Rule-constraint collaborative multi-energy coupling system online operation optimization method

Abstract

Aiming at the problems of low solving efficiency and the like in the online operation optimization of the multi-energy coupling energy supply system, the invention provides a knowledge rule-strict constraint collaborative two-stage offline operation optimization method based on a long / short time domain-coarse / fine time granularity coupling concept, a high-quality offline operation strategy data set is obtained, and the optimization efficiency of the online operation of the multi-energy coupling energy supply system is improved. The data basis required by time sequence prediction model training is met; based on an off-line operation strategy data set and a deep neural network, mining a mapping relation between input characteristics such as wind and light resources and the like and an equipment start-stop state, and constructing an equipment start-stop state prediction model; an initial solution is output by using the start-stop state prediction model, the complexity of an online operation optimization problem is reduced, the solving efficiency is improved, and finally a rule guidance-strict constraint collaborative multi-energy coupling system online operation optimization strategy is established; the optimization precision is considered while the timeliness is ensured, a real-time scheduling decision can be made for wind and light fluctuation and load change, and the operation level of the multi-energy coupling low-carbon power generation system is effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of low-carbon comprehensive energy supply systems, and particularly relates to an online operation optimization method for a rule-constraint collaborative multi-energy coupling system. BACKGROUND

[0002] Low-carbon comprehensive energy supply system modeling is the basis for carrying out operation characteristic analysis, configuration scheme optimization and operation strategy optimization. Due to the diverse types of equipment in the system and the complex energy flow coupling relationship, the number of model variables and constraint conditions is large, and in the process of simulating the annual operation characteristics of the system, the traditional solving method is low in efficiency and is difficult to meet the actual demand. Therefore, how to construct a reasonable and efficient system model and accurately obtain the annual operation characteristics thereof through an efficient solving method is still an important challenge currently faced.

[0003] The operation optimization of a low-carbon comprehensive energy supply system is mainly to simultaneously optimize the start-stop state (discrete variable) and the operation power (continuous variable) of the equipment. However, when a mathematical programming method with strict constraints is used for global optimization, the high-dimensional collaborative solving of discrete variables and continuous variables will cause a “dimension disaster”, resulting in a sharp increase in computational complexity, which is difficult to meet the timeliness requirement of minute-level online optimization.

[0004] Therefore, although the strict constraint optimization method can obtain an offline operation strategy superior to an empirical rule in theory, its computational efficiency bottleneck limits the application scenario: it is more suitable for formulating short-term (such as daily / weekly) optimization schemes, and it is difficult to be directly extended to long-term (such as quarterly / annual) optimization scenarios. In addition, the system operation also faces the optimization target conflict in the time scale, such as the significant time-of-day fluctuation characteristics of key parameters (such as carbon emissions per kilowatt-hour and energy storage SOC change), which causes the inconsistency between short-term operation targets and long-term planning indicators. If rigid long-term indicator constraints are directly introduced into the short-term optimization model, two risks may be caused: constraint conflict, no feasible solution; the model cannot simultaneously meet the short-term operation boundary and the long-term average target. Economic deterioration: in order to forcibly meet the long-term target, some equipment is frequently started and stopped or operated at a high load for a long time, resulting in a sharp increase in operating costs. Ultimately, the system is difficult to generate differentiated and refined operation strategies that adapt to different time period characteristics.

[0005] In order to break through the bottleneck of the above-mentioned computational efficiency and multi-time scale collaboration, the application proposes an online operation optimization efficient solving method driven by offline strategy pre-training. The method aims to extract the global vision and knowledge of offline optimization into an intelligent strategy available online, so as to effectively balance the optimization accuracy while strictly guaranteeing the timeliness of online decision-making, and ultimately comprehensively improve the intelligent operation level of a multi-energy coupling low-carbon energy supply system. SUMMARY

[0006] The application aims to provide a rule-constraint collaborative multi-energy coupling system online operation optimization method to solve the problem that the operation strategy solving for a multi-coupling energy supply system with complex energy flow coupling relationship cannot achieve the goal of improving solving efficiency and obtaining optimal solution.

[0007] The rule-constraint collaborative multi-energy coupling system online operation optimization method comprises the following steps:

[0008] Step 1: Based on the long / short time domain-coarse / fine time granularity coupling concept, a two-stage offline operation optimization method guided by knowledge rules and strict constraints is constructed.

[0009] Step 1.1: Establish a knowledge rule guided solving algorithm;

[0010] Step 1.2: Use the knowledge rule guided solving algorithm to obtain multi-energy coupling system annual long-term operation data for time period slicing and aggregation analysis;

[0011] Step 1.3: Time period slicing and aggregation analysis is performed on the multi-energy coupling system annual long-term operation data to extract system operation boundary and evaluation index constraints in different time periods;

[0012] Step 1.4: Establish a strict constraint optimization algorithm, and based on step 1.3, carry out a strict constraint optimization for a short period to obtain an offline operation strategy data set covering typical working conditions;

[0013] Step 2: Construct a prediction model of energy equipment start-stop state based on the offline operation strategy data set;

[0014] Step 2.1: Based on the LSTM neural network, mine the time sequence correlation between the integer decision variables such as equipment start-stop and charge-discharge in the offline operation strategy data set and external boundary conditions such as wind power and photovoltaic output;

[0015] Step 2.2: Based on the time sequence correlation obtained in step 2.1, construct an energy equipment start-stop state time sequence prediction model to predict the future start-stop state of the energy equipment through future wind and light output time sequence data;

[0016] Step 3: Based on the energy equipment start-stop state time sequence prediction model established in step 2.2, obtain the initial solution of the energy equipment start-stop state;

[0017] Step 4: Establish an online rolling operation optimization efficient solving method based on the start-stop state prediction model;

[0018] Step 4.1: Based on the initial solution provided by the start-stop state prediction model, the dimension of the decision space is reduced through the assignment acceleration method to simplify the optimization problem and improve the solving efficiency of the online operation optimization;

[0019] Step 4.2: Establish a back-substitution dimension reduction solving strategy for the large-scale mixed integer optimization problem based on the initial solution assignment;

[0020] Step 5: Establish the operation strategy of the online operation optimization method of the low-carbon power generation system;

[0021] Step 6: According to the operation strategy established in step 5, the low-carbon power generation system is optimized and scheduled.

[0022] Preferably, it is assumed that the boiler keeps continuous operation, and the thermal load is always not lower than the minimum limit value; when the output power of the power generation subsystem exceeds the power load demand, i.e. there is power surplus, according to the energy conversion efficiency, the energy conversion equipment is charged in turn, and the charging order is the electric energy storage system, the thermal energy storage system, the ammonia energy storage system and the methanol preparation system; when the output power of the power generation subsystem is lower than the power load demand, i.e. there is power shortage, the equipment with high efficiency and low carbon emission is preferentially dispatched to alleviate the power shortage, and the dispatching priority is in turn the electric energy storage discharge, the thermal energy storage releases heat energy to drive the steam turbine to generate electricity, the ammonia energy storage releases ammonia gas into the boiler for co-combustion, and the coal consumption is increased to increase the boiler heat output power; the ramping constraints, the shortest start-stop time constraints of the electric energy storage, the thermal energy storage, the ammonia energy storage and the methanol preparation system are ignored; based on the knowledge rule, a sequential solving algorithm of the multi-energy coupling system guided by the knowledge rule is established;

[0023] The decision variables of the mathematical programming proposition of the multi-energy coupling low-carbon power generation system are the operation powers of each equipment; the objective function is to minimize the total variable operation cost of the system; the constraint conditions include the operation constraints of the subsystem, the power balance constraint, the heat balance constraint, etc., and the mathematical expression of the optimization problem in the strict constraint optimization stage is as follows:

[0024] (1)

[0025]

[0026]

[0027]

[0028] Wherein,

[0029] ——the variable operation cost of the low-carbon energy system at time t;

[0030] ——the total number of steps in the strict constraint optimization stage;

[0031] ——the start-stop state discrete variable of all equipment within the future time steps, which is used to determine whether each equipment is in the running state at each time step;

[0032] — future Running power continuous variable of all devices within one time step;

[0033] — running constraints of subsystem sys;

[0034] — power balance constraints;

[0035] — heat balance constraints.

