Multi-park cross-space-time comprehensive energy consumption multi-objective optimization method and system based on deep learning

Abstract

The invention discloses a multi-park cross-space-time comprehensive energy consumption multi-objective optimization method and system based on deep learning, and the method comprises the steps: carrying out the feature extraction of data in the operation of a multi-park energy system through an extended long-short term memory network model, and constructing a multi-park multivariate energy efficiency evaluation index model according to the extracted multi-dimensional energy consumption features; evaluating the correlation degree of the operation process of the multi-park integrated energy system, and constructing an energy complementary relation matrix; according to the multivariate energy efficiency evaluation index model and the energy complementary relation matrix, constructing a multi-park system multi-target optimization model; solving the optimal solution of the optimization model under the operation constraint through a Transform-based multi-target particle swarm optimization algorithm, and obtaining an energy operation scheduling strategy of the integrated energy system of the plurality of parks; user interaction feedback is adopted in optimization, so that the result better meets the actual demand and target of the user; according to the method, cross-space-time comprehensive energy multi-target optimization of multiple parks can be realized, and the comprehensive energy utilization efficiency of the parks is improved.

Description

Technical Field

[0001] The present invention relates to the field of energy system management technology, and in particular to a multi-objective optimization method and system for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning. Background Art

[0002] Industrial parks are the economic engines of many regions, often housing numerous energy-intensive enterprises that consume significant amounts of energy and emit greenhouse gases. Therefore, industrial parks are an inevitable choice for energy transition. The operation of integrated energy systems in industrial parks is typically characterized by high terminal cooling, heating, and electrical loads, high penetration of new energy sources, high demand for energy consumption, and relatively low energy efficiency. Furthermore, industrial park energy systems account for a significant portion of total electricity consumption in society. Their energy load characteristics are closely linked to industry characteristics and process flows, exhibiting distinct geographical agglomeration effects and regional industrial characteristics. Different types of industrial park integrated energy systems operate in distinct modes, with distinct energy utilization mechanisms. Against the backdrop of the increasing proportion of new energy sources and the continuous improvement of the electricity market, industrial park electricity demand is becoming increasingly diverse and complex, and demands for power supply quality are also increasing. This poses new challenges to the power system, and the issue of optimizing energy consumption to meet grid demands is becoming increasingly prominent.

[0003] However, the existing park energy optimization methods cannot adapt to the energy optimization scheduling needs of parks with increasingly diversified and complex electricity demands, nor can they effectively deal with the complementary characteristics of multiple parks in the time domain and complex multi-objective optimization problems, resulting in a lot of energy waste in the energy distribution process of the park's integrated energy system. Summary of the Invention

[0004] In response to the problems existing in the existing technology, the embodiments of the present invention provide a multi-objective optimization method and system for comprehensive energy consumption across multiple parks, across time and space, based on deep learning. In the optimization solution process, user interactive optimization is adopted, and user opinions and decisions are combined to make the optimization results more consistent with the user's actual needs and goals. This method can achieve comprehensive energy multi-objective optimization across time and space for multiple parks, meet the energy optimization scheduling needs of parks with increasingly diversified and complex electricity demands, optimize the cross-time and space energy distribution of park groups, improve the comprehensive energy efficiency of parks, and reduce computing costs, making the optimization results more consistent with the user's actual needs and goals.

[0005] In the first aspect, an embodiment of the present invention provides a multi-objective optimization method for multi-park cross-temporal and spatial integrated energy consumption based on deep learning, including: extracting features of multivariate heterogeneous data in the operation process of the integrated energy systems of multiple different types of parks through a pre-trained extended long short-term memory network model, and obtaining the multidimensional energy consumption characteristics of the integrated energy systems of each park; constructing a multivariate energy efficiency evaluation index model for different types of parks based on the entropy weight method, hierarchical analysis method, and design structure matrix according to the multidimensional energy consumption characteristics of multiple different types of parks; performing correlation evaluation on the operation process of the integrated energy systems of different types of parks, and analyzing the complementary characteristics of different types of parks in the time domain. , construct a park energy complementary relationship matrix; according to the operation constraints constructed based on the energy consumption characteristics of different types of parks, the preset multiple control mechanisms, the multi-element energy efficiency evaluation index model, and the park energy complementary relationship matrix, construct multiple objective functions for multi-element energy efficiency optimization, and establish a multi-objective optimization model for comprehensive regulation of multi-park energy systems based on the multiple objective functions; adopt a multi-objective particle swarm optimization method based on the Transformer model and user satisfaction feedback interaction to solve the multi-objective optimization model for comprehensive regulation of multi-park energy systems under preset operation constraints, and obtain energy operation scheduling strategies for comprehensive energy systems of multiple different types of parks.

[0006] As an improvement of the above solution, the multi-dimensional energy consumption characteristics include: electricity consumption characteristic data, production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data under different operating modes and different seasons.

[0007] As an improvement of the above-mentioned scheme, the method also includes: screening and classifying various types of data in the multivariate heterogeneous data, and using fuzzy hierarchical analysis method to construct a three-layer energy consumption index evaluation system, wherein the energy consumption index evaluation system includes: target layer indicators, first-level indicators under the target layer indicators, and second-level indicators under the first-level indicators.

[0008] As an improvement to the above scheme, the pre-trained extended long short-term memory network model is used to extract features from multivariate heterogeneous data during the operation of the integrated energy systems of multiple different types of parks to obtain multidimensional energy consumption characteristics of the integrated energy systems of each park, including: extracting time series from the multivariate heterogeneous data during the operation of the integrated energy systems of multiple different types of parks to obtain multidimensional time series data; inputting the multidimensional time series data into the extended long short-term memory network model, wherein the extended long short-term memory network model includes an input layer, an xLSTM layer, and an output layer;

[0009] The input item of the input layer: the multidimensional time series data is defined as:

[0010] Wherein, T represents the time series length of the multidimensional time series data, d represents the feature dimension of each time step, and x t represents the t-th input feature in the multidimensional time series data;

[0011] The xLSTM layer is used to capture the local dynamic characteristics of the multidimensional time series data through XLSTM; in the xLSTM layer, the input feature The intermediate features are obtained through nonlinear transformation Among them, g represents the nonlinear activation function, W x represents the weight matrix of the xLSTM layer, b x Represents the xLSTM layer bias term;

[0012] Update hidden state and cell state:

[0013]

[0014] h t =O t ⊙tanh(C t );

[0015] Among them, σ represents the Sigmoid function, W f , W i , W c represents the weight matrix of the hidden layer, b f , b i , b c represents the bias term of the hidden layer, h t represents the hidden state, C t Represents the state of the memory unit, ⊙ represents element-by-element multiplication, o t Represents the output gate result;

[0016] The hidden state sequence output by the xLSTM layer is

[0017] The self-attention score of the hidden state sequence output by the xLSTM layer is calculated using a Transformer-based multi-head attention mechanism. The self-attention score calculation process of each attention mechanism includes:

[0018] Q=H xLSTM W Q , K=H xLSTM W K , V=H xLSTM W V ; Where Q represents the query vector, K represents the key vector, V represents the value vector, and W Q ,W K ,W VRepresents the weight matrix of a trainable multi-head attention mechanism.

[0019] Self-attention score Among them, d k is the dimension of the key vector K. The superscript T of the key vector K represents the vector transpose.

[0020] The multi-head attention mechanism is integrated to obtain the global feature H attn =MultiHead(Q,K,V)=concat(head1,...,head n )Wo; among them, W O Indicates the transformation matrix of the final linear transformation, head n It is calculated by the self-attention score Attention(Q,K,V), where h represents the number of heads in the multi-head attention mechanism; the hidden state sequence H xLSTM As a local feature, by fusing the local feature H xLSTM and global feature H attn , we get the multi-dimensional energy consumption characteristics: H final =ReLU(W f ·[H xLSTM ;H attn ]+b f ).

[0021] As an improvement to the above scheme, the multivariate energy efficiency evaluation index model of different types of parks is constructed based on the entropy weight method, the hierarchical analysis method, and the design structure matrix according to the multidimensional energy consumption characteristics of multiple different types of parks, including: using the entropy weight method in combination with the hierarchical analysis method, according to the multidimensional energy consumption characteristics of multiple different types of parks, constructing a target layer indicator evaluation system for different types of parks including energy consumption indicators, economic indicators, environmental indicators, and product quality indicators; and based on the target layer indicator evaluation system, after calculating the scores of the secondary indicators, dividing them into different groups according to the seasonal labels, and calculating the weights of the secondary indicators in different seasons; according to the weights of the secondary indicators in different seasons, calculating the scores of the target layer indicators, so as to construct a multivariate energy efficiency evaluation index model for each different type of park; wherein, the calculation process of the indicator scores includes:

[0022] Calculate the characteristic proportion of different areas in the park under each secondary indicator in, It represents the score of the i-th unit in the park on the j-th secondary indicator in season k, and n represents the total number of units;

[0023] According to the characteristic proportion of each secondary indicator under the secondary indicator, the entropy value of each secondary indicator in each season is calculated.

[0024] According to the entropy value of each secondary indicator, calculate the difference coefficient

[0025] The secondary indicators are grouped according to seasons, and the weights of the secondary indicators in different seasons are calculated based on the difference coefficients.

[0026] According to the weight of each secondary indicator, calculate the total score of the weighted secondary indicator layer in different seasons in, It represents the average score of the secondary indicator j of each unit in the park in season k; Indicates the score of the first-level indicator a corresponding to the second-level indicator involved in the calculation in season k;

[0027] According to the scores of the first-level indicators, the total score of the first-level indicator layer is calculated;

[0028] The analytic hierarchy process is used to obtain the weights of the first-level indicators, and based on the weights of the first-level indicators, the scores of the target layer indicators corresponding to the park's energy consumption indicators, economic indicators, environmental indicators, and energy product quality indicators are calculated.

[0029] As an improvement to the above solution, the method further includes: evaluating the importance of different target layer indicators under different park types' standards based on the scores of each target layer indicator in the multivariate energy efficiency evaluation index model for each park, and calculating the park's comprehensive energy efficiency evaluation score; and calculating the park's comprehensive score by designing a structural matrix R based on the park's comprehensive energy efficiency evaluation score;

[0030]

[0031] Among them, the subscript 1 of each element a in the matrix R corresponds to the energy consumption index, 2 corresponds to the economic index, 3 corresponds to the environmental index, and 4 corresponds to the product quality index. The order of the indicators in the subscripts indicates the influence of the previous indicator on the next indicator: In the matrix R, the value of each element indicates the level of mutual influence between the two indicators. When the element value is 0, it means no influence; when the element value is 1, it means weak influence; when the element value is 2, it means medium influence; when the element value is 3, it means strong influence;

[0032] The comprehensive energy efficiency evaluation score of the park is calculated as follows:

[0033] Score=||C s R||1;

[0034] Among them, C s =[Y1, Y2, Y3, Y4] represents the scores of each target layer indicator of the park, and Y1, Y2, Y3, and Y4 represent the scores of energy consumption indicators, economic indicators, environmental indicators, and product quality indicators respectively.

[0035] As an improvement to the above scheme, the operation process of the integrated energy system of different types of parks is evaluated for correlation, the complementary characteristics of different types of parks in the time domain are analyzed, and a park energy complementary relationship matrix is ​​constructed, including: using the correlation coefficient method to analyze the complementary characteristics of different types of parks in the time domain, and constructing a complementary relationship matrix C: wherein the element C in the complementary relationship matrix C is IJ represents the complementarity strength between process I of a certain park and process J of another park; C IJ =w1·(1-|ρ IJ |)+w2·(1-MIC IJ ), w1, w2 represent the preset weight parameters, C IJ represents the MIC value between process I and process J, ρ IJ Represents the Spearman correlation coefficient between process I and process J; the processes include energy, energy transmission, energy storage, and energy use; sort the non-diagonal elements in the complementary relationship matrix, select the element with the largest value, and construct the park energy complementary relationship matrix.

[0036] As an improvement to the above solution, the method further includes: establishing a comprehensive control mechanism model for the operation process of the multi-park energy system based on the energy usage characteristics of different types of parks, the park energy complementarity relationship matrix, and a plurality of selected control mechanisms; the plurality of control mechanisms include: a cogeneration unit capable of simultaneously providing heat and power; a gas boiler unit using liquefied gas, natural gas, or other gas energy as fuel, heating water into steam through a heating furnace to provide the required heat energy for users;

[0037] The comprehensive control mechanism model includes: a dynamic model of wind power generation and photovoltaic power generation, a dynamic model of refrigerators and heat exchangers, and a dynamic model of electrical equipment and energy storage equipment.

[0038] As an improvement to the above solution, the method further includes: constructing a multi-park integrated energy system that must meet multi-dimensional collaborative operation constraints during operation; wherein the operation constraints include:

[0039] Equipment operating limit constraints for the integrated energy system within the park:

[0040] Equipment installation constraints:

[0041] in, Indicates the maximum and minimum configuration numbers of different devices k;

[0042] Equipment operation constraints:

[0043]

[0044] Among them, γ k∈(0,1) is a state variable describing whether device k is running: γ k =0 means the device is not running, γ k =1 means the equipment is operating normally; min is the minimum output power coefficient of the equipment, P t k is the operating power of device k at time t;

[0045] Electric / thermal energy storage operation constraints:

[0046]

[0047] in, Indicates the maximum and minimum values ​​of the remaining capacity of device k; are the charge / discharge / heating power of the electrical energy storage and thermal energy storage of device k at time t, ω char_k 、ω dis_k They represent the charging, discharging / thermal efficiency of the device k, respectively, and are in the range of [0,1];

[0048] Maximum transmission limit constraints for purchased electricity and gas power:

[0049]

[0050] in, In order to set the power limit for purchasing electricity, natural gas and thermal energy from the energy grid, the natural gas purchased is considered to be consumed mainly by CHP units and gas boilers.

