Comprehensive energy system state evaluation method based on multi-source data fusion

Abstract

The invention discloses an integrated energy system state evaluation method based on multi-source data fusion, and relates to the technical field of integrated energy systems. The objective of the invention is to solve the problems of low data redundancy, low observability and large measurement error in state evaluation of the integrated energy system. The method comprises the following specific steps: establishing an electricity-gas coupling measurement model, and mapping a relationship between a measurement vector and a state vector; based on probability distribution of Bayesian learning fitting measurement data, respectively establishing a gas network Gaussian inference model and a power grid Gaussian mixture distribution model; using Monte Carlo sampling to generate a data sample, and screening reasonable data through power flow verification; establishing a data-driven state estimation model based on LSTM, and training and evaluating the performance of the model; through multi-source data fusion and the LSTM network, the method effectively improves the precision of state estimation of the integrated energy system, solves the problems of low measurement data redundancy and unobservable structure, is suitable for state estimation of park-level and building-level integrated energy systems, and has a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated energy systems, and in particular to an integrated energy system state estimation method based on a dynamic weighted algorithm. Background Art

[0002] With the continuous growth of global energy demand and the increasing awareness of environmental protection, problems such as energy resource shortage, environmental degradation and high carbon emissions have become increasingly prominent. In order to solve these problems, the method of interconnecting and complementing multiple energy sources has gradually been applied to various fields such as industry, commerce, and residents. As a system that integrates multiple energy sources such as electricity, heat, and natural gas, the integrated energy system can achieve overall management and coordinated planning of various energy subsystems, thereby significantly improving the development efficiency and utilization rate of energy.

[0003] However, in the actual application of integrated energy systems, their state assessment faces a series of challenges. First, the data redundancy of integrated energy system measurements is very low, and the observability of the system is not high. Since the integrated energy system contains multiple energy sources such as electricity, heat, and natural gas, and there is no unified measurement accuracy standard for the measurement equipment of these energy sources and the data measurement equipment of the power system, the measurement errors of various measurement equipment are large. In addition, there are differences in the sampling time intervals of different measurement equipment, which makes it impossible to accurately evaluate the state of the system at a certain moment in a targeted manner.

[0004] Specifically, the existing integrated energy system state estimation methods have the following main problems: First, the data redundancy is low, which limits the observability of the system and makes it difficult to fully reflect the actual operating status of the system; second, the measurement error is large. Due to the different accuracy of the measuring equipment and the lack of a unified calibration standard, there is a large deviation in the measurement data; third, the time synchronization is poor. The sampling time intervals of different measuring devices are different, which makes it impossible to obtain the system status information at the same time, which brings difficulties to the status assessment.

[0005] In order to solve the above problems, domestic and foreign scholars and engineers have conducted a lot of research and exploration. However, most of the existing technical solutions can only evaluate the status of a single energy system, and cannot accurately estimate the overall status of the integrated energy system. In addition, although some technical solutions consider the coupling relationship of multiple energy systems, they still have deficiencies in data processing and state estimation methods, making it difficult to meet the needs of practical applications.

[0006] Therefore, the present invention proposes a method for evaluating the state of an integrated energy system based on multi-source data fusion, aiming to solve the problems existing in the prior art and improve the accuracy and reliability of the state estimation of the integrated energy system. The method establishes an electric-gas coupled measurement model, uses Bayesian learning to fit the probability distribution of the measurement data, and uses the Monte Carlo sampling algorithm to generate a large amount of data for training the deep learning network, ultimately achieving an accurate evaluation of the state of the integrated energy system. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a comprehensive energy system status assessment method based on multi-source data fusion to solve the technical problems existing in the comprehensive energy system status assessment, such as low data redundancy, low system observability, large measurement error and poor time synchronization.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is a method for evaluating the state of an integrated energy system based on multi-source data fusion, comprising the following steps: Step 1: Establish the measurement model of electric-gas coupling; Step 2: Generate integrated energy system measurement data samples using the Monte Carlo sampling algorithm for subsequent neural network training; Step 3: Establish a data verification mechanism based on the power flow equation, perform power flow verification on the generated measurement data samples, and retain the data that can make the power flow converge for neural network training; Step 4: Establish a data-driven integrated energy system state estimation model based on the long short-term memory network Bi-LSTM (Bi-directional Long Short-Term Memory Network), use the verified measurement data as the input of the neural network, and use the state data as the output of the neural network for training; Step 5: Use the root mean square error, measurement error statistics, and estimation error statistics to evaluate the state assessment model to determine whether the effect of the state assessment model is better than that of directly using the measurement data for evaluation.

[0009] In a preferred solution, the specific steps of Step 1 include: Step 1.1: Establish a mapping relationship between the measurement vector and the state vector of the electric-gas coupling system, wherein the measurement vector includes data of node voltage, power, branch power, and pipeline pressure, and the state vector includes data of gas grid pressure, grid voltage, pipeline flow, and compressor flow; Step 1.2: Perform probability distribution fitting on the measured data of the integrated energy system based on Bayesian learning, including establishing a Gaussian inference model for the gas grid and a Gaussian mixture distribution model for the power grid.

[0010] In a preferred solution, the specific steps of Step 1.1 include: Step 1.1.1: Establish the mapping relationship between the measurement vector and the state vector of the electric-gas coupling system. Assume that the vector injected into the electric-gas coupling system is , the measurement vector is , the state vector is , the mapping relationship established between them is as follows: (1) In the formula, Inject vector and state vector mapping for system nodes; The measurement vector includes the injection vector or branch power, node voltage, and gas grid pressure. The mapping relationship between the measurement vector and the state vector is expressed as follows: (2) (3) In the formula, and They are the measurement functions of gas grid and power grid equipment in the electricity-gas coupling system, respectively; and are the measurement vector sets of the gas grid and the power grid respectively, and are the state vector sets of the gas grid and the power grid respectively, and are the parameters of the gas grid and power grid state estimation models, and are the noise of the gas grid and power grid measurement vectors, is the number of measurement data sets; Next, the Huber robust estimation is used for state assessment. The process is as follows: (4) In the formula, is a function; Step 1.1.2: Establish a probability model that includes measurement noise; Assuming that the data collected from each device are independent of each other, establish the following probability model: (5) In the formula, , , , They are vectors , , , No. elements, is the number of state vectors; Using adjustable parameters of the state estimation model To further express the mapping relationship, based on the above probability model, a comprehensive energy system state estimation probability model is established: (6) (7) (8) In the formula, is the mathematical expectation, is the state estimate of the measurement vector, is the hyperparameter of the measured data, For the The hyperparameters of the measured data.

[0011] In a preferred solution, the specific steps of Step 1.2 include: Step 1.2.1: Establish a Gaussian inference model for the gas grid of the integrated energy system; Measurement data for the gas grid part of the integrated energy system The mean is , the variance is Gaussian distribution, that is: (9) In the formula, is the Gaussian distribution function; When using the Gaussian inference model to calculate the expected measurement data, the mean and variance need to be verified, namely: (10) In the formula, is the uniform distribution function, and They obey the upper and lower bounds of the uniform distribution respectively: (11) In the formula, is the distribution function of the half-normal distribution, for The mean of the half-normal distribution is for The variance of the half-normal distribution of ; assumed , , The values ​​of the three parameters are calculated and the prior distribution of the parameters is calculated. The kernel density estimation method and the central limit theorem are used to determine whether the set parameters are reasonable. The kernel density estimation method is expressed as follows: (12) In the formula, is the kernel density function, is the kernel function, for The number of is the bandwidth of the kernel density function; Use the prior distribution of the set parameters to calculate the posterior distribution of the gas network measurement data and select the most suitable and , further calculate the probability distribution of the original data; Step 1.2.2: Establish a non-Gaussian distribution circle model of the integrated energy system power grid; The Gaussian mixture model and expectation maximization algorithm are combined, the voltage amplitude, current amplitude, branch power and node power are collected by the SCADA system, the voltage and current vectors are measured by the PMU, and the voltage, current amplitude, branch, node power and user power consumption information are collected by the AMI, and the data are fused at multiple time scales. Through the calculation of Gaussian mixture model and expectation maximization algorithm, the non-Gaussian distribution circle model is solved to obtain the distribution parameters of the measurement vector of the integrated energy system power grid: (13) (14) In the formula, is the measurement vector and number, is the mean value after multi-scale fusion of data, , is the identity matrix, is the correlation coefficient of the autoregressive moving average model, is the variance of the grid measurement vector.

