CN115795992A

CN115795992A - Park energy Internet online scheduling method based on virtual deduction of operation situation

Info

Publication number: CN115795992A
Application number: CN202210794743.5A
Authority: CN
Inventors: 朱丹丹; 岑炳成; 周前; 安海云
Original assignee: Electric Power Research Institute of State Grid Jiangsu Electric Power Co Ltd
Current assignee: Electric Power Research Institute of State Grid Jiangsu Electric Power Co Ltd
Priority date: 2022-07-07
Filing date: 2022-07-07
Publication date: 2023-03-14

Abstract

The invention discloses a park energy internet online scheduling method based on operation situation virtual deduction, which comprises the following steps of firstly carrying out park source-network-load-storage operation situation deduction: secondly, constructing a three-stage optimization scheduling model combining a day-ahead off-line scheduling model, an intra-day rolling optimization model and a real-time on-line scheduling model based on the characteristic that the prediction precision is improved along with the approach of a time domain; finally, the problem that information interaction is too frequent and uniform scheduling cannot be effectively achieved in actual operation of the system is solved by a distributed optimization planning method based on an adjustment alternating direction multiplier method, and the influence of wind energy and light energy random acquisition on the PEI is considered to form a final optimization planning result. The method is based on the operation situation virtual deduction, provides support for accurate scheduling by accurately sensing the change trend of the state of the PEI in the dynamic process, improves the accuracy and the calculation rapidity of operation regulation and control, and supports and forms a novel online coordinated scheduling mechanism combining real-time situation perception and super real-time virtual deduction.

Description

Park energy Internet online scheduling method based on virtual deduction of operation situation

Technical Field

The invention relates to a campus energy Internet online scheduling method based on operation situation virtual deduction, which is used for realizing real-time online adjustment of the operation state of a PEI system so as to form a novel online coordinated scheduling mechanism combining PEI multi-type resource real-time situation awareness and super real-time virtual deduction.

Background

Under the construction background of a novel electric power system, high-proportion distributed new Energy and diversified loads are connected to Park Energy Internet (PEI), the PEI multi-type resource device has strong uncertainty and power fluctuation, the PEI multi-type resource device can bear more extreme and severely-changed operating conditions, and higher requirements are provided for the PEI multi-type resource coordination Yuxing safety and the reliable operation. The method is necessary to apply a digital twin model of the PEI multi-type device to carry out inversion calculation and multi-parameter deduction analysis on state data of different sources, establish a mapping relation between the state of PEI multi-level resources and observable characteristic parameters thereof, obtain the distribution condition of the PEI internal multi-physical field parameters and the magnitude of key parameters, realize accurate analysis of the key state and deduction of the coordination mechanism and rule of the PEI multi-type resources under extreme conditions, and provide scientific basis for PEI long-term operation management.

Disclosure of Invention

The purpose of the invention is as follows: in order to overcome the defects in the prior art, the invention provides a park energy Internet online scheduling method based on operation situation virtual deduction.

The technical scheme is as follows: the invention relates to a campus energy Internet online scheduling method based on operation situation virtual deduction, which comprises the following steps of:

(1) The method comprises the following steps of (1) carrying out source-network-load-storage operation situation deduction on in a park: constructing a self-adaptive and self-learning situation deduction strategy, and automatically adjusting model parameters according to an application object and the latest data by the deduction strategy;

(2) Constructing a three-stage optimization scheduling model combining a day-ahead off-line scheduling model, an intra-day rolling optimization model and a real-time on-line scheduling model based on the characteristic that the prediction precision is improved along with the approach of a time domain;

(3) The distributed optimization planning method based on the adjustment alternating direction multiplier method solves the problem that information interaction is too frequent in actual operation of the system and effective unified scheduling cannot be achieved, and the influence of wind energy and light energy random acquisition on PEI is considered to form a final optimization planning result.

Further, the step (1) is realized as follows:

introducing a self-adaptive idea into load prediction, virtually predicting in a recent time period by adopting feedback prediction, predicting future loads by taking a parameter with the best prediction effect as a self-adaptive process model parameter, and forming a prediction closed loop and feedback;

the virtual prediction implementation process comprises two links of adaptive training and prediction: firstly, selecting a day to be predicted, namely a prediction time domain, and determining starting and ending time for prediction and virtual deduction; then, inputting relevant prediction and deduction basic model parameters into a prediction deduction model to perform virtual prediction; secondly, comparing actual values and derived values of historical data in the process of developing virtual prediction, and optimizing and adjusting model parameters; finally, the formed virtual deduction prediction model is used for daily load prediction, and the optimized estimation parameters are used as initial values of next optimization, so that rapid deduction is realized;

for the deduction of the PEI 'source-network-load-storage' operation situation, online monitoring, operation states and external environment data are transmitted to a PEI twin model in real time, and artificial intelligence algorithms such as deep learning and the like are applied to carry out iterative optimization and rolling prediction in the model, so that the park 'source-network-load-storage' independently predicts the future operation states, and real-time interactive closed-loop feedback is utilized to a physical entity PEI system to realize the operation risk autonomous evaluation management; and (3) combining the real-time operation situation inference result of source-network-load-storage with power flow control, performing digital twin simulation analysis, driving PEI to realize self-hardware condition scheduling and pre-scheduling, and realizing on-line analysis decision.