[0036] Preferably, the step 2 is to establish a prediction model of device start-stop and charge-discharge state based on LSTM neural network, the model input is the photovoltaic power, wind power and power demand time series data of future n time steps, and the output is the start-stop state of boiler, electric energy storage, heat energy storage, ammonia energy storage, methanol preparation system and the like within future n time steps.

[0037] The prediction model of device start-stop and charge-discharge state evaluation index is prediction accuracy, overall prediction accuracy and average prediction accuracy of data set, and the calculation method is as follows:

[0038] (2)

[0039] (3)

[0040] Wherein,

[0041] — prediction accuracy of model on start-stop state variable variable in sample i;

[0042] — the number of samples correctly predicted by the model on start-stop state variable i in sample i;

[0043] n — total step number of each sample, for three-day 5-minute level prediction, the value is 864;

[0044] — prediction accuracy of model on start-stop state variable variable in the whole data set;

[0045] — total sample number in data set;

[0046] — average prediction accuracy of the whole data set;

[0047] — total number of prediction variables.

[0048] Preferably, based on the device start-stop state constraints, the step of correcting the initial prediction result of the LSTM neural network is:

[0049] First, the start-stop state variable in the prediction time domain is detected, the start-stop state continuous segment without state switching is recorded, and it is recorded down;

[0050] Secondly, it is judged whether the duration of the continuous segment meets the minimum start-stop time constraint, if it meets, the next time step is corrected; if it does not meet, according to the degree of constraint violation, the state of the violation segment is selected to be switched or the current state is expanded to the adjacent period until all start-stop state continuous segments are corrected.

[0051] Preferably, the step 4 online operation optimization adopts a rolling optimization method, in the actual rolling optimization process, not all scheduling instructions in the future three days are executed in one optimization, but only the scheduling instructions of the first time step in each rolling optimization are executed; at the same time, the device running state after executing the scheduling instructions is transmitted to the next optimization period, through the continuous rolling optimization and execution of the scheduling instructions, the optimal operation of the low-carbon energy system is realized;

[0052] The optimization objective is set to minimize the total variable cost in the future three days, and the decision variable is the start-stop state discrete variable and the running power continuous variable of all devices in the future 864 time steps;

[0053] In the process of rolling optimization, the mathematical expression of each optimization problem is as follows:

[0054] (4)

[0055] Among them,

[0056] ——t moment, variable operation cost of low-carbon energy system;

[0057] ——start-stop state discrete variable of all devices in the future 864 time steps, used to determine whether each device is in running state in each time step;

[0058] ——running power continuous variable of all devices in the future 864 time steps.

[0059] Preferably, the assignment acceleration method in step 4.1 is to directly assign values to the start-stop state discrete variable based on the start-stop plan of each device provided by the start-stop state prediction model, eliminate the start-stop state discrete variable in the decision variable, reduce the dimension of the original mixed integer linear programming problem to linear programming problem, reduce the solving difficulty, and thus reduce the time and resources required for calculation;

[0060] After using the assignment acceleration method, the mathematical expression of the optimization problem becomes:

[0061] (5)

[0062] Wherein, The prediction value of the LSTM prediction model for the start-stop state discrete variable of all devices in the future 864 time steps.

[0063] Compared with the prior art, the technical scheme provided by the present application has the following beneficial effects:

[0064] 1. The multi-coupling energy supply system operation strategy optimization method disclosed by the present application adopts the cooperative optimization of knowledge rule guidance and strict constraint in two stages, which can effectively avoid the no solution risk and economic deterioration problem in short-term optimization while ensuring the achievement of long-term goals, significantly improve the solving efficiency, and finally generate a high-quality offline operation strategy set to provide high-quality data support for subsequent neural network training.

[0065] 2. The multi-coupling energy supply system operation strategy optimization method disclosed by the present application adopts the combination of knowledge rule guidance and strict constraint to form the offline operation strategy of the system, trains the deep neural network model, and finally uses the trained deep neural network model to solve the online operation of the multi-coupling energy supply system. It is an online operation optimization efficient solving method driven by offline strategy pre-training, which guarantees the timeliness while considering the optimization accuracy, and effectively improves the operation level of the multi-energy coupling low-carbon power generation system.

[0066] 3. Based on the offline operation strategy data set and the LSTM neural network, a start-stop state time series prediction model is constructed, and the average prediction accuracy of the time series prediction model can reach 93.2% at a time resolution of three days and 5 minutes, which can provide high-quality start-stop state initial solution for online operation optimization.

[0067] 4. Based on the initial solution provided by the start-stop state time series prediction model, a dimension reduction solving strategy for large-scale mixed integer optimization problems based on initial solution assignment is designed. Compared with direct solving, the assignment acceleration solving method shows significant advantages in calculation efficiency, and the solving time is reduced by more than 90% compared with direct solving. The assignment acceleration solving method realizes the best balance of economy, reliability and carbon emission with longer decision-making time domain and faster calculation speed, effectively improving the overall operation performance of the system. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 The structure diagram of the multi-coupling net zero carbon emission energy supply system related to the embodiments;

[0069] Figure 2 Flow chart of the multi-coupling energy supply system operation strategy optimization method involved in the present application;

[0070] Figure 3 Key indicator distribution map obtained by knowledge rule guided solving;

[0071] Figure 4 Effect picture of the knowledge rule guided-strict constraint two-stage collaborative optimization strategy;

[0072] Figure 5 Initial prediction result correction strategy based on equipment start-stop state constraint;

[0073] Figure 6 Prediction accuracy of the start-stop state prediction model;

[0074] Figure 7 Comparison of the operation scheduling strategy effects of different optimization methods. DETAILED DESCRIPTION

[0075] In order to further understand the content of the present application, the present application will be described in detail in combination with embodiments, and the following embodiments are used to illustrate the present application but not to limit the scope of the present application.

[0076] Embodiment 1

[0077] A rule-constraint collaborative multi-energy coupling system online operation optimization method, first based on a rule-guided-strict constraint two-stage collaborative offline operation optimization method, obtains a high-quality offline operation strategy data set to meet the data basis required for training a time series prediction model; then based on the offline operation strategy data set and a deep neural network, the mapping relationship between input features such as wind and light resources and the equipment start-stop state is mined, and then an equipment start-stop state prediction model is constructed; the initial solution output by the start-stop state prediction model is used to reduce the complexity of the online operation optimization problem and improve the solving efficiency, and finally a rule-guided-strict constraint collaborative multi-energy coupling system online operation optimization strategy is established.