[0012] In the preferred solution, the specific steps of Step 2 of generating the integrated energy system measurement data samples by using Monte Carlo are: The Monte Carlo sampling algorithm is used to generate a large amount of data on the gas grid and power grid of the integrated energy system. The Monte Carlo sampling algorithm constructs a data set through the Markov chain convergence theorem. The state transition process of the Markov chain is expressed by the following formula: (15) In the formula, They are The measurement vector of the moment; Assume that the measurement data at the previous moment is only related to the measurement data at the current moment and the noise of the measurement data The probability distribution of , in this case, the noise in the measured data is proportional to the vector injected into the system is independent and identically distributed, then the probability distribution of the measurement vector Can be distributed by injecting vectors into the system With noise distribution The system's measurement data can be obtained and the generated data can be divided into measurement samples according to the following process and state sample : (16) (17) In the formula, , , and , , The state quantity and measurement quantity confidence level, upper and lower sample limits; The generated data are screened, and the measurement data with a deviation of about 10% are set as data that do not meet the requirements. The state vector is selected at a significance level of 95%, while the measurement component is screened at a significance level of 85%, to obtain the data set of the measurement vector and the data set of the state vector.

[0013] In the preferred solution, the Step 3 specifically uses the generated data to perform flow verification, establishes a flow equation based on the power grid and gas grid of the integrated energy system, performs flow verification on the data, and retains data that can make the flow converge for neural network training. The process of performing flow verification on the data includes: Step 3.1: Establish the flow calculation equation of the gas network, which includes the flow equation of the compressor and the flow equation of the branch pipeline. The compressors include electric drive compressors and fuel drive compressors. The flow equation of the electric drive compressor is as follows: (18) In the formula, The power consumed by the compressor, is the number of compressors, is the pipeline flow rate of the electric drive compressor, is the number of pipelines, For the The node and The ratio of the air pressure at each node; The fuel driven compressor equation is as follows: (19) In the formula, is the ratio of fuel consumption of fuel-driven compressor, is the calorific value of natural gas, is the gas-to-electricity efficiency; The flow equation for the branch pipe is as follows: (20) In the formula, is the pipeline flow rate, For branch The resistance of the pipeline, For the The pressure of each node; The node flow equation is as follows: (twenty one) In the formula, For the The traffic of each node, is the node matrix, for and The column vector composed of The gas network demand flow equation is as follows: (twenty two) In the formula, For the The required flow of each node is is the demand flow matrix; The power generation capacity of gas power plants is as follows: (twenty three) In the formula, is the power generation capacity of the gas power plant, is the number of gas-fired power plants; Step 3.2: Establish the power flow equation of the power grid, including the active power flow equation and the reactive power flow equation, as shown in the following formula: (twenty four) (25) In the formula, and Respectively The active power flowing into each node and the active load power, and Respectively The reactive power and reactive load power flowing into each node, and Respectively The node and The conductance and susceptance between nodes, It is The voltage of the node, For the The node and The voltage phase angle between nodes; Step 3.3: Establish the coupled power flow equation of the integrated energy system. While satisfying the power flow equations of the power grid and gas grid, it is also necessary to satisfy the power balance of the coupled nodes, as shown in the following formula: (26) (27) (28) In the formula, and are the balance coefficients of the power grid and gas grid, and are the grid energy loss and power generation respectively, and They are gas grid energy consumption and gas-fired power generation respectively.

[0014] In the preferred scheme, the specific scheme of Step 4 to establish a data-driven integrated energy system state estimation model based on the long short-term memory network Bi-LSTM is to establish a multi-input-single-output long short-term memory network model based on the mapping relationship between the measurement vector and the state vector of the integrated energy system, use the generated measurement data as the input data of the neural network, and use the operating state data of the integrated energy system that has been verified by the flow as the output data of the neural network, construct a Bi-LSTM data set, use the first 70% of the data set as training samples, and the last 30% as test samples, and use the adaptive moment estimation optimization algorithm to train the neural network. The structure of each LSTM is composed of one or more units with forgettable and memory functions, which are input gate, output gate and forget gate respectively.

[0015] In a preferred solution, the implementation of the long short-term memory network Bi-LSTM function includes the following process: The forget gate controls the content of the cell state of the previous layer. Through the sigmoid activation function, the content of the cell of the previous layer is forgotten and retained. The output of the forget gate is completed by the following formula: (29) The input gate is used to process the input of the current sequence position and determine the information that needs to be updated. This process consists of two parts. One part is the remaining memory information after passing through the forget gate, and the other part is the newly input information. This process is completed by the following formula: (30) (31) After processing through the forget gate and the input gate, the state of the cell can be updated: (32) In the formula, Indicate the information you wish to delete, Indicates newly added information; The output gate is used to determine the output based on the information stored in the cell, that is, to selectively output the information stored in the cell state, similar to the update process of the two parts of the input gate, which is implemented by the following formula: (33) (34) In the formula, Represents the input data at the current moment, , Respectively represent the internal state of the previous moment and the current moment, , Represent the hidden layer states at the previous moment and the current moment respectively, , Respectively and Activation function, , and Respectively represent the calculation variables of the forget gate, input gate and output gate at the current moment, Indicates the unit state of the input at that moment; , , are the weight matrices of the forget gate, input gate, and output gate respectively. is the input unit state weight matrix, is the input unit state bias term, , , They are the output bias vectors of the forget gate, input gate and output gate respectively.

[0016] In the preferred solution, the Step 5 specifically selects the root mean square error , measurement error statistics and the estimated error statistics Evaluate the condition assessment model and determine whether the results of the condition assessment model are better than those of direct assessment using measurement data. The evaluation process includes: (35) (36) (37) In the formula, is the value of the root mean square error, is the number of Bi-LSTM network data sets, For the The next state estimates the true value, For the The next state estimate prediction value, represents the total number of simulations, is the number of measured variables, Indicates The measurement of measured variables, is the standard deviation of the noise, For the The true value of the measured variable, For the The measured variable Sub-estimate.

[0017] In a preferred embodiment, the integrated energy system is a park-level or building-level integrated energy system.

