Multi-energy micro-grid energy regulation and control method and device and storage medium

Abstract

The invention discloses a multi-energy micro-grid energy regulation and control method and device and a storage medium. The method comprises the following steps of: constructing a scene prediction constraint optimization problem by utilizing a trained Warisstein generative adversarial network, and predicting a group of multi-dimensional uncertainty multi-energy micro-grid operation scenes in a future period by following known information at each moment to serve as prediction scenes; forming a pre-scheduling model and a real-time power compensation model based on the prediction scene, and forming a two-stage multi-energy micro-grid real-time energy regulation and control model; a pre-scheduling model obtains a pre-scheduling instruction of the multi-energy micro-grid based on random model prediction control by using the prediction scene information; the real-time power compensation model coordinates schedulable units in the multi-energy micro-grid according to real-time observation of uncertain variables; therefore, the energy in the multi-energy micro-grid is regulated and controlled. Through quick response and adjustment of real-time data, the system can effectively cope with fluctuation caused by uncertainty, thereby realizing accurate regulation and control of the energy of the multi-energy micro-grid.

Description

Technical Field

[0001] The present application relates to the field of energy regulation technology, and in particular to a multi-energy microgrid energy regulation method, device and storage medium. Background Art

[0002] With the continuous growth of global energy demand, the traditional single energy supply model can no longer meet the increasingly complex energy demand, resulting in the transformation of the energy structure towards a more diversified and sustainable direction. As an emerging form of energy system, multi-energy microgrids have gradually become an effective means to solve energy problems by virtue of their ability to comprehensively dispatch and optimize multiple energy forms (such as electricity, heat and cold energy). However, the operation of multi-energy microgrids is affected by many uncertain factors, such as weather, load changes, and energy price fluctuations. The existence of these uncertain factors makes it impossible to accurately regulate the energy of multi-energy microgrids. Summary of the invention

[0003] The main technical problem solved by this application is to provide a multi-energy microgrid energy control method to solve the above-mentioned problems.

[0004] In order to solve the above technical problems, a technical solution adopted in this application is to provide a multi-energy microgrid energy control method, comprising the steps of:

[0005] The trained Wasserstein generative adversarial network is used to construct a scenario prediction constrained optimization problem. At each moment, the known information is followed to predict a set of multi-dimensional uncertain multi-energy microgrid operation scenarios for a period of time in the future as prediction scenarios.

[0006] Based on the prediction scenario, a two-stage multi-energy microgrid real-time energy control model is formed, which is composed of a pre-scheduling model and a real-time power compensation model;

[0007] The pre-scheduling model uses the predicted scenario information to obtain the pre-scheduling instructions of the multi-energy microgrid based on random model predictive control; the real-time power compensation model coordinates the dispatchable units in the multi-energy microgrid according to the real-time observation of uncertain variables, thereby regulating the energy in the multi-energy microgrid.

[0008] The present application also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method when executing the computer program.

[0009] The present application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the described method when executed by a processor.

[0010] The beneficial effect of the present application is that: in the present application, the Wasserstein generative adversarial network is used to predict the operation scenario of the multi-energy microgrid in the future. The Wasserstein generative adversarial network can generate multi-dimensional uncertainty scenarios that conform to the actual situation by learning from historical data. In the multi-energy microgrid, uncertainty mainly comes from the volatility of renewable energy and the dynamic changes in load demand. By modeling these uncertain factors, scenarios for different time periods in the future are generated, which provides an important basis for subsequent energy regulation. A two-stage energy regulation model consisting of pre-scheduling and real-time power compensation. In the pre-scheduling stage, the model needs to reasonably arrange the various subsystems in the energy system according to the predicted multi-energy microgrid scenario to ensure efficient energy distribution. The main task of the pre-scheduling stage is to optimize the energy allocation strategy based on various uncertainties to maximize the operating efficiency and economy of the system. The real-time energy regulation model ensures the power balance of the multi-energy microgrid at each moment by adjusting the output of the dispatchable unit. Through the rapid response and adjustment of real-time data, the system can effectively cope with the fluctuations caused by uncertainty, thereby realizing precise regulation of the energy of the multi-energy microgrid. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a MEMG real-time energy control framework diagram based on data-driven scenario prediction according to an embodiment of the present application;

[0012] Figure 2 It is a diagram of variables involved in a typical configuration of a MEMG and its operation process according to an embodiment of the present application;

[0013] Figure 3 It is a detailed execution process diagram of MEMG real-time energy control according to an embodiment of the present application;

[0014] Figure 4 is a detailed structural diagram of WGAN according to an embodiment of the present application;

[0015] Figure 5 is a graph showing the evolution of Wasserstein distance during WGAN and JSD-GAN training according to an embodiment of the present application;

[0016] Figure 6 It is an autocorrelation coefficient curve diagram of a scene and a historical scene generated according to an embodiment of the present application;

[0017] Figure 7 A CDF curve diagram of a scene and a historical scene generated according to an embodiment of the present application;

[0018] Figure 8 It is a MEMG real-time electricity purchase price curve diagram according to an embodiment of the present application;

[0019] Fig. 9is a power curve diagram of RDG and electrical load and thermal load according to an embodiment of the present application;

[0020] Fig.10 This is a diagram of a MEMG electric power operation scheme according to an embodiment of the present application;

[0021] Fig.11 This is a diagram of a MEMG thermal power operation scheme according to an embodiment of the present application;

[0022] Fig.12 It is a power diagram of electric and thermal imbalance caused by the SMPC and MPC solutions according to an embodiment of the present application;

[0023] Fig.13 This is a comparison diagram of the interaction power between MEMG and the power grid according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to facilitate the understanding of the present application, the present application is described in more detail below in conjunction with the accompanying drawings and specific embodiments. The preferred embodiments of the present application are provided in the accompanying drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described in this specification. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive.

[0025] It should be noted that, unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art of the present application. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in this specification includes any and all combinations of one or more related listed items.

[0026] Figure 1-Figure 13 An embodiment of the multi-energy microgrid energy control method of the present application is shown, including:

[0027] The trained Wasserstein generative adversarial network is used to construct a scenario prediction constrained optimization problem, so as to predict a set of multi-dimensional uncertain multi-energy microgrid operation scenarios in the future period according to the known information at each moment as the prediction scenario;

[0028] Based on the prediction scenario, a two-stage multi-energy microgrid real-time energy control model consisting of pre-scheduling and real-time power compensation is formed;

[0029] In the pre-dispatching stage, the real-time energy control model uses the predicted scenario information to obtain the pre-dispatching instructions of the multi-energy microgrid based on the stochastic model predictive control; in the real-time power compensation stage, the dispatchable units in the multi-energy microgrid are coordinated according to the real-time observation of uncertain variables, thereby regulating the energy in the multi-energy microgrid.