[0051] Inter-park electrical and thermal energy interconnection lines must be within the maximum transmission power constraints:

[0052]

[0053] in is the power transmission power and heat transmission power between park i and park j, are the upper limits of electric energy exchange and thermal energy exchange between park i and park j, respectively;

[0054] Power balance constraints include the supply and demand balance of various energy types, including electricity, heat, gas, and cooling. For a multi-campus integrated energy system, the energy transmission balance between parks can be described using the Energy Hub (EH) model:

[0055]

[0056] Where L is the output matrix of the energy hub, L e ,L g ,L h ,L cis the electricity load, gas load, heating load and cooling load on the user side, η t ,η he are transformer efficiency and CHP unit power generation efficiency, P b , G b 、H b They are the electricity purchase amount, gas consumption of CHP unit and gas boiler, P PV 、P WT are photovoltaic and wind power generation power, are the conversion efficiencies of natural gas to electricity and heat, respectively. ng is the calorific value of natural gas, η ec ,η ac are the conversion efficiencies of electric refrigeration and absorption refrigeration respectively; G mt is the natural gas input, P ec is the electrical energy input power, H ac is the refrigerator input power, P es,d 、P es,c 、P gs,d 、P gs,c 、P hs,d 、P hs,c 、P cs,d 、P cs,c They are the charging and discharging power of electric energy, thermal energy, cold energy and gas respectively;

[0057] Inter-park energy exchange constraints: The exchange power between different parks should meet global balance, that is, the algebraic sum of the exchange power is 0. Specifically, it can be expressed as:

[0058]

[0059] in They are the overall interactive electric power and thermal power between different parks respectively.

[0060] As an improvement of the above scheme, the method also includes: establishing a first objective function f1 with the goal of minimizing total energy consumption; establishing a second objective function f2 with the goal of minimizing energy costs; establishing a third objective function f3 with the goal of minimizing carbon emissions; and establishing a fourth objective function f4 with the goal of system operation quality.

[0061] As an improvement of the above scheme, the multi-objective particle swarm optimization method based on the interaction of the Transformer model and user satisfaction feedback is used to solve the multi-objective optimization model of the comprehensive regulation of the multi-park energy system under the preset operation constraints, and obtain the energy operation scheduling strategy of the comprehensive energy system of multiple different types of parks, including: initialization of the particle swarm configuration related to user orientation: in the initial stage of the particle swarm optimization algorithm, the initialization of the three-dimensional core parameters is completed, and the initialized parameters include: particle space coordinates, motion rate vector and attention weight parameters; initialization of the optimal solution set that meets the user's global optimization expectations; fitness calculation in multi-objective optimization: calculation of the fitness value of each particle on each objective function in the multi-objective optimization model; position encoding: position encoding of each particle to obtain the position information of the embedded input vector of the Transformer model; Transformer model attention mechanism calculation: in the multi-head self-attention mechanism layer, the input vector is mapped to different subspaces, and the attention of each particle is calculated; evolutionary strategy optimization and attention weight Adjustment: Calculate the distance between each particle and the global optimal solution, and optimize the attention weight of each particle; Update particle speed and position: By constructing the coupling relationship between user evaluation adjustment coefficient, cognitive and social learning factors, a speed update equation with feedback gain is established; By introducing a dynamic relaxation factor, a position update equation is established; and according to the individual optimal solution and the global optimal solution of each particle, based on the speed update equation and the position update equation, the particle speed and position are updated; Dynamic management of position constraints: Based on the preset position update strategy, the particle position is boundary checked; Non-dominated solution set maintenance and optimal solution update: For the current particle, its performance is evaluated by the fitness function. If the performance of the current particle is higher than the performance of its historical individual optimal solution, the operation of updating the particle speed and position is re-executed; Global optimal solution set maintenance, using the Pareto dominance relationship to screen non-dominated solutions, the global optimal solution that meets the non-dominated conditions is added to the elite solution set, and the global optimal solution is updated based on the elite retention strategy; Based on the global optimal solution updated in this round, the energy operation scheduling strategy of the integrated energy system of multiple different types of parks is obtained.

[0062] As an improvement to the above scheme, the method also includes: when initializing the particle swarm configuration, supporting the user to weight each objective function to determine the weight of each objective function; and supporting the user to set the expected solution region; in the initialization phase, based on the weight of each objective function and the expected solution region, generating an initial optimal solution set that meets the user's global optimization expectations. In the fitness calculation process of multi-objective optimization, supporting the display of the Pareto solution of each iteration to the user, so that the user can score the Pareto solution and obtain a real-time satisfaction score of the corresponding Pareto solution; or selecting a Pareto solution that meets the user's own preferences; in the evolutionary strategy optimization and attention weight adjustment process, supporting a user feedback mechanism to dynamically adjust the attention weight according to the real-time satisfaction score of the user feedback; in the particle velocity and position update process, constructing a user satisfaction function for calculating user satisfaction for the selected Pareto solution according to the user satisfaction function, and adjusting the weights of each objective function according to the user satisfaction; in the dynamic management of position constraints, supporting a user feedback mechanism to adjust the operation constraints according to the real-time satisfaction score of the user feedback. The global optimal solution of each round of optimization is displayed to the user, and the user's satisfaction score for the global optimal solution is received; whether the satisfaction score of the global optimal solution meets the set termination condition is determined; if the termination condition is not met, the next round of iterative solution is continued; if the termination condition is met, the global optimal solution is output.

[0063] On the second aspect, an embodiment of the present invention provides a multi-park cross-temporal and spatial comprehensive energy consumption multi-objective optimization system based on deep learning, including: a feature extraction module, which is used to extract features of multivariate heterogeneous data in the operation process of the comprehensive energy system of multiple different types of parks through a pre-trained extended long short-term memory network model, and obtain the multidimensional energy consumption characteristics of the comprehensive energy system of each park; an indicator model construction module, which is used to construct a multivariate energy efficiency evaluation index model for different types of parks based on the entropy weight method, hierarchical analysis method, and design structure matrix according to the multidimensional energy consumption characteristics of multiple different types of parks; a complementary relationship matrix construction module, which is used to perform correlation evaluation on the operation process of the comprehensive energy system of different types of parks, and analyze the different types of parks from a time domain perspective. The complementary characteristics of the parks are used to construct a park energy complementary relationship matrix; a multi-objective optimization model construction module is used to construct multiple objective functions of multi-economy energy efficiency optimization according to the operation constraints constructed based on the energy consumption characteristics of different types of parks, the preset multiple control mechanisms, the multi-economy energy efficiency evaluation index model, and the park energy complementary relationship matrix, and establish a multi-objective optimization model for comprehensive regulation of multi-park energy systems based on the multiple objective functions; an optimization solution module is used to adopt a multi-objective particle swarm optimization method based on the Transformer model and user satisfaction feedback interaction to solve the multi-objective optimization model for comprehensive regulation of multi-park energy systems under preset operation constraints, and obtain energy operation scheduling strategies for comprehensive energy systems of multiple different types of parks.

[0064] Compared with the existing technology, the embodiment of the present invention is a multi-objective optimization method and system for multi-park cross-temporal and spatial comprehensive energy consumption based on deep learning. Through a pre-trained extended long short-term memory network model, feature extraction is performed on multivariate heterogeneous data in the operation process of the comprehensive energy system of multiple different types of parks to obtain the multidimensional energy consumption characteristics of the comprehensive energy system of each park; based on the multidimensional energy consumption characteristics of multiple different types of parks, a multivariate energy efficiency evaluation index model is constructed for multiple different types of parks; the operation process of the comprehensive energy system of different types of parks is correlated and a park energy complementary relationship matrix is ​​constructed; based on the multivariate energy efficiency evaluation index model and the park energy complementary relationship matrix, a multi-objective optimization model is constructed; the optimal solution of the multi-objective optimization model under preset operation constraints is solved by a multi-objective particle swarm optimization algorithm to obtain the energy operation scheduling strategy of the comprehensive energy system of multiple different types of parks; the embodiment of the present invention can realize the multi-objective optimization of comprehensive energy across time and space of multiple parks, meet the energy optimization scheduling needs of parks with diversified and complex electricity demand, reduce energy waste of park comprehensive energy systems, optimize the cross-temporal and spatial energy distribution of park groups, and improve the comprehensive energy efficiency of parks. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings used in the implementation methods. Obviously, the drawings described below are only some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0066] Figure 1 This is a flowchart of a multi-objective optimization method for multi-park cross-temporal and spatial comprehensive energy use based on deep learning provided by an embodiment of the present invention;

[0067] Figure 2 This is another flow chart of the multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy consumption based on deep learning provided by an embodiment of the present invention;

[0068] Figure 3 is a schematic diagram of an energy system topology provided by an embodiment of the present invention;

[0069] Figure 4 This is a multi-objective optimization flow chart of a multi-objective particle swarm optimization algorithm based on a Transformer model provided by an embodiment of the present invention;

[0070] Figure 5 This is a structural block diagram of a multi-park, cross-temporal and spatial comprehensive energy utilization multi-objective optimization system based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0072] See Figure 1 , Figure 1 This is a flowchart of a multi-objective optimization method for multi-park cross-temporal and spatial comprehensive energy consumption based on deep learning provided by an embodiment of the present invention. The multi-objective optimization method for multi-park cross-temporal and spatial comprehensive energy consumption based on deep learning specifically includes:

[0073] S11: Using a pre-trained extended long short-term memory network model, we extract features from multi-dimensional heterogeneous data during the operation of integrated energy systems in multiple different types of parks, and obtain the multi-dimensional energy consumption characteristics of the integrated energy systems of each park.

[0074] S12: Based on the multidimensional energy consumption characteristics of multiple different types of parks, a multivariate energy efficiency evaluation index model for different types of parks is constructed using the entropy weight method, hierarchical analysis method, and design structure matrix;

[0075] S13: Conduct a correlation assessment of the operation process of the integrated energy systems of different types of parks, analyze the complementary characteristics of different types of parks in the time domain, and construct a park energy complementary relationship matrix;

[0076] S14: Constructing multiple objective functions for multi-element energy efficiency optimization based on the operation constraints established based on the energy consumption characteristics of different types of parks, the preset multiple control mechanisms, the multi-element energy efficiency evaluation index model, and the park energy complementary relationship matrix, and establishing a multi-objective optimization model for comprehensive regulation of multi-park energy systems based on the multiple objective functions;

[0077] S15: A multi-objective particle swarm optimization method based on the interaction of the Transformer model and user satisfaction feedback is used to solve the multi-objective optimization model for the comprehensive regulation of multi-park energy systems under preset operational constraints, and the energy operation scheduling strategies for the integrated energy systems of multiple different types of parks are obtained.

[0078] For example, another process example of the multi-objective optimization method for multi-park cross-temporal and spatial comprehensive energy consumption based on deep learning in an embodiment of the present invention is as follows: Figure 2As shown in the figure, the specific process includes: S01: Collecting multi-source heterogeneous energy consumption data from multiple parks such as industrial, commercial, and residential parks, including dynamic time series data such as electricity consumption, production capacity, energy transmission, and energy storage, to form a multi-park multi-heterogeneous data set; S02: Designing an energy consumption index evaluation system, defining the four core goals of energy consumption, economy, environment, and energy quality at the target layer, refining the first-level indicators to key dimensions such as energy utilization rate, operating cost, and carbon emission intensity, and expanding the second-level indicators to 31 quantifiable indicators (such as energy consumption per unit area and equipment energy efficiency ratio); S03-S04: Extracting local time series features of energy consumption data (such as equipment operation fluctuations and short-term energy consumption trends) through the xLSTM network model, and combining the Transformer multi-head attention mechanism to mine global correlation features (such as cross-park energy complementarity and long-term energy consumption patterns) to form a collaborative relationship between multiple parks; S05: Constructing indicator calculations for different parks based on the hierarchical analysis method. Calculation method; S06: Construct an indicator design structure matrix, considering the coupling relationship between cross-park indicators (such as the sensitivity of peak power cost in commercial areas to energy storage scheduling in industrial areas); S07: Use the correlation coefficient method to analyze the time domain complementary characteristics of the parks, and construct the park energy complementary relationship matrix of energy supply and demand among multiple parks; S08: Establish a comprehensive regulation mechanism model for multi-park collaboration based on the complementary relationship matrix and energy control theory; S09: Analyze the constraints of the operating constraints that the multi-park energy system must meet, and form an optimization boundary constraint set; S10: Construct multi-objective functions for multi-park energy efficiency optimization such as energy consumption optimization, economic improvement, and low-carbon constraints; S11: Propose a multi-objective particle swarm optimization method based on the Transformer model to solve the multi-element energy efficiency optimization problem; S12: Combine the user satisfaction feedback mechanism in the optimization solution to obtain the energy operation and scheduling strategy of the multi-park integrated energy system under the guidance of user decision-making.

