An optimal scheduling method for integrated energy system based on ALIF-GRU-MTL source-load joint prediction

Abstract

The present invention discloses an integrated energy system optimization scheduling method based on ALIF-GRU-MTL source-load joint prediction. In order to solve the problem that the existing integrated energy system optimization scheduling method is not closely related to the system source-load prediction process, the method comprises the following steps: S1: establishing the basic structure of the integrated energy system and establishing a mathematical model according to the equipment type; S2: performing preprocessing of the source-load data of the integrated energy system, and completing feature engineering by constructing derived features using the original features; S3: obtaining prediction models of different source-load component sequences based on ALIF-GRU-MTL, summing and reconstructing the prediction results of the obtained source-load component sequence model and denormalizing them to obtain the final source-load prediction result; S4: establishing an integrated energy system optimization scheduling model and solving the optimization scheduling model. The invention has the beneficial effects that the use of multi-task learning can better realize parameter sharing between source-load prediction tasks and improve the accuracy of source-load prediction of the integrated energy system.

Description

Technical Field

[0001] The present invention relates to the field of integrated energy system optimization and scheduling, and in particular to an integrated energy system optimization and scheduling method based on ALIF-GRU-MTL source-load joint prediction, providing a method for integrated energy system optimization and scheduling. Background Art

[0002] As the main organizational form and important physical carrier of the energy Internet, the integrated energy system breaks the barriers between traditional energy systems, couples multiple energy sources such as cold, heat, and electricity, can achieve multi-energy coordinated optimization and cascade utilization of energy, improve the consumption of renewable energy, and represents an important commanding height in the future energy field. The multi-load randomness of the integrated energy system is high, and the intermittent and volatile output of wind turbines and photovoltaics is large, resulting in the widespread phenomenon of system abandonment of wind and light. How to reasonably arrange the operation plan of each energy supply equipment to meet the supply and demand balance of the system and achieve the economic operation of the system has important practical significance. Accurate multi-load forecasting and source-end renewable energy output forecasting are important prerequisites for optimizing the scheduling of the integrated energy system, and are an important research direction for the operation and management of the integrated energy system. This can reasonably arrange the work plans of each device in the system to ensure the source-load supply and demand balance within the system. On the basis of accurate source-load joint forecasting, an optimized scheduling model for the integrated energy system is established, and energy storage and multi-energy coupling equipment are used to optimize the system operation status, improve the comprehensive utilization efficiency of energy, reduce energy use costs, and increase the consumption rate of renewable energy, providing a theoretical basis and technical support for energy conservation, emission reduction, and the development of a green economy.

[0003] There are many existing methods for load forecasting of integrated energy systems and renewable energy output forecasting, but most of them only involve method research and rarely combine them with system optimization and scheduling application scenarios.

[0004] A "node load prediction method considering source-load coordination" disclosed in Chinese patent literature, with announcement number CN108985508B, includes: obtaining node load historical data, including actual output of renewable energy on the load bus and actual load of the node; non-parametric estimation fits the renewable energy output curve to restore the theoretical output of renewable energy under power-limiting state; restore the node theoretical load under load regulation state; take the node theoretical load and the theoretical output of renewable energy as input, and the actual node load as output, to train the neural network; obtain the node total load prediction model of source-load coordination; make preliminary predictions of the node theoretical load and the theoretical output of renewable energy based on the support vector machine model; input the preliminary prediction results into the node total load prediction model for prediction, and obtain the final node load prediction value. Its shortcomings are: the existing integrated energy system optimization scheduling method is not closely related to the system source-load prediction process, and the coupling characteristics of the system source-load are not fully explored when predicting the system multi-load and source-end renewable energy output. Summary of the invention

[0005] The present invention is mainly intended to solve the problem that the existing integrated energy system optimization scheduling method is not closely connected with the system source-load prediction process, and the system source-load coupling characteristics are not fully exploited when predicting the system multi-load and source-end renewable energy output. A method for optimizing the scheduling of an integrated energy system based on ALIF-GRU-MTL source-load joint prediction is provided.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] An optimal scheduling method for an integrated energy system based on ALIF-GRU-MTL source-load joint prediction includes the following steps:

[0008] S1: Establish the basic structure of the integrated energy system and build a mathematical model based on the equipment type;

[0009] S2: Preprocess the source-load data of the integrated energy system and construct derived features based on the original features to complete feature engineering;

[0010] S3: Based on ALIF-GRU-MTL, the prediction models of different source-load component sequences are obtained, the prediction results of the obtained source-load component sequence models are summed, reconstructed and denormalized to obtain the final source-load prediction result;

[0011] S4: Establish an optimal scheduling model for the integrated energy system and solve the optimal scheduling model.

[0012] Among them, ALIF refers to Adaptive Local Iterative Filtering, GRU refers to Gate Recurrent Unit, and MTL refers to Multi-Task Learning. Through ALIF-GRU-MTL source-load joint prediction, the source-load prediction of the integrated energy system is closely linked with the optimized scheduling, and the source-load coupling characteristics of the system can be fully considered.

[0013] Preferably, the step S1 comprises the following steps:

[0014] S11: Establish the basic structure of the system according to the actual equipment deployment and energy flow type of the integrated energy system, and clarify the energy flow and coupling relationship between different equipment;

[0015] S12: Divide the system into cold energy flow, thermal energy flow, and electrical energy flow according to the energy flow type;

[0016] S13: Divide the devices in the system into source devices, multi-energy coupling devices, and energy storage devices according to the device type;

[0017] S14: Establish mathematical models for the source device, the multi-energy coupling device, and the energy storage device respectively according to the device operation characteristics.