[0030] In this application, the Wasserstein generative adversarial network is used to predict the operating scenarios of multi-energy microgrids in the future. The Wasserstein generative adversarial network can generate multi-dimensional uncertainty scenarios that conform to the actual situation by learning historical data. In the multi-energy microgrid, uncertainty mainly comes from the volatility of renewable energy and the dynamic changes in load demand. By modeling these uncertain factors, scenarios for different time periods in the future are generated, which provides an important basis for subsequent energy regulation. A two-stage energy regulation model consisting of pre-scheduling and real-time power compensation. In the pre-scheduling stage, the model needs to reasonably arrange the various subsystems in the energy system according to the predicted multi-energy microgrid scenario to ensure efficient energy distribution. The main task of the pre-scheduling stage is to optimize the energy allocation strategy based on various uncertainties to maximize the operating efficiency and economy of the system. The real-time energy regulation model ensures the power balance of the multi-energy microgrid at all times by adjusting the output of the dispatchable unit. Through the rapid response and adjustment of real-time data, the system can effectively cope with the fluctuations caused by uncertainty, thereby realizing the precise regulation of the energy of the multi-energy microgrid.

[0031] This application establishes a MEMG real-time energy control strategy based on data-driven scenario prediction in the RTP environment to mitigate the negative impact of multi-dimensional uncertainty on the operation of MEMG. This strategy extracts features from MEMG historical operation scenarios based on the Wasserstein Generative Adversarial Network (WGAN) to establish an unsupervised representation of the intrinsic statistical distribution of MEMG operation scenarios.

[0032] The trained WGAN is used to construct a scenario prediction constrained optimization problem to predict a set of high-quality MEMG operation scenarios that can accurately characterize the multi-dimensional uncertainty of the future period according to the known information (observation values ​​and point prediction values) at each moment. Based on the predicted scenarios, a two-stage MEMG real-time energy control model consisting of pre-scheduling and real-time power compensation is formed. In the pre-scheduling stage, the scenario prediction information is used to obtain the pre-scheduling instructions of the MEMG based on the stochastic MPC (SMPC), and in the real-time power compensation stage, the dispatchable units in the MEMG are effectively coordinated according to the real-time observation of the uncertain variables, thereby realizing the economic compensation of the unbalanced power of the electric and thermal power.

[0033] In addition, the scene prediction and MEMG energy regulation at each moment make effective use of uncertain variables and system state information updates with the help of the rolling solution mechanism of model predictive control (MPC).

[0034] In order to cope with the challenges brought by the multi-dimensional uncertainty caused by RDG and multi-energy loads to the energy regulation of grid-connected MEMG in RTP environment, the MEMG real-time energy regulation strategy based on data-driven scenario prediction proposed in this application mainly consists of two parts, namely, scenario prediction based on WGAN and real-time energy regulation based on SMPC. Its overall framework is as follows Figure 1 shown. Figure 1 (a) shows the scenario prediction framework based on WGAN, which first trains WGAN to use WGAN to extract features from MEMG historical operation scenarios, thereby adaptively establishing an unsupervised representation of the statistical distribution of MEMG operation scenarios composed of multi-dimensional uncertainties. Then, based on known information (observation values ​​and point prediction values) and combined with the trained WGAN, a scenario prediction constraint optimization problem is constructed. By solving this optimization problem, the input noise vector of the generator can be optimized. The trained generator uses the optimized vector as input to predict a set of high-quality operation scenarios that can accurately characterize the operation uncertainty of MEMG in the future at each moment without any statistical distribution assumptions.

[0035] according to Figure 1 As shown in (b), although MPC can utilize the feedback mechanism contained therein to better achieve real-time control of the system based on the updated system state and uncertain variable prediction information at each moment, the performance of traditional MPC based on point prediction information is heavily dependent on prediction accuracy and cannot effectively solve the performance degradation caused by the inevitable deviation between the point prediction value and the actual value of the uncertain variable.

[0036] As a flexible real-time control framework, MPC can be easily combined with other optimization methods to overcome the limitations of traditional MPC. Therefore, this application combines the MPC framework with stochastic optimization according to the prediction scenario to construct a MEMG real-time energy control model based on SMPC, which can characterize the operation uncertainty through a set of operation scenarios at each moment, thereby effectively covering the possible realization of uncertain variables. The schematic diagram of MEMG real-time energy control based on SMPC is also shown in Figure 1 (b). After the real-time energy control at the current moment is completed, the scenario prediction constraint optimization problem can be reconstructed according to the information update of the uncertain variables, and the possible MEMG operation scenario at the next moment can be re-predicted and used for the MEMG real-time energy control at the next moment. This rolling update solution mechanism helps to further reduce the negative impact of multi-dimensional uncertainty on the operation control of MEMG.

[0037] like Figure 1As shown in (a), in order to realize the prediction of MEMG operation scenarios, WGAN must first be trained based on historical operation scenarios, so that the generator can convert random vectors sampled from simple known distributions into any real scenario with a statistical distribution consistent with the historical operation scenarios of MEMG, thereby establishing an unsupervised representation of the statistical distribution of MEMG operation scenarios. It is worth noting that the above process can also be called scenario generation. For the sake of clarity, this process is uniformly referred to as scenario generation in the following text.

[0038] Figure 2 The configuration of MEMG is reiterated and the variables involved in its operation are indicated. Figure 2 As shown, the variables Ω involved in the operation of MEMG at the tth moment t It can be expressed as a formula.

[0039]

[0040] in, and They represent the electric power of photovoltaic (PV), wind turbine (WT), combined heat and power unit (CHP), electric boiler (EB) and electric load at the tth moment respectively; and They represent the discharge and charging power of the electric energy storage system (ESS) at the tth moment respectively; is the electric power transmitted between MEMG and the main grid at the tth moment; and denote the thermal power of CHP, gas boiler (GB), EB and thermal load at the tth moment respectively; and They represent the heat dissipation and charging power of the thermal storage system (TSS) at the tth moment respectively.