[0079] Among them, the types of parks include but are not limited to: industrial parks, science and technology parks, commercial parks, residential parks, etc. The embodiment of the present invention collects multi-dimensional heterogeneous data involved in the main processes of the operation of the integrated energy systems of various parks (including production processes, energy transmission processes, energy storage processes, and energy consumption processes), such as the four types of data affecting the energy consumption response factors, economic response factors, environmental response factors, and energy product quality response factors of the integrated energy system.

[0080] The electricity users in the industrial park mainly include:

[0081] Thermal power plant: A traditional thermal power plant used for power generation, which mainly uses coal, natural gas, etc. as fuel. Chemical plant: A factory that produces chemical products, which may include branches such as petrochemicals and fine chemicals, and equipment such as reactors and heaters consume a lot of electricity. Precision instrument factory: A factory that manufactures high-precision instruments and equipment, which maintains a constant temperature and humidity environment, and usually has electricity needs for large equipment. Science and technology park: Includes research laboratories, high-tech enterprises, etc., and has high power consumption for equipment and computer equipment. Manufacturing workshop: Such as mechanical processing workshops, metallurgical processing workshops, etc., mechanical equipment, welding equipment and other equipment have high power consumption. Industrial air-conditioning system: An air-conditioning system used to maintain a certain temperature and humidity, especially in the precision manufacturing process, which consumes a lot of electricity. Refrigeration and freezing facilities: Such as the refrigeration facilities of food processing plants and the condensing equipment of chemical plants, which are used for all electrical appliances in storage and production processes.

[0082] In commercial parks, the above four categories of factors are analyzed as follows: at the energy consumption level, the main energy consumption influencing factors include passenger flow and usage density, equipment utilization efficiency, business hours and building energy-saving design; at the economic level, the main economic influencing factors include rental income and sales, store business structure, surrounding passenger flow and transportation convenience, frequency of commercial activities, labor costs and service costs; at the environmental level, the main environmental influencing factors include waste management, public transportation facilities, green buildings and environmental certification, economical use of water and electricity resources, and air pollutant emissions; at the energy product quality level, the main energy product quality influencing factors include service quality, maintenance and updating of facilities, store product quality, safety measures, cleanliness and sanitary environment.

[0083] Major electricity users in business parks include: Shopping malls: Shopping centers and retail stores, where lighting, air conditioning, elevators, escalators, and other facilities consume significant amounts of electricity. Office buildings: Corporate and enterprise offices, where electricity consumption is primarily concentrated in air conditioning, lighting, computers, and servers. Hotels: Accommodation, dining, entertainment facilities, 24-hour air conditioning, hot water supply, and elevators all contribute significantly to electricity consumption. Convenience stores: Refrigeration equipment, air conditioning, lighting, and electric heating equipment are the primary sources of electricity consumption. Restaurants and cafes: Kitchen equipment, refrigeration and freezing equipment, lighting, and air conditioning are significant energy consumers. Cinemas and gyms: Multimedia equipment and large air conditioning systems are essential.

[0084] In residential areas, the above four types of factors are analyzed as follows: at the energy consumption level, the main energy consumption influencing factors include household appliance usage habits, climate and seasonal changes, building insulation performance, community facility electricity management, and the number of households; at the economic level, the main economic influencing factors include residents' income level, property value, operating costs of public service facilities, property management fees, and investment in property management quality infrastructure; at the environmental level, the main economic influencing factors include the treatment and classification of domestic waste, clean energy use, green vegetation coverage, transportation usage, noise and air pollution management; at the energy product quality level, the main energy product quality influencing factors include living environment quality, education and medical service quality, community facility maintenance, property service quality, and community safety.

[0085] The main electricity users in residential areas include: Residential buildings: Household electrical appliances such as air conditioners, refrigerators, washing machines, electric water heaters, and lighting. Residential facilities: Public lighting, elevators, and surveillance systems. Schools: Lighting, air conditioning, laboratory equipment, and multimedia equipment. Municipal entities: Community management centers, subdistrict offices, and other office equipment. Hospitals: Large medical equipment, air conditioning, lighting, and ward electricity demand are high.

[0086] The embodiment of the present invention selects typical parks from different types of parks, including industrial parks, commercial parks, and residential parks to optimize the integrated energy system of multiple parks. By collecting four types of data related to the integrated energy system of industrial parks, commercial parks, and residential parks, namely energy consumption response factors, economic response factors, environmental response factors, and energy product quality response factors, such as the use, energy consumption, energy utilization rate, energy-saving technology (such as waste heat recovery rate, energy-saving transformation investment) of various types of production equipment, power equipment, energy storage equipment, and energy transmission equipment during the operation of the integrated energy system, the output energy products, external factors (such as temperature and humidity, seasons), and other related data are used as energy consumption response factors to collect the operating cost, energy cost, business cost, etc. of the integrated energy system. Data on costs, economic benefits (such as power supply failure rates and power supply service quality), and economic risks (such as market fluctuations and natural disasters) are collected as economic response factors. Data on industrial carbon emissions, transportation carbon emissions, building carbon emissions, renewable energy utilization rates, fossil fuel utilization rates, sulfur dioxide emissions, nitrogen oxide emissions, particulate matter emissions, chemical wastewater emissions, industrial solid waste treatment rates, land use rates, and green coverage rates are collected as environmental response factors. Data on energy consumption efficiency, energy conversion efficiency, equipment consumption rates, failure rates, energy conversion efficiency, equipment consumption rates, customer satisfaction, customer complaint rates, energy load responsiveness, energy reserve levels, and maintenance response times are collected as energy product quality response factors. The above four types of data constitute the multi-faceted and heterogeneous data of the park's integrated energy system.

[0087] Furthermore, the method further comprises:

[0088] The various types of data in the multivariate heterogeneous data are screened and classified, and a three-layer energy consumption index evaluation system is constructed using the fuzzy hierarchical analysis method. The energy consumption index evaluation system includes: target layer indicators, first-level indicators under the target layer indicators, and second-level indicators under the first-level indicators.

[0089] For example, by screening and classifying the above four types of data, the fuzzy hierarchical analysis method is used to construct an energy consumption index evaluation system. For example, a target layer indicator is constructed, and the first-level indicators and second-level indicators are set under it to form a three-level energy consumption index evaluation system.

[0090] The indicators at different levels corresponding to the energy consumption index evaluation system are as follows:

[0091] 1. Energy consumption index (target layer index):

[0092] (1) Energy consumption (primary indicator), and its corresponding secondary indicator: energy consumption per unit area Energy consumption per unit product Energy consumption per unit time

[0093] Among them, E represents the total energy consumption of the integrated energy system, A represents the total usage area of ​​the integrated energy system, P represents the total output of energy products of the integrated energy system, and T represents the total operating time of the integrated energy system.

[0094] (2) Energy utilization rate (primary indicator), and its corresponding secondary indicator: equipment energy efficiency ratio Thermal efficiency Comprehensive energy utilization rate

[0095] Among them, Q out Indicates the output heat or cooling of the integrated energy system, W input represents the input energy of the integrated energy system, Q useful Indicates the effective utilization of heat energy in the integrated energy system, Q input Represents the total thermal energy output of the integrated energy system, E useful Indicates the energy effectively utilized by the integrated energy system, E total Represents the total energy input of the integrated energy system.

[0096] (3) Use of different types of energy (primary indicators), and their corresponding secondary indicators: electricity consumption of the integrated energy system; fossil fuel consumption of the integrated energy system; renewable energy consumption of the integrated energy system (including solar energy, wind energy, and biomass energy consumption, etc.).

[0097] (4) Energy-saving technology (first-level indicator) and its corresponding second-level indicator: waste heat recovery rate Energy-saving transformation investment

[0098] Among them, C investment represents the energy-saving transformation investment cost of the integrated energy system, C total represents the total investment cost of the integrated energy system, Q recovered The recovered heat of the integrated energy system, Q total Represents the total system heat of the integrated energy system.

[0099] (5) External factors (first-level indicators), and their corresponding second-level indicators: temperature information of the integrated energy system; humidity information of the integrated energy system.

[0100] 2. Economic indicators (target level indicators):

[0101] (1) Cost (first-level indicator), and its corresponding second-level indicators: operating cost of the integrated energy system; energy cost of the integrated energy system; operating cost of the integrated energy system.

[0102] (2) Benefit (first-level indicator), and its corresponding second-level indicators: reliability of power supply of the integrated energy system (such as power supply failure rate); service quality of the integrated energy system; technological innovation of the integrated energy system (the technological innovation index score of the integrated energy system can be evaluated by obtaining the technological innovation score input by users);

[0103] (3) Risk (primary indicator), and its corresponding secondary indicators: market fluctuations (market fluctuation risk index scores can be evaluated by quantifying market fluctuations or obtaining market fluctuation scores input by users); natural disasters (natural disaster risk index scores can be evaluated by quantifying different natural disasters and disaster severity).

[0104] 3. Environmental indicators (target layer indicators):

[0105] (1) Carbon emissions (primary indicator), and its corresponding secondary indicators: industrial carbon emissions of the integrated energy system; transportation carbon emissions of the integrated energy system; and building carbon emissions of the integrated energy system.

[0106] (2) Energy structure (primary indicator), and its corresponding secondary indicator: renewable energy utilization rate of the integrated energy system Fossil fuel utilization in integrated energy systems Comprehensive energy utilization rate of integrated energy system

[0107] Among them, E renew Represents renewable energy consumption, E fossil represents the fossil fuel consumption of the integrated energy system, E energy Indicates the electric energy consumption of the integrated energy system, E total Represents the total energy input of the integrated energy system.

[0108] (3) Pollutant emissions (primary indicator), and their corresponding secondary indicators: sulfur dioxide emissions from the integrated energy system; nitrogen oxide emissions from the integrated energy system; particulate matter emissions from the integrated energy system; and chemical wastewater emissions from the integrated energy system.

[0109] (4) Waste treatment (primary indicator), and its corresponding secondary indicators: industrial solid waste treatment rate of the integrated energy system; hazardous waste treatment methods of the integrated energy system (the indicator score of the hazardous waste treatment method can be evaluated by quantifying different hazardous waste treatment methods or obtaining the scores of different hazardous waste treatment methods input by the user);

[0110] Domestic waste recovery rate for integrated energy systems.

[0111] (5) Ecological impact (first-level indicator), and its corresponding second-level indicators: land utilization rate of the integrated energy system; green coverage rate of the integrated energy system.

[0112] 4. Energy product quality indicators (target level indicators):

[0113] (1) Energy consumption efficiency (primary indicator), and its corresponding secondary indicator: unit energy consumption output of the integrated energy system Energy conversion efficiency of the integrated energy system; Equipment consumption rate of the integrated energy system;

[0114] Among them, Q represents the total energy output value of the integrated energy system, and E represents the total energy consumption of the integrated energy system.

[0115] (2) Reliability (first-level indicator), and its corresponding second-level indicator: failure rate Equipment life; system recovery speed;

[0116] Among them, N fail represents the number of power equipment failures in the integrated energy system, T op Indicates the total operating time of the integrated energy system.

[0117] (3) Customer satisfaction (first-level indicator), and its corresponding second-level indicators: customer satisfaction; customer complaint rate.

[0118] (4) Flexibility and adaptability (first-level indicator), and its corresponding second-level indicators: energy supply diversity; energy load responsiveness; energy reserve level.

[0119] It should be noted that the index scores of energy supply diversity / energy load responsiveness / energy reserve level can be evaluated by quantifying energy supply diversity / energy load responsiveness / energy reserve level, or by obtaining the energy supply diversity / energy load responsiveness / energy reserve level scores input by the user.

[0120] (5) After-sales service (first-level indicator), and its corresponding second-level indicators: maintenance response time; service score.

[0121] It should be noted that for the secondary indicators that cannot be quantitatively calculated above (such as energy supply diversity, energy load responsiveness, energy reserve level, hazardous waste treatment methods, service quality, etc.), the scores of the corresponding indicators can be quantified according to the pre-set qualitative analysis or scoring methods of various parks.

[0122] The first weight of the first-level indicator is determined based on the specific attributes and importance of different parks. The value range of the first-level indicator's first weight is [0, 1), and the sum is 1. The second weight of the second-level indicators under the first-level indicator is [0, 1), and the sum is 1.

[0123] Taking the first-level indicators under the target layer as an example, we first use the Saaty 1-9 scale to perform pairwise comparisons and construct a judgment matrix A to represent the importance of each first-level indicator relative to the corresponding target layer:

[0124]

[0125] Among them, element a ij 、a ji ∈A, respectively represents the importance of the i-th first-level indicator under the target layer relative to the j-th first-level indicator, and the importance of the j-th first-level indicator relative to the i-th first-level indicator, where i, j = 1, 2, 3..., n, n represents the number of first-level indicators under the target layer,

[0126] On this basis, the geometric mean method is used to calculate the normalized eigenvalue corresponding to each element in the judgment matrix A, that is, the average value of each column of the matrix A is calculated, and each element in the matrix A is divided by the average value of the corresponding column to obtain the normalized eigenvalue of the corresponding element, forming a normalized judgment matrix A', ensuring that the weight is within the range of [0, 1). Then calculate the sum of each row of the matrix A' W i , and based on the sum of each row of matrix A′W i , and calculate the weight of each first-level indicator in sequence

[0127] It should be noted that the first weight calculation of the first-level indicators of other target layers and the second weight calculation of the second-level indicators under each first-level indicator are similar and will not be repeated here.