[0078] The specific steps are as follows:

[0079] Step 1: based on the long / short time domain-coarse / fine time granularity coupling concept, a two-stage offline operation optimization method of knowledge rule guided-strict constraint collaboration is constructed.

[0080] Step 1.1: establish a knowledge rule guided solving algorithm;

[0081] Step 1.2: use the knowledge rule guided solving algorithm to obtain multi-energy coupling system annual long-term operation data for time period slicing and aggregation analysis;

[0082] Step 1.3: Time period slicing and aggregation analysis of long-term operation data of the multi-energy coupling system, extraction of system operation boundaries and evaluation index constraints in different time periods;

[0083] Step 1.4: Establish a strict constraint optimization algorithm, based on step 1.3, carry out short-term oriented strict constraint optimization, obtain offline operation strategy data set covering typical working conditions;

[0084] Step 2: Construct a prediction model of the start-stop state of energy equipment based on the offline operation strategy data set;

[0085] Step 2.1: Based on LSTM neural network, mine the time sequence correlation between integer decision variables such as equipment start-stop and charge-discharge in offline operation strategy data set and external boundary conditions such as wind power and photovoltaic output;

[0086] Step 2.2: Based on the time sequence correlation obtained in step 2.1, construct a time sequence prediction model of the start-stop state of energy equipment, and predict the start-stop state of future energy equipment through future wind and light output time sequence data;

[0087] Step 3: Based on the time sequence prediction model of the start-stop state of energy equipment established in step 2.2, obtain the initial solution of the start-stop state of energy equipment;

[0088] Step 4: Establish an online rolling operation optimization efficient solution method based on the start-stop state prediction model;

[0089] Step 4.1: Based on the initial solution provided by the start-stop state prediction model, accelerate the decision space through assignment method, simplify the optimization problem, and improve the solution efficiency of online operation optimization;

[0090] Step 4.2: Establish a elimination and dimension reduction solution strategy for large-scale mixed integer optimization problem based on initial solution assignment;

[0091] Step 5: Establish the operation strategy of the online operation optimization method of the low-carbon power generation system;

[0092] Step 6: According to the operation strategy established in step 5, optimize and dispatch the low-carbon power generation system.

[0093] The step 1 is specifically represented by the knowledge rule: assuming that the boiler keeps continuous operation, the thermal load is always not lower than the minimum limit value; when the output power of the power generation system (wind power, photovoltaic, coal-fired power plant) exceeds the power load demand, that is, there is power surplus, according to the energy conversion efficiency, the energy conversion equipment is charged in turn, and the charging order is the electric energy storage system, the thermal energy storage system, the ammonia energy storage system and the methanol preparation system; when the output power of the power generation system is lower than the power load demand, that is, there is power shortage, then the equipment with high efficiency and low carbon emission is preferentially dispatched to alleviate the power shortage, and the dispatching priority is in turn the electric energy storage discharge, the thermal energy storage releases heat energy to drive the steam turbine to generate electricity, the ammonia energy storage releases ammonia gas into the boiler for blending combustion, and the coal consumption is increased to increase the boiler heat output power; the climbing constraints, the shortest start-stop time constraints of the electric energy storage, the thermal energy storage, the ammonia energy storage and the methanol preparation system are ignored.

[0094] Based on the knowledge rule, a knowledge rule guided multi-energy coupling system sequential solving algorithm is established;

[0095] The specific process of the knowledge rule guided multi-energy coupling system sequential solving algorithm is as follows:

[0096] 1) Calculate the output power of the wind power and photovoltaic equipment;

[0097] 2) Assume that the boiler runs as low as possible;

[0098] Firstly, the boiler load is adjusted to reduce the coal consumption as much as possible, and then the carbon emission is reduced, and on this basis, it is judged whether the power load demand can be satisfied by 100%, and the equipment output is adjusted again.

[0099] Since there is a maximum blending ratio limit for ammonia blending combustion in the boiler, in order to reduce the coal consumption, the boiler heat output power needs to be reduced as much as possible. The boiler heat output power should meet the minimum load constraint and the load reduction constraint. The boiler heat output power at this moment is represented as The superscript a1 represents the first assumption, and the subsequent operation constraints of other equipment need to be combined to calculate whether the assumption is reasonable, and the calculation method is as follows:

[0100] (1)

[0101] Among them, represents the rated heat output power of the boiler, kJ / s; represents the minimum load rate of the boiler;

[0102] In order to reduce the coal consumption as much as possible, it is necessary to minimize the heat released by coal combustion , that is, to maximize the heat released by ammonia blending combustion . The maximum heat release of ammonia blending combustion at this moment is represented as , and the calculation method is as follows:

[0103] (2)

[0104] wherein, represents the highest burning ratio of ammonia (in terms of heat release) ; represents the remaining energy of ammonia energy storage, kWh; represents the heat release efficiency of ammonia energy storage, which refers to the efficiency of ammonia converted into heat in the boiler, and certain losses need to be considered on the basis of boiler efficiency;

[0105] obtained , then :

[0106] (3)

[0107] and then calculate the heat consumed by carbon capture at this moment :

[0108] (4)

[0109] wherein, represents the heat consumption required by CCS capture;

[0110] According to the above calculation results, the heat supplied by the boiler to the steam turbine :

[0111] (5)

[0112] 3) Determine whether the boiler heat supply can maintain the normal operation of the steam turbine

[0113] In the first assumption a1, the boiler heat load and coal consumption are as low as possible, but it is still necessary to calculate whether the heat supplied by the boiler to the steam turbine meets the steam turbine operation constraints, mainly the minimum load constraint and the variable load constraint. Because the boiler is assumed to operate at the minimum load in the foregoing, it is not necessary to consider whether the maximum load constraint of the steam turbine, the steam turbine load increasing constraint is met:

[0114] (6)

[0115] wherein, represents the minimum heat input power of the steam turbine, kJ / s; represents the rated heat input power of the steam turbine, kJ / s; represents the minimum load rate of the steam turbine; represents the variable load rate of the steam turbine, % / min;

[0116] If the boiler heat supply meets the steam turbine operation load constraint, it means that the first assumption is reasonable, and thus the second assumption a2 can be entered:

[0117] The thermal energy storage does not need to provide heat for the steam turbine at this moment, so .

[0118] (7)

[0119] The calculation method of the coal-fired power plant's power generation is:

[0120] (8)

[0121] where, represents the steam turbine efficiency; represents the generator efficiency; represents the comprehensive self-use rate;

[0122] The next stage of calculation: 4) Calculate the power surplus.

[0123] If the boiler heat supply does not meet the steam turbine operation load constraint, it means that there is a shortage of heat load for the steam turbine:

[0124] First, the thermal energy storage makes up for this shortage (because the thermal energy stored in the thermal energy storage comes from zero-carbon power from renewable energy, and preferentially using the thermal energy stored in the thermal energy storage helps to reduce carbon emissions):

[0125] (9)

[0126] where, represents the remaining energy of the thermal energy storage, kWh; represents the heat release efficiency of the thermal energy storage, which refers to how much of the thermal energy stored in the molten salt can be converted into high-temperature steam entering the steam turbine (i.e., the heat exchange efficiency considering heat exchange loss), and the value is 98%;

[0127] ① If the thermal energy storage can make up for the shortage of steam turbine load, i.e., , then the values of the following variables are equal to the values of the first assumption:

[0128] (10)

[0129] The calculation method of the coal-fired power plant's power generation is:

[0130] (11)

[0131] The next stage of calculation: 4) Calculate the power surplus.