[0018] Different types of equipment are modeled separately. Based on the analysis of equipment operation characteristics, micro gas turbines and gas boilers are modeled to complete the modeling of source-end equipment; waste heat boilers, absorption refrigerators and electric refrigerators are modeled to complete the modeling of multi-energy coupling equipment; batteries, heat storage devices and cold storage devices are modeled to complete the modeling of energy storage equipment.

[0019] Preferably, the source-end device in step S13 includes a micro gas turbine and a gas boiler, the multi-energy coupling device includes a waste heat boiler, an absorption refrigerator and an electric refrigerator, and the energy storage device includes a battery, a heat storage device and a cold storage device.

[0020] Preferably, the preprocessing of source-load data of the integrated energy system in step S2 comprises the following steps:

[0021] S21: Acquire historical load data and historical output data of the integrated energy system, and acquire time information and meteorological data corresponding to the load data and the output data;

[0022] S22: performing data cleaning on the historical load data, historical output data, time information and meteorological data in step S21, including detecting and processing missing values ​​and abnormal values;

[0023] S23: Perform minimum-maximum normalization on the continuous data in the data processed in step S22, and perform One-Hot encoding on the discrete data in the data processed in step S22.

[0024] Preferably, completing feature engineering according to the original feature structure in step S2 includes the following steps:

[0025] S24: performing coupling characteristic analysis on the integrated energy system data that has undergone data preprocessing in step S23, and selecting meteorological data, time information, historical load data characteristics, and historical output data characteristics with high relevance;

[0026] S25: Use the data features after screening in step S24 to construct derived features to complete feature engineering.

[0027] The coupling characteristics between the historical source-load data, meteorological data and time information in the system are analyzed by drawing a heat map using the Pearson correlation coefficient to measure the magnitude of the correlation between the two variables.

[0028] Preferably, the load data includes cold, heat and electricity load data, the output data includes photovoltaic and wind turbine output data, the time information includes time, week and month, and the meteorological data includes temperature and humidity data, wind speed data, irradiance data and precipitation data.

[0029] A variety of historical load data, historical output data, time information and meteorological data are selected to perform multivariable control to ensure the accuracy and applicability of the function model.

[0030] Preferably, the prediction model for obtaining different source-load component sequences based on ALIF-GRU-MTL in step S3 comprises the following steps:

[0031] S31: Arrange the source-load data of the integrated energy system preprocessed in step S2 into a time series, and decompose them using the ALIF method to obtain component sequences with different frequency bands of historical load data and historical output data;

[0032] S32: The source-load component sequence obtained in step S31 is reconstructed using sample entropy to obtain period, fluctuation and trend sequences, where the period sequences of cold, hot and electric load data and photovoltaic and wind turbine output data are Q 1 , H 1 , P 1 , The fluctuation sequences of cold, hot and electric load data and photovoltaic and wind turbine output data are Q 2 , H 2 , P 2 , The trend sequences of cold, heat, and electricity load data as well as photovoltaic and wind turbine output data are Q 3 , H 3 , P 3 , The superscript i=1, 2, 3 represent the periodic sequence, the fluctuation sequence, and the trend sequence respectively;

[0033] S33: combining the source-load cycle sequence, fluctuation sequence and trend sequence obtained in step S32 with the corresponding meteorological data and time information to obtain a data set, which is divided into a training set and a test set, which are used for training and testing the source-load joint prediction model respectively;

[0034] S34: Input the training set obtained in step S33 into the model for training. Based on GRU, MTL is used to perform multi-task learning training on the cycles, fluctuations and trend sequences of cold, hot and electric loads as well as photovoltaic and wind turbine outputs, to obtain prediction models for different source-load component sequences.

[0035] Preferably, the step S3 of summing up, reconstructing and denormalizing the prediction results of the source-load component sequence model to obtain the final source-load prediction result comprises the following steps:

[0036] S35: Using the prediction model training and prediction obtained in step 34, the data of the previous m days are used to predict the cold, heat, and electricity loads and photovoltaic and wind turbine outputs of 24 points in the next day. The input and output data of the model at time t in the next day are predicted as follows:

[0037]

[0038]

[0039] Where: time(t) = [hour, week, month] is the time information including time, week and month; weather(t) = [tem, hum, ws, ir, rain] is the meteorological characteristics of the source-load time input model using the day-ahead weather forecast data as the forecast time t, which are temperature, humidity, wind speed, irradiance and precipitation; They represent the cold, hot, and electric loads as well as the photovoltaic and wind turbine output characteristics of the same sequence, as shown below:

[0040]

[0041]

[0042]

[0043]

[0044]

[0045] In the formula, the source-load data at the same time t in the previous three days are used to input the model, that is, m is 3, which is the source-load data at the same time t in the previous three days. The abstract representation of the model training process is as follows:

[0046] Y(t)=GRUi(X(t))

[0047] Where: GRUi represents the prediction models of source-load cycle sequence, fluctuation sequence and trend sequence obtained by multi-task learning training for different sequences;

[0048] S36: Use the trained model GRUi to predict the source-load corresponding sequence i at the time k to be predicted on a certain day in the test set. The prediction results of the cold, hot, and electric loads, as well as the photovoltaic and wind turbine output sequences at the time k can be expressed as:

[0049]

[0050] Where: are the cold, hot, electric loads, and photovoltaic and wind turbine output prediction values ​​of sequence i at time k respectively; the cold, hot, electric loads, and photovoltaic and wind turbine output prediction values ​​of each sequence are summed and reconstructed respectively, and the predicted value of source-load at time k after reconstruction can be expressed as:

[0051]

[0052]

[0053] Where: Q(k), H(k), P(k), P PV (k), P WT (k) are the predicted values ​​of cold, hot, electric loads, and photovoltaic and wind turbine output at time k after summing and reconstruction. The source-load prediction values ​​after summing and reconstruction are denormalized to obtain the final source-load prediction values.