[0128] The calculation method for the indicator scores at each level is defined as follows:

[0129] First-level indicator score = ∑(second-level indicator score × second-level indicator's second weight);

[0130] Target layer score = ∑ (first-level indicator score × first-level indicator weight).

[0131] The embodiment of the present invention constructs a multi-park energy consumption index evaluation system through the above-mentioned fuzzy hierarchical analysis method, which can comprehensively reflect the energy consumption of the park in multiple dimensions, flexibly allocate weights according to different park attributes, and ensure that the index evaluation results are more comprehensive and true.

[0132] For a specific industrial park, we perform data preprocessing on the diverse and heterogeneous data of its integrated energy system, including but not limited to data cleaning, data conversion, and data integration, to generate structured data. This structured data is then used to extract features using a pre-trained extended long short-term memory network model to obtain the multidimensional energy usage characteristics of the industrial park's integrated energy system.

[0133] Among them, the multi-dimensional energy consumption characteristics include: electricity consumption characteristic data under different operating modes and different seasons, production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data.

[0134] Then, based on the multi-dimensional energy consumption characteristics of multiple different types of parks and combined with the energy consumption index evaluation system, the scores of indicators at various levels of different parks are calculated, and a multi-element energy efficiency evaluation index model for multiple different types of parks is constructed; the operation process of the integrated energy system of different types of parks is correlated and evaluated, and an energy complementary relationship matrix of the parks is constructed; based on the multi-element energy efficiency evaluation index model and the energy complementary relationship matrix of the parks, a multi-objective optimization model is constructed; the optimal solution of the multi-objective optimization model under the preset operation constraints is solved by a multi-objective particle swarm optimization algorithm, and the energy operation scheduling strategy of the integrated energy system of multiple different types of parks is obtained; the embodiment of the present invention can realize the multi-objective optimization of the integrated energy of multiple parks across time and space, meet the energy optimization scheduling needs of parks with diversified and complex electricity demands, reduce energy waste of the integrated energy system of the parks, optimize the energy distribution across time and space of the park group, and improve the comprehensive energy efficiency of the parks.

[0135] Specifically, the pre-trained extended long short-term memory network model is used to extract features from multivariate heterogeneous data during the operation of the integrated energy systems of multiple different types of parks, thereby obtaining multidimensional energy consumption features of the integrated energy systems of each park, including:

[0136] Time series data is extracted from the multivariate heterogeneous data during the operation of the integrated energy system of multiple different types of parks to obtain multidimensional time series data;

[0137] Inputting the multidimensional time series data into the extended long short-term memory network model; the extended long short-term memory network model comprises an input layer, an xLSTM layer, and an output layer;

[0138] Extracting local features of the multidimensional time series data through the extended long short-term memory network in the extended long short-term memory network model;

[0139] Extracting global features of the multidimensional time series data through a multi-head attention mechanism in the extended long short-term memory network model;

[0140] The local features and the global features are fused through the fusion layer in the extended long short-term memory network model to obtain the multi-dimensional energy consumption characteristics of the integrated energy system of each park.

[0141] For example, by removing missing values, unreasonable outliers, time alignment, fusion, and classification of electricity consumption data, we can obtain several multidimensional time series data. The feature extraction principle of the extended long short-term memory network model is as follows:

[0142] Input layer: Input multidimensional time series data, defined as x t ∈R d ;

[0143] Where T is the length of the time series, d is the characteristic dimension of each time step (for example, the power generation data arranged by sampling time in wind power data, the power generation data arranged by sampling time in photovoltaic power data, the real-time load and power consumption of the regional power grid, etc.), d represents the characteristic dimension of each time step, x t represents the t-th input feature in the multidimensional time series data.

[0144] xLSTM layer (local feature extraction): The local dynamic characteristics of multidimensional time series data are captured through XLSTM;

[0145] Input feature x t After nonlinear transformation, we get:

[0146] Among them, g represents a nonlinear activation function (such as ReLU), W x represents the weight matrix of the xLSTM layer, b x Represents the xLSTM layer bias term;

[0147] Update hidden state and cell state:

[0148]

[0149] h t =O t ⊙tanh(C t );

[0150] Among them, σ represents the Sigmoid function, W f , W i , W c represents the weight matrix of the hidden layer, b f , b i , b c represents the bias term of the hidden layer, h t represents the hidden state, C t Represents the state of the memory unit, ⊙ represents element-by-element multiplication, ot Represents the output gate result.

[0151] Output hidden state sequence:

[0152] The extended long short-term memory (xLSTM) model, combined with the Transformer multi-head attention mechanism, is used to extract and analyze the multidimensional energy usage characteristics of the multi-dimensional heterogeneous data of the park's integrated energy system. This embodiment of the present invention leverages the xLSTM model's ability to capture long-term dependencies and the Transformer's global attention capability. The xLSTM excels at capturing long-term and short-term dependencies in sequences, while the Transformer multi-head attention mechanism introduces a multi-head self-attention layer (Multi-HeadAttention) to extract global correlations in sequences. This enriches the feature representation of the park's multi-dimensional energy usage characteristics and allows for processing complex time series and multidimensional data.

[0153] Multi-head attention mechanism (global feature extraction): The output of the xLSTM layer is used as the input of the multi-head attention layer to perform self-attention calculation: Input H xLSTM Generate query vector, key vector, and value vector through linear transformation:

[0154] Q=H xLSTM W Q ,K=H xLSTM W K ,V=H xLSTM W V Among them, W Q ,W K ,W V Represents the weight matrix of a trainable multi-head attention mechanism.

[0155] Calculate the attention score: Among them, d k is the dimension of the key vector, and the superscript T indicates the vector transpose.

[0156] Multi-head attention fusion: extracting global correlation by computing multiple groups of attention mechanisms in parallel: H attn =MultiHead(Q,K,V)=concat(head1,...,head n )Wo; where multiple attention groups are obtained from the above attention scores, W O Indicates the transformation matrix of the final linear transformation, head n The above attention score is calculated, where h represents the number of heads in the multi-head attention mechanism. The output is the global feature H extracted by the multi-head attention layer. attn .

[0157] Fusion layer (fusion of local and global features): by fusing local features (xLSTM output H xLSTM ) and global features (multi-head attention output H attn ), and obtain the final feature representation: Specifically, the fully connected layer is used to further compress the fused features to obtain the multi-dimensional energy usage features: H final =ReLU(W f ·[H xLSTM ;H attn ]+b f ).

[0158] For example, electricity consumption characteristic data, production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data can be extracted through features.

[0159] Among them, electricity consumption characteristic data include the following six categories: the first electricity consumption characteristic classified as high-load industrial type: continuous high electricity load data (for example, the electricity load continuously exceeds the first set threshold), especially during peak production periods, the cooling demand increases in summer, and the load may increase in winter due to the use of heating equipment; the second electricity consumption characteristic classified as low-load service type: electricity load data with low electricity load (for example, the electricity load is lower than the second set threshold), mainly concentrated in office, lighting, air conditioning and other equipment, and the use of air conditioning will fluctuate significantly in summer and winter; the third electricity consumption characteristic classified as the type with obvious peak-to-valley difference: electricity load data with significant fluctuations in electricity load within a day (for example, fluctuations exceeding the third set threshold), with obvious peak and trough periods of electricity consumption, and the peak-to-valley difference between summer and winter is even greater, which is affected by equipment such as air conditioners; the fourth electricity consumption characteristic classified as the type of high-load seasonal fluctuation: extremely high electricity load in specific seasons (for example, the total electricity load in a quarter exceeds the fourth set threshold), electricity consumption increases significantly in summer or winter, and decreases significantly in spring and autumn. The fifth electricity consumption characteristic of enterprises classified as intermittent operation: the electricity load is intermittent (for example, the electricity load in a quarter shows cyclical changes), and there may be cyclical fluctuations in winter and summer, but not as significant as those in the continuous operation category; the sixth electricity consumption characteristic of enterprises classified as energy-saving operation: the use of energy-saving technologies and strategies to maintain a low electricity load throughout the year, with small fluctuations and relatively stable load, and may rely on energy-saving equipment or renewable energy (for example, the continuous electricity load within a year is lower than the second threshold, and the daily fluctuation is less than the third threshold, etc.).

[0160] Taking the industrial park as an example, the production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data are as follows:

[0161] Capacity characteristic data: power generation efficiency, equipment start-up and shutdown frequency, and seasonal variations of gas-fired power generation equipment; thermal efficiency, fuel consumption, start-up and shutdown frequency, and heating time period of boilers; gas output, consumption efficiency, and equipment operating hours of gas-fired equipment; power generation, seasonal wind speed variations, and wind power efficiency of wind turbines; power generation efficiency, sunshine hours, and seasonal fluctuations of photovoltaic panels.

[0162] Energy transmission characteristic data: grid transmission loss, transmission stability, peak-valley current changes; thermal energy transmission loss, pipeline insulation effect, transmission pressure fluctuations; gas transmission loss, gas pipeline pressure, flow stability; wind power transmission loss, transmission line stability, load fluctuations; photovoltaic power transmission loss, transmission stability, photovoltaic system output fluctuations.

[0163] Energy storage characteristic data: charging and discharging frequency of energy storage equipment, battery storage capacity, and response speed during periods of electricity price fluctuations; heat storage capacity, heat dissipation rate, and thermal energy regulation frequency of thermal storage equipment; gas reserve capacity, gas leakage rate, and safety of gas storage equipment; capacity, response speed, and wind power storage utilization rate of energy storage devices; capacity, discharge efficiency, and photovoltaic energy storage response time of energy storage equipment.

[0164] Energy consumption characteristic data: fluctuations in electricity demand for production equipment and R&D equipment, load rates in different time periods, and electricity consumption of office equipment; demand for thermal energy in the production process, seasonal heating demand, and heat distribution in each production unit; gas demand, seasonal gas usage, and operating frequency of each production process; auxiliary power supply demand and backup power supply utilization frequency; electricity demand for low-power equipment and lighting, and the proportion of photovoltaic power usage.

[0165] Taking a business park as an example, the production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data are as follows:

[0166] Capacity characteristic data: power generation of gas generators, operating frequency of power generation equipment, and seasonal demand changes; heat production efficiency of the heating system, fuel consumption of the heating boiler, and heating time; output efficiency of gas equipment, gas supply, and equipment start and stop frequency; power generation efficiency of wind turbines, wind speed fluctuations, and seasonal changes in wind power systems; daily power generation of photovoltaic systems, seasonal sunshine changes, and photovoltaic panel efficiency.

[0167] Energy transmission characteristic data: grid transmission stability, peak-valley current changes, and transmission efficiency; heating pipeline transmission loss, insulation effect, and pipeline pressure fluctuations; gas pipeline pressure, transmission loss, and flow fluctuations; wind energy transmission loss, load fluctuations, and transmission stability; photovoltaic transmission loss, transmission stability, and output fluctuations.

[0168] Energy storage characteristic data: charging and discharging capacity of energy storage equipment, battery response speed, and adjustment during periods of electricity price fluctuations; thermal energy storage capacity, heat dissipation loss, and heat storage adjustment frequency; gas storage capacity, gas leakage rate, and reserve response time; energy storage device capacity, response time, and wind energy adjustment frequency; energy storage device charging capacity, response speed, and nighttime utilization rate.

[0169] Energy consumption characteristic data: load demand fluctuations of shopping malls, office buildings, hotels, etc., peak electricity consumption periods, and basic loads during non-working hours; heating demand of hotels and office buildings in winter, and temperature control needs of various commercial areas; gas demand for commercial kitchen equipment, hotel hot water supply demand, and gas consumption fluctuation rate; backup power demand, and the proportion of wind power in total power supply; lighting demand, and the proportion of low-power equipment used during non-peak hours.

[0170] Taking a residential park as an example, the production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data are as follows:

[0171] Capacity characteristic data: no internal power generation equipment, power purchase amount from the power grid, peak and valley electricity price distribution; central heating boiler heat production efficiency, winter heating time; natural gas supply, gas equipment operating efficiency, seasonal demand; small-scale photovoltaic power generation, sunshine hours, photovoltaic system efficiency.

[0172] Energy transmission characteristic data: grid transmission loss, community power distribution efficiency, load balance; thermal energy transmission loss, heating pipeline insulation performance, pressure fluctuation; gas pipeline pressure, loss, flow control; photovoltaic power transmission efficiency and output stability.

[0173] Energy storage characteristic data: energy storage equipment capacity, electricity price adjustment response, charging and discharging frequency; thermal storage equipment capacity, thermal storage efficiency, winter adjustment frequency; community gas reserve capacity, gas leakage rate, response speed; community energy storage equipment capacity, energy storage equipment response.

[0174] Energy consumption characteristic data: fluctuations in residents' demand for household appliances, seasonal changes in electricity consumption, and nighttime basic load; winter heating demand and heating load distribution in residential buildings; residents' daily cooking needs and seasonal gas demand; lighting in public areas and power supply demand for low-load facilities.

