Dynamic modeling method for integrated energy system equipment based on mechanism and residual correction

Abstract

The invention discloses an integrated energy system equipment dynamic modeling method based on mechanism and residual correction, which comprises the following steps of: obtaining a physical structure and equipment of an integrated energy system, and respectively constructing a first principle model of each equipment; building a scheduling-oriented closed-loop control system model of the equipment based on the first principle model of each equipment, and building a scheduling-oriented linear transfer function model of the equipment through a system identification method; generating equipment variable working condition scheduling instruction data by adopting a pseudo-random multi-level signal, taking the data as input of a first principle model and a transfer function model of the equipment, and sampling to obtain an input and output parameter data set of each piece of equipment; constructing and training based on the operation parameter data set to obtain an MLP residual correction model; and connecting the transfer function model and the MLP residual correction model in series to construct a hybrid drive model framework. The modeling precision and the calculation efficiency of the integrated energy system equipment are improved, and a model basis is provided for economic dispatching of the integrated energy system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of integrated energy system modeling, and particularly relates to a dynamic modeling method for integrated energy system equipment based on mechanism and residual correction. BACKGROUND

[0002] Integrated energy system (IES) has the advantages of renewable energy consumption, high energy utilization rate, and multi-energy complementation, and has become a research hotspot in the reform and transformation of the energy field. However, the dynamic responses of the energy forms such as cold, heat and electricity in the system are significantly different, the operating efficiency of the equipment has strong nonlinearity, and the economic dispatching depends on accurate and lightweight dynamic variable operating condition models of the equipment.

[0003] At present, the modeling methods for integrated energy system equipment mainly include mechanism method, data-driven method and hybrid modeling method. The mechanism method mainly establishes a dynamic variable operating condition model based on physical principle equations of each device, which has clear physical meaning and high accuracy. However, this method has problems such as difficulty in high-precision modeling, long model solving time, and is difficult to apply to optimization scheduling. The data-driven method only relies on operation data to obtain the fitting operation characteristics of the device, and the modeling method is simple, but lacks clear physical meaning. The existing hybrid modeling method mainly focuses on non-linear efficiency correction and ignores the dynamic characteristics of the equipment. SUMMARY

[0004] The purpose of the present application is to provide a dynamic modeling method for integrated energy system equipment based on mechanism and residual correction, which uses a mechanism-driven module to characterize the dynamic characteristics of the integrated energy system equipment, and uses a data-driven module to compensate for the non-linear errors of the simulation and simulation of the mechanism-driven module, solving the problem that the existing modeling method for integrated energy system equipment cannot balance accuracy and computational efficiency.

[0005] The technical scheme of the present application is as follows:

[0006] (1) Obtain the physical structure and equipment of the integrated energy system, and construct a first principle model for each device as a device operation benchmark model;

[0007] (2) Construct a closed-loop control system model for each device facing scheduling based on the first principle model of each device, and obtain a linear transfer function model through system identification, which is used to characterize the dynamic characteristics of the device;

[0008] (3) Construct a data-driven model training required operating parameter data set based on the first principle model and the linear transfer function model;

[0009] (4) Constructing a multi-layer perception model based on the operating parameter dataset to correct the nonlinear error between the transfer function model and the first principle model;

[0010] (5) Connecting the transfer function model and the multi-layer perception model in series to form a hybrid driving model, and modeling and simulating each device.

[0011] Further, in step (1), the first principle model of each device is constructed as follows: using a modular modeling method, the device is divided into several typical modules according to the working principle and link of the device, and the first principle model of each device is established based on the physical laws of mass conservation, energy conservation and momentum conservation, which is used as the device operation reference model.

[0012] Further, in step (2), the linear transfer function model of the device oriented to scheduling is established as follows: a closed-loop control system model of the device oriented to scheduling is constructed according to the first principle model, and a linear transfer function model of the device oriented to scheduling is obtained by system identification; wherein, the linear transfer function model formula is as follows:

[0013] ;

[0014] Wherein, G(s) is the transfer function, Y(s) is the Laplace transform of the system output signal, U(s) is the Laplace transform of the system input signal, s is the complex frequency domain variable, a0-an are the denominator polynomial coefficients determined by the system physical structure and properties, b0-bn are the numerator polynomial coefficients, n and m are the system order and the numerator polynomial order respectively, and n≥m represents the causality, i.e. the output cannot precede the input.