[0018] The comprehensive energy system state estimation method based on the dynamic weighted algorithm provided by the present invention has the following beneficial effects: 1. The present invention solves the technical problems existing in the comprehensive energy system status assessment, such as low data redundancy, low system observability, large measurement error and poor time synchronization; 2. The design of the multi-source data fusion solution of the present invention improves the data redundancy and system observability by fusing data from different devices and time scales, making the system status more transparent and controllable, thereby further improving the accuracy of the state estimation of the integrated energy system; 3. The solution of the present invention applies multi-source data fusion technology to the state assessment of the integrated energy system for the first time. By integrating data from different devices and time scales, the redundancy of data and the observability of the system are improved. This is a new attempt in the field of integrated energy system state assessment. 3. The use of Gaussian inference and Gaussian mixture model in the present invention adopts Bayesian learning method to fit the probability distribution of measurement data, thereby improving the accuracy and robustness of data processing; 4. The present invention innovatively combines Gaussian inference and Gaussian mixture model to process the measurement data of the integrated energy system, and performs probability distribution fitting on the measurement data through Bayesian learning method, thereby improving the accuracy and robustness of data processing, which is also a new technological breakthrough in the field of data processing; 5. The present invention adopts Monte Carlo sampling and LSTM model, generates a large number of data samples through Monte Carlo sampling algorithm, and uses LSTM model for state estimation, which improves the accuracy and efficiency of state evaluation and provides strong support for the management and optimization of integrated energy systems; 6. The solution of the present invention combines the Monte Carlo sampling algorithm with the LSTM model for the first time for the state estimation of the integrated energy system. A large number of data samples are generated by the Monte Carlo sampling algorithm and trained using the LSTM model, which improves the accuracy and efficiency of the state assessment. This is a new innovation in the state assessment method; 7. The present invention also realizes accurate identification and prediction of complex operating states of the integrated energy system by constructing a multi-dimensional feature space and combining it with a deep learning algorithm, providing strong technical support for intelligent management of the system and further promoting efficient, safe and green operation of the integrated energy system; 8. The multi-dimensional feature extraction and deep learning fusion strategy of the present invention effectively improves the sensitivity and timeliness of system anomaly detection, lays a solid foundation for fault warning and rapid response, and demonstrates the future trend of intelligent operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with the accompanying drawings and implementation examples: Figure 1 The present invention is based on the data-driven integrated energy system state estimation model; Figure 2 This is a long short-term memory network model diagram of the present invention; Figure 3 This is a kernel density estimation diagram of the branch flow of node 5 of the gas pipeline network in Example 2 of the present invention; Figure 4 3 is a comparison diagram of the root mean square error between the existing network and the network of the present invention in Example 3 of the present invention; Figure 5 is the error rate of the estimation of the voltage of each electrical network node by different models in Embodiment 4 of the present invention; Figure 6 It is the error rate of estimation of the state of each gas network node by different models in Example 4 of the present invention. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is further described below in conjunction with the embodiments and drawings: Example 1 like Figure 1~2 As shown, a comprehensive energy system state assessment method based on multi-source data fusion includes the following steps: Step 1: Establish the measurement model of electric-gas coupling; Step 2: Generate integrated energy system measurement data samples using the Monte Carlo sampling algorithm for subsequent neural network training; Step 3: Establish a data verification mechanism based on the power flow equation, perform power flow verification on the generated measurement data samples, and retain the data that can make the power flow converge for neural network training; Step 4: Establish a data-driven integrated energy system state estimation model based on the long short-term memory network Bi-LSTM (Bi-directional Long Short-Term Memory Network), use the verified measurement data as the input of the neural network, and use the state data as the output of the neural network for training; Step 5: Use the root mean square error, measurement error statistics, and estimation error statistics to evaluate the state assessment model to determine whether the effect of the state assessment model is better than that of directly using the measurement data for evaluation.

[0021] In this embodiment, the specific steps of Step 1 include: Step 1.1: Establish a mapping relationship between the measurement vector and the state vector of the electric-gas coupling system, wherein the measurement vector includes data of node voltage, power, branch power, and pipeline pressure, and the state vector includes data of gas grid pressure, grid voltage, pipeline flow, and compressor flow; Step 1.2: Perform probability distribution fitting on the measured data of the integrated energy system based on Bayesian learning, including establishing a Gaussian inference model for the gas grid and a Gaussian mixture distribution model for the power grid.

[0022] Furthermore, the specific steps of Step 1.1 include: Step 1.1.1: Establish the mapping relationship between the measurement vector and the state vector of the electric-gas coupling system. Assume that the vector injected into the electric-gas coupling system is , the measurement vector is , the state vector is , the mapping relationship established between them is as follows: (1) In the formula, Inject vector and state vector mapping for system nodes; The measurement vector includes the injection vector or branch power, node voltage, and gas grid pressure. The mapping relationship between the measurement vector and the state vector is expressed as follows: (2) (3) In the formula, and They are the measurement functions of gas grid and power grid equipment in the electricity-gas coupling system, respectively; and are the measurement vector sets of the gas grid and the power grid respectively, and are the state vector sets of the gas grid and the power grid respectively, and are the parameters of the gas grid and power grid state estimation models, and are the noise of the gas grid and power grid measurement vectors, is the number of measurement data sets; The least squares estimation is very susceptible to residuals in the process of solving state estimation. If the residuals are too large, the estimation accuracy will be reduced. Huber can The function reduces the influence of excessive residual on the estimation accuracy. The process of using Huber robust estimation for state assessment is as follows: (4) In the formula, is a function; Step 1.1.2: Establish a probability model including measurement noise; The electric-pneumatic coupling system contains a large amount of operating data, but the data measured by the measuring device has certain errors, low data redundancy, and the non-uniform sampling interval of the data set will also bring difficulties to state estimation. Bayesian learning is suitable for systems with low observability and can effectively solve the problems of inaccurate data and small scale of data sets. Bayesian learning uses probability to characterize the relationship between different variables. Assuming that the data collected from each device are independent of each other, the following probability model is established: (5) In the formula, , , , They are vectors , , , No. elements, is the number of state vectors; In order to characterize the relationship between the measurement vector and the state vector, the adjustable parameters of the state estimation model are used. To further express the mapping relationship, based on the above probability model, a comprehensive energy system state estimation probability model is established: (6) (7) (8) In the formula, is the mathematical expectation, is the state estimate of the measurement vector, is the hyperparameter of the measured data, For the The hyperparameters of the measured data.

[0023] Furthermore, the specific steps of Step 1.2 include: Step 1.2.1: Establish a Gaussian inference model for the gas grid of the integrated energy system; Measurement data for the gas grid part of the integrated energy system The mean is , the variance is Gaussian distribution, that is: (9) In the formula, is the Gaussian distribution function; When using the Gaussian inference model to calculate the expected measurement data, the mean and variance need to be verified, namely: (10) In the formula, is the uniform distribution function, and They obey the upper and lower bounds of the uniform distribution respectively: (11) In the formula, is the distribution function of the half-normal distribution, for The mean of the half-normal distribution is for The variance of the half-normal distribution of ; assumed , , The values ​​of the three parameters are calculated and the prior distribution of the parameters is calculated. The kernel density estimation method and the central limit theorem are used to determine whether the set parameters are reasonable. The kernel density estimation method is expressed as follows: (12) In the formula, is the kernel density function, is the kernel function, for The number of is the bandwidth of the kernel density function; If the parameter and The distribution of is close to Gaussian distribution, then the prior distribution of the set parameters is used to calculate the posterior distribution of the gas network measurement data, and finally the most suitable and ,During the selection process, the posterior distribution sampling of the gas network ,measurement data is performed multiple times. By comparing the kernel density ,functions of the original measurement data and the predicted data, and ,selecting the kernel density function by comparing the peak and shape of the data, the ,probability distribution of the original data is further calculated; Step 1.2.2: Establish a non-Gaussian distribution circle model of the integrated energy system power grid; In order to represent the Gaussian distribution function of the measurement vector of the node voltage, the Gaussian mixture model and the expectation maximization algorithm are combined. The SCADA (Supervisory Control And Data Acquisition) system is used to collect voltage amplitude, current amplitude, branch power and node power. The data collection cycle is 3 to 5 seconds. The PMU (Power Management Unit) is used to measure the voltage and current vectors. The data collection cycle is 5 to 30 milliseconds. The AMI (Alternate Mark Inversion) is used to collect voltage, current amplitude, branch, node power and user power consumption information. The data collection cycle is 15 minutes. Due to the different data collection cycles of various devices, the data needs to be fused at multiple time scales. Through the calculation of Gaussian mixture model and expectation maximization algorithm, the non-Gaussian distribution circle model is solved to obtain the distribution parameters of the measurement vector of the integrated energy system power grid: (13) (14) In the formula, is the measurement vector and number, is the mean value after multi-scale fusion of data, , is the identity matrix, is the correlation coefficient of the autoregressive moving average model, is the variance of the grid measurement vector.