[0054] Preferably, the establishment of the comprehensive energy system optimization scheduling model in step S4 includes the following steps:

[0055] S41: using the prediction model based on ALIF-GRU-MTL to obtain different source-load component sequences in step S3, the cold, heat and electricity load data and the predicted values ​​of photovoltaic and wind turbine output data for 24 time periods in the future day are obtained, and energy market information such as natural gas prices and electricity prices and system equipment operating parameters are imported;

[0056] S42: Using the source-load prediction results and equipment operating parameters in step S41, combined with the equipment mathematical model established in step S1, the objective function of the optimization scheduling model is established according to the economy, environmental protection and safety of the integrated energy system operation. The cycle of the optimization scheduling is one day and the time interval is 1 hour; S43: Quantifying the two goals of economy and environmental protection into costs to establish a unified objective function, as follows:

[0057]

[0058] Where: C IES represents the total daily operating cost of the system; C grid (t) is the cost of purchasing electricity from the external power grid; C gas (t) Cost of purchasing natural gas for the system; C operation (t) is the operation and maintenance cost of each device in the system; C environment (t) is the environmental penalty fee of the system, where both cost and fee are in RMB;

[0059] S44: The economic efficiency of the system operation in step S43 is mainly determined by the system's electricity and gas purchase costs and the operating and maintenance costs of each device in the system. The calculation formula for the system's external power grid electricity purchase cost is as follows:

[0060] C grid (t) = P grid (t).c grid (t)

[0061] Where: P grid (t) is the electric power input from the external power grid to the integrated energy system during period t, in kW; c grid (t) is the electricity price in period t, in RMB / kWh; the cost of purchasing natural gas for the integrated energy system consists of the consumption of micro gas turbines and gas boilers. The specific calculation formula is as follows:

[0062] C gas (t) = C MT (t)+C GB (t)

[0063] Where: C MT (t) is the cost of natural gas consumed by the micro gas turbine, C GB (t) is the cost of natural gas consumed by the gas boiler, where the cost unit is RMB; the system operation and maintenance cost is proportional to the unit output power of each device;

[0064] S45: The environmental friendliness of the system operation in step S43 is mainly determined by the system's pollution emission control cost and the penalty for abandoning solar and wind power, wherein the pollution emission control cost is determined by the amount of natural gas burned and the amount of electricity purchased from the external power grid. The goal of the dispatch model also considers the consumption of renewable energy and establishes a penalty for abandoning solar and wind power. The calculation formula is as follows:

[0065] C environment (t) = C pollution (t)+C pv (t)+C wt (t)

[0066] Where: C pollution (t) is the pollution emission cost of the system; C pv (t) is the penalty for abandoning light; C wt (t) is the penalty for wind curtailment, where both cost and penalty are in yuan.

[0067] Preferably, solving the optimization scheduling model in step S4 comprises the following steps:

[0068] S46: Using the source-load prediction result of step S41 and the equipment operating parameters, combined with the comprehensive energy system equipment model established in step S1, the system constraints are established, including cold, hot and electric energy flow balance constraints, equipment output constraints and energy storage constraints;

[0069] S47: The integrated energy system optimization scheduling model is composed of the system optimization scheduling objective function established in steps S43, S44, and S45 and the constraint conditions established in step S46;

[0070] S48: Use the YALMIP toolbox to call CPLEX to solve the optimization scheduling model, and output the equipment output and daily operating cost of the system optimization scheduling in each period.

[0071] The beneficial effects of the present invention are:

[0072] (1) This method fully considers the coupling characteristics of multiple loads and renewable energy output of the integrated energy system. The multi-task learning method can better realize the parameter sharing between source-load prediction tasks, improve the accuracy of source-load prediction of the integrated energy system, and lay the foundation for the optimal scheduling of the integrated energy system.

[0073] (2) This method overcomes the problem that the existing integrated energy system optimization and scheduling methods are not closely linked to the system source-load prediction process. It closely links the application scenarios with the source-load prediction results, and comprehensively considers the economy, environmental protection and safety of the system operation during the system optimization and scheduling process. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a schematic diagram of the source-load joint prediction model based on the ALIF-GRU-MTL integrated energy system in the present invention.

[0075] Figure 2 It is a schematic diagram of the structure of the comprehensive energy system of the present invention.

[0076] Illustration: 1- source device, 2- multi-energy coupling device, 3- heat energy flow, 4- load data, 5- energy storage device, 6- cold energy flow, 7- electric energy flow, 8- output data. DETAILED DESCRIPTION

[0077] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0078] This example is a comprehensive energy system of an industrial park in the north. The load types included in this energy system include cooling load, heating load, and electric load. The renewable energy sources at the source end include wind turbines and photovoltaics. This method closely links the source-load prediction and optimal scheduling of the comprehensive energy system, and considers the source-load coupling characteristics of the system, taking the optimal scheduling method of the comprehensive energy system as an example.