Further, the construction process of the day-ahead offline scheduling model in the step (2) is as follows:

the objective function is to maximize the daily revenue of PEI:

in the formula, F ₁ Representing the total profit of the PEI all day, wherein delta T represents the optimization time granularity, and T represents the optimization total time period number; t is the current time period;

the price of the internal electricity selling of the PEI and the user is expressed, is set by the PEI operator,

and P _t ^F The price and quantity of the purchased electricity respectively signed by the PEI and the electricity supplier,

and P _t ^S Respectively, the electricity price and the electricity quantity of the PEI and the real-time market transaction, when the PEI purchases electricity from the market in real time, P _t ^S Is positive; when it sells electricity, it is negative, F _cut.t 、F _tr.t And F _sh.t Respectively represent the compensation costs of reducible, translatable and convertible loads; f _TVPP Represents the total power generation cost of the PEI and the output P of the PEI _TVPP.t Correlation;

the constraint conditions are as follows:

and electricity selling price constraint:

and

the lower limit value and the upper limit value of the price of the electricity sold by the PEI user are shown,

the average electricity selling price of the PEI in one day is shown, and the electricity selling price of the PEI is ensured to be within a certain range;

considering the degree of satisfaction of the user, the amount of reduction and the number of times of reduction by which the load can be reduced are restricted:

in the formula N ₁ Represents the maximum value of the reduction times for reducing the load in one day,

and

a lower limit value and an upper limit value indicating that the amount of load reduction can be reduced;

transferable load constraint:

assume an initial run period of the translatable load is [ t ] ₁ ,t ₂ ]The time interval after translation is [ t ] _1- ,t ₂₊ ]Because some transferable load equipment cannot be started or stopped frequently, in order to prevent the electric equipment from being transferred into a plurality of dispersed time periods, the transfer time and the transfer power of the equipment are restricted;

representing the minimum run time of the device after the transfer,

and

a lower limit value and an upper limit value representing the transfer load amount, respectively;

translatable load restraint:

setting the time interval of original energy of load capable of translating as [ t ₃ ,t ₄ ]The power consumption range after the transfer is [ t ] _3- ,t ₄₊ ]The constraints for the translatable load are as follows:

and

a lower limit value and an upper limit value respectively representing a translatable load;

and (3) power balance constraint:

further, the construction process of the rolling optimization model in day in step (2) is as follows:

the goals for intra-day roll optimization are as follows:

in the formula, t0 is the optimization starting time in a day; Δ t is the granularity of the optimization time in day; d is the number of the optimized cycles; iq (T) is the starting and stopping state of the unit Q at the time T and is a variable of 0-1; pq (T) is the penalty cost for the unit Q at time T.

Further, the real-time online scheduling model building process in step (2) is as follows:

taking the minimum production cost of the PEI as an objective function:

F _TVPP representing the output cost of the lower PEI in the whole scheduling cycle, wherein t is the current operation time period; f _ge.j.t Represents the operating cost of the jth small gas turbine, F _w.t Representing the cost of electricity generated by the wind turbine, F _s.t Representing the cost of photovoltaic power generation, F _en.t Represents the operating cost of the segment battery;

the constraint conditions include:

and power balance constraint:

operating constraints of the gas turbine:

and (3) generating power constraint:

in the formula

An upper limit value representing the generated power of the jth gas turbine;

ramp rate constraint of the gas turbine:

in the formula

And

representing the lower limit and the upper limit of the climbing power of the jth unit;

output restraint of the fan and the photovoltaic unit:

in the formula

And

representing the lower limit and the upper limit of the actual output of the wind turbine generator in the period t,

and

representing the lower limit and the upper limit of the photovoltaic power generation power in the t period.

Further, the step (3) is realized as follows:

adding a regular term in a standard ADMM algorithm to ensure convergence when processing operators y and z specifically comprises the following steps:

in the formula: mu is more than 2, and the two-norm term where mu is located is a regular term;

a general distributed planning and scheduling model based on a PSR-ADMM algorithm is constructed and used for solving the distributed optimization problem of the electro-thermal-pneumatic PEI:

establishing an augmented Lagrange function for an electro-thermal-gas system:

L＝L _e +L _h +L _g (27)

the above equations (27), (28), (29) are the energy distribution subproblem, the thermal subproblem and the natural subproblem, respectively, where: A. b, C and D correspond to decomposition indexes of reliability, energy conservation, economy and environmental protection; p (-) and U (-) respectively represent the coupling variable and the coordinating variable of each subproblem, i represents the ith coupling device of the PEI, and the coupling devices comprise a cogeneration device, a gas turbine, an electric gas-converting device and an electric boiler.

Compared with the prior art, the invention has the beneficial effects that: compared with the traditional optimized scheduling method, the method is based on the operation situation virtual deduction, provides support for accurate scheduling by accurately sensing the change trend of the state of the PEI in the dynamic process, thereby improving the accuracy and the calculation rapidity of operation regulation and control, and supports and forms a novel online coordinated scheduling mechanism combining real-time situation awareness and super real-time virtual deduction.

Drawings

Fig. 1 is a flow chart of virtual deduction adaptation;

FIG. 2 is a diagram of a typical application of a "source-grid-load-store" situation deduction in a campus;

FIG. 3 is a flow chart of PEI multi-time scale optimization;

FIG. 4 is a schematic diagram of an example structure of the park energy Internet;

FIG. 5 is a fuzzy time-of-use energy price graph;

FIG. 6 is a comparison graph of the electrical load optimization of

scenarios

1 and 2;

FIG. 7 is a comparison graph of air load optimization conditions for

scenes

1 and 2;

FIG. 8 is a comparison chart of the thermal load optimization situation of

scenarios

1 and 2;

FIG. 9 is a comparison graph of the scores of the indicators and the comprehensive benefit scores in each scene;

FIG. 10 is a graph of residual convergence;

FIG. 11 is a diagram of a load side day ahead-day intra-optimization scheduling result;

FIG. 12 is a diagram of the power side day ahead-day optimized scheduling results;

FIG. 13 is a graph of load side day ahead-day in-real time optimal scheduling results;

FIG. 14 is a diagram of power side day ahead-day in-real time optimization scheduling results;

FIG. 15 is a graph of load power at different time scales;

FIG. 16 is a graph of fan output over different time scales.