[0024] Furthermore, the specific steps of Step 2 for generating the integrated energy system measurement data samples using Monte Carlo are: The Monte Carlo sampling algorithm is used to generate a large amount of data on the gas grid and power grid of the integrated energy system. The Monte Carlo sampling algorithm constructs a data set through the Markov chain convergence theorem. The state transition process of the Markov chain is expressed by the following formula: (15) In the formula, They are The measurement vector of the moment; Assume that the measurement data at the previous moment is only related to the measurement data at the current moment and the noise of the measurement data The probability distribution of , in this case, the noise in the measured data is proportional to the vector injected into the system is independent and identically distributed, then the probability distribution of the measurement vector Can be distributed by injecting vectors into the system With noise distribution The system's measurement data can be obtained and the generated data can be divided into measurement samples according to the following process and state sample : (16) (17) In the formula, , , and , , The state quantities and measurement quantity confidence level, upper and lower sample limits; The generated data are screened. In the actual measured data, the limit of abnormal data is generally above 5%. In order to ensure the quality of the data, the measured data with a deviation of about 10% is set as data that does not meet the requirements. The state vector is selected at a significance level of 95% and the measurement component is screened at a significance level of 85% to obtain the data set of the measurement vector and the data set of the state vector.

[0025] Furthermore, the Step 3 specifically uses the generated data to perform flow verification, establishes a flow equation based on the power grid and gas grid of the integrated energy system, performs flow verification on the data, and retains data that can make the flow converge for neural network training. The process of performing flow verification on the data includes: Step 3.1: Establish the flow calculation equation of the gas network, which includes the flow equation of the compressor and the flow equation of the branch pipeline. The compressors include electric drive compressors and fuel drive compressors. The flow equation of the electric drive compressor is as follows: (18) In the formula, The power consumed by the compressor, is the number of compressors, is the pipeline flow rate of the electric drive compressor, is the number of pipelines, For the The node and The ratio of the air pressure at each node; The fuel driven compressor equation is as follows: (19) In the formula, is the ratio of fuel consumption of fuel-driven compressor, is the calorific value of natural gas, is the gas-to-electricity efficiency; The flow equation of the branch pipe is as follows: (20) In the formula, is the pipeline flow rate, For branch The resistance of the pipeline, For the The pressure of each node; The node flow equation is as follows: (twenty one) In the formula, For the The traffic of each node, is the node matrix, for and The column vector composed of The gas network demand flow equation is as follows: (twenty two) In the formula, For the The required flow of each node is is the demand flow matrix; The power generation capacity of gas power plants is as follows: (twenty three) In the formula, is the power generation capacity of the gas power plant, is the number of gas-fired power plants; Step 3.2: Establish the power flow equation of the power grid, including the active power flow equation and the reactive power flow equation, as shown in the following formula: (twenty four) (25) In the formula, and Respectively The active power flowing into each node and the active load power, and Respectively The reactive power and reactive load power flowing into each node, and Respectively The node and The conductance and susceptance between nodes, It is The voltage of the node, For the The node and The voltage phase angle between nodes; Step 3.3: Establish the coupled power flow equation of the integrated energy system. While satisfying the power flow equations of the power grid and gas grid, it is also necessary to satisfy the power balance of the coupled nodes, as shown in the following formula: (26) (27) (28) In the formula, and are the balance coefficients of the power grid and gas grid, and are the grid energy loss and power generation respectively, and They are gas grid energy consumption and gas-fired power generation respectively.

[0026] Furthermore, the specific scheme of the Step 4 for establishing a data-driven integrated energy system state estimation model based on the long short-term memory network Bi-LSTM is to establish a multi-input-single-output long short-term memory network model based on the mapping relationship between the measurement vector and the state vector of the integrated energy system, use the generated measurement data as the input data of the neural network, and use the operating state quantity data of the integrated energy system that has been verified by the flow as the output data of the neural network, construct a Bi-LSTM data set, use the first 70% of the data set as training samples, and the last 30% as test samples, and use the adaptive moment estimation optimization algorithm to train the neural network. The structure of each LSTM is composed of one or more units with forgettable and memory functions, which are input gate, output gate and forget gate respectively.

[0027] Furthermore, the implementation of the Bi-LSTM function of the long short-term memory network includes the following process: The forget gate controls the content of the cell state of the previous layer. Through the sigmoid activation function, the content of the cell of the previous layer is forgotten and retained. The output of the forget gate is completed by the following formula: (29) The input gate is used to process the input of the current sequence position and determine the information that needs to be updated. This process consists of two parts. One part is the remaining memory information after passing through the forget gate, and the other part is the newly input information. This process is completed by the following formula: (30) (31) After processing through the forget gate and the input gate, the state of the cell can be updated: (32) In the formula, Indicate the information you wish to delete, Indicates newly added information; The output gate is used to determine the output based on the information stored in the cell, that is, to selectively output the information stored in the cell state, similar to the update process of the two parts of the input gate, which is implemented by the following formula: (33) (34) In the formula, Represents the input data at the current moment, , Respectively represent the internal state of the previous moment and the current moment, , Represent the hidden layer states at the previous moment and the current moment respectively, , Respectively and Activation function, , and Respectively represent the calculation variables of the forget gate, input gate and output gate at the current moment, Indicates the unit state of the input at that moment; , , are the weight matrices of the forget gate, input gate, and output gate respectively. is the input unit state weight matrix, is the input unit state bias term, , , They are the output bias vectors of the forget gate, input gate and output gate respectively.

[0028] Furthermore, the Step 5 specifically selects the root mean square error , measurement error statistics and the estimated error statistics Evaluate the condition assessment model and determine whether the results of the condition assessment model are better than those of direct assessment using measurement data. The evaluation process includes: (35) (36) (37) In the formula, is the value of the root mean square error, is the number of Bi-LSTM network data sets, For the The next state estimates the true value, For the The next state estimate prediction value, represents the total number of simulations, is the number of measured variables, Indicates The measurement of measured variables, is the standard deviation of the noise, For the The true value of the measured variable, For the The measured variable Sub-estimate.

[0029] Furthermore, the integrated energy system is a park-level or building-level integrated energy system.

[0030] Example 2 In another preferred embodiment, based on the above embodiment 1, the state estimation is performed by taking the park-level integrated energy system as an example. The park is composed of an integrated energy system consisting of a 13-node gas grid and a 30-node grid coupling. The system includes 4 gas compressor nodes, including two gas compressor nodes and two electric compressor nodes. The system coupling nodes include 10 gas turbine nodes and two electric drive compressor nodes. The configuration of the integrated energy system is shown in Table 1, Table 2, and Table 3, respectively. The state estimation model of the integrated energy system based on data-driven is as follows Figure 1 As shown: Table 1 Gas power plant coupling nodes

[0031] Table 2 Compressor coupling nodes

[0032] Table 3 Measurement device node configuration

[0033] A comprehensive energy system state assessment method based on multi-source data fusion, comprising: Step 1: In order to effectively estimate the state of the integrated energy system, an electric-gas coupled measurement model is established; Step 2: Based on the fluctuation of the gas and power grid states, the measurement probability distribution of the integrated energy system is fitted based on Bayesian learning; Step 3: In order to improve the estimation accuracy of the deep learning network, the Monte Carlo sampling algorithm is used to generate the measurement data samples of the integrated energy system; Step 4: In order to ensure the rationality of the generated data, the power flow equation of the integrated energy system is constructed using the generated data for verification; Step 5: Establish a data-driven model for integrated energy system state estimation based on a bidirectional long short-term memory network.

[0034] Step 6: Use indicators such as root mean square error and calculation time to evaluate the performance of the state assessment method proposed in the present invention.

[0035] This example selects a park-level integrated energy system in Hebei as a research object for verification, and the specific implementation method is as follows: The specific implementation steps of step 1 are as follows: Step 1.1: Establish the mapping relationship between the measurement vector and the state vector of the electric-gas coupling system; The measured variables in the park-level integrated energy system mainly include node power, branch power, pipeline pressure, node flow, etc.; the state variables mainly include node voltage, node pressure, pipeline flow, compressor flow, etc. The node injection vector directly affects the node state vector, and there is a certain functional relationship between the two. With the state vector The relationship between them is shown as follows: (1) Furthermore, the node's measurement vector It is composed of the node's injection vector, branch power, gas grid pressure, node voltage, etc. Therefore, there is a mapping relationship between the node's measurement vector and state vector.