[0079] like Figure 1 and Figure 2 As shown, a comprehensive energy system optimization scheduling method based on ALIF-GRU-MTL source-load joint prediction includes the following steps:

[0080] S1: Establish the basic structure of the integrated energy system and build a mathematical model based on the equipment type;

[0081] S2: Preprocess the source-load data of the integrated energy system and complete feature engineering based on the original feature construction;

[0082] S3: Based on ALIF-GRU-MTL, the prediction models of different source-load component sequences are obtained, the prediction results of the obtained source-load component sequence models are summed, reconstructed and denormalized to obtain the final source-load prediction result;

[0083] S4: Establish an optimal scheduling model for the integrated energy system and solve the optimal scheduling model.

[0084] Among them, ALIF refers to Adaptive Local Iterative Filtering, GRU refers to Gate Recurrent Unit, and MTL refers to Multi-Task Learning. Through ALIF-GRU-MTL source-load joint prediction, the source-load prediction of the integrated energy system is closely linked with the optimized scheduling, and the source-load coupling characteristics of the system can be fully considered.

[0085] Step S1 includes the following steps:

[0086] S11: Establish the basic structure of the system according to the actual equipment deployment and energy flow type of the integrated energy system, and clarify the energy flow and coupling relationship between different equipment;

[0087] S12: Divide the system into cold energy flow 6, heat energy flow 3, and electric energy flow 7 according to the energy flow type;

[0088] S13: Divide the devices in the system into source end devices 1, multi-energy coupling devices 2, and energy storage devices 5 according to device types;

[0089] S14: Establish mathematical models for the source device, the multi-energy coupling device, and the energy storage device respectively according to the device operation characteristics.

[0090] Different types of equipment are modeled separately. Based on the analysis of equipment operation characteristics, micro gas turbines and gas boilers are modeled to complete the modeling of source-end equipment; waste heat boilers, absorption refrigerators and electric refrigerators are modeled to complete the modeling of multi-energy coupling equipment; batteries, heat storage devices and cold storage devices are modeled to complete the modeling of energy storage equipment.

[0091] In step S13, the source-end equipment includes a micro gas turbine and a gas boiler, the multi-energy coupling equipment includes a waste heat boiler, an absorption refrigerator and an electric refrigerator, and the energy storage equipment includes a battery, a heat storage device and a cold storage device.

[0092] Modeling of various source-end devices, multi-energy coupling devices and energy storage devices is carried out to realize the system's multi-load, fully explore the system's source-load coupling characteristics, and ensure the accuracy and applicability of the function model.

[0093] The integrated energy system source-load data preprocessing in step S2 includes the following steps:

[0094] S21: Obtain historical load data 4 and historical output data 8 of the integrated energy system, and obtain time information and meteorological data corresponding to the load data and output data;

[0095] S22: performing data cleaning on the historical load data, historical output data, time information and meteorological data in step S21, including detecting and processing missing values ​​and abnormal values;

[0096] S23: Perform minimum-maximum normalization on the continuous data in the data processed in step S22, and perform One-Hot encoding on the discrete data in the data processed in step S22.

[0097] In step S2, feature engineering is completed based on the original feature construction, including the following steps:

[0098] S24: performing coupling characteristic analysis on the integrated energy system data that has undergone data preprocessing in step S23, and selecting meteorological data, time information, historical load data characteristics, and historical output data characteristics with high relevance;

[0099] S25: Use the data features after screening in step S24 to construct derived features to complete feature engineering.

[0100] The coupling characteristics between the historical source-load data, meteorological data and time information in the system are analyzed by drawing a heat map using the Pearson correlation coefficient to measure the magnitude of the correlation between the two variables.

[0101] Load data includes cold, heat and electricity load data; output data includes photovoltaic and wind turbine output data; time information includes time, week and month; meteorological data includes temperature and humidity data, wind speed data, irradiance data and precipitation data.

[0102] A variety of historical load data, historical output data, time information and meteorological data are selected to perform multivariable control to ensure the accuracy and applicability of the function model.

[0103] The prediction model of different source-load component sequences obtained based on ALIF-GRU-MTL in step S3 includes the following steps:

[0104] S31: Arrange the source-load data of the integrated energy system preprocessed in step S2 into a time series, and decompose them using the ALIF method to obtain component sequences with different frequency bands of historical load data and historical output data;

[0105] S32: The source-load component sequence obtained in step S31 is reconstructed using sample entropy to obtain period, fluctuation and trend sequences, where the period sequences of cold, hot and electric load data and photovoltaic and wind turbine output data are Q 1 , H 1 , P 1 , The fluctuation sequences of cold, hot and electric load data and photovoltaic and wind turbine output data are Q 2 , H 2 , P 2 , The trend sequences of cold, heat, and electricity load data as well as photovoltaic and wind turbine output data are Q 3 , H 3 , P 3 , The superscript i=1, 2, 3 represent the periodic sequence, the fluctuation sequence, and the trend sequence respectively;

[0106] S33: combining the source-load cycle sequence, fluctuation sequence and trend sequence obtained in step S32 with the corresponding meteorological data and time information to obtain a data set, which is divided into a training set and a test set, which are used for training and testing the source-load joint prediction model respectively;

[0107] S34: Input the training set obtained in step S33 into the model for training. Based on GRU, MTL is used to perform multi-task learning training on the cycles, fluctuations and trend sequences of cold, hot and electric loads as well as photovoltaic and wind turbine outputs, to obtain prediction models for different source-load component sequences.