Detailed Description

The present invention will be further described with reference to the accompanying drawings.

The invention provides a campus energy Internet online scheduling method based on operation situation virtual deduction, which specifically comprises the following steps of:

step 1: and a park source-network-load-storage operation situation deduction.

The situation deduction is to deduce the situation by mining the multidimensional perception of potential relations and operation rules in a 'source-network-load-store' information data set, so as to predict the trend change rule in a certain period in the future, and is the key for realizing supervision under the situation perception. The multi-element load in PEI has complex change mode and different rules. In order to accurately infer the PEI condition and carry out situation deduction, a comprehensive and perfect prediction process needs to be constructed to meet practical requirements, and the important approach is to construct a self-adaptive and self-learning situation deduction strategy, wherein the deduction strategy automatically adjusts model parameters according to an application object and the latest data.

The idea of adaptation is embodied in an artificial neural network that can create independent operations without external control and learn from self-failures, external world observations and experience. When conditions change, a corresponding adjustment is made. The self-adaptive idea is introduced into load prediction, feedback type prediction is adopted to virtually predict a recent time interval, parameters with the best prediction effect are used as self-adaptive process model parameters to predict future loads, and a prediction closed loop and feedback are formed.

The virtual prediction implementation process is divided into two links of adaptive training and prediction, as shown in fig. 1, the basic principle is to adjust other parameters in advance within a value range according to the current initial values of default parameters, if the adjustment results in a better virtual deduction result, the adjustment is accepted, otherwise, the adjustment is rejected, and the process is circulated until the optimal values of the parameters are static and unchanged. Firstly, selecting a day to be predicted, namely a prediction time domain, and determining starting and ending time for prediction and virtual deduction; then, inputting relevant prediction and deduction basic model parameters into a prediction deduction model to perform virtual prediction; secondly, in the process of developing virtual prediction, comparing actual values and derived values of historical data, optimizing and adjusting model parameters, gradually improving precision and reducing calculation time; and finally, the formed virtual deduction prediction model is used for daily load prediction, and the optimized estimation parameters are used as initial values of next optimization, so that the calculation complexity of parameter adjustment is reduced, and quick deduction is realized.

For the PEI 'source-network-load-store' operation situation deduction, the basic characteristic is that the feedback of a physical entity is controlled through a decision instruction based on artificial intelligence. Data such as online monitoring, operation states, external environments and the like are transmitted to a PEI twin model in real time, iterative optimization and rolling prediction are carried out in the model by applying artificial intelligence algorithms such as deep learning and the like, so that future operation states of a park are independently predicted, and real-time interactive closed-loop feedback is carried out on the future operation states to a physical entity PEI system, and operation risk independent evaluation management is realized; and (3) combining a source-network-load-storage real-time operation situation inference result with power flow control, performing digital twin simulation analysis, driving the PEI to realize self-hardware condition scheduling and pre-scheduling, and realizing on-line analysis decision. A specific situation deduction application is shown in fig. 2.

Step 2: and constructing a three-stage optimized scheduling model combining a day-ahead offline scheduling model, an intra-day rolling optimization model and a real-time online scheduling model based on the characteristic that the prediction precision is improved along with the approach of a time domain.

As the multi-type network coupling and the source load interaction in the PEI are continuously deepened, the online safety state of the PEI is difficult to identify in real time, and the traditional physical mechanism modeling method is difficult to solve the problem of large-scale coordinated optimization source network load storage. Based on a PEI digital twin system, state data can be sensed in real time, and artificial intelligent algorithms such as deep reinforcement learning and transfer learning are embedded on the basis of supporting off-line simulation. The online safe stable state intelligent evaluation and weak link identification form a virtual deduction of operation situation, the data-driven optimization method can continuously interact with the environment, the optimal operation strategy is formed through autonomous learning, the adaptability of PEI multi-link collaborative operation is enhanced, the source load matching degree is improved, and the utilization of renewable energy sources is promoted. Therefore, offline learning and online decision are combined, the calculation efficiency of the source-network-load-storage online decision can be effectively improved, and active online analysis and autonomous intelligent adjustment are finally realized. In order to reduce the influence of the intermittent output of renewable energy sources such as wind power and photovoltaic and random factors such as load fluctuation on PEI operation scheduling, a three-stage optimization scheduling framework combining off-line before day, in-day adjustment and real-time online is constructed based on the characteristic that the prediction precision is improved along with the approach of a time domain, as shown in FIG. 3.

(1) And (5) a day-ahead offline scheduling model.

And (4) taking 1h as time granularity on the previous day of optimization, and determining a unit output and load reduction plan of 24h in the future. According to the yield prediction, the load prediction and the electricity price information of renewable energy sources such as wind energy, photovoltaic energy and the like, an optimal offline availability model is established one day in advance to maximize the daily income of PEI:

in the formula F ₁ Representing the total profit of the PEI all day, wherein delta T represents the optimization time granularity, and T represents the optimization total time period number; t is the current time period;

denotes PEI withThe internal electricity selling price of the user is set by the PEI operator,

and P _t ^S Electricity price and quantity of electricity for PEI's trading with the real-time market, respectively, P when PEI purchases electricity from the market in real time _t ^S Is positive; when it sells electricity, it is negative, F _cut.t 、F _tr.t And F _sh.t Respectively represent the compensation costs of reducible, translatable and convertible loads; f _TVPP Represents the total power generation cost of the PEI and the output P of the PEI _TVPP.t And (6) correlating.