[0036] The least squares estimation is very susceptible to residuals in the process of solving state estimation. If the residuals are too large, the estimation accuracy will be reduced. Huber can The function reduces the impact of excessive residuals on estimation accuracy, so Huber robust estimation is used for state assessment. The process is as follows: (4) The above formula can reflect the accuracy of each sensor. However, since the sampling time of the measuring equipment is not uniform, the measured data cannot be directly used for state estimation and needs further conversion.

[0037] Step 1.2: Build a probability model that includes measurement noise; Since there are certain errors in the measurement data, a probability model including the measurement data noise is established: (5) Since the focus is on analyzing the relationship between the measurement vector and the state vector, the parameters of the state estimation model can be used To express the mapping relationship, Simplified to , and further establish a probabilistic measurement model for state estimation of the integrated energy system: (6) (7) (8) The specific implementation steps of step 2 are as follows: Step 2.1: Establish a Gaussian inference model for the gas grid of the integrated energy system; The measured value of the gas pipeline is composed of pipeline pressure and node flow. By performing Gaussian inference on the measured data of the gas network, the most ideal Gaussian distribution curve is obtained. Figure 3 This is the kernel density estimation diagram of the branch flow of node 5 in the gas pipeline network. The distribution curve of the optimal prediction value is obtained by fitting the distribution of the measured data multiple times, and the Gaussian distribution of the optimal prediction value is selected as the distribution of the branch measurement data.

[0038] Step 2.2: Establish a non-Gaussian distribution circle model of the integrated energy system power grid; The measured value of the power grid is composed of the power of nodes and branches. The Gaussian distribution of the voltage of the power grid nodes is obtained through the Gaussian mixture model and the expectation maximization algorithm. Due to the different sampling time intervals of the measuring equipment, the measured data needs to be aggregated to obtain a new data set. The distribution curve of the optimal predicted value is obtained by multiple fitting of the measured data distribution, and the Gaussian distribution of the optimal predicted value is selected as the distribution of the measurement data of the branch.

[0039] The specific implementation steps of step 3 are as follows: In order to solve the problem of measurement data deviation, the model has been used to obtain the Gaussian distribution of the data. In order to better train the neural network, the Monte Carlo sampling algorithm is used to generate a large amount of data. The data generated in this way cannot guarantee the quality of each data, so it is necessary to screen the data. The state quantity data and measurement quantity data are screened according to the significance level of 95% and 85% respectively to obtain the data set after data screening.

[0040] The specific implementation steps of step 4 are as follows: In order to ensure the rationality of each data set in the data set, a power flow equation is established based on the power grid and gas grid of the integrated energy system, and the data is checked for power flow. The data that can make the power flow converge is retained for the training of the neural network. The verification process is as follows: First, establish the flow calculation equation of the gas network, which includes the flow equation of the compressor and the flow equation of the branch pipeline. The compressors include electric drive compressors and fuel drive compressors. The flow equation of the electric drive compressor is as follows: (18) In the formula, The power consumed by the compressor, is the number of compressors, is the pipeline flow rate of the electric drive compressor, is the number of pipelines, For the The node and The ratio of the air pressure at each node.

[0041] The fuel driven compressor equation is as follows: (19) In the formula, is the ratio of fuel consumption of fuel-driven compressor, is the calorific value of natural gas, The gas-to-electricity efficiency.

[0042] The flow equation of the branch pipe is as follows: (20) In the formula, is the pipeline flow rate, For branch The resistance of the pipeline, For the The pressure of a node.

[0043] The node flow equation is as follows: (twenty one) In the formula, For the The traffic of each node, is the node matrix, for and Column vector composed of .

[0044] The gas network demand flow equation is as follows: (twenty two) In the formula, For the The required flow of each node is is the demand flow matrix.

[0045] The power generation capacity of gas power plants is as follows: (twenty three) In the formula, is the power generation capacity of the gas power plant, is the number of gas-fired power plants.

[0046] Secondly, establish the power flow equation of the power grid, including the active power flow equation and the reactive power flow equation, as shown in the following equation: (twenty four) (25) In the formula, and Respectively The active power flowing into each node and the active load power, and Respectively The reactive power and reactive load power flowing into each node, and Respectively The node and The conductance and susceptance between nodes, It is The voltage of the node, For the The node and The voltage phase angle between the nodes.

[0047] Finally, the coupled power flow equation of the integrated energy system is established. While satisfying the power flow equations of the power grid and gas grid respectively, it is also necessary to satisfy the power balance of the coupled nodes, as shown in the following equation: (26) (27) (28) In the formula, and are the balance coefficients of the power grid and gas grid, and are the grid energy loss and power generation respectively, and They are gas grid energy consumption and gas-fired power generation respectively.

[0048] The specific implementation steps of step 5 are as follows: A data-driven model based on LSTM is established. The first 70% of the corrected data set is used as the training set, and the last 30% is used as the test set. The corrected measurement data is used as the input of the network, and the state data is used as the output of the network to train the neural network. As shown in Table 4, the state estimation calculation time of different neural networks is shown. The single estimation and total estimation are respectively estimated for one and all measurement data, and the single network and total network training are respectively trained on the data of the nodes of a single network and all networks (including the electricity network and the gas network) of the integrated energy system.

[0049] Table 4 State estimation calculation time of different neural networks

[0050] The specific implementation steps of step 6 are as follows: In order to compare the performance of different methods in the state estimation of integrated energy systems, the root mean square error is used as an indicator to evaluate the prediction deviation of the model, and the measurement error statistics are used. and the estimated error statistics To evaluate whether the estimated value is practical, the expression is as follows: (36) (37) In the formula, represents the total number of simulations, is the number of measured variables, Indicates The measurement of measured variables, is the standard deviation of the noise, For the The true value of the measured variable, For the The measured variable Sub-estimate.

[0051] when Compare When is large, it means that the estimation effect of the integrated energy system state is better than the measurement effect, otherwise the direct measurement effect will be better, as shown in Table 5, which is a comparison of the root mean square error, measurement error statistics and estimation error statistics of different models in the total network training.

[0052]

[0053] In summary, it is shown that a comprehensive energy system state assessment method based on multi-source data fusion can effectively improve the accuracy of state estimation of the park-level comprehensive energy system.

[0054] Example 3 In another preferred embodiment, based on the above embodiments 1 and 2, a building-level integrated energy system is used as an example for state estimation. The integrated energy system includes an integrated energy system consisting of a 16-node gas grid and a 24-node grid coupled. The system includes two gas compressor nodes, including a gas compressor node and an electric compressor node. The system coupling node includes a gas turbine node. The measurement device node configuration is shown in Table 6: Table 6 Measurement device node configuration

[0055] A comprehensive energy system state assessment method based on multi-source data fusion, comprising: Step 1: To effectively estimate the state of the integrated energy system, a measurement model of electrical and gas coupling is established; Step 2: In view of the state fluctuations of the gas grid and the power grid, the Bayesian learning method is used to fit the probability distribution of the measurement data of the integrated energy system; Step 3: In order to improve the estimation accuracy of the deep learning network, the Monte Carlo sampling algorithm is used to generate measurement data samples of the integrated energy system; Step 4: In order to verify the rationality of the generated data, the power flow equations of the power grid and gas grid are established for data verification; Step 5: Establish a data-driven model for integrated energy system state estimation based on long short-term memory network.

[0056] Step 6: The performance of the proposed state assessment method is evaluated through indicators such as root mean square error and calculation time.

[0057] This example selects a building-level integrated energy system in Liaoning as the research object, and the specific implementation method is as follows: The specific implementation steps of step 1 are as follows: Step 1.1: Establish the mapping relationship between the measurement vector and the state vector of the electric-gas coupling system.

[0058] The measurement variables in the building-level integrated energy system include node and branch power, pipeline pressure and flow, etc. The state variables mainly include node voltage and pressure, etc. Since the state of the node is affected by the injection amount of the node, there is a mapping relationship between the two. Furthermore, the measurement vector of the node is composed of the injection vector of the node, branch power, gas network pressure, etc., so there is a mapping relationship between the measurement vector and the state vector, as shown below: (38) In the formula, The mapping relationship between the node measurement vector and the injection vector.