[0132] ② If it is not satisfied, the boiler needs to increase the load to maintain the operation of the steam turbine, and at this time the heat supplied by the boiler to the steam turbine is:

[0133] (12)

[0134] Similarly, in order to reduce coal consumption as much as possible, it is necessary to minimize the heat released by coal combustion , that is, maximize the heat released by ammonia combustion . The maximum heat release of ammonia combustion at this moment is represented as , and the calculation method is as follows:

[0135] (13)

[0136] where, unknown, need to be converted.

[0137] known:

[0138] (14)

[0139] (15)

[0140] where, represents the heat consumption required by CCS capture, accounting for a proportion of the heat released by coal combustion;

[0141] So:

[0142] (16)

[0143] (17)

[0144] Substitute and into , which can be converted to:

[0145] (18)

[0146] After is found, according to , and , respectively , and can be found.

[0147] The calculation method of the power generation of the coal-fired power plant at this moment is:

[0148] (19)

[0149] 4) Calculate the power surplus

[0150] The power consumption required by the CCS device at this moment is:

[0151] (20)

[0152] where, represents the proportion of the electricity consumption required by CCS in the heat released by coal combustion;

[0153] Thus, the electricity surplus at this moment can be calculated:

[0154] (21)

[0155] wherein, represents the output power of the photovoltaic power generation system, kW; represents the output power of the wind power generation system, kW; represents the power generation capacity of the coal-fired power plant, kW; represents the power demand, kW; represents the electricity consumption required by the CCS device, kW;

[0156] 5) If the electricity surplus ≥ 0

[0157] The power storage does not need to discharge at this moment, so the discharge power .

[0158] Based on the electricity surplus, the charging power of the power storage is determined:

[0159] (22)

[0160] wherein, represents the electricity surplus at this moment, kW; represents the installed capacity of the power conversion module of the power storage, kW; represents the installed capacity of the energy storage module of the power storage, kWh; represents the charging efficiency of the power storage;

[0161] According to the energy change relationship of the power storage, the remaining energy of the power storage at this moment is updated, and the formula is as follows:

[0162] (23)

[0163] wherein, represents the discharge power of the power storage; represents the discharge efficiency of the power storage;

[0164] The thermal storage responds based on the output of the coal-fired power plant and the remaining electricity surplus:

[0165] (24)

[0166] The discharge power of the thermal storage is the output of the thermal storage in the coal-fired power plant:

[0167] (25)

[0168] wherein, represents the discharge efficiency of the thermal storage;

[0169] Based on the power surplus after the charging of the electric storage is completed, the charging power of the thermal storage is determined:

[0170] (26)

[0171] wherein, represents the installed capacity of the power conversion module of the thermal storage, kW; represents the installed capacity of the energy storage module of the thermal storage, kWh; represents the residual energy of the thermal storage, kWh; represents the charging efficiency of the thermal storage;

[0172] According to the energy change relationship of the thermal storage, the residual energy of the thermal storage at the present time is updated, and the formula is as follows:

[0173] (27)

[0174] wherein, represents the discharge power of the thermal storage, kW; represents the discharge efficiency of the thermal storage;

[0175] The discharge power of the ammonia storage is the output of the ammonia storage in the coal-fired power plant:

[0176] (28)

[0177] The discharge power of the ammonia storage is the output of the ammonia storage in the coal-fired power plant:

[0178] (29)

[0179] wherein, represents the discharge efficiency of the ammonia storage;

[0180] Based on the power surplus after the charging of the electric storage and the thermal storage is completed, the charging power of the ammonia storage is determined:

[0181] (30)

[0182] wherein, represents the installed capacity of the power conversion module of the ammonia storage, kW; represents the installed capacity of the energy storage module of the ammonia storage, kWh; represents the residual energy of the ammonia storage, kWh; represents the charging efficiency of the ammonia storage (i.e. the ammonia production efficiency);

[0183] According to the energy change relationship of ammonia storage energy, update the residual energy of ammonia storage energy at this moment, the formula is as follows:

[0184] (31)

[0185] wherein, represents the discharge power of ammonia storage energy, kW;

[0186] Based on the power surplus after the charging of electricity, heat and ammonia storage energy is completed, determine the charging power of synthetic methanol:

[0187] (32)

[0188] wherein, represents the installed capacity of methanol preparation system, kW;

[0189] The values of the following variables are equal to the values of the second assumption:

[0190] (33)

[0191] The amount of abandoned electricity at this moment is:

[0192] (34)

[0193] There is no power shortage at this moment, so .

[0194] 6) If the power surplus <0

[0195] It indicates that there is a power shortage:

[0196] (35)

[0197] Because there is a power shortage at this moment, there is no excess power to charge the electric storage energy, the heat storage energy, the ammonia storage energy and the synthetic methanol device, and the charging power of the four devices is 0:

[0198] (36)

[0199] The electric storage energy is preferentially compensated for the energy shortage at each moment. Its discharge power is:

[0200] (37)

[0201] According to the energy change relationship of electric storage energy, update the residual energy of electric storage energy at this moment, refer to the formula ).

[0202] The remaining energy shortage is compensated by the heat storage energy first (the discharge efficiency of the heat storage energy is higher than that of the ammonia storage energy), and the heat storage energy is discharged through a steam turbine, so the maximum load that the steam turbine can reach at this moment is calculated first:

[0203] (38)

[0204] The heat release amount of the heat storage energy is calculated as follows:

[0205] (39)

[0206] The meaning is that the load compensated by the heat storage energy needs to meet the maximum load constraint of the steam turbine, the remaining energy constraint of the heat storage energy, and the power shortage constraint (the power generation amount of the heat load increased by the heat storage energy exactly meets the power shortage).

[0207] The discharge amount of the heat storage energy is calculated, and the remaining energy of the heat storage energy at this moment is updated according to the energy change relationship of the heat storage energy.

[0208] The new energy shortage is the old energy shortage minus the share compensated by the electric storage energy and the heat storage energy:

[0209] (40)

[0210] (1) If the steam turbine cannot increase the load any more, that is:

[0211] The values of the following variables are equal to the values of the second assumption (a2):

[0212] (41)

[0213] The power generation amount of the coal-fired power plant at this moment is:

[0214] (42)

[0215] There is no power curtailment at this moment, so

[0216] The power shortage at this moment is:

[0217] (43)

[0218] (2) If the steam turbine can still increase the load, that is:

[0219] The boiler needs to increase the load as much as possible to meet the user load. Let the load increased by the boiler be , and the corresponding change amount of the heat load provided by the coal in the boiler be , and the change amount of the heat load provided by the ammonia storage energy be ​The amount of change in heat consumption required by CCS is The amount of change in power consumption required by CCS is The amount of change in heat load of the boiler entering the steam turbine is The calculation method is as follows:

[0220] (44)

[0221] The variables need to meet the following constraints:

[0222] ①The amount of change in boiler load needs to meet the maximum load constraint and the load increase constraint:

[0223] (45)

[0224] ②The amount of change in heat load of the boiler entering the steam turbine needs to meet the power shortage at this moment and the maximum load constraint of the steam turbine:

[0225] (46)

[0226] ③The ammonia energy storage completely releases heat, or reaches the highest blending combustion ratio:

[0227] (47)

[0228] Based on the above, the above constraints can be sorted into constraints related only to , :

[0229] (48)

[0230] Under the condition of meeting , the maximum value of is obtained, which is the optimization target.

