic optimization system based on load forecasting as described in claim 7, characterized in that: In the prediction module, the specific process of predicting user load is: using the Informer model, a deep learning model for time series prediction, to predict user load and obtain short-term load characteristic variables and deployable resource state variables.

9. A multi-type user resource scheduling dynamic optimization system based on load forecasting as claimed in claim 8, characterized in that: The process of predicting user load by the Informer model is as follows: short-term load characteristic variables and deployable resource state variables are input into a preset Informer encoder for linear transformation of feature vectors to obtain a characterization query matrix, a characterization key matrix, and a characterization value matrix with equal column lengths, and the three matrices are processed by a multi-head probabilistic sparse self-attention mechanism to output the user load prediction result, that is, the short-term load prediction value and the active / reactive output that can be provided by the deployable resources are obtained.

10. A multi-type user resource scheduling dynamic optimization system based on load forecasting as claimed in claim 7, characterized in that: In the day-ahead scheduling module, the preset day-ahead scheduling model takes the minimum total scheduling cost of user resources as the objective function, and the objective function FDA is Wherein, T represents the total decision cycle, which is 24 hours; t represents the time period; μloss,t represents the unit line loss cost in time period t; Ploss,t represents the power line loss of the entire network in time period t; Cess,t represents the cost of energy storage participating in the dispatch in time period t; Cmt,t represents the cost of micro gas turbine participating in the dispatch in time period t; Coltc,t represents the reactive power regulation cost of the transformer on-load tap changer in time period t; It represents the cost of class A demand response decomposed into period t.

11. A multi-type user resource scheduling dynamic optimization system based on load forecasting as claimed in claim 7, characterized in that: In the intraday scheduling module, the constructed user resource intraday dynamic scheduling optimization model constructs an objective function with the goal of minimizing the comprehensive cost within a single decision cycle within the day, the node voltage deviation and the resources called within the decision cycle. The constraints of the constructed objective function include at least flow constraints and power constraints.

12. A multi-type user resource scheduling dynamic optimization system based on load forecasting as claimed in claim 7, characterized in that: In the dynamic optimization module, if the obtained optimal scheduling plan within the day does not meet the preset control target, it returns to the day-ahead scheduling module to readjust the scheduling cost of user resources.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the dynamic optimization method for multi-type user resource scheduling based on load prediction as described in any one of claims 1 to 6 are implemented.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the steps of the dynamic optimization method for resource scheduling of multiple types of users based on load prediction are implemented as described in any one of claims 1-6.

15. A computer program product comprising software code, characterized in that The program in the software code executes the steps of the dynamic optimization method for resource scheduling of multiple types of users based on load prediction as described in any one of claims 1-6.

Images