[0059] The least squares estimation is very susceptible to residuals in the process of solving state estimation. If the residuals are too large, the estimation accuracy will be reduced. Huber can The function reduces the impact of excessive residuals on estimation accuracy, so Huber robust estimation is used for state assessment. The process is as follows: (4) Due to the different measurement accuracy of different measuring equipment, the quality of the measured data is low. Directly using the measured data for state assessment will produce large deviations, so further processing is required.

[0060] Step 1.2: Build a probability model that includes measurement noise; In order to reduce the impact of the error of the measured data on the evaluation results, a probability model is established to generate data. Due to the uncertainty of the error of the measuring equipment, a probability model of noise is first established. The state estimation measurement model of the integrated energy system is established using the probability model of noise, as shown in the following formula: (5) (6) (7) (8) The specific implementation steps of step 2 are as follows: Step 2.1: Establish a Gaussian inference model for the gas grid of the integrated energy system; The measurement data of the gas network part of the building-level integrated energy system obeys the Gaussian distribution. Before establishing the Gaussian inference model, it is necessary to verify the mean and variance of the measurement data, and then preset the upper and lower limits of the uniform distribution and the standard deviation of the half-normal distribution. The kernel density estimation method is used to determine whether the preset parameters are reasonable. After multiple analyses of the distribution of the measurement data, the Gaussian distribution of the optimal prediction value is selected as the distribution of the branch measurement data.

[0061] Step 2.2: Establish a non-Gaussian distribution circle model of the integrated energy system power grid; Since the parameters of the power grid may change greatly in a short period of time, and the collection cycles of different devices are different during the data measurement process, the data collection cycle of the SCADA system is 3 to 5 seconds, the collection cycle of the PMU is 5 to 30 seconds, and the data collection cycle of the AMI is 5 to 30 milliseconds. Therefore, it is necessary to perform multi-time scale fusion on multi-source data, and use the Gaussian mixture model and the expectation-maximization algorithm to solve the non-Gaussian distribution circle model to obtain the distribution parameters of the power grid measurement vector of the building-level integrated energy system.

[0062] The specific implementation steps of step 3 are as follows: In order to solve the problem of deviation in measurement data, the above work has used the model to derive the Gaussian distribution characteristics of the data. In order to improve the training effect of the neural network, a large amount of sample data needs to be generated by using the Monte Carlo sampling algorithm. However, these generated data are not necessarily of good quality, so data screening is performed according to different confidence levels, and the significance levels of 95% and 85% are set respectively to effectively screen the state quantity and measurement quantity data. Through this process, a screened data set can be obtained to ensure that the data used meets the requirements of model training in terms of quality and reliability, thereby improving the accuracy and effectiveness of subsequent analysis.

[0063] The specific implementation steps of step 4 are as follows: In order to ensure the rationality of the screened data, a flow equation is established on the basis of the gas grid and the power grid, and the data is verified to ensure the rigor of the neural network training process. First, the flow calculation equation of the gas grid is established, including the electric drive compressor flow equation, the fuel drive compressor flow equation, the branch pipeline flow equation, the node flow equation and the gas grid demand flow equation. Then, the flow calculation equation of the power grid is established, including the active power flow equation and the reactive power flow equation. Finally, the coupling flow equation of the building-level integrated energy system is established to meet the power balance of the coupling nodes.

[0064] The specific implementation steps of step 5 are as follows: After the data set was verified by the flow, a data-driven model based on Bi-LSTM was established. The first 70% of the data set was used as the training set, and the last 30% was used as the test set. The training was carried out under the conditions of single estimation, total estimation, single network and total network, respectively. The calculation time of different neural networks for the comprehensive energy system status assessment was recorded. The results showed that Bi-LSTM had the shortest calculation time under different environments, and the calculation time would increase with the increase of training nodes and data. Compared with CNN (Convolutional Neural Network) with faster calculation time, Bi-LSTM reduced the calculation time by about 40%.

[0065] The specific implementation steps of step 6 are as follows: Select Root Mean Square Error , measurement error statistics and the estimated error statistics The state assessment model is evaluated, and it is determined whether the result of the state assessment model is better than the effect of directly using the measurement data for evaluation. The evaluation process is shown in the following formula. The root mean square error comparison between the existing network and the network of the present invention is as follows: Figure 4 The evaluation results are shown in Table 7: (35) (36) (37).

[0066]

[0067] In summary, an integrated energy system state assessment method based on multi-source data fusion can effectively solve the problems of low redundancy of measurement data and unobservable structure, and improve the accuracy of state estimation of building-level integrated energy systems.

[0068] Example 4 In another preferred embodiment, based on the above embodiments 1, 2, and 3, the state estimation is performed by taking the building-level integrated energy system in different regions and seasons as an example, and the state evaluation is performed on 4 building-level integrated energy systems in Guangdong, Jiangsu, Shandong, and Liaoning in March, June, September, and December respectively. The four building-level integrated energy systems include 18-node gas grids and 36-node power grids. Each system includes 4 gas compressor nodes, including two gas compressor nodes and two electric compressor nodes. The system coupling node includes 2 gas turbine nodes. The measurement device node configuration is shown in Table 8: Table 8 Measurement device node configuration

[0069] A comprehensive energy system state assessment method based on multi-source data fusion, comprising: Step 1: In order to achieve accurate state estimation of the integrated energy system, a measurement model that comprehensively considers the interaction between the electrical network and the gas network is established; Step 2: In view of the state fluctuations of the gas grid and the power grid during operation, the Bayesian learning method is used to fit the probability distribution of the measurement data of the integrated energy system; Step 3: Generate a large amount of data using the Monte Carlo sampling algorithm for training the deep learning network; Step 4: Use the generated data to construct the power flow equation of the integrated energy system to verify the rationality of the data; Step 5: Establish a data-driven integrated energy system state estimation model based on Bi-LSTM.

[0070] Step 6: Use various indicators such as root mean square error and computational efficiency to compare and analyze, and conduct a comprehensive performance evaluation of the developed state assessment method.

[0071] This example selects a building-level integrated energy system in Liaoning as the research object, and the specific implementation method is as follows: The specific implementation steps of step 1 are as follows: Step 1.1: Establish the mapping relationship between the measurement vector and the state vector of the electric-gas coupling system; Each integrated energy system includes a measurement vector, a state vector and an injection vector. The injection vector directly affects the state vector of the system, and the measurement vector includes the injection vector. Therefore, there is a mapping relationship between the measurement vector and the state vector. The mapping relationship is expressed by the following formula.

[0072] (39) In the formula, Indicates measurement vector, Indicates State vector.

[0073] The least squares estimation is very susceptible to residuals in the process of solving state estimation. If the residuals are too large, the estimation accuracy will be reduced. Huber can The function reduces the impact of excessive residuals on estimation accuracy, so Huber robust estimation is used for state assessment. The process is as follows: (4) Step 1.2: Build a probability model that includes measurement noise; Because the accuracy of the sensor is not high, there will be a certain amount of noise in the measurement data. Therefore, a probability model of data noise is established, and a probability measurement model of the integrated energy system is further established: (5) (6) (7) (8) The specific implementation steps of step 2 are as follows: Step 2.1: Establish a Gaussian inference model for the gas grid of the integrated energy system; In the gas grid part of the building-level integrated energy system, the measurement data is considered to follow a Gaussian distribution. Before building a Gaussian inference model, the mean and variance of the measurement data need to be verified. At the same time, the upper and lower limits of the uniform distribution and the standard deviation of the half-normal distribution need to be set. The rationality of these preset parameters is evaluated through the kernel density estimation method, and the optimal Gaussian distribution is selected as the final distribution of the branch measurement data by comparing the distribution of the actual measurement data multiple times. The modeling method of each building-level integrated energy system is the same in different seasons.

[0074] Step 2.2: Establish a non-Gaussian distribution circle model of the integrated energy system power grid; Since grid parameters may change significantly in a short period of time and different devices have different data collection cycles, for example, the cycle of the SCADA system is 3 to 5 seconds, the PMU is 5 to 30 seconds, and the AMI is 5 to 30 milliseconds, it is necessary to fuse multi-source data at different time scales; for this purpose, the Gaussian mixture model and the expectation-maximization algorithm can be combined to solve the non-Gaussian distribution circle model, thereby obtaining the distribution parameters of the grid measurement vector of the building-level integrated energy system.