[0041] Ω t All variables except and All other variables are controllable variables, so these four variables are the source of MEMG operation uncertainty. Based on this, a complete operation scenario that can capture the uncertainty of MEMG operation can be defined as , which consists of multi-dimensional uncertain variables and each variable is a time series of length T.

[0042]

[0043] Where T is the time range of the MEMG running scenario.

[0044] During the operation, the MEMG operator has recorded a series of historical operation scenarios that can characterize the actual dynamic behavior of MEMG. The four uncertain variables that constitute the MEMG operation scenario are affected by different factors. For example, PV power generation is mainly affected by the intensity of solar radiation, while WT power generation is mainly affected by wind speed. Therefore, it can be considered that these four variables are independent of each other. In order to efficiently extract features from historical scenarios, this application migrates the channel concept (i.e., RGB channel) in the field of color images to the MEMG operation scenario. In this way, the MEMG operation scenario composed of multidimensional time series can be jointly represented as a whole with four channels and each channel is a time series of length T. Moreover, since each channel is relatively independent in the generative learning process, the mutual influence between different variables in the feature extraction process can be minimized.

[0045] In addition, this application also designs a corresponding WGAN based on the characteristics of MEMG operation scenarios to adaptively construct an unsupervised representation of the statistical distribution of historical operation scenarios without any explicit statistical assumptions. The network uses Wasserstein distance as the training target and adopts a one-dimensional convolutional layer suitable for extracting time series features from time series and generating time series in the network structure. The detailed principle of WGAN and the network structure adopted will be described in detail later.

[0046] according to Figure 1 As shown in (a), WGAN consists of two neural networks, called generator and discriminator. The generator is trained to convert random latent vectors sampled from a simple known distribution (such as Gaussian distribution or uniform distribution) into sufficiently realistic data (i.e., consistent with the statistical distribution of historical data), while the discriminator is trained to distinguish whether the data is true (samples derived from historical data are true data) or false (samples converted by the generator are false data). When the alternating competitive training between the generator and the discriminator reaches equilibrium, the discriminator will no longer be able to distinguish whether the data comes from historical data or generated data, and the WGAN is considered to have been trained.

[0047] The goal of MEMG scenario generation is to generate diverse scenarios that can correctly capture the statistical distribution of historical running scenarios. Wasserstein distance is an intuitive criterion for measuring the difference between two statistical distributions. Therefore, in order to perform scenario generation tasks in a reliable manner, the training process of WGAN can be prompted to establish an unsupervised representation of a distribution with the minimum Wasserstein distance from the statistical distribution of historical scenarios. Therefore, in the training of WGAN, this application uses Wasserstein distance as the value function of the game between the generator and the discriminator, so that when the game between them reaches equilibrium, a generator can be obtained so that the statistical distribution of the generated scenario is as close as possible to the statistical distribution of the historical scenario, thereby better completing the scenario generation task. The training goal of WGAN is shown in the formula:

[0048]

[0049] Where G and D represent the generator and discriminator respectively; x and P x Represent the historical operation scenarios of MEMG and the corresponding statistical distribution respectively; z is the input implicit vector of the generator and z~P z ;P z As a simple known distribution that is easy to sample, this application uses P z Set to a uniform distribution in [-1,1].

[0050] Based on the formula, the generator and the discriminator compete through a minimum-maximum game. The generator is trained to deceive the discriminator into treating the generated scene as a real scene, where the output of the generator is represented as G(z) and G(z)~P G , here P G Represents the statistical distribution of the generated scene. The discriminator and the generator are trained alternately to distinguish P x and P G , where the output of the discriminator is an indication of whether the scene is real or fake, and a larger output value indicates that the scene is more real from the perspective of the discriminator.

[0051] In summary, the generator loss function value L G The smaller the value, the more realistic the scene generated by the generator is, and the discriminator loss function value L D The smaller it is, the stronger the discriminator’s ability to distinguish between real scenes and generated scenes is. Based on this, the loss functions of the generator and discriminator can be defined as shown in equation (4) and equation (5) respectively:

[0052]

[0053]

[0054] Among them, E represents the empirical mean during the training process.

[0055] When the parameters of the discriminator are given, Equation (6) can be interpreted as P x and P G The Wasserstein distance between them is shown in formula (6).

[0056]

[0057] Among them, θ (D) Represents the parameters of the discriminator.

[0058] The alternating competitive training of the discriminator and the generator allows them to evolve continuously until the discriminator can no longer distinguish between the scenes generated by the generator and the real scenes, which also means that P x and P G The Wasserstein distance between them converges to the minimum value as the game between the generator and the discriminator reaches equilibrium. At this point, it can be considered that the generator has learned the true statistical distribution of the MEMG operation scenario, and its parameter θ (G) can be fixed and used to generate sufficiently realistic MEMG operation scenarios. In addition, the evolution curve of Wasserstein distance during WGAN training can also be used as a visualization criterion for its training convergence.

[0059] For MEMG running scenario prediction, the predicted scenario must not only have the same statistical characteristics as the real scenario, but also follow specific information. Assuming the current time is k-1, the information that can be obtained includes the observed values ​​of the uncertain variables at the past k-1 moments And point prediction information at future time k~T Therefore, the operation scenario prediction task of MEMG is to predict a set of operation scenarios at future time k~T based on the above given information to accurately capture the possible dynamics and changes of MEMG operation uncertainty.

[0060] The trained generator can generate any operation scenario that maintains the same mode as the operation of MEMG by converting the random implicit vector sampled from the uniform distribution of [-1,1]. By optimizing the input vector of the generator, a specific scenario can be generated according to specific information to accurately characterize the uncertainty of MEMG operation in a specific period in the future. To this end, this application constructs a scenario prediction optimization problem based on the trained WGAN and the observed values ​​and point prediction values ​​of uncertain variables to obtain the optimal input vector of the generator, thereby realizing MEMG operation scenario prediction. This method can flexibly perform scenario prediction without limiting the prediction range, and its execution architecture is as follows: Figure 1 The optimization problem shown in (a) and used for MEMG operation scenario prediction is constructed as follows:

[0061] The part of the scenario predicted at the k-1th moment in the past k-1 moments must be as close as possible to the observed value of the uncertain variable at the k-1th moment in the past, and the part at the future k~T moments must reflect the real operating characteristics of MEMG and cannot conflict with the point prediction information. Based on the above ideas, the MEMG operating scenario prediction optimization problem can be defined as follows.