[0015] Further, in step (3), the training dataset of the data-driven model is obtained as follows: pseudo-random multi-level signals are generated to generate device scheduling instruction data, which is applied to the first principle model and the transfer function model of each device as input, and the corresponding output data is obtained, and finally the input parameter and output parameter parameter set is obtained by sampling.

[0016] Further, in step (4), the multi-layer perception model constructed according to the dataset includes: an input layer, one or more hidden layers and an output layer, wherein the hidden layer uses a nonlinear activation function for processing, and the output layer generates an output value for correcting the nonlinear error between the transfer function model and the first principle model.

[0017] Further, in step (5), the constructed hybrid driving model includes a mechanism driving module and a data driving module, wherein the data driving module takes the output of the mechanism driving module as one of the inputs, adopts an MLP model to fit the residual error of the transfer function model and the first principle dynamic model; the output of the data driving module is taken as a residual correction term, and is added to the output of the mechanism driving module to obtain the output of the hybrid driving model.

[0018] The dynamic modeling system of the comprehensive energy system equipment based on mechanism and residual correction, comprises the following:

[0019] The first principle module is used for obtaining the physical structure and equipment of the comprehensive energy system, and constructing a first principle model of each equipment as a device operation benchmark model.

[0020] The closed-loop control system module is used for constructing a closed-loop control system model of each equipment facing scheduling based on the first principle model of each equipment, and obtaining a linear transfer function model through system identification, which is used for describing the dynamic characteristics of the equipment.

[0021] The data set module is used for constructing a running parameter data set required for training the data driving model based on the first principle model and the linear transfer function model.

[0022] The multi-layer perception machine module is used for constructing a multi-layer perception machine model based on the running parameter data set, and realizing the correction of the nonlinear error between the transfer function model and the first principle model.

[0023] The hybrid driving module is used for connecting the transfer function model and the multi-layer perception machine model in series to form a hybrid driving model, and modeling and simulating analysis are performed for each equipment.

[0024] The computer readable storage medium stores a computer program, and when the program is executed by a processor, the method of any one of the above embodiments is realized.

[0025] The electronic device comprises a memory and a processor, and the memory stores a computer program, and when the program is executed by the processor, the method of any one of the above embodiments is realized.

[0026] Beneficial effects: Compared with the prior art, the present application has the following remarkable advantages: the present application proposes a dynamic variable working condition modeling method of mechanism and data fusion, which on the one hand solves the problems of large modeling difficulty and long solving time of pure mechanism modeling method, and on the other hand makes up for the lack of poor interpretability of pure data-driven modeling method. For the system with large time scale difference between devices such as comprehensive energy system, the proposed hybrid modeling method can quickly and accurately perform model simulation and solving, and can significantly reduce the nonlinear deviation between the system output and the reference when performing large-scale variable working condition operation. The method of the present application has good generalization and portability, can adapt to modeling of various comprehensive energy system devices, and shortens the system modeling complexity. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a flowchart of the present application;

[0028] Figure 2 is a heat pump heating cycle structure diagram of the present application;

[0029] Figure 3 is a first principle model structure diagram of the heat pump of the present application;

[0030] Figure 4 is a closed loop control structure diagram of the heat pump of the present application;

[0031] Figure 5 is a structure diagram of the data-driven module (MLP model) in the hybrid driving framework of the present application;

[0032] Figure 6 is a heat pump simulation comparison diagram of the hybrid driving model based on mechanism and residual correction, the first principle model, the transfer function model, and the nonlinear autoregressive neural network model (NARX) based on out-of-band source input of the present application, wherein (a) in the figure is a comparison curve diagram of the output heat power of the heat pump of each model, and (b) in the figure is a comparison curve diagram of the consumed electric power of the heat pump of each model. DETAILED DESCRIPTION

[0033] The technical solutions of the present application will be further described below in combination with the drawings.