The constraint conditions include:

and (3) restricting electricity selling price:

and

the lower limit value and the upper limit value of the electricity selling price of the PEI user are shown,

the average electricity selling price of the PEI in one day is shown, and the electricity selling price of the PEI is guaranteed to be within a certain range.

Considering the degree of satisfaction of the user, the amount and number of times of reduction of the reducible load are restricted:

in the formula N ₁ Represents the maximum number of times of reduction in which the load can be reduced during a day,

and

the lower limit value and the upper limit value indicate that the amount of load reduction can be reduced.

Transferable load constraint:

assume an initial run period of a translatable load is t ₁ ,t ₂ ]The time interval after translation is [ t ] _1- ,t ₂₊ ]Because some transferable load equipment can not be started or stopped frequently, the transfer time and the transfer power of the equipment are restricted to prevent the electric equipment from being transferred into a plurality of scattered time periods.

Indicating the minimum run time of the device after the transfer,

and

the lower limit value and the upper limit value of the transfer load amount are respectively indicated.

Translatable load restraint:

setting the time interval of original energy of load capable of translating as [ t ₃ ,t ₄ ]The power consumption range after the transfer is [ t ] _3- ,t ₄₊ ]The constraints for the translatable loads are as follows: :

and

respectively representing a lower limit value and an upper limit value of the translatable load.

And power balance constraint:

(2) And (4) rolling the optimization model in days.

Rolling optimization within a day takes 15min as time granularity; according to the latest prediction information, the daily production plan of each unit must be optimized and adjusted so as to reduce the daily operation cost of the system and the fine cost of starting and stopping each unit to the maximum extent.

During the intraday roll optimization process, renewable energy sources, such as wind energy, solar energy, and load data, are updated according to the predicted time. In order to avoid the repeated starting of the unit due to deviation from the day-ahead scheduling plan, the unit starting and stopping punishments are taken into the system operation cost, and a day-ahead optimization model is created to minimize the total cost. Under the condition of meeting the system operation constraint, the output plan of each unit in the prediction time domain is optimally adjusted, but the actual execution is only the operation plan in the control time domain. Meanwhile, the prediction time domain and the control time domain are updated according to the optimized time granularity, namely 15min rolling, an intra-day optimization model is calculated according to the updated wind power photovoltaic output and load rolling prediction information, a rolling operation plan is formed, and the next plan is implemented. And after repeated iteration and updating, the plan of all time periods is updated. The goals for intra-day roll optimization are as follows.

In the formula: t0 is the optimization starting time in a day; Δ t is the granularity of the optimized time in day; d is the number of the optimization cycles; iq (T) is the starting and stopping state of the unit Q at the time T and is a variable of 0-1; pq (T) is the penalty cost for the unit Q at time T.

(3) And (5) real-time online scheduling model.

In the real-time online optimization stage, the deviation of the daily operation is adjusted mainly based on the ultra-short term 'source load' operation situation prediction so as to meet the time-varying requirement of the actual operation state of the system. The real-time adjustment takes 5min as an interval, and a day operation plan based on an MPC frame is made in order to reduce the total amount of equipment adjustment at the next moment as much as possible according to the real-time prediction information of wind power, photovoltaic output and load. Each period of the daily operation plan comprises a plurality of real-time adjustment stages after the real-time adjustment is completed, the optimization stage is returned within one day to form the next part of the plan, and the final output plan of the system is formed by adjusting the output in the optimization period of the new stage in real time and updating the output in a rolling manner in any time domain.

After the day-ahead offline scheduling model is solved, an optimization result with the maximum PEI yield as a target can be obtained, PEI output in an upper-layer scheduling result is taken as a constraint condition, and the determination of an objective function is to take the lowest PEI production cost as a target, namely:

F _TVPP representing the output cost of the lower PEI in the whole scheduling cycle, wherein t is the current operation time period; f _ge.j.t Represents the operating cost of the jth small gas turbine, F _w.t Representing the cost of electricity generated by the wind turbine, F _s.t Representing the cost of photovoltaic power generation, F _en.t Representing the operating cost of the segment battery.

Renewable energy output cost:

taking the output cost of the wind turbine as an example, the power generation cost F _w.t The expression is as follows:

the power generation cost of the fan consists of two parts，

Represents the actual output cost of the fan,

and representing the penalty cost of wind curtailment of the fan. In the formula P _w.t Represents the predicted power of the fan during the period t,

representing the actual generated power of the fan during the period t,

the electricity price and the electricity generation delta P for the electricity selling price of the real-time market _w.t Representing the air reject amount for the t period. b is a mixture of _w The linear relation coefficient of the actual generated power of the fan and the generating cost is related to the generating characteristic of the wind turbine generator. The photovoltaic power generation cost is also composed of the output cost and the light abandoning cost.

The storage battery generating and discharging cost is as follows:

F _en.t ＝a _en |P _en.t | ² +b _en |P _en.t |+c _en (10)

in the formula P _en.t Represents the battery capacity in the time period of t, when P _en.t When it is positive, the accumulator generates electricity, when P is _en.t When it is negative, the accumulator stores energy, a _en 、b _en And c _en The power parameter of the storage battery is related to the consumption characteristic of the storage battery.

Gas turbine power generation cost:

F _ge.j.t ＝a _ge.j P _ge.j.t ² +b _ge.j P _ge.j.t +c _ge.j (11)

P _ge.j.t represents the generated power of the jth gas turbine in the period t, a _ge.j 、b _ge.j And c _ge.j The characteristic parameter of the power generation consumption of the jth gas turbine is related to the performance and the energy consumption of the gas turbine.