[0062]

[0063] z∈Z (8)

[0064]

[0065] Among them, P obse (G(z)) and P pred (G(z)) represents the part of G(z) before k-1 moments and in the future k~T moments respectively; γ is a weight parameter to determine the influence of the discriminator output on the objective function; Z represents the domain of the vector z; and P pred Lower and upper bounds for (G(z)).

[0066] The meaning of the objective function is: Formula (7) is a cost function used to measure P obse (G(z)) and s obse The degree of closeness between them; Formula (9) is a penalty term to help ensure that the predicted scene is as realistic as possible. Since the input of the generator is a random implicit vector sampled from a uniform distribution of [-1,1], the domain Z of z is a hypercube bounded by [-1,1]. In addition, in order to prevent the predicted scene from conflicting with the point prediction information, the predicted scene should not fall outside the prediction interval defined by the point prediction. Based on this consideration, a constraint is constructed, where and It can be calculated according to formula (10) and formula (11) respectively.

[0067]

[0068]

[0069] Among them, α is a predefined point prediction confidence level, which plays a role in controlling the fluctuation range of the prediction scenario and α>1.

[0070] The optimization problem contains highly non-convex generators and discriminators, so there are a large number of local optimal solutions to this optimization problem. By iteratively solving the optimization problem with different initial values ​​z, a set of local optimal solutions can be obtained. Then, the optimal solution set is input into the generator to predict a set of different and realistic MEMG operation scenarios for the future k~T moments, as shown in the following formula.

[0071]

[0072] Among them, z * represents the set of local optimal solutions; represents the i-th local optimal solution; m is the number of times the optimization problem is iterated. By adjusting m, the number of prediction scenarios can be easily adjusted; S represents the prediction scenario set.

[0073] In order to solve the optimization problem efficiently, the inequality constraints can be incorporated into the objective function as shown in the following equation.

[0074]

[0075] Among them, κ and ν are weight parameters and κ=ν.

[0076] In addition, the pre-optimization problem shown in the above formula can be solved first to obtain a better initial value z for iteration, thereby improving the quality of the local optimal solution.

[0077]

[0078] z∈Z(15)

[0079]

[0080] Among them, s ini For The samples are randomly sampled from , and the α in the formula is slightly smaller than that set in the formula.

[0081] After predicting a set of operation scenarios that can characterize the uncertainty of MEMG operation in the future k~T moments at the k-1th moment, the pre-scheduling model of MEMG at the kth moment can be constructed under the SMPC framework, where the objective function is shown in the formula.

[0082]

[0083] Where E(·) represents the expected value of each cost under all predicted MEMG operation scenarios; and They represent the electricity purchase cost of MEMG from the grid, the natural gas purchase cost, and the recession costs of ESS and TSS in the pre-dispatch model at the kth moment.

[0084] Each term in the formula can be calculated according to the following formula. The following formula calculates the expected cost of electricity purchased by MEMG from the power grid. The MEMG real-time energy control model of this application is developed in the RTP environment. The cost of electricity purchased by MEMG from the power grid is directly deducted from the profit of electricity sales. The following formulas respectively calculate the expected natural gas procurement cost of MEMG and the expected decline cost of ESS and TSS.

[0085]

[0086]

[0087]

[0088]

[0089] Where s is the scene index; N s Indicates the total number of MEMG running scenarios; π s is the probability of occurrence of the sth scenario; and They represent the purchase price and sales price of electricity from the grid by MEMG at the tth moment respectively; and They represent the power purchased and sold by MEMG to the grid at the tth moment in the sth scenario; Δt represents the duration of a moment; λ gas is the price of natural gas; and Respectively represent the electric power and thermal power output of CHP and GB at the tth moment in the sth scenario; η CHP and η GB are the efficiency of natural gas conversion into electricity and heat for CHP and GB respectively; ρ ESS is the decline cost coefficient of ESS; and They represent the discharge and charging power of ESS at the tth moment in the sth scenario; ρ TSS is the recession cost coefficient of TSS; and They represent the heat release and heat charging power of TSS at the tth moment in the sth scenario, respectively. The objective function of the MEMG pre-scheduling model must satisfy the operation constraints and power balance constraints of all devices in the MEMG.

[0090] (1) Transmission power constraint: In order to ensure the safe and reliable operation of the main grid, the transmission power of the interconnection line between MEMG and the main grid must be limited to a certain range, and the purchase and sale of electricity from the MEMG to the main grid cannot occur at the same time, as shown in the following formula. In addition, the interaction power between MEMG and the main grid can be expressed as follows.

[0091]

[0092]

[0093]

[0094] in, Indicates the upper limit of the power transmitted between MEMG and the main power grid; is a binary variable used to avoid the simultaneous purchase and sale of electricity by MEMG at time t in the sth scenario; It represents the transmission power between MEMG and the main grid at the tth moment in the sth scenario.

[0095] (2) CHP operation constraints:

[0096]

[0097]

[0098] in, and They represent the lower and upper limits of CHP output power, respectively; b is the thermal-electric ratio of CHP; It represents the thermal power output of CHP at the tth moment in the sth scenario.

[0099] (3) ESS operation constraints:

[0100]

[0101]

[0102]

[0103]

[0104]

[0105] Among them, Q s,t represents the amount of electricity stored in ESS at the tth moment in the sth scenario; and They represent the lower and upper limits of ESS charging and discharging power respectively; is a binary variable used to avoid simultaneous charging and discharging of ESS at time t in the sth scenario; and They are the lower and upper limits of the amount of electricity that can be stored in the ESS; is the operating power of ESS at the tth moment in the sth scenario.

[0106] (4)TSS operation constraints:

[0107]

[0108]

[0109]

[0110]

[0111]

[0112] Among them, H s,t represents the heat stored in TSS at the tth moment in the sth scenario; and They represent the lower and upper limits of TSS charging and discharging thermal power respectively; is a binary variable used to avoid simultaneous charging and discharging of TSS at time t in the sth scenario; and are the lower and upper limits of the amount of heat that can be stored in TSS, respectively; is the operating power of TSS at the tth moment in the sth scenario.

[0113] (5)EB operation constraints:

[0114]

[0115]

[0116] in, and are the lower and upper limits of EB output thermal power respectively; and They represent the thermal power output and electrical power consumed by EB at the tth moment in the sth scenario respectively; η EB is the electrical-to-heat efficiency of EB.

[0117] (6) GB constraints:

[0118]

[0119] in, and They represent the lower and upper limits of GB output thermal power respectively.