[0108] In step S3, the prediction results of the source-load component sequence model are summed, reconstructed and denormalized to obtain the final source-load prediction result, which includes the following steps:

[0109] S35: Using the prediction model training and prediction obtained in step 34, the data of the previous m days are used to predict the cold, heat, and electricity loads and photovoltaic and wind turbine outputs of 24 points in the next day. The input and output data of the model at time t in the next day are predicted as follows:

[0110]

[0111]

[0112] Where: time(t) = [hour, week, month] is the time information including time, week and month; weather(t) = [tem, hum, ws, ir, rain] is the meteorological characteristics of the source-load time input model using the day-ahead weather forecast data as the forecast time t, which are temperature, humidity, wind speed, irradiance and precipitation; They represent the cold, hot, and electric loads as well as the photovoltaic and wind turbine output characteristics of the same sequence, as shown below:

[0113]

[0114]

[0115]

[0116]

[0117]

[0118] In the formula, the source-load data at the same time t in the previous three days are used to input the model, that is, m is 3, which is the source-load data at the same time t in the previous three days. The abstract representation of the model training process is as follows:

[0119] Y(t)=GRUi(X(t))

[0120] Where: GRUi represents the prediction models of source-load cycle sequence, fluctuation sequence and trend sequence obtained by multi-task learning training for different sequences;

[0121] S36: Use the trained model GRUi to predict the source-load corresponding sequence i at the time k to be predicted on a certain day in the test set. The prediction results of the cold, hot, and electric loads, as well as the photovoltaic and wind turbine output sequences at the time k can be expressed as:

[0122]

[0123] Where: are the cold, hot, electric loads, and photovoltaic and wind turbine output prediction values ​​of sequence i at time k respectively; the cold, hot, electric loads, and photovoltaic and wind turbine output prediction values ​​of each sequence are summed and reconstructed respectively, and the predicted value of source-load at time k after reconstruction can be expressed as:

[0124]

[0125]

[0126] Where: Q(k), H(k), P(k), P PV (k), P WT (k) are the predicted values ​​of cold, hot, electric loads, and photovoltaic and wind turbine output at time k after summing and reconstruction. The source-load prediction values ​​after summing and reconstruction are denormalized to obtain the final source-load prediction values.

[0127] The establishment of the comprehensive energy system optimization scheduling model in step S4 includes the following steps:

[0128] S41: using the prediction model of different source-load component sequences obtained based on ALIF-GRU-MTL in step S3 to obtain the cold, heat and electricity load data and the predicted values ​​of photovoltaic and wind turbine output data for 24 time periods in the future day, and importing energy market information such as natural gas prices and electricity prices and system equipment operating parameters;

[0129] S42: Using the source-load prediction results and equipment operating parameters in step S41, combined with the equipment mathematical model established in step S1, the objective function of the optimization scheduling model is established according to the economy, environmental protection and safety of the integrated energy system operation. The cycle of the optimization scheduling is one day and the time interval is 1 hour; S43: Quantifying the two goals of economy and environmental protection into costs to establish a unified objective function, as follows:

[0130]

[0131] Where: C IESrepresents the total daily operating cost of the system; C grid (t) is the cost of purchasing electricity from the external power grid; C gas (t) Cost of purchasing natural gas for the system; C operation (t) is the operation and maintenance cost of each device in the system; C environment (t) is the environmental penalty fee of the system, where both cost and fee are in RMB;

[0132] S44: The economic efficiency of the system operation in step S43 is mainly determined by the system's electricity and gas purchase costs and the operation and maintenance costs of each device in the system. The calculation formula for the system's external power grid electricity purchase cost is as follows:

[0133] C grid (t) = P grid (t).c grid (t)

[0134] Where: P grid (t) is the electric power input from the external power grid to the integrated energy system during period t, in kW; c grid (t) is the electricity price in period t, in RMB / kWh; the cost of purchasing natural gas for the integrated energy system consists of the consumption of micro gas turbines and gas boilers. The specific calculation formula is as follows:

[0135] C gas (t) = C MT (t)+C GB (t)

[0136] Where: C MT (t) is the cost of natural gas consumed by the micro gas turbine, C GB (t) is the cost of natural gas consumed by the gas boiler, where the cost unit is RMB; the specific calculation formula is as follows:

[0137]

[0138]

[0139] Where: P MT (t) is the electric power output of the micro gas turbine during period t (kW); η MT (t) is the power generation efficiency of the micro gas turbine corresponding to the period t (%); H GB (t) is the thermal power output of the gas boiler during period t (kW); η GB is the efficiency of the gas boiler (%); c gas (t) is the price of natural gas in period t (yuan / m3); HI is the lower calorific value of natural gas, which is taken as a constant of 9.8 (kWh / m3); the system operation and maintenance cost is proportional to the unit output power of each device, and the specific calculation formula is as follows:

[0140]

[0141] Where: C operation (t) is the operation and maintenance cost of the system; β PV , β WT , β MT , β GB , β WHB , β EC , β AC , β BS , β HS , β CS They are the unit output power operation and maintenance costs (yuan / kWh) of photovoltaic, wind turbine, micro gas turbine, gas boiler, waste heat boiler, electric refrigerator, absorption refrigerator, battery, heat storage device, and cold storage device. The system operation and maintenance costs are proportional to the unit output power of each device.