The constraint conditions include:

and power balance constraint:

operating constraints of the gas turbine:

and (3) generating power constraint:

in the formula

Represents an upper limit value of the generated power of the jth gas turbine.

Ramp rate constraint of the gas turbine:

in the formula

And

representing the lower limit and the upper limit of the climbing power of the jth unit.

Output restraint of the fan and the photovoltaic unit:

in the formula

And

represents the lower limit and the upper limit of the actual output of the wind turbine generator in the period t,

and

represents the lower limit and the upper limit of the photovoltaic power generation power in the t period.

And (3) battery restraint:

the constraint conditions of the storage battery are consistent with the constraint conditions of the energy storage equipment.

And step 3: and a park energy Internet online optimization scheduling solution.

In order to deal with a PEI complex scheduling model, the existing research mainly adopts a centralized method, and ignores the problem that the information interaction of the actual operation of each electric-thermal-pneumatic subsystem is too frequent and the scheduling cannot be effectively unified, so that a distributed optimization planning method based on a PSR-ADMM (power system-adaptive planning method) is provided in the section to solve the problem, and the influence of random acquisition of wind energy, optical energy and the like on PEI is researched to form a final optimization planning result.

And (3) processing wind and light output uncertainty: for the system power balance problem caused by the randomness of wind power and photovoltaic output, the system power balance is expressed in a probability form by adopting a probability finite planning theory, and a probability model is converted into a deterministic expression as follows:

since wind photovoltaics are randomly varied, equation (16) can be converted into the probability form:

in the formula: β is the confidence level. Carrying out determination treatment to convert the following steps:

in the formula: f _W,t (. Is) a combined probability distribution function of wind power output and photovoltaic output in a time period t, and can be according to output random variables thereof

And

the probability density function of (2) can be solved by adopting a dichotomy, and the following results are obtained:

in the formula:

is F _W,t Inverse function of (·).

By converting the above equation (18) into equation (19), the power balance probability problem is also converted into an uncertain certainty optimization problem.

Fuzzy time-sharing energy price: according to tax issues, the time-shared energy pricing mechanism divides 24 hours per day into three categories: the corresponding energy prices are reduced in turn at the peak, the gentle and the low valleys, and the user is stimulated to autonomously 'peak clipping and valley filling'. The conventional peak-valley average time-sharing energy price is a fixed price, the change frequency of the time-sharing energy price in different types of peak-valley average can be increased by using a fuzzy mathematical theory, the time-sharing electricity price with different prices per hour is formed, namely the fuzzy time-sharing energy price, and the user is stimulated to actively participate in response through refined pricing, so that the peak clipping and valley filling benefits are improved.

The specific method of fuzzy time-sharing energy pricing is as follows:

according to the quoted price; "electrothermal" for different load curves, the peak membership for each cycle is determined by a half-trapezoidal fuzzy membership function to convert "electrothermal". The mathematical expression for the probability that each point on each load curve belongs to a peak fall period is: :

the upper formula is a larger semi-trapezoidal membership function, and m and n are respectively an upper limit and a lower limit of the load; δ (t) — time period t load value in the load curve.

Each cycle is divided into four types: peak, off-peak and off-peak periods. Each period has a duration of at least 2 hours.

A base price should be determined for each type of term. The energy prices per hour in a time period fluctuate over a preset base price, and the fluctuation can be described as a gaussian membership function:

in the formula: r is the number of four classes; a is _r 、b _r 、c _r Respectively, the coefficients of the gaussian membership functions.

According to the basic energy price and the energy price fluctuation function of each period, a mathematical model for determining the fuzzy time-sharing energy price is as follows:

in the formula: c _f (t)、C _fp (t)、C _pg (t)、C _g (t) fuzzy time-sharing energy prices at peak hours, peak-average hours, average hours and low hours, respectively; c _base,f 、C _base,fp 、C _base,pg 、C _base,g Are the basis of the energy price in four time periods respectively; f _f (t)、F _fp (t)、F _pg (t)、F _g (t) is a fluctuating function of the energy prices for four periods, respectively.

The general distributed scheduling model based on the PSR-ADMM algorithm is as follows: the optimal scheduling model of the park energy Internet with the IDR is generally calculated uniformly by adopting a centralized method, is only suitable for a single energy system, and is not suitable for optimal scheduling of PEI (polyetherimide) with mutually independent and interconnected energy subsystems, so that the PEI optimal scheduling problem considering the fuzzy time-sharing energy price is solved by adopting a distributed algorithm.

The ADMM thought introduces a coordination variable U to decouple the coupling relation between subsystems, and can effectively solve the problem of single-region or multi-region distribution optimization with the same characteristics. Its standard form is:

in the formula: x, y, z are separable operators, respectively; A. b and C are parameters;

respectively their respective objective functions; x, Y and Z are variable sets of three separable operators respectively;

an augmented Lagrange function for introducing a Lagrange multiplier λ and an iteration step ρ.

The solution procedure for the ADMM standard form is:

the above formula shows that, in the (k + 1) th iteration, the calculation of the operators y and z is not fair, and the convergence is difficult to be ensured by the standard ADMM algorithm, and for this reason, the regular term is added to ensure the convergence when the operators y and z are processed, which is specifically as follows:

in the formula: mu is greater than 2, and the two-norm term where mu is located is a regular term.