[0231] It is noted that the constraint is to select the minimum value from two values, in order to speed up the solution, it is converted into two branches, according to whether the final calculation result of the two branches is reasonable, it is judged which branch should be selected.

[0232] Branch (1): The ammonia energy storage releases heat in the boiler according to the highest blending combustion ratio

[0233] According to the first term of , that is:

[0234] (49)

[0235] By moving the term , we can get:

[0236] (50)

[0237]

[0238] Substitute into , we can get:

[0239] (51)

[0240] In order to meet the user load as much as possible, need to be as large as possible, so:

[0241] (52)

[0242] Thus we can get , compare it with the second term of , if is less than or equal to the second term, then the branch is established, otherwise go to branch (2).

[0243] Branch (2): ammonia energy storage sends all the remaining heat to the boiler

[0244] According to the second term of , that is:

[0245] (53)

[0246] So the inequality can be converted to:

[0247] (54)

[0248] In order to meet the user load as much as possible, need to be as large as possible, so:

[0249] (55)

[0250] The numerical value of the following variables at this moment is equal to the numerical value of the second assumption (a2) plus the change:

[0251]

[0252] (56)

[0253] According to the energy change relationship of ammonia energy storage, update the remaining energy of ammonia energy storage at this moment, refer to the formula ).

[0254] The power generation of the coal-fired power plant at this moment:

[0255] (57)

[0256] There is no power surplus at this moment, so .

[0257] The power shortage at this moment is:

[0258] (58)

[0259] 7) Calculate the next moment, until all the moments of the year are calculated

[0260] The strict constraint solving method maps the system operation constraints, device characteristics and objective function into mathematical expressions, thereby constructing a mathematical programming proposition, and solving it with the help of CPLEX, Gurobi solver.

[0261] As a priority, the decision variables of the mathematical programming proposition of the multi-energy coupled low-carbon power generation system are the operating powers of each device; the objective function is to minimize the total variable operating cost of the system; and the constraint conditions cover the operating constraints of the subsystem, power balance constraints, heat balance constraints, etc. The mathematical expression of the optimization problem in the strict constraint optimization stage is as follows:

[0262] (59)

[0263]

[0264]

[0265]

[0266] wherein,

[0267] — The variable operating cost of the low-carbon energy system at time t;

[0268] — The total number of steps in the strict constraint optimization stage;

[0269] — The start-stop state discrete variables of all devices within the future time steps, used to determine whether each device is in an operating state at each time step;

[0270] — The operating power continuous variables of all devices within the future time steps;

[0271] — The operating constraints of the subsystem sys;

[0272] — The power balance constraints;

[0273] - heat balance constraints.

[0274] The step 2 is to establish a prediction model of the start-stop and charge-discharge state of the equipment based on the LSTM neural network. The model input is the photovoltaic power, wind power and power demand time series data of the future n time steps, and the output is the start-stop state of the boiler, electric energy storage, thermal energy storage, ammonia energy storage and methanol preparation system in the future n time steps.

[0275] Preferably, the evaluation index of the prediction model of the start-stop and charge-discharge state of the equipment is the prediction accuracy (Sample Accuracy), the overall prediction accuracy of the data set (Dataset Accuracy) and the average prediction accuracy (Mean Accuracy), which are calculated as follows:

[0276] (60)

[0277]

[0278] Wherein,

[0279] - the prediction accuracy of the model for the start-stop state variable variable in sample i;

[0280] - the number of samples correctly predicted by the model for the start-stop state variable i in sample i;

[0281] n - the total number of steps of each sample, which is 864 for a three-day 5-minute level prediction;

[0282] - the prediction accuracy of the model for the start-stop state variable variable in the entire data set;

[0283] - the total number of samples in the data set;

[0284] - the average prediction accuracy of the entire data set;

[0285] - the total number of prediction variables.

[0286] Further, since the prediction results of the start-stop variable by the LSTM neural network prediction model have certain randomness, the equipment may run in violation of the start-stop constraint when it runs according to the predicted start-stop state, such as the time from shutdown to start-up being less than the minimum start-stop time. Therefore, the initial prediction result of the LSTM neural network needs to be corrected based on the device start-stop state constraint. First, the start-stop state variable in the prediction time domain is detected, the continuous segment of the start-stop state without state switching is recorded, and the record is recorded; second, it is judged whether the duration of the continuous segment meets the minimum start-stop time constraint, if it meets, the next time step is corrected; if it does not meet, according to the degree of violation of the constraint, the state of the violation segment is selected to be switched or the current state is expanded to the adjacent period, until all the continuous segments of the start-stop state are corrected.

[0287] The step 4 online operation optimization adopts a rolling optimization mode. In the actual rolling optimization process, not all scheduling instructions for the next three days are executed in one optimization, but only the scheduling instructions for the first time step are executed. At the same time, the running state of the equipment after executing the scheduling instructions is transmitted to the next optimization period, and through the continuous rolling optimization and execution of the scheduling instructions, the optimal operation of the low-carbon energy system is realized.

[0288] The optimization objective is set to minimize the total variable cost for the next three days, and the decision variable is the start-stop state discrete variable and the running power continuous variable of all equipment in the total of 864 time steps in the next three days. In the process of rolling optimization, the mathematical expression of each optimization problem is as follows:

[0289] (61)

[0290] Among them,

[0291] —t time, variable operation cost of low-carbon energy system;

[0292] —start-stop state discrete variable of all equipment in the future 864 time steps, used to determine whether each equipment is in a running state in each time step;

[0293] —running power continuous variable of all equipment in the future 864 time steps.

[0294] As preferred, the assignment acceleration method in step 4.1 is to directly assign the start-stop state discrete variable based on the future start-stop plan of each device provided by the start-stop state prediction model, eliminate the start-stop state discrete variable in the decision variable, reduce the dimension of the original mixed integer linear programming problem to linear programming problem, reduce the difficulty of solving, and thus reduce the time and resources required for calculation. After using the assignment acceleration method, the optimization of the discrete variable will not be needed, and the mathematical expression of the optimization problem becomes:

[0295] (62)

[0296] wherein, the predicted value of the start-stop state discrete variable of all devices within the future 864 time steps by the LSTM prediction model.

[0297] Embodiment 2. This embodiment takes a multi-coupling net-zero carbon emission energy supply system as shown in Figure 1 Embodiment 2. This embodiment takes a multi-coupling net-zero carbon emission energy supply system as shown in

[0298] The renewable energy power generation subsystem is used to generate electricity by using renewable energy, output electricity based on electricity demand, and input excess electricity to the energy storage device, and includes a wind power generation device and a photovoltaic power generation device, wherein the installed capacity of the photovoltaic power generation subsystem is 2169 MW, and the installed capacity of the wind power generation subsystem is 3267 MW.