[0120] (7) Power balance constraints:

[0121]

[0122]

[0123] in, and They represent the power of PV power generation, WT power generation, electric load and thermal load at the tth moment in the sth scenario respectively.

[0124] In summary, the objective function is constructed based on the scene prediction information from k to T obtained at the k-1th moment, and the MEMG pre-scheduling model based on SMPC is constrained by the above. By solving the model, the scheduling instructions of MEMG at the kth moment in all scenarios can be obtained. Based on the idea of ​​random optimization, the expected value of the scheduling instructions in all scenarios can be calculated to obtain the pre-scheduling instructions of MEMG at the kth moment.

[0125] The exchange power between MEMG and the main grid Taking as an example, the pre-scheduling instruction at the kth moment can be expressed as the following formula. The calculation method of pre-scheduling decisions for other controllable devices is similar to this and will not be repeated here.

[0126]

[0127] in, It represents the pre-scheduling instruction of power transmission between MEMG and the main grid at the kth moment.

[0128] Although scenario prediction can improve the performance of characterizing the uncertainty of MEMG operation, there is still an inevitable deviation between the predicted information of uncertain variables and their true values. When the time reaches the kth moment, the MEMG pre-scheduling decision obtained according to the scenario prediction will cause a mismatch between the energy supply and demand in the MEMG to a certain extent. For this reason, this application constructs a real-time power compensation model to compensate for the prediction error and ensure the real-time power balance of MEMG. The goal of MEMG real-time power compensation is to adjust the operation decision at the kth moment in the most economical way to compensate for the prediction error while satisfying all its operating constraints. Therefore, the objective function of the MEMG real-time power compensation model can be defined as the following formula.

[0129]

[0130] in, and They represent the electricity purchase cost, natural gas purchase cost and the decay cost of ESS and TSS caused by power compensation of MEMG at the kth moment, respectively, and they can be calculated according to the following formulas.

[0131]

[0132]

[0133]

[0134]

[0135] in, and They represent the power purchase and sales adjustment of MEMG from the power grid at the kth moment respectively; represents the output power adjustment of CHP at the kth moment; is the output thermal power adjustment of GB at the kth moment; and They represent the discharge and charging power adjustment of ESS at the kth moment respectively; and They respectively represent the heat release and heat charging power adjustment of TSS at the kth moment.

[0136] The real-time power compensation model makes decision adjustments to the operation of MEMG based on the pre-scheduling decision, so the real-time power compensation decision must first meet the operation constraints and power balance constraints of all devices in MEMG, and the decision adjustment amount cannot exceed the imbalance of the electric and thermal power at that moment. After obtaining the actual power of RDG and multi-energy load at the kth moment, the imbalance of the electric and thermal power in MEMG at that moment can be calculated according to the following formula.

[0137]

[0138]

[0139] Where ΔP k and Δh k They represent the unbalanced amount of electric heating power at the kth moment respectively; and They represent the pre-scheduling instructions of CHP output power, EB consumption power and ESS operation power at the kth moment respectively; and They represent the actual power of electric load, PV and WT generation at the kth moment respectively; and They represent the pre-scheduling instructions of CHP, GB and EB output thermal power and TSS operating power at the kth moment respectively; is the actual power of the heat load at the kth moment.

[0140] The constraints that the MEMG real-time power compensation model needs to meet are as follows:

[0141] (1) Transmission power adjustment constraint: The following formula restricts the adjustment amount of power purchased and sold by MEMG to the grid to not exceed the power imbalance, and at the same time restricts the power purchase and sales adjustments to not occur at the same time; and the adjusted transmission power between MEMG and the grid still cannot exceed the transmission power limit of the interconnection line.

[0142]

[0143]

[0144]

[0145] in, It is a binary variable used to prevent MEMG from adjusting the purchase and sale of electricity power at the same time at the kth moment.

[0146] (2) CHP adjustment constraints: The adjustment amount of CHP output power cannot exceed the power imbalance amount, and the equipment capacity constraints cannot be violated after adjustment.

[0147]

[0148]

[0149]

[0150] in, Represents the output thermal power adjustment of CHP at the kth moment.

[0151] (3) ESS adjustment constraints:

[0152]

[0153]

[0154]

[0155]

[0156]

[0157] Among them, Q k represents the amount of electricity stored in ESS at the kth moment; is a binary variable used to prevent the ESS from adjusting the charging and discharging power simultaneously at the kth moment.

[0158] (4)TSS adjustment constraints:

[0159]

[0160]

[0161]

[0162]

[0163]

[0164] Among them, H k represents the heat stored in TSS at the kth moment; is a binary variable used to prevent TSS from adjusting the charging and discharging powers simultaneously at the kth moment.

[0165] (5) EB adjustment constraints:

[0166]

[0167]

[0168]

[0169] in, and They represent the output thermal power and consumed electrical power adjustment of EB at the kth moment respectively.

[0170] (6) GB adjustment constraints:

[0171]

[0172]

[0173] (7) Power balance constraint: The operating electric power and thermal power adjustments of all controllable devices at the kth moment must match the corresponding electric power and thermal power imbalances, as shown in the following equations.

[0174]

[0175]

[0176] In summary, the objective function of the real-time power compensation model of MEMG at the kth moment is an optimization problem with the above constraints. By solving this model, the power adjustment decisions of all adjustable devices can be obtained, and then superimposed on the pre-dispatch instructions to obtain the final control decision for the actual operation of MEMG (taking the exchange power between MEMG and the power grid and the CHP output power as an example).

[0177]

[0178]

[0179] For the sake of clarity, Figure 3 The detailed execution process of MEMG real-time energy control is summarized.

[0180] This application evaluates and verifies the effectiveness of the data-driven MEMG operation scenario prediction method and real-time energy control strategy proposed in this application through numerical simulation. The WGAN-based MEMG operation scenario prediction method is developed using the Pytorch toolkit in the Python environment. The MEMG real-time energy control model is written and solved using the Gurobi solver in the Python environment.