[0034] As shown in Figure 1 , the embodiment of the present application provides a dynamic modeling method of comprehensive energy system equipment based on mechanism and residual correction. The comprehensive energy system equipment usually operates in variable working condition according to the dispatching instruction, so considering the dynamic characteristics of the equipment, constructing the variable working condition dynamic characteristic model of the equipment is helpful for economic and accurate dispatching. The method includes the following steps:

[0035] S1: Obtain the physical structure and equipment of the comprehensive energy system, and construct the first principle model of each device as the device operation reference model;

[0036] S2: Construct a closed-loop control system model for each device based on the first-principles model of each device, and obtain a linear transfer function model through system identification to characterize the dynamic characteristics of the device.

[0037] S3: Construct the dataset of runtime parameters required for training the data-driven model based on the first-principles model and the linear transfer function model;

[0038] S4: Construct a multilayer perceptron (MLP) model based on the running parameter dataset to correct the nonlinear error between the transfer function model and the first-principles model;

[0039] S5: Connect the transfer function model and the MLP model in series to form a hybrid driving modeling framework, and perform modeling and simulation analysis for each device.

[0040] The method for constructing dynamic variable operating condition models of integrated energy system equipment provided by this invention constructs first-principles models of each piece of equipment in the integrated energy system as the equipment operation baseline model; then, based on the equipment first-principles models, a closed-loop control system model for scheduling is constructed, and the equipment transfer function model for scheduling is obtained through system identification; in addition, pseudo-random multi-level signals are used to generate equipment variable operating condition operation scheduling command data, which is input into the first-principles models and transfer function models to obtain the equipment operating parameter dataset, and an MLP residual correction model is constructed and trained based on the operating parameter dataset; finally, the transfer function model and the MLP residual correction model are cascaded to construct a hybrid drive modeling framework for the equipment, and a high-precision and lightweight dynamic variable operating condition model is constructed for each piece of equipment, providing a model foundation for the economic scheduling of the system.

[0041] In S1, the physical structure and equipment of the integrated energy system are obtained, and the first-principles models of each device are constructed, including:

[0042] Different integrated energy systems have different physical structures and equipment, and their internal composition is relatively complex, involving different physical and chemical reaction processes. Therefore, after determining the system structure, a modular modeling method is adopted. The equipment is divided into several typical modules according to the working principle and process. First-principles models of each device are established based on physical laws such as conservation of mass, energy, and momentum, and these models serve as the benchmark models for equipment operation.

[0043] The linear transfer function model expression for device scheduling in S2 is as follows:

[0044] ;

[0045] in, For the dispatching instructions of integrated energy system equipment, to synthesize the actual output of the energy system equipment.

[0046] The training data set of the data-driven model in S3 is obtained in the following way:

[0047] In order to simulate the daily operation data of each device in the integrated energy system, a ten-level pseudo-random multi-level signal (PRMS) with random repetition duration is used to generate device output power set value data, which is applied as input to the first principle model and transfer function model of each device to obtain the corresponding output data. Finally, the input parameters and output parameters are sampled to obtain the parameter set.

[0048] The MLP model expression constructed according to the data set in S4 is:

[0049] ;

[0050] wherein, , , are the weight matrices of the input layer, the k hidden layers, and the output layer neurons; , , are the bias vectors of the input layer, the k hidden layers, and the output layer neurons; is a nonlinear activation function (such as ReLU, Sigmoid, or tanh); is the last hidden layer neuron set vector; is the training input feature data set; is the kth hidden layer neuron set vector; is the output feature.