[0299] The energy storage subsystem is used to convert excess electricity into carbon-free energy and heat energy, and includes an ammonia energy storage device and a heat energy storage device. The ammonia energy storage device includes: a water electrolysis hydrogen production device for electrolyzing water to produce hydrogen; an air separation nitrogen production device for extracting nitrogen from air; an ammonia synthesis device for producing ammonia from hydrogen and nitrogen; and an ammonia storage tank for storing ammonia. The heat energy storage device includes: a low-temperature storage tank for storing low-temperature molten salt; a high-temperature storage tank for storing high-temperature molten salt; an electric heater for heating low-temperature molten salt by using excess electricity from the renewable energy power generation subsystem, and transferring the heated high-temperature molten salt to the high-temperature storage tank; a steam generator for heat exchange with the high-temperature molten salt in the high-temperature storage tank to form steam for use by the coal-fired power generation subsystem, and transferring the low-temperature molten salt formed after heat exchange to the low-temperature storage tank; and an electric energy storage device for directly storing excess electricity. The power conversion module of the electric energy storage system is configured to be 495 MW, the energy storage module is configured to be 3507 MWh, the power conversion module of the heat energy storage system is 863 MW, the energy storage module is configured to be 5951 MWh, and the power conversion module and the energy storage module of the ammonia energy storage system are 50 MW and 50 MWh, respectively.

[0300] The coal-fired power generation system is used for burning green energy and coal to generate steam or steam generated by heat energy to supplement the power demand when the output power of the renewable energy power generation system cannot meet the power demand. The coal-fired power generation system comprises a boiler for burning ammonia and coal to form water vapor and deliver the water vapor to a steam turbine; the steam turbine is used for converting the heat energy of the steam generated by the boiler and the steam generator into kinetic energy; the generator is used for generating electricity by means of the kinetic energy of the steam turbine and outputting the electricity. The installed capacity of the coal-fired power plant is 185 MW, and the steam turbine is additionally expanded by 232 MW.

[0301] The carbon capture subsystem is used for capturing part of the CO2 generated by the coal-fired power generation system; the carbon dioxide consumption subsystem is used for preparing low-carbon energy by using the CO2 captured by the carbon capture subsystem and the externally purchased CO2. The carbon capture subsystem comprises an absorption tower for absorbing part of the CO2 generated during the combustion of the boiler to form carbon absorption liquid and deliver the carbon absorption liquid to a carbon absorption liquid storage tank; the carbon absorption liquid storage tank is used for storing the carbon absorption liquid.

[0302] The analysis tower is used for separating gaseous CO2 from the carbon absorption liquid; the compressor is used for compressing the gaseous CO2; and the CO2 storage tank is used for storing the compressed gaseous CO2.

[0303] The carbon dioxide consumption subsystem is used for preparing low-carbon energy by consuming the CO2 generated during the power generation of the coal-fired power generation system and the externally purchased CO2. The low-carbon energy prepared by the carbon dioxide consumption subsystem is methanol, and the methanol preparation system is configured to have a capacity of 1371 MW.

[0304] Firstly, based on the rule guidance-strict constraint two-stage collaborative offline operation optimization method, a high-quality offline operation strategy data set is obtained to meet the data basis required for training the time series prediction model; then based on the offline operation strategy data set and the deep neural network, the mapping relationship between the input features such as wind and light resources and the device start-stop state is mined, and a device start-stop state prediction model is constructed; the initial solution output by the start-stop state prediction model is used to reduce the complexity of the online operation optimization problem and improve the solving efficiency, and finally a rule guidance-strict constraint collaborative multi-energy coupling system online operation optimization strategy is established. The specific steps are as follows:

[0305] Step 1: Based on the long / short time domain-coarse / fine time granularity coupling concept, a two-stage offline operation optimization method guided by knowledge rules and strictly constrained is constructed.

[0306] Step 1.1: Establish a knowledge rule guided solution algorithm;

[0307] Step 1.2: Use the knowledge rule guided solution algorithm to obtain multi-energy coupling system annual long-term operation data for time period slicing and aggregation analysis;

[0308] Step 1.3: Perform time period slicing and aggregation analysis on the multi-energy coupling system annual long-term operation data to extract system operation boundaries and evaluation index constraints at different time periods;

[0309] Step 1.4: Establish a strict constraint optimization algorithm based on step 1.3 to carry out short-term oriented strict constraint optimization and obtain an offline operation strategy data set covering typical operating conditions;

[0310] Step 2: Construct a prediction model for the start-stop state of energy equipment based on the offline operation strategy data set;

[0311] Step 2.1: Based on the LSTM neural network, mine the time sequence correlation between the integer decision variables such as equipment start-stop and charge-discharge in the offline operation strategy data set and external boundary conditions such as wind power and photovoltaic output;

[0312] Step 2.2: Based on the time sequence correlation obtained in step 2.1, construct a time sequence prediction model for the start-stop state of energy equipment to predict the start-stop state of future energy equipment through future wind and light output time sequence data;

[0313] Step 3: Based on the energy equipment start-stop state time sequence prediction model established in step 2.2, obtain the initial solution of the energy equipment start-stop state;

[0314] Step 4: Establish an online rolling operation optimization efficient solution method based on the start-stop state prediction model;

[0315] Step 4.1: Based on the initial solution provided by the start-stop state prediction model, use the assignment acceleration method to reduce the decision space and simplify the optimization problem, thereby improving the solution efficiency of online operation optimization;

[0316] Step 4.2: Establish a solution strategy for large-scale mixed integer optimization problems based on the initial solution assignment;

[0317] Step 5: Establish the operation strategy of the online operation optimization method of the low-carbon power generation system;

[0318] Step 6: According to the operation strategy established in step 5, optimize and dispatch the low-carbon power generation system.

[0319] In step 1, the knowledge rule is specifically represented as follows: assuming that the boiler is continuously operated, the thermal load is always not lower than the minimum limit value; when the output power of the power generation system (wind power, photovoltaic, coal-fired power plant) exceeds the power load demand, i.e. there is power surplus, according to the energy conversion efficiency, the energy conversion equipment is sequentially charged, and the charging order is the electric energy storage system, the thermal energy storage system, the ammonia energy storage system and the methanol preparation system; when the output power of the power generation system is lower than the power load demand, i.e. there is power shortage, then the equipment with high efficiency and low carbon emission is preferentially dispatched to alleviate the power shortage, and the dispatching priority is in turn the electric energy storage discharge, the thermal energy storage releases heat energy to drive the steam turbine to generate electricity, the ammonia energy storage releases ammonia gas into the boiler for blending combustion, and the coal consumption is increased to increase the boiler heat output power. The climbing constraints, the shortest start-stop time constraints of the electric energy storage, the thermal energy storage, the ammonia energy storage and the methanol preparation system are ignored.

[0320] Based on the knowledge rule, a knowledge rule guided multi-energy coupling system sequential solving algorithm is established, as shown in Figure 2 The knowledge rule guided multi-energy coupling system sequential solving algorithm is shown in the specific process.

[0321] Example 3

[0322] The effect of the rule-guided-strict-constraint two-stage collaborative off-line operation optimization method is verified, and the continuous 48-hour operation data of the multi-coupling net-zero carbon emission energy supply system is selected as the verification sample. The annual long-term operation data obtained by the knowledge rule guided solving method is combined, and the February operation data is daily sliding with a time window of two days, generating 27 sets of continuous two-day operation data sets, and performing aggregation analysis. The distribution of key indicators such as degree electricity cost, methanol output, degree electricity carbon emission, load satisfaction rate and abandoned wind / light rate is shown in Figure 3 .

[0323] Based on the average value of the key indicators, the operation boundary conditions and performance evaluation index constraints suitable for strict constraint optimization are extracted, as shown in Table 1.