[0075] The specific implementation steps of step 3 are as follows: In order to solve the problem of deviation in the measured data, the Gaussian distribution characteristics of the data have been derived through the model. In order to improve the training effect of the neural network, a large amount of sample data is generated by using the Monte Carlo sampling algorithm; since the quality of these data may vary, the state quantity and measurement quantity data are screened according to the significance level of 99% and 85%, thereby obtaining a screened data set to ensure that the data quality and reliability meet the requirements of model training and improve the accuracy and effectiveness of subsequent analysis.

[0076] The specific implementation steps of step 4 are as follows: To ensure the rationality of the screened data, a power flow equation is established on the basis of the gas grid and the power grid for data verification and to enhance the rigor of the neural network training. First, the power flow calculation equation of the gas grid is established, including the electric-driven compressor, fuel-driven compressor, branch pipeline, node flow and gas grid demand flow equation. Then, the active and reactive power power flow equations of the power grid are established, and finally, the coupled power flow equation of the building-level integrated energy system is constructed to ensure the power balance of the coupled nodes.

[0077] The specific implementation steps of step 5 are as follows: After the power flow verification is completed, a data-driven model based on Bi-LSTM is established, and the first 70% of the data set is used as a training set and the last 30% as a test set; training is carried out under the conditions of single estimation, total estimation, single network and total network, and the calculation time of different neural networks in the integrated energy system status assessment is recorded; the results show that Bi-LSTM has the shortest calculation time in each environment, and increases with the increase of training nodes and data volume. Compared with CNN with faster calculation time, Bi-LSTM reduces the calculation time by about 35%.

[0078] The specific implementation steps of step 6 are as follows: Root mean square error , measurement error statistics and the estimated error statistics The state assessment model is evaluated and the final method for the state assessment of the integrated energy system is determined. The evaluation process is shown in the following formula. The error rate of voltage estimation of each power network node by different models is as follows: Figure 5 As shown in the figure, the error rate of each gas network node state estimation is as follows: Figure 6 The evaluation results are shown in Table 9: (35) (36) (37).

[0079]

[0080] In summary, an integrated energy system state assessment method based on multi-source data fusion can effectively solve the problems of low redundancy of measurement data and unobservable structure, improve the accuracy of state estimation of building-level integrated energy systems, and has good performance in state estimation of building-level integrated energy systems in different regions and seasons.

[0081] In summary, the present invention provides a method for evaluating the state of an integrated energy system based on multi-source data fusion, which solves the technical problems existing in the state evaluation of an integrated energy system, such as low data redundancy, low system observability, large measurement error and poor time synchronization. The present invention establishes an electric-gas coupled measurement model, fits the measurement probability distribution of the integrated energy system based on Bayesian learning, generates measurement data samples of the integrated energy system using Monte Carlo, constructs the tidal flow equation of the integrated energy system for data verification, and establishes a data-driven integrated energy system state estimation model based on LSTM. This series of steps together constitutes a complete and efficient state evaluation system. In terms of model construction, the scheme adopts a variety of models, such as Gaussian inference model, Gaussian mixture distribution model and LSTM model. The application of these models in the field of integrated energy system state evaluation is unprecedented. Through the combined use of these models, the scheme realizes accurate estimation of the state of the integrated energy system, showing a high degree of creativity. The present invention uses multi-source data fusion, The use of Gaussian inference and Gaussian mixture models and the combination of Monte Carlo sampling and LSTM models effectively solve the problems of low redundancy of measurement data of the integrated energy system, unobservable structure and large measurement error, and improve the accuracy and efficiency of state evaluation; In general, the present invention uses probability distribution to process measurement data by proposing a Gaussian inference model based on Bayesian learning and a non-Gaussian distribution circle model, thereby improving the quality of measurement data. At the same time, in view of the problem of low observability of measurement data, a data-driven integrated energy system state estimation model is established, which effectively reduces the state estimation error and realizes the state estimation of the global network of the integrated energy system. Finally, in view of the problem of low data redundancy, the Bi-LSTM model is used to estimate the state of the integrated energy system, thereby improving the robustness and accuracy of state estimation; In addition, the scheme also has good scalability and adaptability, and can flexibly cope with integrated energy systems of different scales and structures, providing strong technical support and guarantee for the future construction of smart grids and energy Internet.

Claims

1. A comprehensive energy system status assessment method based on multi-source data fusion, characterized in that: The method comprises the following steps: Step 1: Establish the measurement model of electric-gas coupling; Step 2: Generate integrated energy system measurement data samples using the Monte Carlo sampling algorithm for subsequent neural network training; Step 3: Establish a data verification mechanism based on the power flow equation, perform power flow verification on the generated measurement data samples, and retain the data that can make the power flow converge for neural network training; Step 4: Establish a data-driven integrated energy system state estimation model based on the long short-term memory network Bi-LSTM, use the verified measurement data as the input of the neural network, and use the state data as the output of the neural network for training; Step 5: Use the root mean square error, measurement error statistics, and estimation error statistics to evaluate the state assessment model to determine whether the effect of the state assessment model is better than that of directly using the measurement data for evaluation.

2. The method for evaluating the state of an integrated energy system based on multi-source data fusion according to claim 1 is characterized in that: The specific steps of Step 1 include: Step 1.1: Establish a mapping relationship between the measurement vector and the state vector of the electric-gas coupling system, wherein the measurement vector includes data of node voltage, power, branch power, and pipeline pressure, and the state vector includes data of gas grid pressure, grid voltage, pipeline flow, and compressor flow; Step 1.2: Perform probability distribution fitting on the measured data of the integrated energy system based on Bayesian learning, including establishing a Gaussian inference model for the gas grid and a Gaussian mixture distribution model for the power grid.

3. The method for evaluating the state of an integrated energy system based on multi-source data fusion according to claim 2 is characterized in that: The specific steps of Step 1.1 include: Step 1.1.1: Establish the mapping relationship between the measurement vector and the state vector of the electric-gas coupling system. Assume that the vector injected into the electric-gas coupling system is , the measurement vector is , the state vector is , the mapping relationship established between them is as follows: （1）； In the formula, Inject vector and state vector mapping for system nodes; The measurement vector includes the injection vector or branch power, node voltage, and gas grid pressure. The mapping relationship between the measurement vector and the state vector is expressed as follows: （2）； （3）； In the formula, and They are the measurement functions of gas grid and power grid equipment in the electricity-gas coupling system, respectively; and are the measurement vector sets of the gas grid and the power grid respectively, and are the state vector sets of the gas grid and the power grid respectively, and are the parameters of the gas grid and power grid state estimation models, and are the noise of the gas grid and power grid measurement vectors, is the number of measurement data sets; Next, the Huber robust estimation is used for state assessment. The process is as follows: （4）； In the formula, is a function; Step 1.1.2: Establish a probability model that includes measurement noise; Assuming that the data collected from each device are independent of each other, establish the following probability model: （5）； In the formula, , , , They are vectors , , , No. elements, is the number of state vectors; Using adjustable parameters of the state estimation model To further express the mapping relationship, based on the above probability model, a comprehensive energy system state estimation probability model is established: （6）； （7）； （8）； In the formula, is the mathematical expectation, is the state estimate of the measurement vector, is the hyperparameter of the measured data, For the The hyperparameters of the measured data.