[0051] The hybrid-driven model framework based on the transfer function model and the MLP model in S5 is: the hybrid-driven integrated energy system equipment model framework is composed of a mechanism-driven module and a data-driven module in series. The data-driven module takes the output of the mechanism-driven module as one of the inputs, and uses the MLP model to fit the residual error of the transfer function model and the first principle dynamic model. The output of the data-driven module is superimposed with the output result of the mechanism-driven module as a correction term, which is the final output of the entire hybrid-driven model. The output expression of the hybrid model is:

[0052] ;

[0053] wherein, , and are the final output vector of the hybrid-driven model, the output vector of the mechanism-driven module, and the residual error vector obtained by the data-driven module, respectively; , input vectors of the mechanism-driven module and the data-driven module, respectively; a parameter set of the mechanism-driven module, obtained by data identification; a parameter set of the data-driven module, trained by a residual data set of the transfer function model output and the first-principle dynamic model output.

[0054] Based on the above scheme, in order to verify the effectiveness and superiority of the method of the present application, the following specific embodiments are used to further verify the above-mentioned comprehensive energy system equipment hybrid driving dynamic modeling method, as follows:

[0055] S1 constructs a first-principle model of the equipment as a benchmark model for the operation of the equipment: in this embodiment, a water source heat pump is taken as the research object, and its physical structure and heating principle are as shown in Figure 2 The main equipment includes a compressor, an evaporator, a condenser, an expansion valve and the like, and the working principle of heating is that the low-temperature and low-pressure refrigerant enters the evaporator, absorbs heat from the low-temperature heat source and evaporates into a gaseous state; then, after being compressed by the compressor into a high-temperature and high-pressure gaseous state, the refrigerant enters the condenser to condense and release heat to raise the temperature of the feed water; after passing through the throttle valve, the refrigerant becomes a low-temperature and low-pressure liquid again and enters the evaporator to start a new working cycle. Specifically, the first-principle model of the water source heat pump is constructed by using the lumped parameter method as shown in Figure 3 The specific expressions are as follows:

[0056] The relationship between the refrigerant flow rate and the compressor speed is as follows:

[0057] ;

[0058] wherein, is the refrigerant flow rate; is the compressor speed, rpm; is the compressor theoretical gas delivery, m 3 ; is the volume coefficient; is the compressor suction specific volume, m 3 / kg.

[0059] The compressor input power is as follows:

[0060] ;

[0061] wherein, represents the polytropic index of the compression process; is the total efficiency of the compressor, which is the ratio of the indicated power and the electric power consumption; , respectively represent the discharge pressure and the suction pressure, Pa, and the ratio of the two is the compression ratio.

[0062] The evaporator adopts a forced heat exchanger, and the heat transfer process is driven by a water pump circulation. The heat exchange equations of the water supply side and the refrigerant side are respectively:

[0063]

[0064] wherein, m w represents the mass of water in the evaporator, kg; m w represents the flow rate of water in the evaporator, kg / s; , T w in and T w out represent the inlet and outlet temperatures of hot water in the evaporator, respectively, ℃; , h in and h out represent the specific enthalpy of the refrigerant at the inlet and outlet of the evaporator, respectively, J / kg.

[0065] The heat exchange amount on the evaporator side is:

[0066]

[0067] wherein, h e represents the heat transfer coefficient on the evaporator side, W / (m 2 ·K); A e represents the heat exchange area on the evaporator side, m 2 ; T e represents the average temperature of the refrigerant on the evaporator side, ℃.

[0068] The modeling process of the condenser is similar to that of the evaporator.

[0069] The refrigerant flow process in the expansion valve can be regarded as an isenthalpic process, and the refrigerant flow rate and the inlet and outlet enthalpy values satisfy the following formula.

[0070]

[0071] wherein, K represents the flow coefficient determined by the structural parameters of the expansion valve; A min represents the minimum flow area controlled by the valve, m 2 ; ρ represents the density of two-phase fluid, kg / m 3 .

[0072] S2 constructs a linear transfer function model of the heat pump facing the scheduling: first, based on the first principle model constructed by S1, the model related parameters are shown in Table 1, and the closed-loop control system of the heat pump is constructed to study the dynamic relationship between the heat power set value and the actual output heat power. The power closed-loop control structure of the heat pump is shown in Figure 4 The closed-loop control system of the heat pump adopts the PID control method, which changes the mass flow rate of the refrigerant by adjusting the speed of the compressor, and then changes the heat exchange amount in the condenser, so that the output heat power of the heat pump tracks the change of the power set value.