[0324] Table 1 Operation boundary and evaluation index constraints provided by knowledge rule guided solving for short-term strict constraint optimization

[0325] Operation boundary or evaluation index constraint Numerical value Wind and light curtailment rate ≤10.3% Methanol production (tons) ≥1184 Carbon emission per kilowatt-hour (g / kWh) ≤46 Cost per kilowatt-hour (yuan / kWh) ≤1.01 Load satisfaction rate ≥98.7% Initial SOC of electric energy storage system 27% Initial SOC of thermal energy storage system 82%

[0326] Figure 4The operation strategy comparison analysis of the two methods in the 48-hour operation cycle is shown. The knowledge rule guided solution adopts a time-sharing solution strategy, and realizes fast solution by optimizing the equipment operation state hour by hour. However, this method lacks a cross-period optimization mechanism, resulting in large-scale wind and light curtailment after the energy storage device is fully charged during the day. In contrast, the knowledge rule guided-strict constraint two-stage collaborative optimization method optimizes the charging and discharging strategy of the energy storage through mathematical programming to plan the whole cycle operation. It uses the remaining energy of the electric energy storage and the thermal energy storage system to prepare methanol during the night load valley period, reserves buffer space for renewable energy consumption during the day, and significantly reduces wind and light curtailment.

[0327] As shown in Table 2, the knowledge rule guided-strict constraint two-stage collaborative optimization reduces the wind and light curtailment rate from 19.1% to 4.4%, while the methanol production increases by 57%, the carbon dioxide net emission is optimized from 27 tons to-55 tons, and the system achieves negative carbon emission. In terms of economy, thanks to the methanol preparation benefit, the system degree electricity cost decreases from 0.891 yuan / kWh to 0.816 yuan / kWh, with a decrease of 8.4%.

[0328] Table 2 Comparison of evaluation indexes of knowledge rule guided solution and two-stage collaborative optimization

[0329] Evaluation index Knowledge rule guided solution Two-stage collaborative optimization Decrease or increase Wind and light curtailment rate 19.1% 4.4% ↓77% Methanol production (tons) 1613 2530 ↑57% Net carbon dioxide emission (tons) 27 -55 / Carbon emission per kilowatt-hour (g / kWh) 0.9 -1.8 / Cost per kilowatt-hour (yuan / kWh) 0.891 0.816 ↓8.4% Load satisfaction rate 100% 99.1% ↓0.9%

[0330] Example 4

[0331] Since the prediction results of the LSTM neural network prediction model for start-stop variables have certain randomness, there may be a situation that the equipment runs according to the predicted start-stop state, which violates the start-stop constraint, such as the time from stopping to starting being less than the minimum start-stop time. Therefore, the initial prediction results of the LSTM neural network need to be corrected based on the start-stop state constraint. The specific process is shown in Figure 5 First, the start-stop state variables in the prediction time domain are detected, and the continuous segments of the start-stop state without state switching are recorded and recorded. Second, it is judged whether the duration of the continuous segment meets the minimum start-stop time constraint. If it meets, the next time step is corrected. If it does not meet, according to the degree of violation of the constraint, the violated segment is selected to switch the state or expand the current state to the adjacent period until all the continuous segments of the start-stop state are corrected.

[0332] The average prediction accuracy is used as the core monitoring index to train the LSTM neural network, and a high-accuracy start-stop state prediction model is successfully constructed. The prediction analysis result of the prediction model on the data set is as follows Figure 6The data set is the running data of 24 hours, sampled at 5-minute intervals, a total of 288 sample points. To improve the rationality of the prediction results, a correction strategy is used to optimize the prediction results. The prediction results show that the average prediction accuracy of the equipment start-stop state is more than 95%.

[0333] Example 5

[0334] The performance differences of three solving methods for online operation optimization problems, i.e. direct solving, assignment acceleration solving, and day-ahead-intra-day-real-time hierarchical solving, under typical operating conditions are compared and evaluated. The performances of the three methods in terms of operation scheduling strategy, economy, and system reliability are analyzed to verify the improvement effect of the assignment acceleration solving method compared with other methods. The settings of the decision time domain, time resolution, and rolling period of the three solving methods are shown in Table 3. To ensure the fairness of the comparison of the solving methods, the time resolution of each layer of the hierarchical solving is uniformly set to 5 minutes, which is the same as the time resolution of the assignment acceleration solving and direct solving methods.

[0335] Table 3 Parameter settings of the three solving methods

[0336] Solution method Decision time domain Time resolution Rolling period Direct solution 72 hours 5 minutes 5 minutes Assignment solution 72 hours 5 minutes 5 minutes Day-ahead-intra-day-real-time hierarchical solution Day-ahead: 72 hours Intra-day: 6 hours Real-time: 30 minutes Day-ahead: 5 minutes Intra-day: 5 minutes Real-time: 5 minutes Day-ahead: 24 hours Intra-day: 15 minutes Real-time: 5 minutes

[0337] Among them, the day-ahead-intra-day-real-time hierarchical solving method, the optimization function settings of each level are as follows: the day-ahead optimization is used to develop the start-stop plan of the boiler, ammonia energy storage, and methanol preparation system to guide the lower layer intra-day optimization; the intra-day optimization further develops the start-stop plan of the thermal energy storage system based on the day-ahead optimization to guide the lower layer real-time optimization; the real-time optimization stage optimizes the operation output of all subsystems based on the start-stop plan provided by the intra-day optimization.

[0338] The operation strategy comparison of the three different online operation optimization methods is shown in Figure 7 The hierarchical solving method has a shorter real-time decision time domain. To achieve the short-term optimal objective function, it frequently calls the energy storage system to prepare methanol during the day, resulting in insufficient remaining power at night and unable to meet 100% of the user load, causing power shortages; the direct solving method does not frequently call the energy storage system to prepare methanol, and there is still remaining power at night to meet the user load, but due to the longer time consumption of direct optimization, it cannot develop a scheduling strategy in time during some periods, resulting in wind and light curtailment or load shortages; the assignment acceleration solving method has a long enough decision time domain and fast enough calculation speed, so it balances economy and reliability and can meet 100% of the user load.

[0339] Table 4 compares the key evaluation indicators of the three methods under the regular operation scenario. The hierarchical solution method has the highest benefit, reaching 2.18 million yuan, which is mainly due to its relatively short-sighted scheduling strategy, which frequently calls the energy storage system to produce methanol during the day to maximize benefits. The short-sightedness of the strategy also results in a load satisfaction rate of only 92.52%, and when the night load cannot be fully met, the coal-fired power plant has to run at high load, emitting a large amount of carbon dioxide, resulting in a total carbon emission greater than zero throughout the day. In contrast, the direct solution method reduces the benefit to some extent, but improves the load satisfaction rate, and because the decision-making time domain is longer, the scheduling strategy does not need to increase the coal power output at night to meet the electricity demand, and the carbon emissions are greatly reduced, achieving the goal of negative carbon emissions with a total carbon emission less than zero throughout the day. The assignment acceleration solution achieves the best balance between economic, reliability and carbon emissions, not only 100% meets the electricity load, but also achieves the goal of negative carbon emissions, and its benefit is higher than that of the direct optimization method.