[0181] (1) Data description and simulation setup

[0182] The time scale of RTP release is generally 30 minutes, so this application sets the time range of the MEMG operation scenario to 1 day and has a time resolution of 30 minutes (i.e., T=48). Therefore, the MEMG operation scenario is a tensor with 4 channels (i.e., PV power generation, WT power generation, electric load, and thermal load) and each channel is a time series of length 48 corresponding to an uncertain variable. Given the dimension of the MEMG operation scenario, a suitable WGAN structure is designed to perform feature learning from historical scenarios to achieve MEMG operation scenario generation and prediction. Figure 4 The network structure of the designed WGAN is presented in detail. Figure 4 In the figure, N represents the number of samples contained in each mini-batch during training; Conv1d-(a,b,c,d,e) represents a one-dimensional convolutional layer with hyperparameters a,b,c,d,e, where the hyperparameters are the number of input channels, the number of output channels, the kernel size, the stride, and the padding length, respectively; PReLU() represents the activation function PReLU; LeakyReLU(0.2) represents the LeakyReLU activation function with a negative slope of 0.2.

[0183] Since each channel is a time series, the network structure uses a one-dimensional convolutional layer suitable for extracting time series features from time series and generating time series. The test results show that the network structure used can be effectively trained to convergence, such as Figure 5 shown.

[0184] Algorithm 1 gives the detailed training process of WGAN. Both the generator and the discriminator are trained using RMSprop and the learning rate η G and η D Set to 10 respectively -4 and 5×10 -5 The mini-batch size N is set to 64. discriSet to 5, which means that the discriminator will be trained 5 times for every generator training in each training cycle. In order to make WGAN meet Lipschitz continuity and promote its stable training, the discriminator parameters are pruned during training and the pruned parameter φ is set to 0.02. All other training parameters are set to the default parameters of the Pytorch toolkit. The input implicit vector of the generator is sampled from a uniform distribution of [-1,1].

[0185]

[0186]

[0187] (2) Scene Generation Performance Evaluation

[0188] In order to achieve reliable MEMG operation scenario prediction, we must first train WGAN well, so as to construct an unsupervised representation of the statistical distribution of MEMG operation scenarios to generate sufficiently realistic operation scenarios that can reflect the operation characteristics of MEMG. To this end, we first evaluate the performance of MEMG operation scenario generation based on WGAN. The Wasserstein distance can provide an intuitive standard for measuring the difference between two statistical distributions. Therefore, in order to illustrate the effectiveness of using WGAN to perform scenario generation tasks, Figure 3-5 The evolution of the Wasserstein distance between the statistical distribution of historical and generated scenes during WGAN training is shown and compared with the traditional GAN ​​(Jensen-Shannon Divergence GAN, JSD-GAN) using Jensen-Shannon divergence. According to the requirements of JSD-GAN, a Sigmoid activation function is added to the last layer of the discriminator and an Adam optimizer is used. Apart from this, the network structure and training parameters of JSD-GAN are consistent with those of WGAN. Figure 5 As can be seen from the figure, as the alternating competitive training of the generator and the discriminator reaches equilibrium, the Wasserstein distance of WGAN gradually decreases and converges to a minimum value close to 0 after about 150 training cycles. At this point, the statistical distribution of the scenes generated by WGAN is almost the same as the statistical distribution of the historical scenes, thus obtaining a well-trained generator that can generate sufficiently realistic scenes. In contrast, the Wasserstein distance of JSD-GAN is always large and cannot converge, which means that the scenes generated by JSD-GAN are not realistic enough, and the statistical distribution of the generated scenes is quite different from that of the historical scenes.

[0189] After WGAN training, the generator was used to generate the same number of MEMG running scenarios as in the training set. Then, the scenario generation performance of the proposed method was further evaluated through a series of quantitative analyses and compared with two benchmark methods. The two benchmark methods are described in detail as follows:

[0190] Benchmark 1: The statistical model-based approach performs scenario generation tasks. Here, 30 scenarios are generated for each uncertain variable, and then these scenarios need to be combined to form a complete MEMG operation scenario, which means that a total of 30 scenarios are generated. 4 Afterwards, the scenario reduction method based on K-means was used to reduce the scenarios to 1040 (consistent with the number of historical running scenarios in the training set).

[0191] Benchmark 2: The WGAN-based single-variable scenario generation method performs the scenario generation task, and the number of generated scenarios and the processing method are consistent with those in Benchmark 1.

[0192] The following analysis and comparison are performed on the scenarios generated by the method proposed in this application and the two benchmark methods as well as the historical scenarios in terms of temporal correlation and statistical characteristics to verify the effectiveness of the proposed method.

[0193] Time series correlation analysis: The time series correlation of time series data is often analyzed using the autocorrelation coefficient. The autocorrelation coefficient R(τ) of a sample with a time lag τ can be calculated according to the formula.

[0194]

[0195] Among them, S t Represents the value of the generated scenario or historical scenario time series at the tth moment; μ and σ 2 represent the mean and variance of the time series respectively.

[0196] Figure 6 From a statistical point of view (i.e., the average value of the autocorrelation coefficients of all scenarios), the autocorrelation coefficient curves of the historical scenarios and the scenarios generated by the method proposed in this application, Benchmark 1 and Benchmark 2 are compared. As can be seen from the figure, the autocorrelation coefficients of all uncertain variables in the scenarios generated by the method proposed in this application have similar curves to those of the historical scenarios, which shows that the method proposed in this application can maintain good performance in capturing the timing characteristics of all uncertain variables. In contrast, Benchmarks 1 and 2 can only correctly capture the timing characteristics of some variables, but have poor performance in capturing the timing characteristics of other variables. The above results show that the method proposed in this application can effectively learn and reflect the inherent timing characteristics in the historical operation scenarios of MEMG.

[0197] Statistical Analysis: To verify the statistical similarity between the generated and historical scenarios, their statistical properties were analyzed by CDF. Figure 7 The CDF curves of the historical scenarios and the scenarios generated by the proposed method, Benchmark 1, and Benchmark 2 are shown. As can be seen from the figure, the scenarios generated by the proposed method have almost the same marginal distribution as the historical scenarios. However, the CDF curve of the scenario generated by the benchmark method deviates significantly from the CDF curve of the historical scenarios. This shows that the performance of the proposed method is better than the benchmark method and can well capture the intrinsic statistical characteristics of the MEMG historical operation scenarios.

[0198] Table 1 lists the comparison between the mean and standard deviation of the scenarios generated by the method proposed in this application and the historical scenarios to further demonstrate the statistical similarity between them. It can be seen that the corresponding means and standard deviations are very close, which once again proves that the method proposed in this application can generate sufficiently realistic MEMG operation scenarios.