[0175] Specifically, according to the multi-dimensional energy consumption characteristics of multiple different types of parks, a multivariate energy efficiency evaluation index model for different types of parks is constructed based on the entropy weight method, the hierarchical analysis method, and the design structure matrix, including:

[0176] Obtain energy impact information of integrated energy systems of multiple different types of parks, and perform feature extraction on the energy impact information to obtain multi-dimensional environmental characteristics, multi-dimensional energy product quality characteristics, and multi-dimensional economic characteristics of the integrated energy systems of multiple different types of parks;

[0177] The multidimensional energy consumption characteristics, the multidimensional environmental characteristics, the multidimensional energy product quality characteristics, and the multidimensional economic characteristics of each different type of park are evaluated based on a preset energy consumption index evaluation system, and the entropy weight method, the hierarchical analysis method, and the design structure matrix are used to calculate the index scores to construct a multivariate energy efficiency evaluation index model for each different type of park;

[0178] The scores of the secondary indicators in the multivariate energy efficiency evaluation index model are obtained by quantifying the characteristic quantities of the multidimensional energy consumption characteristics, the multidimensional environmental characteristics, the multidimensional energy product quality characteristics, and the multidimensional economic characteristics of different seasons and different electricity users;

[0179] The score of the first-level indicator in the multivariate energy efficiency evaluation index model is calculated based on the scores of the second-level indicators attached thereto.

[0180] It should be noted that the feature extraction of multi-dimensional environmental features, multi-dimensional energy product quality features, and multi-dimensional economic features is also implemented by sampling and combining the xLSTM model with the Transformer multi-head attention mechanism. The principle of feature extraction can refer to the principle of extracting multi-dimensional energy consumption features, and is not specifically limited in the embodiments of the present invention. In the process of constructing the multi-dimensional energy efficiency evaluation index model, the embodiment of the present invention introduces the entropy weight method to make full use of the information provided by objective data to determine the objective weight and eliminate subjective influences. Based on the multi-dimensional environmental features, multi-dimensional energy product quality features, and multi-dimensional economic features obtained after feature extraction, the scores of indicators at different levels of each park are calculated according to the energy consumption index evaluation system constructed above. It should be noted that if the indicator cannot be calculated quantitatively, the score of the indicator can be determined according to the preset percentage scoring rule.

[0181] After completing the calculation of the scores of the secondary indicators, the park is divided into different groups according to the season labels, and the weights of the secondary indicators in different seasons are calculated. First, the characteristic proportions of different areas in the park under each secondary indicator are calculated.

[0182] in, It represents the score of the i-th unit in the park for the j-th secondary indicator in season k (for example, the energy consumption score per unit area of ​​the office building unit in the industrial park in spring), and n represents the total number of units.

[0183] Based on this, the entropy values ​​of each secondary index in this season are calculated And calculate the coefficient of variation The weight of each secondary indicator Calculate the total score of the weighted secondary indicator layer in different seasons

[0184] in, This represents the average score of indicator j for each unit in the park in season k (e.g., the average energy consumption score per unit area for each unit in the target industrial park in spring). Similarly, the total score for the first-level indicator layer a can be calculated (e.g., the energy consumption score for the industrial park).

[0185] On this basis, the weight of the first-level indicator can be calculated by the same logic, and the scores of the target park's energy consumption indicator target layer, economic indicator target layer, environmental indicator target layer, and energy product quality indicator target layer can be calculated based on this.

[0186] The park's comprehensive score can be calculated based on the score matrix of each target layer. Similarly, a design structure matrix R can be designed to define the importance of each target layer relative to the corresponding park's comprehensive energy system:

[0187]

[0188] Among them, the subscripts 1, 2, 3, and 4 of the elements in the matrix represent the energy consumption index target layer, economic index target layer, environmental index target layer, and energy product quality index target layer respectively. The order of the index codes in the subscripts represents the impact of the previous index on the next index. Each element is the level of mutual influence between two indicators: the possible values ​​of the elements in the matrix are 0, (±)1, (±)2, and (±)3; among them, 0 means no influence, 1 means weak influence, 2 means medium influence, and 3 means strong influence. (+) is a positive influence and (-) is a negative influence. For example, a higher energy consumption index may increase costs and reduce economic indicators (negative influence), but it is friendly to environmental indicators (positive influence). Different parks have different evaluation results. The park's comprehensive score Score=||C s R||1. Among them, C s =[Y1, Y2, Y3, Y4], where Y1, Y2, Y3, and Y4 represent the index scores of each target layer of the park respectively.

[0189] Specifically, the correlation evaluation of the operation process of the integrated energy systems of different types of parks and the construction of the park energy complementary relationship matrix include:

[0190] Spearman correlation analysis was performed on the time series characteristics of production capacity, transmission, storage, and consumption of different types of parks, and a correlation coefficient matrix was constructed between the production capacity, transmission, storage, and consumption processes of different types of parks.

[0191] The maximum information coefficient analysis was conducted on the time series characteristics of production capacity, transmission, storage, and consumption of different types of parks, and a symmetric correlation matrix was constructed between the production capacity process, transmission process, storage process, and consumption process of different types of parks.

[0192] A park energy complementary relationship matrix is ​​constructed based on the correlation coefficient matrix and the symmetric association matrix.

[0193] In an embodiment of the present invention, after extracting the multi-dimensional energy usage characteristics, the correlation between the production capacity, transmission, storage, and consumption processes of different parks is further analyzed and evaluated. The complementary characteristics of the production capacity, transmission, storage, and consumption processes in the time domain during the operation of the integrated energy systems of multiple different types of parks are analyzed. By finding multiple groups of process combinations with the largest negative correlation coefficients, a park energy complementary relationship matrix for the energy production, transmission, storage, and consumption processes of multiple parks is constructed. The specific process is as follows:

[0194] For the production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data of a certain park, the corresponding one-dimensional first time series features are extracted, and each first time series feature is normalized. Then, the Spearman correlation coefficient analysis method is used to combine the normalized time series features of all parks in pairs, and the Spearman correlation coefficient of the two time series feature combinations is calculated. The closer the absolute value of the Spearman correlation coefficient is to 1, the stronger the correlation between the two time series features; the closer it is to 0, the weaker the correlation and the stronger the independence. For these two time series feature combinations, the feature pairs with negative Spearman correlation coefficients are extracted, and then sorted by absolute value. The feature pairs with the largest absolute value of the Spearman correlation coefficient are selected first and stored in the correlation coefficient matrix R. S In the same way, the other two time series feature combinations are constructed to construct the correlation coefficient matrix R S .

[0195] Similarly, for the time series characteristics obtained after normalization of the production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data of a certain park, the maximum information coefficient (MIC) analysis is used to calculate the MIC value between the two time series characteristics, which can measure the correlation between the two time series characteristics and is used to represent the correlation between the processes corresponding to the two time series characteristics (such as energy production, transmission, storage, and use processes), so that a symmetric correlation matrix between the production capacity process, energy transmission process, energy storage process, and energy consumption process of different types of parks can be constructed; it should be noted that the use of maximum information coefficient (MIC) analysis to calculate the MIC value between two variables belongs to the existing technology in this field and will not be explained in detail here.

[0196] For the production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data of the park, the time series characteristics corresponding to all data are traversed and combined in pairs, the MIC value of the two time series feature combinations is calculated, and the symmetric correlation matrix M∈R is constructed. m×m .

[0197] in, X I ,X J They represent the I-th and J-th time series features respectively, and m represents the total number of time series features.

[0198] Based on the interpretability criterion of information theory, the feature complementarity criterion is defined as:

[0199] Among them, θ represents the preset threshold, taking θ∈[0.1,0.3], and screening The time series feature pairs are taken as candidate complementary feature sets. The MIC value lower than θ indicates that there is a weak correlation but complementary information between the two time series features, and their MIC values ​​are stored in the MIC matrix R M middle.

[0200] The correlation coefficient method is used to analyze the complementary characteristics of different types of parks in the time domain, and the complementary relationship matrix C is constructed. The elements C in the matrix C are IJ It represents the complementarity strength between a process I in a certain park and a process J in another park. It can be used as the basis for multi-park collaborative optimization and as the basis for establishing a comprehensive regulation mechanism model for the operation process of multi-park energy systems.

[0201] Definition C IJ =w1·(1-|ρ IJ |)+w2·(1-MIC IJ ), w1, w2 represent the preset weight parameters, C IJ represents the MIC value between process I and process J, that is, the MIC value between the time series characteristics corresponding to any two processes in the process of energy production, transmission, storage, and use; ρ IJ It represents the Spearman correlation coefficient between process I and process J, that is, the Spearman correlation coefficient between the time series characteristics corresponding to any two processes in the energy production, transmission, storage, and use process.

[0202] Sort the off-diagonal elements in matrix C and select the combination with the largest value (i.e. the process pair with the strongest complementarity), that is, select the element with the largest value, and construct the park energy complementary relationship matrix. If processes I and J have a high negative correlation, then C IJ The larger the value, the stronger the complementarity; if there is no significant correlation or it is positively correlated, then C IJ The combinations with larger negative correlation coefficients are filled into the complementary relationship matrix, so that in the subsequent optimization model, these processes with complementary characteristics can be used to collaboratively optimize the system.

[0203] For example, by analyzing three parks and four processes (capacity, transmission, storage, and use), the complementary relationship matrix C can be expressed as:

[0204] The subscripts A, B, and C of the elements in the matrix C represent three different types of parks respectively. AB,prod Element C AB,trans Element C AB,store and element C AB,use These represent the complementary strengths between the production capacity, transmission, storage, and consumption processes of Park A and Park B, respectively. This is analogous and will not be repeated here. It should be noted that the complementary strength between a process in one park and the four processes in another park can be obtained by summing the MIC values ​​between a process in one park and the four processes in another park.

[0205] Specifically, the multi-objective optimization model is constructed based on the multivariate energy efficiency evaluation index model and the park energy complementary relationship matrix, including:

[0206] Furthermore, the method further comprises:

[0207] According to the energy consumption characteristics of different types of parks, based on the park energy complementary relationship matrix and the selected multiple control mechanisms, a comprehensive regulation mechanism model for the operation process of the multi-park energy system is established;

[0208] The various control mechanisms include: a combined heat and power unit capable of providing both heat and electricity; a gas boiler unit using liquefied gas, natural gas or other gas energy as fuel, heating water into steam through a heating furnace to provide the required heat energy to users;

[0209] The comprehensive control mechanism model includes: a dynamic model of wind power generation and photovoltaic power generation, a dynamic model of refrigerators and heat exchangers, and a dynamic model of electrical equipment and energy storage equipment.

[0210] The energy complementary relationship matrix of the parks is used as the comprehensive control mechanism of the operation process of the multi-park integrated energy system to construct the energy system topology between different types of parks, such as Figure 3 As shown, each park is directly connected to the regional power grid and natural gas grid, and purchases energy from the upper-level energy network; power lines and heat pipelines are laid between parks to achieve end-to-end energy transmission between parks; each park will also use energy equipment to meet its own electricity, heat, cooling and gas needs, thereby achieving an overall energy supply and demand balance in the integrated energy system.

[0211] Among them, the dynamic models of the main equipment of the park's integrated energy system are as follows:

[0212] Combined heat and power (CHP) units can provide both heat and electricity. The corresponding dynamic model is constructed based on the relationship between the electrical power and thermal power supply:

[0213]

[0214] in, The electric power output of the CHP unit; is the output thermal power of CHP; is the natural gas power input to CHP; G CHP is the gas consumption of CHP; H gas It is the low calorific value of natural gas; and Represent the unit's electrical power and thermal power efficiency respectively.

[0215] Gas boilers use liquefied gas, natural gas and other gas energy as fuel, and heat water into steam through a heating furnace to provide users with the required thermal energy. The corresponding dynamic model based on its output thermal power is as follows:

[0216]

[0217] Among them, Q GB (t) is the output thermal power of GB unit; H gas is the lower calorific value of natural gas; η GB is the heating efficiency of the boiler; G GB (t) is the natural gas consumption of the boiler; Q GB,max Indicates the upper limit of heating power.

[0218] The output power characteristics of wind power generation are determined by the cut-in wind speed, cut-out wind speed, rated power, and rated wind speed. The corresponding dynamic model is constructed as follows:

[0219]

[0220] Among them, P WT and are the actual output power and rated output power of the wind turbine respectively; v and v N are the actual wind speed and the rated wind speed respectively; v cut-in and v cut-out are the cut-in wind speed and cut-out wind speed of the fan respectively.

[0221] The amount of photovoltaic power generation is related to temperature, light intensity, and the performance parameters of the photovoltaic system itself. The corresponding dynamic model is constructed as follows:

[0222]

[0223] Among them, P PV and are the actual and rated output power of photovoltaic power generation respectively; f PV is the photovoltaic derating factor; S is the light intensity of the photovoltaic in the actual environment. Under standard working conditions, the light intensity S STC Defined as 1000w / m2; k is the photovoltaic power temperature coefficient; T C is the temperature of the photovoltaic unit in the actual environment; T a is the actual ambient temperature; v is the wind speed at the photovoltaic panel installation location; T STC This is the standard test working environment temperature.

[0224] Refrigerator is divided into electric refrigerator and absorption refrigerator, and the corresponding dynamic model is constructed as follows:

[0225]

[0226] Among them, C EC and P EC are the cooling power and electrical power consumed by the electric refrigerator, respectively, η EC is the cooling coefficient of the electric refrigerator; C AC and H AC are the cooling power and heat power consumed by the absorption refrigerator, η AC is the cooling coefficient of the absorption chiller.