[0340] Table 4 Comparison of evaluation indicators of different solution methods under the regular operation scenario

[0341] Evaluation index Hierarchical solution Direct solution Assignment accelerated solution Total variable cost (ten thousand yuan, negative value represents benefit) -218.32 -124.67 -148.34 Wind and light curtailment rate 0.00% 0.55% 1.03% Load satisfaction rate 92.52% 97.40% 100.00% Net carbon emission (tons) 1415.0 -255.4 -496.4

Claims

1. A method for online operation optimization of a rule-constraint coordinated multi-energy coupled system, characterized in that, Includes the following steps: Step 1: Based on the concept of coupling long / short time domain and coarse / fine time granularity, construct a two-stage offline operation optimization method guided by knowledge rules and coordinated by strict constraints; Step 1.1: Establish a knowledge rule-guided solution algorithm; Step 1.2: Use knowledge rules to guide the solution algorithm and obtain the long-term annual operation data of the multi-energy coupled system for time-segmented slice aggregation analysis; Step 1.3: Perform time-segmented aggregation analysis on the long-term annual operation data of the multi-energy coupled system to extract the system operation boundaries and evaluation index constraints for different time periods; Step 1.4: Establish a strict constraint optimization algorithm. Based on Step 1.3, conduct strict constraint optimization for short-term conditions and obtain an offline operation strategy dataset covering typical working conditions. Step 2: Construct a predictive model for the start-up and shutdown status of energy equipment based on the offline operation strategy dataset; Step 2.1: Based on LSTM neural network, mine the temporal correlation between integer decision variables such as device start-up / stop and charging / discharging in offline operation strategy data and external boundary conditions such as wind power and photovoltaic power output; Step 2.2: Based on the temporal correlation obtained in Step 2.1, construct a temporal prediction model for the start-up and shutdown status of energy equipment, and predict the start-up and shutdown status of future energy equipment through future wind and solar power output time-series data; Step 3: Based on the timing prediction model of energy equipment start-up and shutdown states established in Step 2.2, obtain the initial solution of energy equipment start-up and shutdown states; Step 4: Establish an efficient solution method for online rolling operation optimization based on the start-stop state prediction model; Step 4.1: Based on the initial solution provided by the start-stop state prediction model, the decision space is reduced in dimensionality by the assignment acceleration method, which simplifies the optimization problem and improves the solution efficiency of online optimization. Step 4.2: Establish a solution strategy for large-scale mixed integer optimization problems based on initial solution assignment; Step 5: Establish an operational strategy for optimizing the online operation of low-carbon power generation systems; Step 6: Optimize the scheduling of the low-carbon power generation system based on the operation strategy established in Step 5.

2. The online operation optimization method for a rule-constraint cooperative multi-energy coupled system according to claim 1, characterized in that: Assuming the boiler operates continuously and the heat load never falls below the minimum limit; when the output power of the power generation system exceeds the power load demand, i.e., there is a power surplus, the energy conversion equipment is charged sequentially according to its energy conversion efficiency, in the order of electric energy storage system, thermal energy storage system, ammonia energy storage system, and methanol production system; when the output power of the power generation system is lower than the power load demand, i.e., there is a power shortage, the equipment with high efficiency and low carbon emissions is prioritized to alleviate the power shortage, with the priority order being: electric energy storage discharge, thermal energy storage releasing heat to drive the turbine to generate electricity, ammonia energy storage releasing ammonia to be co-fired in the boiler, and increasing coal consumption to increase the boiler's thermal output power; the ramp-up constraints and shortest start-up and shutdown time constraints of the electric energy storage, thermal energy storage, ammonia energy storage, and methanol production systems are ignored; based on the aforementioned knowledge rules, a knowledge rule-guided sequential solution algorithm for multi-energy coupled systems is established; The decision variables for the mathematical programming problem of a multi-energy coupled low-carbon power generation system are the operating power of each device; the objective function is to minimize the total variable operating cost of the system; and the constraints cover subsystem operating constraints, power balance constraints, heat balance constraints, etc. The mathematical expression of the optimization problem in the strict constraint optimization stage is as follows: (1)； ； ； ； in, —The variable operating cost of a low-carbon energy system at time t; —Strictly constrain the total number of steps in the optimization phase; --future Discrete variables representing the start / stop status of all devices within a time step are used to determine whether each device is in operation at each time step. --future The operating power of all devices within a time step is a continuous variable. —Operational constraints of the subsystem sys; —Power balance constraints; —Heat balance constraint.

3. The online operation optimization method for a rule-constraint cooperative multi-energy coupled system according to claim 1, characterized in that: In step 2, a prediction model for the start-up, shutdown, and charging / discharging status of equipment is established based on an LSTM neural network. The model input is the time series data of photovoltaic power, wind power, and electricity demand for the next n time steps, and the output is the start-up and shutdown status of equipment such as boilers, electric energy storage, thermal energy storage, ammonia energy storage, and methanol production systems within the next n time steps. The evaluation metrics for the prediction model of equipment start-up, shutdown, and charging / discharging status are prediction accuracy, overall prediction accuracy of the dataset, and average prediction accuracy, calculated as follows: (2)； (3)； in, —In sample i, the model's prediction accuracy for the start-stop state variable; —In sample i, the number of samples in which the model correctly predicts the start-stop state variable i; n—the total number of steps for each sample, which is 864 for a three-day, five-minute prediction. —The model's prediction accuracy for the start-stop state variable across the entire dataset; —The total number of samples in the dataset; —The average prediction accuracy of the entire dataset; —The total number of predictor variables.

4. The online operation optimization method for a rule-constraint cooperative multi-energy coupled system according to claim 3, characterized in that: The steps for correcting the initial prediction results of the LSTM neural network based on device start / stop state constraints are as follows: First, the start and stop state variables in the prediction time domain are detected, and the continuous segments of start and stop states without state transitions are recorded. Secondly, determine whether the duration of the continuous segment meets the minimum start-stop time constraint. If it does, then correct the next time step. If it does not, then depending on the degree of constraint violation, choose to switch the state of all violating segments or extend the current state to the adjacent time period until all start-stop state continuous segments have been corrected.

5. The online operation optimization method for a rule-constraint cooperative multi-energy coupled system according to claim 1, characterized in that: In step 4, the online operation optimization adopts a rolling optimization method. In the actual rolling optimization process, not all scheduling instructions for the next three days in one optimization are executed, but only the scheduling instructions for the first time step of each rolling optimization are executed. At the same time, the equipment operating status after the scheduling instructions are executed is passed to the next optimization cycle. Through continuous rolling optimization and execution of scheduling instructions, the optimal operation of the low-carbon energy system is achieved. The optimization objective is to minimize the total variable cost over the next three days. The decision variables are the discrete variables of the start-stop status of all equipment and the continuous variables of the operating power of all equipment within a total of 864 time steps over the next three days. In the rolling optimization process, the mathematical expression for each step of the optimization problem is as follows: (4)； in, —The variable operating cost of a low-carbon energy system at time t; —Discrete variables representing the start / stop status of all devices over the next 864 time steps, used to determine whether each device is in operation at each time step; —The continuous variable of operating power of all devices over the next 864 time steps.

6. The online operation optimization method for a rule-constraint cooperative multi-energy coupled system according to claim 5, characterized in that: The assignment acceleration method in step 4.1 is based on the future start-up and stop plans of each device provided by the start-up and stop state prediction model. It directly assigns values ​​to the discrete variables of start-up and stop states, eliminates the discrete variables of start-up and stop states in the decision variables, reduces the original mixed integer linear programming problem to a linear programming problem, reduces the difficulty of solving the problem, and thus reduces the time and resources required for computation. After using the assignment acceleration method, the mathematical expression of the optimization problem becomes: (5)； in, —Predicted values ​​of the discrete variables of the start-up and shutdown status of all devices over the next 864 time steps using the LSTM prediction model.

Images