[0073] ​​​Table 1. Parameters related to the first principle model of water source heat pump

[0074] ;

[0075] In this embodiment, the power set value step response experiment of the closed-loop control system of the heat pump is carried out, and by selecting the step response data, the closed-loop transfer function model of the heat pump heat power set value to the actual heat power of the heat pump is obtained by means of system identification method:

[0076]

[0077] wherein, is the actual heat power provided by the heat pump, kW; is the heat power set value of the heat pump, kW.

[0078] The electric power consumed by the heat pump is:

[0079]

[0080] wherein, is the electric power consumed by the heat pump, kW; is the energy efficiency ratio of the heat pump, which changes with the heat power set value of the heat pump.

[0081] S3: Based on the first principle model of S1 and the transfer function model of S2, the running parameter data set required for training the data-driven model is constructed: In this embodiment, in order to simulate various daily operation data of the heat pump, ten-level PRMS power set value data with random repetition duration are generated, which are applied to the closed-loop control system model and the transfer function model of the heat pump as input, and the corresponding output data are obtained, and the parameter set of input parameters and output parameters is obtained by sampling.

[0082] All the processed data sets are normalized to reduce the negative impact of the order of magnitude difference on the prediction accuracy of the MLP. Finally, the data set is divided into three parts of training data, validation data and test data, and the proportion is 70%, 20% and 10% respectively. The training and validation data will be used for training and validation of the MLP, and the test data will be used to evaluate the performance of the trained MLP.

[0083] S4: Train and construct the MLP model based on the data set of S3: In this embodiment, the MLP model is as follows: Figure 5As shown, it is an MLP model containing a 64x32 double-hidden fully connected layer, and the activation function is ReLU. The training learning rate of the model is set to 0.001, the optimization batch size is 64, and the loss function is mean square error (MSE). After training, the test loss function value of the heat pump output heat power residual correction model remains convergent to 0.003, and the training loss function value converges to 0.024; the test loss function value of the heat pump consumed electric power residual correction model remains convergent to 0.0021, and the training loss converges to 0.0013.

[0084] S5 uses a hybrid driving model framework to model and simulate the device: in this embodiment, the hybrid driving model based on mechanism and residual correction proposed in the application is compared with the first principle model, the transfer function model, and the nonlinear autoregressive neural network model (NARX) based on out-of-band source input in terms of effect. Figure 6 The simulation calculation results of the output heat power and the consumed electric power of the heat pump under different modeling methods are compared in the figure, wherein (a) in the figure shows the comparison results of the output heat power of the heat pump under different modeling methods, and (b) in the figure shows the comparison results of the consumed electric power of the heat pump under different modeling methods.

[0085] As shown in Figure 6 The NARX model and the hybrid driving model are closer to the first principle model than the transfer function model, but the hybrid driving model greatly shortens the calculation time compared with the NARX model, and successfully realizes a lighter and more accurate process simulation.

[0086] In order to compare the simulation efficiency and simulation accuracy of the four models, the water source heat pump is simulated for 24 hours based on the four models. The calculation time of each model, the root mean square error (RMSE) of the transfer function model, the NARX model and the hybrid driving model relative to the first principle model are shown in Table 2. From the evaluation results, it can be seen that the hybrid driving model has the shortest calculation time and the highest simulation accuracy, and the calculation time is reduced by 85.72% and 95.56% compared with the first principle model and the NARX model, respectively. The RMSE of the output heat power and the consumed electric power of the hybrid driving model compared with the transfer function model is reduced by 50.76% and 89.30%, respectively, and the simulation accuracy is equivalent to that of the NARX model. Therefore, the hybrid driving model can more accurately simulate the operation process of the water source heat pump.

[0087] Table 2 Performance evaluation results of each model

[0088] .