On the basis of the algorithm, a general distributed planning and scheduling model based on a PSR-ADMM algorithm is constructed and used for solving the distributed optimization problem of the electro-thermal-pneumatic PEI. First, an augmented Lagrange function for the electro-thermal-gas system is established by the algorithm:

L＝L _e +L _h +L _g (27)

The specific solving process is as follows:

1) Initializing coordination variables

Lagrange multiplier λ and number of iterations k =0;

2) Starting the (k + 1) th iteration, solving the minimized electron problem equation (2-30) to obtain:

3) Updating the coordination variable and Lagrange multiplier:

in the formula:

respectively representing coupling variables of the coupling device of the ith coupling device in the electric-heat-gas combined system, and needing energy efficiency conversion in the actual treatment process; n (i) represents the number of coupling systems in the coupling equipment in the ith coupling device, such as an electric heating boiler of the electric heating coupling equipment, n (i) takes 2 in the updating of the coordination variable,

take 0.

4) Minimizing the thermodynamic subproblem equation (33), adding a regularization term when solving, resulting in:

5) And (3) minimizing a natural gas subproblem equation (34), and adding a regular term during solving to obtain:

6) Updating the coordination variable and Lagrange multiplier:

7) If the following conditions are met:

an optimal planning solution is considered to be obtained in this iteration, where P _i ^k+1 Representing the coupling variable of the (i) th coupling device in the (k + 1) th iteration in each subsystem,

is a coordination variable corresponding to the coupling variable; epsilon ₁ 、ε ₂ Respectively coupling residual and original residual, σ _p 、σ _d Is a relative stop threshold; otherwise, go back to 2) to continue the iterative computation.

In order to verify the influence of the fuzzy time-sharing energy price on the optimized operation effect of the PEI, 3 typical scenes are set, the price type comprehensive demand response is considered, and the scenes are as follows:

scene 1: the method aims at minimizing the operation cost, and considers the time-sharing energy price;

scene 2: the fuzzy time-sharing energy price is considered with the aim of minimizing the operation cost.

Scene 3: based on a multi-target model with reliability, energy conservation, economy and environmental protection, the fuzzy time-sharing energy price is considered.

The system architecture for performing the park energy internet offline optimization and online scheduling example analysis is shown in fig. 4, wherein the upper left part is an electric network, the lower left part is an air supply network, and the right part is a heat network.

Fuzzy time-sharing energy price: based on the typical daily load data in winter, the peak-valley membership degree is calculated, and the peak-valley membership degree is obtained as shown in table 1 by taking the time-of-use electricity price as an example.

TABLE 1 Peak-to-valley membership of Electricity prices

Then, each time period is classified by using the calculated peak-to-valley membership value, and the classification results in table 4 below are obtained as shown in table 2.

TABLE 2 electricity price time period division

Finally, combining the fuzzy time-sharing energy price model defined in the formula (38), the energy time-sharing fuzzy price divided according to the 24-hour and 1-hour time is shown in fig. 5, compared with the traditional time-sharing peak-valley electricity price, the time granularity of the fuzzy time-sharing electricity price is smaller, the time granularity is more flexible according to the fluctuation change of the peak and valley, the change trend can be flexibly adjusted according to the load curve, and therefore, the user is further guided to mine and adjust the potential by 'peak clipping and valley filling'.

Analyzing a multi-scene optimization result of the park energy Internet: the PEI electricity-gas-heat load combined optimization is shown in the figure 6 and the figure 7. The comparison shows that on the basis of fuzzy price, the load transfer force of the scene 2 is obviously improved, the average reduction amplitude of the scene 2 reaches 3.61% in the electricity utilization peak time period of 17-20h, the load is reduced by high electricity price, the load is increased by low electricity price, the peak clipping and valley filling are realized, and the electricity utilization cost is also saved; during the peak period 15-18h, the natural gas consumption of scenario 2 is much lower than scenario 1, with an average drop of 4.42%. The demand is shifted to the rest of the valley period to achieve the same peaking and filling effect as the energy load in terms of natural gas demand, and the heat load demand distribution is also significantly improved. In addition, through the comparative analysis of fig. 8, the total cost in the daily scheduling period of scene 1 is 1.82 ten thousand yuan, and scene 2 is 1.74 ten thousand yuan, which accounts for 4.23% of the total cost, and considering the comprehensive response to the fuzzy time-sharing energy price demand, the peak-valley difference of the system valley can be reduced, and the economic benefit can be remarkably improved.

And performing multi-dimensional evaluation on the three scenes to obtain the comprehensive benefits of optimized operation of the PEI in different scenes, as shown in FIG. 9. As can be seen from fig. 9, the comprehensive benefit score ranking scene 3> scene 2> scene 1, both scene 1 and scene 2 use the minimum running cost as the objective function, the economic indicators are higher than scene 3, and the comprehensive benefit scores are not separated by scene 3 because the economic indicator weight ratio is higher; and the scene 3 considering the multi-objective optimization has the highest score on other indexes except the economic index, so that the difference of the scene 3 on the economic index is made up, and the comprehensive benefit is the best. Therefore, the optimal operation of the PEI can be realized by the optimal scheduling method comprehensively considering a plurality of targets.

Fig. 10 shows the residual convergence logarithmic curve of the PSR-ADMM algorithm, and it can be noted that even residuals do not yet satisfy the convergence requirement although the original residuals satisfy the convergence requirement 27 times. In order to verify the effectiveness of the proposed method, a comparative analysis was performed on the optimization results of the distributed algorithm and the centralized algorithm, as shown in table 3. It can be seen that the computational performance of the centralized algorithm is significantly insufficient compared to the distributed algorithm. Compared with the improved ADMM algorithm, the algorithm reduces iteration times and improves convergence speed due to the addition of the regular term during solving, so that the calculation performance of the PSR-ADMM algorithm is verified, and a correct and fast-convergence optimal solution is obtained when the electric-thermal-gas PEI distributed optimal power flow is solved.