4. The method for evaluating the state of an integrated energy system based on multi-source data fusion according to claim 2 is characterized in that: The specific steps of Step 1.2 include: Step 1.2.1: Establish a Gaussian inference model for the gas grid of the integrated energy system; Measurement data for the gas grid part of the integrated energy system The mean is , the variance is Gaussian distribution, that is: （9）； In the formula, is the Gaussian distribution function; When using the Gaussian inference model to calculate the expected measurement data, the mean and variance need to be verified, namely: （10）； In the formula, is the uniform distribution function, and They obey the upper and lower bounds of the uniform distribution respectively: （11）； In the formula, is the distribution function of the half-normal distribution, for The mean of the half-normal distribution is for The variance of the half-normal distribution of ; assumed , , The values ​​of the three parameters are calculated and the prior distribution of the parameters is calculated. The kernel density estimation method and the central limit theorem are used to determine whether the set parameters are reasonable. The kernel density estimation method is expressed as follows: （12）； In the formula, is the kernel density function, is the kernel function, for The number of is the bandwidth of the kernel density function; The posterior distribution of the gas network measurement data is calculated using the prior distribution of the set parameters, and the most suitable and , further calculate the probability distribution of the original data; Step 1.2.2: Establish a non-Gaussian distribution circle model of the integrated energy system power grid; The Gaussian mixture model and expectation maximization algorithm are combined, the voltage amplitude, current amplitude, branch power and node power are collected by the SCADA system, the voltage and current vectors are measured by the PMU, and the voltage, current amplitude, branch, node power and user power consumption information are collected by the AMI, and the data are fused at multiple time scales. Through the calculation of Gaussian mixture model and expectation maximization algorithm, the non-Gaussian distribution circle model is solved to obtain the distribution parameters of the measurement vector of the integrated energy system power grid: （13）； （14）； In the formula, is the measurement vector and number, is the mean value after multi-scale fusion of data, , is the identity matrix, is the correlation coefficient of the autoregressive moving average model, is the variance of the grid measurement vector.

5. The method for evaluating the state of an integrated energy system based on multi-source data fusion according to claim 1 is characterized in that: The specific steps of Step 2 using Monte Carlo to generate comprehensive energy system measurement data samples are: The Monte Carlo sampling algorithm is used to generate a large amount of data on the gas grid and power grid of the integrated energy system. The Monte Carlo sampling algorithm constructs a data set through the Markov chain convergence theorem. The state transition process of the Markov chain is expressed by the following formula: （15）； In the formula, They are The measurement vector of the moment; Assume that the measurement data at the previous moment is only related to the measurement data at the current moment and the noise of the measurement data The probability distribution of , in this case, the noise in the measured data is proportional to the vector injected into the system is independent and identically distributed, then the probability distribution of the measurement vector Can be distributed by injecting vectors into the system With noise distribution The system's measurement data can be obtained and the generated data can be divided into measurement samples according to the following process and state sample : （16）； （17）； In the formula, , , and , , The state quantity and measurement quantity confidence level, upper and lower sample limits; The generated data are screened, and the measurement data with a deviation of about 10% are set as data that do not meet the requirements. The state vector is selected at a significance level of 95%, while the measurement component is screened at a significance level of 85%, to obtain the data set of the measurement vector and the data set of the state vector.

6. The method for comprehensive energy system status assessment based on multi-source data fusion according to claim 1 is characterized in that: The Step 3 specifically uses the generated data to perform flow verification, establishes a flow equation based on the power grid and gas grid of the integrated energy system, performs flow verification on the data, and retains data that can make the flow converge for neural network training. The process of performing flow verification on the data includes: Step 3.1: Establish the flow calculation equation of the gas network, which includes the flow equation of the compressor and the flow equation of the branch pipeline. The compressors include electric drive compressors and fuel drive compressors. The flow equation of the electric drive compressor is as follows: （18）； In the formula, The power consumed by the compressor, is the number of compressors, is the pipeline flow rate of the electric drive compressor, is the number of pipelines, For the The node and The ratio of the air pressure at each node; The fuel driven compressor equation is as follows: （19）； In the formula, is the ratio of fuel consumption of fuel-driven compressor, is the calorific value of natural gas, is the gas-to-electricity efficiency; The flow equation for the branch pipe is as follows: （20）； In the formula, is the pipeline flow rate, For branch The resistance of the pipeline, For the The pressure of each node; The node flow equation is as follows: （21）； In the formula, For the The traffic of each node, is the node matrix, for and The column vector composed of The gas network demand flow equation is as follows: （22）； In the formula, For the The required flow of each node is is the demand flow matrix; The power generation capacity of gas power plants is as follows: （23）； In the formula, is the power generation capacity of the gas power plant, is the number of gas-fired power plants; Step 3.2: Establish the power flow equation of the power grid, including the active power flow equation and the reactive power flow equation, as shown in the following formula: （24）； （25）； In the formula, and Respectively The active power flowing into each node and the active load power, and Respectively The reactive power and reactive load power flowing into each node, and Respectively The node and The conductance and susceptance between nodes, It is The voltage of the node, For the The node and The voltage phase angle between nodes; Step 3.3: Establish the coupled power flow equation of the integrated energy system. While satisfying the power flow equations of the power grid and gas grid, it is also necessary to satisfy the power balance of the coupled nodes, as shown in the following formula: （26）； （27）； （28）； In the formula, and are the balance coefficients of the power grid and gas grid, and are the grid energy loss and power generation respectively, and They are gas grid energy consumption and gas-fired power generation respectively.

7. The method for evaluating the state of an integrated energy system based on multi-source data fusion according to claim 1 is characterized in that: The specific scheme of the Step 4 for establishing a data-driven integrated energy system state estimation model based on the long short-term memory network Bi-LSTM is to establish a multi-input-single-output long short-term memory network model based on the mapping relationship between the measurement vector and the state vector of the integrated energy system, use the generated measurement data as the input data of the neural network, and use the operating state data of the integrated energy system that has been verified by the flow as the output data of the neural network, construct a Bi-LSTM data set, use the first 70% of the data set as training samples, and the last 30% as test samples, and use the adaptive moment estimation optimization algorithm to train the neural network. The structure of each LSTM consists of one or more units with forgettable and memory functions, which are input gate, output gate and forget gate respectively.

8. The method for evaluating the state of an integrated energy system based on multi-source data fusion according to claim 7 is characterized in that: The implementation of the Bi-LSTM function of the long short-term memory network includes the following processes: The forget gate controls the content of the cell state of the previous layer. Through the sigmoid activation function, the content of the cell of the previous layer is forgotten and retained. The output of the forget gate is completed by the following formula: （29）； The input gate is used to process the input of the current sequence position and determine the information that needs to be updated. This process consists of two parts. One part is the remaining memory information after passing through the forget gate, and the other part is the newly input information. This process is completed by the following formula: （30）； （31）； After processing through the forget gate and the input gate, the state of the cell can be updated: （32）； In the formula, Indicate the information you wish to delete, Indicates newly added information; The output gate is used to determine the output based on the information stored in the cell, that is, to selectively output the information stored in the cell state, similar to the update process of the two parts of the input gate, which is implemented by the following formula: （33）； （34）； In the formula, Represents the input data at the current moment, , Respectively represent the internal state of the previous moment and the current moment, , Represent the hidden layer states at the previous moment and the current moment respectively, , Respectively and Activation function, , and Respectively represent the calculation variables of the forget gate, input gate and output gate at the current moment, Indicates the unit state of the input at that moment; , , are the weight matrices of the forget gate, input gate, and output gate respectively. is the input unit state weight matrix, is the input unit state bias term, , , They are the output bias vectors of the forget gate, input gate and output gate respectively.

9. The method for comprehensive energy system status assessment based on multi-source data fusion according to claim 1 is characterized in that: The Step 5 specifically selects the root mean square error , measurement error statistics and the estimated error statistics Evaluate the condition assessment model and determine whether the results of the condition assessment model are better than those of direct assessment using measurement data. The evaluation process includes: （35）； （36）； （37）； In the formula, is the value of the root mean square error, is the number of Bi-LSTM network data sets, For the The next state estimates the true value, For the The next state estimate prediction value, represents the total number of simulations, is the number of measured variables, Indicates The measurement of measured variables, is the standard deviation of the noise, For the The true value of the measured variable, For the The measured variable Sub-estimate.

10. A method for evaluating the state of an integrated energy system based on multi-source data fusion according to any one of claims 1 to 9, characterized in that: The integrated energy system is a park-level or building-level integrated energy system.

Images