[0227] Based on the process of heat energy exchange between the heat exchanger and the external heat distribution network, the corresponding dynamic model is constructed as follows:

[0228] H he =η he H b ;

[0229] Where H he is the output thermal power of the heat exchanger; η he Indicates the conversion efficiency of the heat exchanger; H b The heat power purchased by the park from the external heat distribution network.

[0230] Air conditioning is the main load equipment in commercial parks and residential parks. The corresponding dynamic model is constructed as follows:

[0231]

[0232] Among them, P AC is the air conditioner operating power η AC is the coefficient of performance of the air conditioning equipment, P cool is the power of the cooling coil, C o represents the specific heat capacity of air, T in Indicates the initial temperature of the cooling process, T out is the final temperature after cooling; P varableis the power of the variable frequency drive equipment, S Ac is the cooling air supply flow rate, T cool is the temperature maintained by the cooling coil, μ is the energy conversion coefficient of the cooling coil, COP is the performance coefficient of the cooling coil, and k1, k2, and k3 are the relevant coefficients for power calculation of the variable frequency drive equipment.

[0233] Based on the charging and discharging process of electric vehicles, the corresponding dynamic model is constructed as follows:

[0234] State of charge of electric vehicles when charging:

[0235] The state of charge of an electric vehicle during discharge:

[0236] in, are the rated charging power and actual charging efficiency of electric vehicles respectively; Q EV is the total battery capacity of the electric vehicle. are the rated discharge power and actual discharge efficiency of the electric vehicle respectively; Δt represents the unit time length.

[0237] Energy storage equipment includes: power storage equipment (batteries, capacitors), heat storage equipment, etc. The corresponding dynamic model is constructed as follows:

[0238] Energy storage status of the power storage device:

[0239] Where x∈{es,hs,cs,gs} represents different types of energy storage, es, hs, cs, gs represent electricity storage, heat storage, cold storage and gas storage respectively; δ x represents the self-consumption rate; η x,c and η x,d The charging and discharging efficiency of different types of energy storage; and are the charging and discharging powers of the energy storage during period t, respectively.

[0240] Construct dynamic models and operational constraints for production equipment, transmission equipment, storage equipment, and energy-consuming equipment in integrated energy systems of different types of parks. The multi-park integrated energy system must meet the following multi-dimensional collaborative operational constraints during operation:

[0241] (1) Equipment operating limit constraints of the integrated energy system within the park, including:

[0242] Equipment installation constraints:

[0243] in, Indicates the maximum and minimum configuration numbers of different devices k.

[0244] Equipment operation constraints:

[0245]

[0246] Among them, γ k ∈(0,1) is a state variable describing whether device k is running: γ k =0 means the device is not running, γ k =1 means the equipment is operating normally; min It is the minimum value of the output power coefficient of the equipment. t k is the operating power of device k at time t.

[0247] Electric / thermal energy storage operation constraints:

[0248]

[0249] in, Indicates the maximum and minimum values ​​of the remaining capacity; are the charging and discharging / heating power of electrical energy storage and thermal energy storage, ω char_k 、ω dis_k Represent the charge and discharge / heat efficiency respectively, and the range is [0,1].

[0250] (2) Maximum transmission limit constraints for purchased electricity and gas power;

[0251]

[0252] in, In order to set the power limit for purchasing electricity, natural gas and thermal energy from the energy grid, the natural gas purchased is considered to be consumed mainly by CHP units and gas boilers.

[0253] Inter-park electrical and thermal energy interconnection lines must be within the maximum transmission power constraints:

[0254]

[0255] in is the power of electric energy and heat energy transmission between park i and park j, are the upper limits of the interaction of electric energy and thermal energy between park i and park j, respectively; are the lower limits of the interaction of electric energy and thermal energy between park i and park j.

[0256] (3) Power balance constraint: This includes the supply and demand balance of various types of energy, including electricity, heat, gas, and cooling. For a multi-park integrated energy system, the energy transmission balance between parks can be described using the Energy Hub (EH) model:

[0257]

[0258] Where L is the output matrix of the energy hub, L e ,L g ,L h ,L c is the electricity load, gas load, heating load and cooling load on the user side, η t ,η he are transformer efficiency and CHP unit power generation efficiency, P b , G b 、H b They are the electricity purchase amount, gas consumption of CHP unit and gas boiler, P PV 、P WT are photovoltaic and wind power generation power, are the conversion efficiencies of natural gas to electricity and heat, respectively. ng is the calorific value of natural gas, η ec ,η ac are the conversion efficiencies of electric refrigeration and absorption refrigeration respectively; G mt is the natural gas input, P ec is the electrical energy input power, H ac is the refrigerator input power, They are the charging and discharging power of electric energy, thermal energy, cold energy and gas respectively.

[0259] (4) Inter-park energy interaction constraints: The interaction power between different parks should satisfy the global balance, that is, the algebraic sum of the exchange power is 0, which can be expressed as follows:

[0260]

[0261] in They are the overall interactive electrical and thermal power between different parks respectively.

[0262] Based on the energy system topology and the dynamic model, multiple objective functions are constructed and the multi-objective optimization problem is defined as follows:

[0263] (1) Energy consumption: The first objective function f1 that minimizes the total energy consumption, including the steady-state energy consumption E steady And the non-steady-state energy consumption E caused by uncertain factors such as wind and solar uncertain .

[0264]

[0265] Among them S i To assign weights based on historical data or forecast models to ensure that the results are both economical and robust.

[0266] (2) Economic efficiency: The second objective function f2 that minimizes energy costs, including the investment cost C of the planned equipment inv, interaction cost between the park and the power grid C coe , natural gas purchase cost Environmental cost C en and equipment operation and maintenance costs C om , ε i The percentage of different typical days in a year.

[0267]

[0268] (3) Environmental impact: the third objective function f3 that minimizes carbon emissions;

[0269]

[0270] in, is the actual carbon emissions of the park area, are the purchased electric power, CHP output power and heat generation power at time t, ε e , ε h is the emission coefficient per unit of electricity and heat production.

[0271] (4) System quality: Ensure the operational stability and reliability of the system, measured by the energy supply reliability cost, and construct the fourth objective function f4:

[0272]

[0273] Among them, C P1 represents the penalty price per unit of wind and solar power curtailment; L des Indicates the amount of wind and solar power curtailment; C P2 represents the unit penalty price caused by interrupted load; L loss Indicates the amount of loss caused by system load interruption.

[0274] Then the multi-objective optimization model can be expressed as:

[0275] Multi-objective optimization function: f1, f2, f3, f4; sb.to g(x)≤0, h(x)=0.

[0276] Among them, g(x) and h(x) are the operational constraints mentioned above, including inequality constraints and equality constraints.

[0277] The embodiment of the present invention proposes a Transformer-based Multi-Objective Particle Swarm Optimization (TBMOPSO) optimization algorithm for the multi-objective planning model of multi-element energy efficiency optimization of different types of parks, and introduces the multi-head attention mechanism into the multi-objective particle swarm algorithm, so that the multi-objective optimization can more effectively explore the complex solution space. In addition, TBMOPSO will be combined with the interactive user satisfaction mechanism during optimization, so that the optimization process can better reflect user preferences and gradually improve the personalized effect. Through real-time feedback from users, the algorithm can dynamically adjust the search direction of particles, so as to find solutions that better meet user expectations in the Pareto optimal solution set, such as Figure 4 shown.

[0278] Specifically, the multi-objective particle swarm optimization method based on the interaction of the Transformer model and user satisfaction feedback is used to solve the multi-objective optimization model for the comprehensive regulation of multi-park energy systems under preset operation constraints, and the energy operation scheduling strategies of the comprehensive energy systems of multiple different types of parks are obtained, including:

[0279] (1) Initialization and user-guided particle swarm configuration: In the initial stage of the particle swarm optimization algorithm, the initialization of the three-dimensional core parameters is required: the particle space coordinates X it , motion rate vector V it and the attention weight parameter W it . This mechanism innovatively integrates the user preference guidance function and provides two interactive initialization strategies: a multi-objective priority evaluation system that allows users to clarify the relative importance of each objective function through weighted scoring, assign weights to each objective function, and subsequently use the weighted objective function for fitness calculation; a solution space anchor positioning technology that supports users to directly preset the desired solution area. By integrating these preference information, the algorithm can generate an initial solution set that meets the user's global optimization expectations during the initialization phase, including the individual optimal solution of the particle and the initial value of the global optimal solution, to achieve adaptive distribution of individual solutions in the solution space:

[0280]

[0281] Among them, it is the particle number; d is the dimension of the problem, Different features of particles represented as embedded user ratings.

[0282] (2) Fitness calculation in Pareto-based multi-objective optimization: To evaluate the quality of each particle, its fitness value F needs to be calculated. it , including the fitness value f of the particle on the kth objective function itk :

[0283] F it =[f it1 ,f it2 ,…,f itk ];

[0284] It should be noted that in this embodiment of the present invention, a weighted objective function is used as the target, and the fitness of the particles with respect to each weighted objective function is calculated using the Pareto algorithm to obtain a Pareto solution. The technology for calculating the fitness of particles using the Pareto algorithm to obtain a Pareto solution is prior art and will not be described in detail here.

[0285] The Pareto solution of each iteration is displayed to the user, who can score the solution, obtain a real-time satisfaction score for the corresponding solution, or select the solution that best suits their preference.

[0286] (3) Position encoding: Since the Transformer itself cannot perceive the order, it is necessary to introduce position encoding for each particle to embed the position information of the input sequence into the model:

[0287]

[0288] Among them, pos represents the position, ir represents the dimension index of it of the corresponding particle, d model Represents the dimension of the input feature matrix.

[0289] (4) Transformer attention mechanism calculation: The core architecture of the Transformer model consists of two modules: the encoder and the decoder. Each module contains several identical layers, and each stacked layer in the encoder contains two core sub-layers. The first is the multi-head self-attention mechanism (Multi-Head Self-Attention) layer that captures the long-distance dependencies within the series. The input vector is mapped to different subspaces through three matrices (Q, K, V), and then the matrices Q, K, and V are divided into multiple "heads", and each head calculates attention independently. Through parallel calculation, semantic information at different positions is paid attention to at the same time: the attention calculation process can be seen above. The specific formula is as follows:

[0290]

[0291] Among them, W it Q 、W it K 、W it VRepresents a learnable weight matrix. The query vector Q is responsible for dynamically calculating the weight distribution between positions in the input sequence, the key vector K is used to encode the contextual dependencies across positions, and the value vector V carries the semantic weight reconstruction of the corresponding features.

[0292] The Feed-Forward Network (FNN) structure layer implements a nonlinear projection transformation on the position feature representation through a learnable parameter space mapping matrix and bias vector. This is specifically manifested as a combination of nonlinear activation functions between two layers of affine transformation, achieving dimensional expansion and compressed reorganization of features:

[0293] FFN(x)=max(0,xR1+b1)R2+b2;

[0294] Among them, FFN(x) represents the mapping output of the feedforward neural network to the input feature quantity x, R1 is the weight matrix of the first layer linear transformation, b1 is the bias term of the first layer linear transformation, R2 and b2 are the weight matrix and bias of the second layer linear transformation respectively.

[0295] The decoder is also stacked with N identical layers, including three sub-layers: Masked Multi-Head Self-Attention, Encoder-Decoder Attention, and Feedforward Neural Network (FFN).

[0296] The output layer includes linear transformation + Softmax, which maps the final output of the decoder to a probability distribution through a linear layer.

[0297] (5) Evolutionary strategy optimization and attention weight adjustment: Calculating particle X it The distance D between the global optimal solution gBest obtained in the next step it (X it ,gBest), and continuously adjust its value through mutual iteration to flexibly optimize the attention weight W it :

[0298] W it =updateAttentionWeight(W it ,D it );

[0299] At the same time, the user feedback mechanism is integrated into it, and the attention weight is dynamically adjusted to adjust the priority of each objective function and guide the particles to gather in the direction of the solution that satisfies the user:

[0300]

[0301] Among them, γ is the user feedback sensitivity coefficient, U h is the historical preference feature of the current user (determined based on the user's historical satisfaction score), and MLP(.) is the nonlinear mapping implemented by the multi-layer perceptron. S u Rate user satisfaction in real time (0-10 scale).

[0302] (6) Update particle speed and position: The MOPSO framework includes a dynamic feedback correction mechanism based on user preferences. By constructing the coupling relationship between the user evaluation adjustment coefficient ω(t) and the cognitive-social learning factor (c1, c2), a speed v with feedback gain is established. it (t) Update equation:

[0303] v it (t+1)=ω(t)·v it (t)+c1·r1·(pbest it -x it (t))+c2·r2·(gbest-x it (t))+η·ΔF u ;

[0304] Among them, pbest it For particle X it The individual optimal solution, η represents the user feedback gain coefficient, ΔF u Represents the user's quality evaluation gradient of the candidate solution. r1 and r2 are the core random factors of the PSO algorithm, which dynamically balance the influence of individual and group experience. ΔF u Further expansion is as follows:

[0305]

[0306] Among them S u Score user satisfaction in real time (0-10 scale), is the 4-dimensional preference vector (weights of energy consumption / economy / environment / quality indicators) constructed by the hierarchical analysis method, α and β are the feedback coefficients adjusted by online learning, is the trend gradient based on historical feedback.