TABLE 3 comparative analysis of distributed and centralized optimization results

And (3) analyzing the scheduling results of different time scales: in the day scheduling stage, the prediction accuracy of the fan output and the load prediction power is obviously improved compared with that before the day, and in order to adapt to the change of the prediction accuracy, the day variable unit needs to be readjusted on the basis of day scheduling. In the real-time scheduling, the output of the distributed unit at the power supply side also needs to be changed, the class 2 loads can be reduced to meet the supply and demand balance of the system, and 3 scheduling models are designed for comparative analysis.

Scheduling model 1: and (5) scheduling the model day ahead. Scheduling once every hour, 24 scheduling periods all day, comprehensively scheduling each component unit of a demand side and a power supply side in the virtual power plant, determining scheduling amounts of transferable loads and transferable loads in price type demand response and incentive type demand response, and further determining a start-stop plan of a conventional unit and the output and calling states of energy storage equipment.

Scheduling model 2: day ahead-day in-day scheduling model. Optimizing once every 15min, totally 96 scheduling periods in a whole day, further determining the control quantity of the class 1 and the class 2 capable of reducing the load and the sets with poor adjusting capability in the wind turbine generator and the conventional set in the day stage on the basis of the scheduling model 1, and correcting the day result based on comparison and analysis of the day-ahead result.

Scheduling model 3: day ahead-day inside-real time scheduling model. And the optimized time granularity is 288 scheduling time intervals every 5min in the whole day, and the scheduling result in the day before and in the day is corrected to obtain the scheduling amount capable of reducing the class 2 load and the output of the fan and the conventional unit.

According to the scheduling result of the model 1, the optimal scheduling confidence coefficient in the day-ahead is set as: beta is a ₂ ＝0.95，γ ₂ =0.95. The confidence of the day-ahead-day-inside-real-time scheduling model is as follows: beta is a ₃ ＝0.98，γ ₃ =0.98. The scheduling results for model 2 are shown in fig. 11 and 12. The total time is divided into 96 periods in the scheduling in the day, and the class 1 can reduce the load in 24-37 periods, namely the load is reduced by the scheduling of the virtual power plant in the early peak period, and then the partial load is reduced in the afternoon and the evening in the 43-86 period. Class 2 can shed loads in three peak hours, 32-36, 45-55, and 67-85, i.e., early peak, mid-day peak, and late peak hours. In the peak period, the power consumption of the user is large, the load is reduced, the power resource can be saved, the state of power supply shortage in the peak period is relieved, and the balance of a power system is maintained. In the day power supply side scheduling, the output of the storage battery is consistent with that of the day ahead, the starting and stopping states of the conventional unit are unchanged, the output of the fan is scheduled based on the latest prediction data of the fan on the day time scale, and in order to maintain the balance of supply and demand of the system, the output of the conventional unit changes the scheduling result along with the final output of the fan.

The three time scale scheduled structures before day-within day-in real time are shown in fig. 13 and 14. It can be known that wind power, photovoltaic and load prediction errors exist, the unit combination scheme of the day-ahead-day optimized scheduling formed by the method cannot meet the demand and supply balance constraint in the PEI, and in order to guarantee important load supply, the real-time scheduling plan can enter a day-ahead scheduling plan which is not executed in a subsequent time period when necessary. In the real-time scheduling stage, only class 2 on the load side can reduce the load and receive scheduling, and the scheduling is received in periods of 94-102, 133-168 and 201-245 respectively, so that the electric load of a user side is cut off. In the day and real-time scheduling, the type 1 reducible load and the type 2 reducible load can both respond under the conditions of peak power utilization period and insufficient supply at the power supply side, but the type 2 reducible load scheduling cost is higher, the load can be quickly cut off under the scheduling of the control center, the response speed is high, uncertainty does not exist, the method is continuously applied to the real-time scheduling, and when other controlled loads cannot meet the balance requirements of a system and standby power, the load can be continuously reduced. Besides the load peak avoiding function, the 2-level load reduction can be used as a standby to stabilize the wind energy fluctuation on a small time scale, reduce the starting and stopping times of the conventional unit, keep the running supply and demand balance of the system and guarantee the power supply of important loads.

With the improvement of wind power prediction precision, the method is beneficial to reducing the configuration requirement of controllable 'source-load-storage' resources under the constraint of system operation. In the traditional scheduling strategy, in order to avoid the difference between the intraday and real-time scheduling result and the day-ahead scheduling state, 2 larger conventional units with faster acceleration are set as a rotary standby unit during the scheduling plan before the day so as to ensure that sufficient standby is provided to guarantee the supply of load. In three different time scale schedules, the schedule plan of the storage battery is consistent with the schedule stage before the day. The conventional unit is a controllable unit, and when the starting state and the stopping state are determined, in order to maintain the balance of supply and demand of the virtual power plant, the output of the fan is changed along with different time scales. Along with the approach of time, the wind power prediction accuracy is improved, the wind power prediction error is reduced, the system stability is increased, the system constraint condition can be more accurately met, and the configuration requirements of the power generation side and the load side are maintained.