Claims

1. A dynamic modeling method for integrated energy system equipment based on mechanism and residual correction, characterized in that, Includes the following steps: (1) Obtain the physical structure and equipment of the integrated energy system, and construct a first-principles model for each device as the benchmark model for device operation; (2) Construct a closed-loop control system model for each device based on the first-principles model of each device, and obtain a linear transfer function model through system identification to characterize the dynamic characteristics of the device; (3) Construct the dataset of running parameters required for training the data-driven model based on the first-principles model and the linear transfer function model; (4) Construct a multilayer perceptron model based on the running parameter dataset to correct the nonlinear error between the transfer function model and the first-principles model; (5) Connect the transfer function model and the multilayer perceptron model in series to form a hybrid driving model, and perform modeling and simulation analysis for each device.

2. The method for dynamic modeling of integrated energy system equipment based on mechanism and residual correction according to claim 1, characterized in that, In step (1), the construction of the first principle model of each device is specifically as follows: adopting a modular modeling method, the device is divided into several typical modules according to the working principle and process of the device, and the first principle model of each device is established based on the physical laws of mass conservation, energy conservation and momentum conservation, and used as the benchmark model for device operation.

3. The method for dynamic modeling of integrated energy system equipment based on mechanism and residual correction according to claim 1, characterized in that, In step (2), the linear transfer function model for equipment scheduling is established as follows: A closed-loop control system model for equipment scheduling is constructed based on the first-principles model, and the linear transfer function model for equipment scheduling is identified using system identification methods; the formula for the linear transfer function model is as follows: Where G(s) is the transfer function, Y(s) is the Laplace transform of the system output signal, U(s) is the Laplace transform of the system input signal, s is a complex frequency domain variable, a0-an are the coefficients of the denominator polynomial, which are determined by the physical structure and properties of the system, b0-bn are the coefficients of the numerator polynomial, n and m are the order of the system and the order of the numerator polynomial, respectively, and n≥m indicates causality, that is, the output cannot lead the input.

4. The method for dynamic modeling of integrated energy system equipment based on mechanism and residual correction according to claim 1, characterized in that, In step (3), the training dataset for the data-driven model is obtained as follows: pseudo-random multi-level signals are used to generate device scheduling instruction data, which are then applied as input to the first-principles model and transfer function model of each device to obtain the corresponding output data. Finally, the parameter set of input parameters and output parameters is obtained through sampling.

5. The method for dynamic modeling of integrated energy system equipment based on mechanism and residual correction according to claim 1, characterized in that, In step (4), the multilayer perceptron model constructed based on the dataset includes: an input layer, one or more hidden layers, and an output layer, wherein the hidden layers are processed using a nonlinear activation function, and the output layer generates output values ​​to correct the nonlinear error between the transfer function model and the first-principles model.

6. The method for dynamic modeling of integrated energy system equipment based on mechanism and residual correction according to claim 1, characterized in that, In step (5), the constructed hybrid driving model includes a mechanism-driven module and a data-driven module. The data-driven module uses the output of the mechanism-driven module as one of the inputs and uses an MLP model to fit the residuals of the transfer function model and the first-principles dynamic model. The output of the data-driven module is used as a residual correction term and added to the output of the mechanism-driven module to obtain the output of the hybrid driving model.

7. A dynamic modeling system for integrated energy system equipment based on mechanism and residual correction, characterized in that, Including the following: First Principles Module: Used to obtain the physical structure and equipment of the integrated energy system, and to build a first principle model for each device as a benchmark model for device operation; Closed-loop control system module: used to build a scheduling-oriented closed-loop control system model for each device based on the first-principles model of each device, and obtain a linear transfer function model through system identification to characterize the dynamic characteristics of the device; Dataset module: Used to build the dataset of runtime parameters required for training data-driven models based on first-principles models and linear transfer function models; Multilayer perceptron module: Used to build a multilayer perceptron model based on the running parameter dataset, and to correct the nonlinear error between the transfer function model and the first-principles model; Hybrid drive module: Used to connect the transfer function model and the multilayer perceptron model in series to form a hybrid drive model, which is used for modeling, simulation and analysis of each device.

8. A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any one of claims 1-6.

9. An electronic device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, implements the method as described in any one of claims 1-6.

Images