[0199] Table 1 Mean and standard deviation of each variable in the MEMG operation scenario

[0200]

[0201] The above results analysis shows that the method proposed in this application can generate complete and diverse MEMG operation scenarios that can maintain the timing and statistical characteristics of MEMG historical operation scenarios. In the benchmark method, the operation scenario corresponding to each uncertain variable is generated separately, so they need to be combined to form a complete MEMG operation scenario. This method limits the number of scenarios generated for each variable and should not be too large, otherwise it will cause the number of complete MEMG operation scenarios to explode and cause a huge computational burden. However, if the number of scenarios generated for each variable is insufficient, the operation uncertainty of MEMG cannot be effectively captured. Therefore, the method proposed in this application not only outperforms the benchmark method but also effectively avoids the computational burden caused by reducing a large number of operation scenarios in the benchmark method. Therefore, it can be considered that the method proposed in this application can more efficiently generate MEMG operation scenarios composed of multiple uncertain variables, which also lays a good foundation for performing MEMG operation scenario prediction at each moment.

[0202] Analysis of MEMG real-time energy regulation results

[0203] (1) Simulation settings

[0204] Table 2 lists the specification parameters of all devices in the MEMG studied in the simulation experiment. The time range of energy regulation is 1 day and the time scale is consistent with the time scale of the MEMG operation scenario, which is 30 minutes (also consistent with the time scale of the RTP electricity price release adopted), so Δt = 0.5 hours and T = 48. RTP data comes from the power market of a certain region. The market operator will predict the electricity price at all future times based on the power demand data in the market at each moment and publish the predicted value. Therefore, this application assumes that RTP is known and the power purchase price data used in the simulation experiment is as follows Figure 8 As shown. In addition, the electricity purchase price of MEMG from the grid is generally higher than the electricity sales price, so the electricity sales price is set to 0.9 times the electricity purchase price. The natural gas purchase price is set to 0.055AUD / kWh, and the decline cost coefficients of ESS and TSS are set to 0.01AUD / kWh and 0.012AUD / kWh respectively. Fig. 9 The actual RDG and electric heating load power curves used in the experiment are shown.

[0205] Table 2 Specifications of all devices in MEMG

[0206]

[0207] (2) Results analysis and discussion

[0208] Fig.10 and Fig.11 The electrical power and thermal power operation schemes of MEMG before and after real-time power compensation are shown respectively. Fig.10 (a) and Fig.11 As can be seen in (a), due to the inevitable deviation between the predicted information and the actual value of the uncertain variables, directly using the pre-scheduling instructions cannot meet the power balance constraints of the MEMG, thus generating imbalances in both electrical power and thermal power, such as Fig.10 The 48th moment in (a) and Fig.11 (a) is the second moment. Fig.12 It can also be seen that the instructions obtained based on the SMPC pre-scheduling model will lead to the generation of electric and thermal imbalance power at every moment, which reflects the necessity of the real-time power compensation model proposed in this application in maintaining the safe and stable operation of MEMG. After performing real-time power compensation according to the actual value of the uncertain variable at each moment, the final operation plan of MEMG can be obtained, such as Fig.10 (b) and Fig.11(b) As shown in the figure. It can be observed that under the action of the real-time power compensation model, the operation decisions of the equipment in the MEMG and the interactive power between the MEMG and the power grid are optimally adjusted and the power balance constraints are satisfied. For example, at the 48th moment, due to the low electricity price at this time, the MEMG increased the purchase of electricity from the power grid to compensate for the power shortage. It can also be seen from the figure that in the period of low electricity prices, the energy demand of the MEMG is mainly met by purchasing electricity from the power grid. For example, at the 1st and 2nd moments, the MEMG purchased a large amount of electricity from the power grid, of which a part is stored in the ESS for discharge in the subsequent period of higher electricity prices, and the other part is produced through EB to meet the heat load demand. In the period of higher electricity prices, in addition to the internally configured RDG, the energy demand of the MEMG is mainly met more economically by using CHP to produce electricity and heat energy using natural gas. This reflects the good performance of the MEMG real-time energy control strategy proposed in this application in reducing operating costs.

[0209] Fig.12 The absolute values ​​of the electric and thermal unbalanced power at each moment caused by the pre-dispatch instructions obtained by the scenario-based SMPC and the traditional MPC based on point prediction are shown. It can be clearly observed from the figure that the electric and thermal unbalanced power generated by the MPC-based strategy is higher than that of the SMPC-based strategy in most periods. The cumulative unbalanced electricity of the SMPC and MPC methods are 972.19 kWh and 1220.35 kWh, respectively, and the unbalanced heat is 286.69 kWh and 722.19 kWh, respectively, which shows that the scenario predicted by the method proposed in this application is significantly better than point prediction in characterizing the uncertainty of MEMG operation.

[0210] If the actual power of RDG and multi-energy load is known at all times, ideal scheduling can be performed to obtain the ideal MEMG operation plan and operation cost. In addition, the MEMG operation plan that meets the operation constraints can be obtained by performing real-time power compensation on the pre-scheduling instructions obtained by MPC based on point prediction. Fig.13The comparison of the interactive power between MEMG and the power grid obtained by the SMPC method and the MPC method as well as the ideal situation is shown. It can be seen that the interactive power curve obtained by the SMPC-based strategy is closer to the ideal situation and the cumulative power differences between the SMPC strategy and the MPC strategy and the ideal situation are 821.62kWh and 1294.58kWh respectively. After calculation, the operating costs of MEMG under the ideal situation and the SMPC and MPC strategies are 495.92AUD, 503.35AUD and 517.16AUD respectively. Compared with the ideal situation, the SMPC-based strategy and the MPC-based strategy have increased the operating costs by 1.50% and 4.28% respectively. The operating costs generated by the strategy proposed in this application are closer to the ideal situation. The above results further confirm the effectiveness of the MEMG real-time energy control strategy proposed in this application, and the MEMG operation scenario prediction method developed in this application can predict high-quality operation scenarios in real time, thereby accurately characterizing the operation uncertainty of MEMG.