[0142] S45: The environmental friendliness of the system operation in step S43 is mainly determined by the system's pollution emission control cost and the penalty for abandoning solar and wind power. The pollution emission control cost is determined by the amount of natural gas burned and the amount of electricity purchased from the external power grid. The goal of the dispatch model also considers the consumption of renewable energy and establishes a penalty for abandoning solar and wind power. The calculation formula is as follows:

[0143] C environment (t) = C pollution (t)+C pv (t)+C wt (t)

[0144] Where: C pollution (t) is the pollution emission cost of the system; C pv (t) is the penalty for abandoning light; C wt (t) is the wind curtailment penalty, where both the cost and penalty units are RMB; the specific calculation formula is as follows:

[0145]

[0146]

[0147]

[0148] Where: C pollution (t) is the pollution emission cost; k represents the different pollutants emitted; β k Costs required to treat pollutant k; is the emission coefficient of the power grid pollutant k; is the emission coefficient of the micro gas turbine pollutant k; is the emission coefficient of gas boiler pollutant k; P PV(t) is the actual photovoltaic output value in period t (kW), is the actual photovoltaic power consumption value (kW) during period t, c pv is the penalty cost of the unabsorbed photovoltaic power (yuan / kW); P WT (t) is the actual output value of the fan during period t (kW), is the actual consumption value of the fan during period t (kW), c wt is the penalty cost for the unabsorbed power of the wind turbine (yuan / kW).

[0149] Solving the optimization scheduling model in step S4 includes the following steps:

[0150] S46: Using the source-load prediction results of step S41 and the equipment operating parameters, combined with the comprehensive energy system equipment model established in step S1, the system constraints are established, including cold, hot and electric energy flow balance constraints, equipment output constraints and energy storage constraints; wherein the cold energy flow balance constraint is:

[0151]

[0152] Where: Q(t) is the system cooling load; Q EC (t) is the cooling power of the electric refrigerator; Q AC (t) is the cooling power of the absorption refrigerator; and are the cooling state and cooling power of the cold storage device respectively; and They are the cold storage state and cold storage power of the cold storage device, and the cold storage and cooling states are 0-1 variables;

[0153] The thermal energy flow balance constraint is:

[0154]

[0155] Where: H(t) is the heat load of the system; H GB (t) is the thermal power output of the gas boiler; H WHB (t) is the thermal power output of the waste heat boiler; H AC (t) is the thermal power consumed by the absorption refrigerator; and are the heat release state and heat release power of the heat storage device respectively; and are the heat storage state and heat storage power of the heat storage device, respectively. The heat storage and heat release states are 0-1 variables;

[0156] The power flow balance constraint is:

[0157]

[0158] Where: P(t) is the electrical load of the system; P grid (t) inputting electric power of the integrated energy system into the external power grid; is the actual absorption value of photovoltaic power; is the actual consumption value of the fan; P MT (t) is the power generation of the micro gas turbine; P EC (t) is the electrical power consumed by the electric refrigerator; and are the discharge state and discharge power of the battery respectively; and Charging state and charging power, charging and discharging states are 0-1 variables;

[0159] Equipment output constraints:

[0160]

[0161]

[0162]

[0163]

[0164]

[0165] Where: The maximum electric power input into the integrated energy system to the external power grid; Indicates the rated heating power of the gas boiler (kW); Indicates the rated heating power of the waste heat boiler (kW); Indicates the rated cooling power of the absorption chiller (kW); Indicates the rated cooling power of the electric refrigerator (kW);

[0166] Energy storage constraints, taking batteries as an example, heat storage devices and cold storage devices are similar and will not be described in detail. During the operation of batteries, charging and discharging cannot be performed at the same time, and there are certain charging and discharging power constraints. The specific constraints are as follows:

[0167]

[0168]

[0169]

[0170] Where: and They represent the charge and discharge status of the battery in time period t as 0-1 variables; and Respectively represent the upper and lower limits of the battery charging power; Respectively represent the upper and lower limits of the battery discharge power. In addition to the above constraints, the battery must also meet the upper and lower capacity constraints. At the same time, the energy stored in the battery must remain unchanged after the operation cycle ends. The details are as follows:

[0171]

[0172] SOC BS (0) = SOC BS (twenty four)

[0173] Where: and Respectively represent the upper and lower limits of battery storage energy; SOC BS (0) and SOC BS (24) respectively represent the energy storage status of the battery at the start time period 0 of the operation cycle and the end time period 24 of the operation cycle;

[0174] S47: The integrated energy system optimization scheduling model is composed of the system optimization scheduling objective function established in steps S43, S44, and S45 and the constraint conditions established in step S46;

[0175] S48: Use the YALMIP toolbox to call CPLEX to solve the optimization scheduling model, and output the equipment output and daily operating cost of the system optimization scheduling in each period.

[0176] It should be understood that this embodiment is only used to illustrate the present invention and is not used to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application.