The load output ratio after scheduling in different time scales is shown in fig. 15, in load scheduling, the difference between the intra-day scheduling result and the real-time scheduling result is not large with respect to the real-time scheduling result, and the current scheduling result and the real-time load power have some differences mainly because of the error of the load prediction power in different time scales and the difference of load scheduling. The output of the fan at different time scales is shown in fig. 16, the output of the fan is similar to the load scheduling result, compared with the scheduling in the day and the day, the output of the fan in the real-time scheduling has a few small changes, the real-time scheduling can reduce errors caused by predicted values, the consumption of renewable power generation is facilitated, and the stability of the system is improved.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The present invention is not limited to the above embodiments, and any modifications, equivalent substitutions, improvements, etc. within the spirit and principle of the present invention are included in the scope of the claims of the present invention as filed.

Claims

1. A park energy Internet online scheduling method based on operation situation virtual deduction is characterized by comprising the following steps:

(1) The operation situation deduction of 'source-network-load-storage' in the garden: constructing a self-adaptive and self-learning situation deduction strategy, and automatically adjusting model parameters according to an application object and the latest data by the deduction strategy;

(3) The distributed optimization planning method based on the adjustment alternative direction multiplier method solves the problem that information interaction is too frequent in actual operation of a system and effective unified scheduling cannot be achieved, and the influence of wind energy and optical energy random acquisition on PEI is considered to form a final optimization planning result.

2. The campus energy internet online scheduling method based on the virtual deduction of the operating situation as claimed in claim 1, wherein the step (1) is implemented as follows:

introducing a self-adaptive idea into load prediction, virtually predicting in a recent time period by adopting feedback prediction, predicting future load by taking a parameter with the best prediction effect as a self-adaptive process model parameter, and forming a prediction closed loop and feedback;

the virtual prediction implementation process comprises two links of adaptive training and prediction: firstly, selecting a day to be predicted, namely a prediction time domain, and determining start-stop time for prediction and virtual deduction; then, inputting relevant prediction and deduction basic model parameters into a prediction deduction model to perform virtual prediction; secondly, comparing the actual value and the derived value of the historical data in the process of developing virtual prediction, and optimizing and adjusting model parameters; finally, the formed virtual deduction prediction model is used for daily load prediction, and the optimized estimation parameters are used as initial values of next optimization, so that rapid deduction is realized;

3. The campus energy internet online scheduling method based on the virtual deduction of the operation situation as claimed in claim 1, wherein the construction process of the day-ahead offline scheduling model in the step (2) is as follows:

the objective function is to maximize the daily revenue of PEI:

in the formula, F ₁ The total profit of the PEI is shown, delta T shows the granularity of the optimization time, and T shows the number of the optimization total time; t is the current time period;

indicating the price of the internal electricity sold by the PEI and the user, as established by the PEI operator,

and P _t ^S Electricity price and quantity of electricity for PEI's trading with the real-time market, respectively, P when PEI purchases electricity from the market in real time _t ^S Is positive; when it sells electricity, it is negative, F _cut.t 、F _tr.t And F _sh.t Respectively representing the compensation costs of reducible, translatable and convertible loads; f _TVPP Represents the total power generation cost of the PEI and the output P of the PEI _TVPP.t Correlation;

the constraint conditions are as follows:

and electricity selling price constraint:

and

indicating average power sold by PEI during a dayPrice, namely ensuring that the electricity selling price of the PEI is within a certain range;

and

transferable load constraint:

assume an initial run period of a translatable load is t ₁ ,t ₂ ]The time interval after translation is [ t ] _1- ,t ₂₊ ]Because some transferable load equipment cannot be started or stopped frequently, in order to prevent the electric equipment from being transferred into a plurality of scattered time periods, the transfer time and the transfer power of the equipment are restricted;

representing the minimum run time of the device after the transfer,

and

translatable load restraint:

and

and power balance constraint:

4. the campus energy internet online scheduling method based on the operation situation virtual deduction as claimed in claim 1, wherein the construction process of the rolling optimization model in step (2) is as follows:

the goals for intra-day roll optimization are as follows:

in the formula, t0 is the optimization starting time in a day; Δ t is the granularity of the optimization time in day; d is the number of the optimized cycles; iq (T) is the starting and stopping state of the unit Q at the time T and is a variable from 0 to 1; pq (T) is the penalty cost for the unit Q at time T.

5. The campus energy internet online scheduling method based on the virtual deduction of the operating situation as claimed in claim 1, wherein the real-time online scheduling model in step (2) is constructed as follows:

taking the minimum production cost of the PEI as an objective function:

F _TVPP representing the output cost of the lower PEI in the whole scheduling cycle, wherein t is the current operation time period; f _ge.j.t Represents the operating cost of the jth small gas turbine, F _w.t Representing the cost of electricity generated by the wind turbine, F _s.t Representing the cost of electricity generation by the photovoltaic, F _en.t Represents the operating cost of the segment battery;

the constraint conditions include:

and (3) power balance constraint:

operating constraints of the gas turbine:

and (3) generating power constraint:

in the formula

An upper limit value representing the power generation power of the jth gas turbine;

ramp rate constraint of the gas turbine:

in the formula

And

output restraint of the fan and the photovoltaic unit:

in the formula

And

and

6. The campus energy internet online scheduling method based on the virtual deduction of the operating situation as claimed in claim 1, wherein the step (3) is implemented as follows:

establishing an augmented Lagrange function for the electro-thermo-pneumatic system:

L＝L _e +L _h +L _g (27)

the above equations (27), (28), (29) are the energy distribution subproblem, the thermal subproblem and the natural subproblem, respectively, where: A. b, C and D correspond to decomposition indexes of reliability, energy conservation, economy and environmental protection; p (-) and U (-) represent the coupling variables and the coordinating variables of each subproblem respectively, i represents the ith coupling device of PEI, and the coupling devices comprise a cogeneration device, a gas turbine, a power conversion gas and an electric boiler.