[0307] This mechanism achieves the synergy of two optimization effects. It enhances the exploration density of high-potential areas in the solution space through the gradient correction term, constructs the orthogonal projection operator in the user's preferred direction, and constrains the convergence path of particle swarm evolution.

[0308] The position update equation adopts a constrained convex combination form, and the adaptive adjustment of the exploration-exploitation trade-off is achieved by introducing a dynamic relaxation factor λ(t):

[0309]

[0310] Where ⊙ represents the feasible solution space under the constraints of the Voronoi diagram Projection operator, N ε (gbest) defines an expanded set of globally optimal neighborhoods. This architecture effectively reduces the probability of premature convergence, improving the preference fit of the solution set while ensuring Pareto front coverage.

[0311] (7) Dynamic management of position constraints: Based on the preset position update strategy, the particle position is checked to ensure that it is always within the feasible domain defined by the algorithm:

[0312]

[0313] Among them, l (j) ,u (j) is the original boundary of the j-th dimension decision variable, u (j) is the boundary relaxation amount, ∈ is the safety margin (take 1% of the rated capacity of the equipment) S u ∈[0,10] is the user's real-time satisfaction score. δ(S u ) represents a relaxation calculation function related to the user's real-time satisfaction score, which can be customized and is not limited in the embodiments of the present invention.

[0314] At the same time, by introducing the user interaction feedback mechanism, the operation constraints are dynamically adjusted, that is, Achieve a balance between the controllability of the optimization process and the interpretability of the results, thereby improving user acceptance of the solution set.

[0315] (8) Maintenance of non-dominated solution set and update of optimal solution: Update of individual optimal solution. For the current particle, its performance is evaluated by the fitness function. If it is better than the historical individual optimal solution, the update operation is performed:

[0316]

[0317] Among them, F i is the fitness value of all particles, the symbol > represents the Pareto dominance relationship, X (t) represents the current candidate solution at the tth iteration of the optimization algorithm.

[0318] (9) Global optimal solution set maintenance: Use the Pareto dominance relationship to screen non-dominated solutions (i.e., Pareto optimal solutions, global optimal solutions), add particle solutions that meet the non-dominated conditions to the elite solution set, and update the global optimal solution based on the elite retention strategy:

[0319]

[0320] Among them G best is the global optimal solution set, Xi / j is an individual solution in the candidate solution set, and F is the set of all candidate solutions.

[0321] (10) User interaction feedback and judgment of the number of iterations: Based on the global optimal solution of this round of updates, the energy operation scheduling strategy of the integrated energy system of multiple different types of parks is obtained, including obtaining the satisfaction score of the global optimal solution of this round of updates input by the user; based on the satisfaction score, it is judged whether the preset termination condition is met; if so, the global optimal solution of this round of updates is used as the energy operation scheduling strategy of the integrated energy system of multiple different types of parks; if not, the multi-objective particle swarm optimization algorithm is used to perform the next round of iterative solution on the multi-objective optimization model until the termination condition is met.

[0322] In this embodiment of the present invention, when initializing the particle swarm configuration, users are allowed to clarify the relative importance of each goal through weighted scoring, and are supported to directly preset the desired solution region. By integrating this preference information, an initial set of global optimal solutions that meets the user's global optimization expectations can be generated during the initialization phase, achieving adaptive distribution of individual solutions in the solution space:

[0323] The Pareto solution of each iteration of the fitness calculation is also displayed to the user. The user can score the solution and obtain a real-time satisfaction score of the corresponding Pareto solution; or select the solution that best suits their preference.

[0324] During the process of evolutionary strategy optimization and attention weight adjustment, a user feedback mechanism is introduced to dynamically adjust the attention weight according to the real-time satisfaction score of user feedback, so as to dynamically adjust the priority of each goal and guide particles to gather towards the solution that satisfies the user.

[0325] Constructing a user satisfaction function during particle velocity and position update Used to calculate user satisfaction to dynamically guide the search process, where S k is the user's preference coefficient for the kth objective function (for example, the preference coefficient is the weight of the kth objective function), f k ′ is the normalized objective function value, which is used to set the optimization preference. Specifically, the system periodically displays a radar chart of the Pareto solution set, and the user selects and marks a Pareto solution with high satisfaction. The system then calculates the user satisfaction of the Pareto solution based on the user satisfaction function. The user satisfaction is divided into intervals according to preset values ​​and converted into a satisfaction gradient with a scale of 0-10. The weight of the objective function is dynamically adjusted according to the satisfaction gradient. For example, if the satisfaction gradient of the objective function increases, the weight of the objective function will be increased accordingly according to the mapping relationship between the preset satisfaction gradient and the weight. If the user selects a low-emission solution multiple times, the weight of the objective function with the goal of minimizing carbon emissions will be increased by a certain proportion.

[0326] In the dynamic management of position constraints, a user feedback mechanism is introduced. Based on the real-time satisfaction scores of user feedback, the operation constraints are dynamically adjusted to achieve a balance between the controllability of the optimization process and the interpretability of the results, thereby improving the user's recognition of the solution set and obtaining the user-entered satisfaction score for the optimal solution;

[0327] The Pareto optimal solution (i.e., the global optimal solution) of each round of optimization is presented to the user. Based on the satisfaction score of the user's feedback, the score is compared with the set termination condition. If the termination condition is not met, the next round of iteration is continued; if the termination condition is met, the algorithm is terminated and the final optimal solution is output.

[0328] This embodiment of the present invention utilizes user-interactive optimization during the optimization process, integrating user input with decision-making to guide the search process, providing real-time feedback, and setting optimization preferences based on a multi-element energy efficiency evaluation index system, ensuring that the optimization results are more aligned with the user's actual needs and goals. By focusing on areas of interest, the algorithm can reduce searches in areas less likely to have good solutions, thereby reducing computational costs and ensuring that the optimization results are more aligned with the user's actual needs and goals.

[0329] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0330] A multi-dimensional energy efficiency evaluation index system was constructed by comprehensively considering energy consumption indicators, economic indicators, environmental indicators and product quality indicators. At the same time, each indicator and its sub-indicators as well as the relationship between them were taken into consideration to ensure the comprehensiveness and accuracy of the evaluation results.

[0331] Deep learning technologies (such as xLSTM and Transformer models) are used to extract and classify electricity consumption data, realizing data-driven energy production and usage feature classification. This can capture high-order features of different frequency components, extract periodic electricity consumption change characteristics, and provide accurate data support for energy management.

[0332] By analyzing the complementary characteristics of different types of parks in the time domain, a complementary relationship matrix of the multi-park energy system operation process was established, and the coordinated optimization of the multi-park energy system was achieved, which helps to improve the energy utilization efficiency of the entire region, reduce energy costs, and minimize environmental impact.

[0333] The Transformer-based multi-objective particle swarm optimization (TBMOPSO) method, combined with an attention mechanism, effectively explores complex solution spaces and balances conflicts between different optimization objectives, improving both the efficiency and quality of solutions to multi-objective optimization problems. An interactive user satisfaction mechanism is also introduced, allowing users to provide real-time feedback on optimization results. The algorithm then dynamically adjusts its search strategy based on this feedback. This user-centric approach ensures that optimization results are more aligned with users' actual needs and goals, enhancing the adaptability and practicality of the optimization algorithm.

[0334] See also Figure 5 , Figure 5 The embodiment of the present invention provides a structural block diagram of a multi-park cross-temporal and spatial comprehensive energy consumption multi-objective optimization system based on deep learning. The multi-objective optimization system for comprehensive energy systems based on deep learning includes: a feature extraction module 11, which is used to extract features of multivariate heterogeneous data in the operation process of comprehensive energy systems of multiple different types of parks through a pre-trained extended long short-term memory network model, and obtain the multidimensional energy consumption characteristics of the comprehensive energy systems of each park; an indicator model construction module 12, which is used to construct a multivariate energy efficiency evaluation indicator model for different types of parks based on the multidimensional energy consumption characteristics of multiple different types of parks based on the entropy weight method, hierarchical analysis method, and design structure matrix; a complementary relationship matrix construction module 13, which is used to perform correlation evaluation on the operation process of the comprehensive energy system of different types of parks, and analyze The complementary characteristics of different types of parks in the time domain are used to construct the park energy complementary relationship matrix; the multi-objective optimization model construction module 14 is used to construct multiple objective functions of multi-economy energy efficiency optimization according to the operation constraints constructed based on the energy consumption characteristics of different types of parks, the preset multiple control mechanisms, the multi-economy energy efficiency evaluation index model, and the park energy complementary relationship matrix, and establish a multi-objective optimization model for comprehensive regulation of multi-park energy systems based on the multiple objective functions; the optimization solution module 15 is used to adopt a multi-objective particle swarm optimization method based on the Transformer model and user satisfaction feedback interaction to solve the multi-objective optimization model for comprehensive regulation of multi-park energy systems under preset operation constraints, and obtain the energy operation scheduling strategy of the comprehensive energy system of multiple different types of parks.

[0335] It should be noted that the working process of each module in the multi-park cross-temporal and spatial comprehensive energy consumption multi-objective optimization system based on deep learning described in the embodiment of the present invention can refer to the working process of the multi-park cross-temporal and spatial comprehensive energy consumption multi-objective optimization method based on deep learning described in the above embodiment, and the technical effect achieved is also the same as the multi-objective optimization method of the comprehensive energy system based on deep learning described in the above embodiment, which will not be repeated here.

[0336] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, various improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy consumption based on deep learning, characterized by: include: By using a pre-trained extended long short-term memory network model, we extract features from multi-dimensional heterogeneous data during the operation of integrated energy systems in multiple different types of parks, and obtain the multi-dimensional energy consumption characteristics of the integrated energy systems of each park. According to the multi-dimensional energy consumption characteristics of different types of parks, a multivariate energy efficiency evaluation index model for different types of parks is constructed based on the entropy weight method, hierarchical analysis method, and design structure matrix; Conduct a correlation evaluation of the operation process of the integrated energy systems of different types of parks, analyze the complementary characteristics of different types of parks in the time domain, and construct a park energy complementary relationship matrix; Based on the operating constraints constructed based on the energy consumption characteristics of different types of parks, the preset multiple control mechanisms, the multi-element energy efficiency evaluation index model, and the park energy complementary relationship matrix, multiple objective functions for multi-element energy efficiency optimization are constructed, and a multi-objective optimization model for comprehensive regulation of multi-park energy systems is established based on the multiple objective functions; A multi-objective particle swarm optimization method based on the interaction of the Transformer model and user satisfaction feedback is used to solve the multi-objective optimization model of the comprehensive regulation of multi-park energy systems under preset operation constraints, and the energy operation scheduling strategies of the integrated energy systems of multiple different types of parks are obtained.

2. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 1 is characterized in that: The multi-dimensional energy consumption characteristics include: electricity consumption characteristic data, production capacity characteristic data, energy transmission characteristic data, energy storage characteristic data, and energy consumption characteristic data under different operating modes and different seasons.

3. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 1 is characterized in that: The method further comprises: The various types of data in the multivariate heterogeneous data are screened and classified, and a three-layer energy consumption index evaluation system is constructed using the fuzzy hierarchical analysis method. The energy consumption index evaluation system includes: target layer indicators, first-level indicators under the target layer indicators, and second-level indicators under the first-level indicators.

4. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 1 is characterized in that: The pre-trained extended long short-term memory network model is used to extract features from multi-dimensional heterogeneous data during the operation of the integrated energy systems of multiple different types of parks, and obtain the multi-dimensional energy consumption characteristics of the integrated energy systems of each park, including: Time series data is extracted from the multivariate heterogeneous data during the operation of the integrated energy system of multiple different types of parks to obtain multidimensional time series data; Inputting the multidimensional time series data into the extended long short-term memory network model, wherein the extended long short-term memory network model comprises an input layer, an xLSTM layer, and an output layer; The input items of the input layer are: The multidimensional time series data is defined as: x t ∈R d ; Wherein, T represents the time series length of the multidimensional time series data, d represents the feature dimension of each time step, and x t represents the t-th input feature in the multidimensional time series data; The xLSTM layer is used to capture the local dynamic characteristics of the multidimensional time series data through XLSTM; In the xLSTM layer, the input feature x t The intermediate features are obtained through nonlinear transformation Among them, g represents the nonlinear activation function, W x represents the weight matrix of the xLSTM layer, b x Represents the xLSTM layer bias term; Update hidden state and cell state: h t =O t ⊙tanh(C t ); Among them, σ represents the Sigmoid function, W f , W i , W c represents the weight matrix of the hidden layer, b f , b i , b c represents the bias term of the hidden layer, h t represents the hidden state, C t Represents the state of the memory unit, ⊙ represents element-by-element multiplication, o t Represents the output gate result; The hidden state sequence output by the xLSTM layer is Calculate the self-attention score of the hidden state sequence output by the xLSTM layer through the Transformer-based multi-head attention mechanism; The calculation process of the self-attention score of each attention mechanism includes: Q=H xLSTM W Q ,K=H xLSTM W K ,V=H xLSTM W V ; Among them, Q represents the query vector, K represents the key vector, V represents the value vector, and W Q ,W K ,W V Represents the weight matrix of the trainable multi-head attention mechanism; Self-attention score Among them, d k is the dimension of the key vector K, and the superscript T of the key vector K represents the vector transpose; The multi-head attention mechanism is integrated to obtain the global feature H attn =MultiHead(Q,K,V)=concat(head1,...,head n )I; Among them, W O Indicates the transformation matrix of the final linear transformation, head n Calculated by the self-attention score Attention(Q,K,V), h represents the number of heads in the multi-head attention mechanism; The hidden state sequence H xLSTM As a local feature, by fusing the local feature H xLSTM and global feature H attn , we get the multi-dimensional energy consumption characteristics: H final =ReLU(W f ·[H xLSTM ;H attn ]+b f ).

5. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning as claimed in claim 3 is characterized in that: According to the multi-dimensional energy consumption characteristics of multiple different types of parks, a multivariate energy efficiency evaluation index model for different types of parks is constructed based on the entropy weight method, hierarchical analysis method, and design structure matrix, including: Using the entropy weight method combined with the analytic hierarchy process, we constructed a target-level indicator evaluation system for different types of parks, including energy consumption indicators, economic indicators, environmental indicators, and product quality indicators, based on the multidimensional energy consumption characteristics of multiple different types of parks. Based on the target-level indicator evaluation system, after calculating the scores of the secondary indicators, we divided them into different groups according to seasonal labels, and calculated the weights of the secondary indicators in different seasons. Based on the weights of the secondary indicators in different seasons, we calculated the scores of the target-level indicators to construct a multivariate energy efficiency evaluation indicator model for each different type of park. The calculation process of the indicator score includes: Calculate the characteristic proportion of different areas in the park under each secondary indicator in, It represents the score of the i-th unit in the park on the j-th secondary indicator in season k, and n represents the total number of units; According to the characteristic proportion of each secondary indicator under the secondary indicator, the entropy value of each secondary indicator in each season is calculated. According to the entropy value of each secondary indicator, calculate the difference coefficient The secondary indicators are grouped according to seasons, and the weights of the secondary indicators in different seasons are calculated based on the difference coefficients. According to the weight of each secondary indicator, calculate the total score of the weighted secondary indicator layer in different seasons in, It represents the average score of the secondary indicator j of each unit in the park in season k; According to the scores of the first-level indicators, the total score of the first-level indicator layer is calculated; The analytic hierarchy process is used to obtain the weights of the first-level indicators, and based on the weights of the first-level indicators, the scores of the target layer indicators corresponding to the park's energy consumption indicators, economic indicators, environmental indicators, and energy product quality indicators are calculated.

6. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 5 is characterized in that: The method further comprises: Based on the scores of each target layer indicator in the multivariate energy efficiency evaluation index model of each park, the importance of different target layer indicators is evaluated under the standards of different types of parks, and the comprehensive energy efficiency evaluation score of the park is calculated; According to the comprehensive energy efficiency evaluation score of the park, the comprehensive score of the park is calculated by designing the structure matrix R; Among them, the subscript 1 of each element a in the matrix R corresponds to the energy consumption index, 2 corresponds to the economic index, 3 corresponds to the environmental index, and 4 corresponds to the product quality index. The order of the indicators in the subscripts indicates the influence of the previous indicator on the next indicator: In the matrix R, the value of each element indicates the level of mutual influence between the two indicators. When the element value is 0, it means no influence; when the element value is 1, it means weak influence; when the element value is 2, it means medium influence; when the element value is 3, it means strong influence; The comprehensive energy efficiency evaluation score of the park is calculated as follows: Score=‖C s R‖1; Among them, C s =[Y1,Y2,Y3,Y4] represents the scores of each target layer indicator of the park, and Y1, Y2, Y3, and Y4 represent the scores of energy consumption indicators, economic indicators, environmental indicators, and product quality indicators respectively.

7. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 1 is characterized in that: The above-mentioned process of evaluating the correlation of the operation process of the integrated energy system of different types of parks, analyzing the complementary characteristics of different types of parks in the time domain, and constructing the park energy complementary relationship matrix includes: The correlation coefficient method is used to analyze the complementary characteristics of different types of parks in the time domain and construct the complementary relationship matrix C: Among them, the element C in the complementary relationship matrix C IJ It represents the complementarity strength between process I of a certain park and process J of another park; C IJ =w1·(1-|ρ IJ |)+w2·(1-MIC IJ ), w1, w2 represent the preset weight parameters, C IJ represents the MIC value between process I and process J, ρ IJ represents the Spearman correlation coefficient between process I and process J; the processes include energy, energy transmission, energy storage, and energy use; The non-diagonal elements in the complementary relationship matrix are sorted, the element with the largest value is selected, and the park energy complementary relationship matrix is ​​constructed.

8. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 7 is characterized in that: The method further comprises: According to the energy consumption characteristics of different types of parks, based on the park energy complementary relationship matrix and the selected multiple control mechanisms, a comprehensive regulation mechanism model for the operation process of the multi-park energy system is established; The various control mechanisms include: a combined heat and power unit capable of providing both heat and electricity; a gas boiler unit using liquefied gas, natural gas or other gas energy as fuel, heating water into steam through a heating furnace to provide the required heat energy to users; The comprehensive control mechanism model includes: a dynamic model of wind power generation and photovoltaic power generation, a dynamic model of refrigerators and heat exchangers, and a dynamic model of electrical equipment and energy storage equipment.

9. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 1 is characterized in that: The method further comprises: The construction of a multi-park integrated energy system requires the fulfillment of multi-dimensional coordinated operational constraints. These operational constraints include: Equipment operating limit constraints for the integrated energy system within the park: Equipment installation constraints: in, Indicates the maximum and minimum configuration numbers of different devices k; Equipment operation constraints: Among them, γ k ∈(0,1) is a state variable describing whether device k is running: γ k =0 means the device is not running, γ k =1 means the equipment is operating normally; min is the minimum output power coefficient of the equipment, P t k is the operating power of device k at time t; Electric / thermal energy storage operation constraints: in, Indicates the maximum and minimum values ​​of the remaining capacity of device k; are the charge / discharge / heating power of the electrical energy storage and thermal energy storage of device k at time t, ω char_k 、ω dis_k They represent the charging, discharging / thermal efficiency of the device k, respectively, and are in the range of [0,1]; Maximum transmission limit constraints for purchased electricity and gas power: in, To determine the power ceiling for purchasing electricity, natural gas, and thermal energy from the energy grid, the natural gas purchased is considered to be consumed primarily by CHP units and gas boilers; Inter-park electrical and thermal energy interconnection lines must be within the maximum transmission power constraints: in is the power transmission power and heat transmission power between park i and park j, are the upper limits of electric energy exchange and thermal energy exchange between park i and park j, respectively; Power balance constraints include the supply and demand balance of various energy types, including electricity, heat, gas, and cooling. For a multi-campus integrated energy system, the energy transmission balance between parks can be described using an energy hub model: Where L is the output matrix of the energy hub, L e ,L g ,L h ,L c is the electricity load, gas load, heating load and cooling load on the user side, η t ,η he are transformer efficiency and CHP unit power generation efficiency, P b , G b 、H b They are electricity purchase, gas consumption of CHP units and gas consumption of gas boilers. P PV 、 P WT are photovoltaic and wind power generation power, are the conversion efficiencies of natural gas to electricity and heat, respectively. ng is the calorific value of natural gas, η ec ,η ac are the conversion efficiencies of electric refrigeration and absorption refrigeration respectively; G mt is the natural gas input, P ec is the electrical energy input power, H ac is the refrigerator input power, P es,d 、P es,c 、P gs,d 、P gs,c 、P hs,d 、P hs,c 、P cs,d 、P cs,c They are the charging and discharging power of electric energy, thermal energy, cold energy and gas respectively; Inter-park energy exchange constraints: The exchange power between different parks should meet global balance, that is, the algebraic sum of the exchange power is 0. Specifically, it can be expressed as: in They are the overall interactive electric power and thermal power between different parks respectively.

10. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 9, characterized in that: The method further comprises: Establish the first objective function f1 with the goal of minimizing total energy consumption; Establishing a second objective function f2 with the goal of minimizing energy cost; Establishing the third objective function f3 with the goal of minimizing carbon emissions; A fourth objective function f4 is established with the system operation quality as the goal.

11. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 10, characterized in that: The multi-objective particle swarm optimization method based on the interaction of the Transformer model and user satisfaction feedback is used to solve the multi-objective optimization model for the comprehensive regulation of multi-park energy systems under preset operation constraints, and obtain energy operation scheduling strategies for the comprehensive energy systems of multiple different types of parks, including: Initialize the particle swarm configuration related to user guidance: In the initial stage of the particle swarm optimization algorithm, complete the initialization of the three-dimensional core parameters, including: particle space coordinates, motion rate vector and attention weight parameter; initialize the optimal solution set that meets the user's global optimization expectations; Fitness calculation in multi-objective optimization: Calculate the fitness value of each particle on each objective function in the multi-objective optimization model; Position encoding: Perform position encoding on each particle to obtain the position information of the embedded input vector of the Transformer model; Transformer model attention mechanism calculation: In the multi-head self-attention mechanism layer, the input vector is mapped to different subspaces and the attention of each particle is calculated; Evolutionary strategy optimization and attention weight adjustment: Calculate the distance between each particle and the global optimal solution and optimize the attention weight of each particle; Update particle speed and position: By building a coupling relationship between the user evaluation adjustment coefficient and the cognitive and social learning factors, a speed update equation with feedback gain is established. By introducing a dynamic relaxation factor, a position update equation is established. Based on the individual optimal solution of each particle and the global optimal solution, the particle speed and position are updated based on the speed update equation and the position update equation. Dynamic management of position constraints: Based on the preset position update strategy, the particle position is checked for boundaries; Maintenance of non-dominated solution sets and updating of optimal solutions: For the current particle, its performance is evaluated through the fitness function. If the performance of the current particle is higher than that of its historical individual optimal solution, the operation of updating the particle speed and position is re-executed; Global optimal solution set maintenance: Use the Pareto dominance relationship to screen non-dominated solutions, add the global optimal solutions that meet the non-dominated conditions to the elite solution set, and update the global optimal solution based on the elite retention strategy; Based on the global optimal solution updated in this round, the energy operation scheduling strategy of the integrated energy system of multiple different types of parks is obtained.

12. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 11, characterized in that: The method further comprises: When initializing the particle swarm configuration, users are supported to perform weighted scoring on each objective function to determine the weight of each objective function; and users are supported to set the expected solution region; In the initialization phase, an initial set of optimal solutions that meets the user's global optimization expectations is generated based on the weights of each objective function and the expected solution region.

13. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 11, characterized in that: The method further comprises: During the fitness calculation process in multi-objective optimization, the Pareto solution of each iteration is displayed to the user, allowing the user to score the Pareto solution and obtain a real-time satisfaction score for the corresponding Pareto solution; or select a Pareto solution that meets the user's own preferences; During the evolutionary strategy optimization and attention weight adjustment process, a user feedback mechanism is supported to dynamically adjust the attention weight based on the real-time satisfaction ratings of user feedback; During the particle velocity and position update process, a user satisfaction function is constructed to calculate user satisfaction for the selected Pareto solution according to the user satisfaction function, and the weights of various objective functions are adjusted according to the user satisfaction; In the dynamic management of location constraints, a user feedback mechanism is supported to adjust operation constraints based on real-time satisfaction scores from user feedback.

14. The multi-objective optimization method for multi-park cross-temporal and cross-spatial comprehensive energy use based on deep learning according to claim 11, characterized in that: The method further comprises: The global optimal solution of each round of optimization is displayed to the user, and the user's satisfaction score for the global optimal solution is received; whether the satisfaction score of the global optimal solution meets the set termination condition is determined; if the termination condition is not met, the next round of iterative solution is continued; if the termination condition is met, the global optimal solution is output.

15. A multi-park cross-temporal and spatial comprehensive energy utilization multi-objective optimization system based on deep learning, characterized by: include: The feature extraction module is used to extract features from multi-dimensional heterogeneous data during the operation of the integrated energy systems of multiple different types of parks using a pre-trained extended long short-term memory network model, thereby obtaining the multi-dimensional energy consumption characteristics of the integrated energy systems of each park; The indicator model construction module is used to construct a multivariate energy efficiency evaluation indicator model for different types of parks based on the multidimensional energy consumption characteristics of multiple different types of parks using the entropy weight method, hierarchical analysis method, and design structure matrix; The complementary relationship matrix construction module is used to evaluate the correlation of the operation process of the integrated energy system of different types of parks, analyze the complementary characteristics of different types of parks in the time domain, and construct the park energy complementary relationship matrix; A multi-objective optimization model construction module is used to construct multiple objective functions for multi-element energy efficiency optimization based on the operating constraints constructed based on the energy consumption characteristics of different types of parks, multiple preset control mechanisms, the multi-element energy efficiency evaluation index model, and the park energy complementary relationship matrix, and establish a multi-objective optimization model for comprehensive regulation of multi-park energy systems based on the multiple objective functions; The optimization solution module is used to adopt a multi-objective particle swarm optimization method based on the Transformer model and user satisfaction feedback interaction to solve the multi-objective optimization model of the comprehensive control of multi-park energy systems under preset operation constraints, and obtain the energy operation scheduling strategy of the comprehensive energy systems of multiple different types of parks.

Images