Claims

1. A comprehensive energy system optimization scheduling method based on ALIF-GRU-MTL source-load joint prediction, characterized in that: The following steps are involved: S1: Establish the basic structure of the integrated energy system and build a mathematical model based on the equipment type; S2: Preprocess the source-load data of the integrated energy system and use the original features to construct derived features to complete feature engineering; S3: Based on ALIF-GRU-MTL, the prediction models of different source-load component sequences are obtained, and the prediction results of the obtained source-load component sequence model are summed, reconstructed and denormalized to obtain the final source-load prediction results; the pre-processed source-load data of the integrated energy system are arranged into time series, and the ALIF method is used to decompose them to obtain component sequences with different frequency bands of historical load data and historical output data; the obtained component sequences are reconstructed by sample entropy to obtain cycle, fluctuation and trend sequences; the obtained cycle, fluctuation and trend sequences are combined with the corresponding meteorological data and time information to obtain a data set, which is divided into a training set and a test set; The obtained training set is input into the model for training. Based on GRU, MTL is used to perform multi-task learning training on the cycles, fluctuations and trend sequences of cold, hot and electric loads as well as photovoltaic and wind turbine outputs, and the prediction models of different source-load component sequences are obtained. S4: Establish an optimal scheduling model for the integrated energy system and solve the optimal scheduling model.

2. According to claim 1, a comprehensive energy system optimization scheduling method based on ALIF-GRU-MTL source-load joint prediction is characterized in that: The step S1 comprises the following steps: S11: Establish the basic structure of the system according to the actual equipment deployment and energy flow type of the integrated energy system, and clarify the energy flow and coupling relationship between different equipment; S12: Divide the system into cold energy flow, thermal energy flow, and electrical energy flow according to the energy flow type; S13: Divide the devices in the system into source devices, multi-energy coupling devices, and energy storage devices according to the device type; S14: Establish mathematical models for the source device, the multi-energy coupling device, and the energy storage device respectively according to the device operation characteristics.

3. The method for optimizing and scheduling an integrated energy system based on ALIF-GRU-MTL source-load joint prediction according to claim 2 is characterized in that: The source-end equipment in step S13 includes a micro gas turbine and a gas boiler, the multi-energy coupling equipment includes a waste heat boiler, an absorption refrigerator and an electric refrigerator, and the energy storage equipment includes a battery, a heat storage device and a cold storage device.

4. According to claim 1, a comprehensive energy system optimization scheduling method based on ALIF-GRU-MTL source-load joint prediction is characterized in that: The preprocessing of the source-load data of the integrated energy system in step S2 includes the following steps: S21: Acquire historical load data and historical output data of the integrated energy system, and acquire time information and meteorological data corresponding to the load data and the output data; S22: performing data cleaning on the historical load data, historical output data, time information and meteorological data in step S21, including detecting and processing missing values ​​and abnormal values; S23: Perform minimum-maximum normalization on the continuous data in the data processed in step S22, and perform One-Hot encoding on the discrete data in the data processed in step S22.

5. The method for optimizing and scheduling an integrated energy system based on ALIF-GRU-MTL source-load joint prediction according to claim 1 is characterized in that: The step S2 includes the following steps of completing feature engineering according to the original feature structure: S24: performing coupling characteristic analysis on the integrated energy system data preprocessed in step S23, and selecting meteorological data, time information, historical load data characteristics, and historical output data characteristics with high correlation; S25: Use the data features after screening in step S24 to construct derived features to complete feature engineering.

6. A comprehensive energy system optimization scheduling method based on ALIF-GRU-MTL source-load joint prediction according to claim 4 or 5, characterized in that: The load data includes cold, heat and electricity load data, the output data includes photovoltaic and wind turbine output data, the time information includes time, week and month, and the meteorological data includes temperature and humidity data, wind speed data, irradiance data and precipitation data.

7. The method for optimizing and scheduling an integrated energy system based on ALIF-GRU-MTL source-load joint prediction according to claim 6 is characterized in that: The prediction model for obtaining different source-load component sequences based on ALIF-GRU-MTL in step S3 includes the following steps: S31: Arrange the source-load data of the integrated energy system preprocessed in step S2 into a time series, and decompose them using the ALIF method to obtain component sequences with different frequency bands of historical load data and historical output data; S32: The source-load component sequence obtained in step S31 is reconstructed using sample entropy to obtain period, fluctuation and trend sequences, where the period sequences of cold, hot and electric load data and photovoltaic and wind turbine output data are Q 1 , H 1 , P 1 , The fluctuation sequences of cold, hot and electric load data and photovoltaic and wind turbine output data are Q 2 , H 2 , P 2 , The trend sequences of cold, heat, and electricity load data as well as photovoltaic and wind turbine output data are Q 3 , H 3 , P 3 , The superscript i=1, 2, 3 represent the periodic sequence, the fluctuation sequence, and the trend sequence respectively; S33: combining the source-load cycle sequence, fluctuation sequence and trend sequence obtained in step S32 with the corresponding meteorological data and time information to obtain a data set, which is divided into a training set and a test set, which are used for training and testing the source-load joint prediction model respectively; S34: Input the training set obtained in step S33 into the model for training. Based on GRU, MTL is used to perform multi-task learning training on the cycles, fluctuations and trend sequences of cold, hot and electric loads as well as photovoltaic and wind turbine outputs, to obtain prediction models for different source-load component sequences.