[0211] This application establishes a MEMG real-time energy control strategy based on data-driven scenario prediction in the RTP environment to reduce the negative impact of multi-dimensional uncertainty on the operation of MEMG. The strategy first designs a WGAN suitable for the characteristics of MEMG operation scenarios, which is used to extract features from MEMG historical operation scenarios, thereby adaptively establishing an unsupervised representation of the statistical characteristics of MEMG operation scenarios. Then, based on the trained WGAN and combined with known information, a scenario prediction optimization problem is established to optimize the input of the generator, thereby predicting a set of high-quality MEMG operation scenarios at each moment, and achieving an accurate characterization of the uncertainty of MEMG operation in the future period. Finally, based on the scenario prediction results, a MEMG real-time energy control strategy with a two-stage structure is established, which first obtains the MEMG pre-scheduling instructions based on SMPC, and then economically compensates for the unbalanced electric and thermal power through a real-time power compensation model. The effectiveness of the strategy proposed in this application has been fully verified through simulation experiments, and the main conclusions are summarized as follows: (1) The proposed scenario generation method can generate MEMG operation scenarios that follow the edge distribution and temporal correlation of historical scenarios and outperforms the benchmark method, which lays a good foundation for the construction of MEMG operation scenario prediction methods; (2) The developed scenario prediction method can effectively predict high-quality operation scenarios within different prediction time ranges and can accurately characterize the uncertainty of MEMG operation in the future period; (3) The MEMG real-time energy control strategy constructed based on the prediction scenario can effectively alleviate the negative impact of multi-dimensional uncertainty on the operation of MEMG, and the resulting MEMG operation cost is significantly reduced compared with the traditional MPC method.

[0212] The above are merely embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structural transformations made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A multi-energy microgrid energy control method, characterized in that: Includes steps: The trained Wasserstein generative adversarial network is used to construct a scenario prediction constrained optimization problem. At each moment, the known information is followed to predict a set of multi-dimensional uncertain multi-energy microgrid operation scenarios for a period of time in the future as prediction scenarios. Based on the prediction scenario, a two-stage multi-energy microgrid real-time energy control model is formed, which is composed of a pre-scheduling model and a real-time power compensation model; The pre-scheduling model uses the predicted scenario information to obtain the pre-scheduling instructions of the multi-energy microgrid based on random model predictive control; the real-time power compensation model coordinates the dispatchable units in the multi-energy microgrid according to the real-time observation of uncertain variables, thereby regulating the energy in the multi-energy microgrid.

2. The multi-energy microgrid energy control method according to claim 1, characterized in that: The Wasserstein generative adversarial network includes a generator and a discriminator; the generator is used to convert a random latent vector sampled from a simple known distribution into realistic data, and the discriminator is used to distinguish whether the data is true or false; when the alternating competitive training between the generator and the discriminator reaches equilibrium, the discriminator will no longer be able to distinguish whether the data comes from historical data or generated data, indicating that the Wasserstein generative adversarial network has been trained.

3. The multi-energy microgrid energy control method according to claim 2, characterized in that: The training objective of the Wasserstein generative adversarial network is shown in formula (3): Where G and D represent the generator and discriminator respectively; x and P x Represent the historical operation scenarios of the multi-energy microgrid and the corresponding statistical distribution respectively; z is the input implicit vector of the generator and z~P z ;P z is a known distribution; The loss functions of the generator and discriminator can be defined as shown in equation (4) and equation (5) respectively: Among them, E represents the empirical mean during the training process.

4. The multi-energy microgrid energy control method according to claim 3, characterized in that: The Wasserstein generative adversarial network operation scenario prediction optimization problem can be defined as shown in the following formula: z∈Z(8) Among them, P obse (G(z)) and P pred (G(z)) represents the output G(z) of the generator at k-1 moments before and k to T moments in the future; γ is a weight parameter to determine the influence of the discriminator output on the objective function; Z represents the domain of the vector z; and P pred Lower and upper bounds for (G(z)).

5. The multi-energy microgrid energy control method according to claim 4, characterized in that: The objective function of the pre-scheduling model is as follows: Wherein, E(·) represents the expected value of each cost under all predicted operation scenarios of the multi-energy microgrid; and They respectively represent the electricity purchase cost from the power grid, the natural gas purchase cost, and the decay cost of the electric energy storage system and the thermal energy storage system of the multi-energy microgrid in the pre-dispatch model at the kth moment.

6. The multi-energy microgrid energy control method according to claim 5, characterized in that: The objective function of the pre-dispatching model satisfies the operation constraints and power balance constraints of all equipment in the multi-energy microgrid, including the transmission power constraints of the interconnection line between the multi-energy microgrid and the main grid, the operation constraints of the cogeneration unit, the operation constraints of the electric energy storage system, the operation constraints of the thermal energy storage system, the operation constraints of the electric boiler, the gas boiler constraints and the power balance constraints.

7. The multi-energy microgrid energy control method according to claim 6, characterized in that: The objective function of the real-time power compensation model can be defined as follows: in, and They represent the electricity purchase cost, natural gas purchase cost, and the decay cost of the electric energy storage system and the thermal energy storage system generated by the multi-energy microgrid at the kth moment due to power compensation; The real-time power compensation model makes decision adjustments to the operation of the multi-energy microgrid based on the decision of the pre-scheduling model, so the decision of real-time power compensation satisfies the operation constraints and power balance constraints of all equipment in the multi-energy microgrid, and the decision adjustment amount cannot exceed the imbalance of the electric and thermal power at that moment; after obtaining the actual power of renewable distributed generation and multi-energy load at the kth moment, the imbalance of the electric and thermal power in the multi-energy microgrid at that moment can be calculated respectively according to the following formula: Where ΔP k and Δh k They represent the unbalanced amount of electric heating power at the kth moment respectively; and They represent the pre-dispatching instructions of the output power of the cogeneration unit, the power consumed by the electric boiler and the operating power of the electric energy storage system at the kth moment respectively; and They represent the actual power of electric load, photovoltaic power generation and wind turbine power generation at the kth moment respectively; and They represent the pre-dispatching instructions of the combined heat and power unit, the output thermal power of the gas boiler and the electric boiler, and the operating power of the thermal energy storage system at the kth moment respectively; is the actual power of the heat load at the kth moment.

8. The multi-energy microgrid energy control method according to claim 7, characterized in that: The constraints that the real-time power compensation model needs to meet include: transmission power adjustment constraints, cogeneration unit adjustment constraints, electric energy storage system adjustment constraints, thermal energy storage system adjustment constraints, electric boiler adjustment constraints, gas boiler adjustment constraints and power balance constraints; the objective function of the real-time power compensation model is constructed and an optimization problem with the above constraints; the power adjustment decisions of all adjustable devices are obtained by solving them, and then superimposed on the pre-dispatching instructions to obtain the final control decision for the actual operation of the multi-energy microgrid.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images