8. The method for optimizing and scheduling an integrated energy system based on ALIF-GRU-MTL source-load joint prediction according to claim 7 is characterized in that: The step S3 of summing up, reconstructing and denormalizing the prediction results of the source-load component sequence model to obtain the final source-load prediction result includes the following steps: S35: Using the prediction model training and prediction obtained in step 34, the data of the previous m days are used to predict the cold, heat, and electricity loads and photovoltaic and wind turbine outputs of 24 points in the next day. The input and output data of the model at time t in the next day are predicted as follows: Where: time(t) = [hour, week, month] is the time information including time, week and month; weather(t) = [tem, hum, ws, ir, rain] is the meteorological characteristics of the source-load input model at time t using the day-ahead weather forecast data, which are temperature, humidity, wind speed, irradiance and precipitation. They represent the cold, hot, and electric loads as well as the photovoltaic and wind turbine output characteristics of the same sequence, as shown below: In the formula, the source-load data at the same time t in the previous three days are used to input the model, that is, m is 3, which is the source-load data at the same time t in the previous three days. The abstract representation of the model training process is as follows: Y(t)=GRUi(X(t)) (4) Where: GRUi represents the prediction models of source-load cycle sequence, fluctuation sequence and trend sequence obtained by multi-task learning training for different sequences; S36: Use the trained model GRUi to predict the source-load corresponding sequence i at the time k to be predicted on a certain day in the test set. The prediction results of the cold, hot, and electric loads, as well as the photovoltaic and wind turbine output sequences at the time k are expressed as follows: Where: are the cold, hot, electric loads, and photovoltaic and wind turbine output prediction values ​​of sequence i at time k respectively; the cold, hot, electric loads, and photovoltaic and wind turbine output prediction values ​​of each sequence are summed and reconstructed respectively, and the predicted value of source-load at time k after reconstruction is expressed as: Where: Q(k), H(k), P(k), P PV (k), P WT (k) are the predicted values ​​of cold, hot, electric loads, and photovoltaic and wind turbine output at time k after summing and reconstruction. The source-load prediction values ​​after summing and reconstruction are denormalized to obtain the final source-load prediction values.

9. The method for optimizing and scheduling an integrated energy system based on ALIF-GRU-MTL source-load joint prediction according to claim 6, characterized in that: The step S4 of establishing the comprehensive energy system optimization scheduling model includes the following steps: S41: using the prediction model based on ALIF-GRU-MTL to obtain different source-load component sequences in step S3, the cold, heat and electricity load data and the photovoltaic and wind turbine output data forecast values ​​for 24 time periods in the future day are obtained, and the natural gas price and electricity price energy market information and system equipment operation parameters are imported; S42: Using the source-load prediction results and equipment operating parameters in step S41, combined with the equipment mathematical model established in step S1, the objective function of the optimization scheduling model is established according to the economy, environmental protection and safety of the integrated energy system operation. The cycle of the optimization scheduling is one day and the time interval is 1 hour; S43: Quantifying the two goals of economy and environmental protection into costs to establish a unified objective function, as follows: Where: C IES represents the total daily operating cost of the system; C grid (t) is the cost of purchasing electricity from the external power grid; C gas (t) Cost of purchasing natural gas for the system; C operation (t) is the operation and maintenance cost of each device in the system; C environment (t) Penalty fees for the environmental performance of the system; S44: The economic efficiency of the system operation in step S43 is mainly determined by the system's electricity and gas purchase costs and the operating and maintenance costs of each device in the system. The calculation formula for the system's external power grid electricity purchase cost is as follows: C grid (t)=P grid (t).c grid (t) (8) Where: P grid (t) is the electric power input from the external power grid into the integrated energy system during period t; c grid (t) is the electricity price in period t; the cost of purchasing natural gas for the integrated energy system is composed of the consumption of micro gas turbines and gas boilers. The specific calculation formula is as follows: C gas (t)=C MT (t)+C GB (t) (9) Where: C MT (t) is the cost of natural gas consumed by the micro gas turbine, C GB (t) is the cost of natural gas consumed by the gas boiler; the system operation and maintenance costs are proportional to the unit output power of each device; S45: The environmental friendliness of the system operation in step S43 is mainly determined by the system's pollution emission control cost and the penalty for abandoning solar and wind power, wherein the pollution emission control cost is determined by the amount of natural gas burned and the amount of electricity purchased from the external power grid. The goal of the dispatch model also considers the consumption of renewable energy and establishes a penalty for abandoning solar and wind power. The calculation formula is as follows: C environment (t)=C pollution (t)+C pv (t)+C wt (t) (10) Where: C pollution (t) is the pollution emission cost of the system; C pv (t) is the penalty for abandoning light; C wt (t) Penalty for wind abandonment.

10. The method for optimizing and scheduling an integrated energy system based on ALIF-GRU-MTL source-load joint prediction according to claim 9, characterized in that: The step S4 of solving the optimization scheduling model comprises the following steps: S46: using the source-load prediction result of step S41 and the equipment operating parameters, combined with the comprehensive energy system equipment model established in step S1, to establish system constraints, including cold, hot and electric energy flow balance constraints, equipment output constraints and energy storage constraints; S47: The integrated energy system optimization scheduling model is composed of the system optimization scheduling objective function established in steps S43, S44, and S45 and the constraint conditions established in step S46; S48: Solve the optimization scheduling model and output the equipment output and daily operation cost of the equipment participating in the system optimization scheduling in each time period.

Images