Multi-source low-temperature waste heat adaptive grid-connection optimization method based on electric-thermal cooperative scheduling

Abstract

The application discloses a multi-source low-temperature waste heat self-adaptive grid-connection optimization method based on electric-thermal cooperative scheduling, relates to the technical field of energy management, and comprises the following steps: according to a dynamic weight coefficient, a node marginal electricity price and a real-time power generation instruction, an economic objective function and a carbon emission objective function are established, multi-objective cooperative optimization is carried out, and a power setting value of a power generation device is generated; power generation operation is carried out based on the power setting value of the power generation device, hierarchical control instructions of a heat pump and a heat exchanger are generated, equipment operation state parameters are collected to construct a digital twin, and an actual heating load value is generated through a compensation algorithm; and the deviation between the actual heating load value and a heat network load prediction curve is used to update global federated learning model parameters.The application realizes high-dimensional space-time modeling of a nonlinear coupling relationship between economy and a carbon emission target, and breaks through the technical limitation that it is difficult to respond to real-time fluctuations of an electricity price and a carbon price.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management, in particular to a multi-source low-temperature waste heat adaptive grid-connected optimization method based on electric-thermal collaborative scheduling. BACKGROUND

[0002] With the deepening of the low-carbon transformation of energy structure, the electric-thermal coupled utilization mode increasingly highlights its energy efficiency improvement and carbon emission reduction potential in industrial processes and regional energy supply scenarios. In recent years, heat pump technology, organic Rankine cycle (ORC) and absorption heat exchange equipment have been widely used for low-temperature waste heat grade upgrading and grid-connected heating, combined with power system operation requirements, gradually forming an electric-thermal collaborative regulation architecture. On this basis, some research introduces data-driven methods to optimize the scheduling strategy, uses time series models or neural networks to predict heat load, and combines electricity price signals for economic scheduling. At the same time, the introduction of carbon trading mechanism makes the carbon emission cost explicit, promoting the scheduling model to evolve from a single economic target to an economic-low-carbon dual target.

[0003] The core problems of the existing technical bottlenecks are concentrated in two aspects: first, heat load prediction is generally based on local historical data to build independent models, and the information island phenomenon between each heat source unit is serious, which leads to insufficient generalization ability of the model under load mutation or climate disturbance; although there have been attempts to introduce distributed learning mechanisms, the gradient aggregation process does not fully consider the spatio-temporal heterogeneity of different plant load patterns, lacks dynamic weighted fusion of key features, and thus the global prediction performance is limited. The second is that in the multi-objective optimization level, traditional methods often use linear weighting or fixed preference to set the priority of economic and carbon emission targets, and fail to dynamically associate real-time electricity price signals and carbon costs, leading to the decision result deviating from the actual operation optimal trajectory. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a multi-source low-temperature waste heat adaptive grid-connected optimization method based on electric-thermal collaborative scheduling to solve the problems of insufficient heat load prediction accuracy and poor dynamic adaptability of multi-objective optimization in the prior art.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] The application provides a multi-source low-temperature waste heat adaptive grid-connection optimization method based on electric-thermal cooperative scheduling, which comprises the following steps: obtaining real-time power generation instructions of a superior power grid, and extracting node marginal electricity price and real-time carbon quota data stream; based on the real-time power generation instructions, combining historical heat network data to calculate a gradient vector of a heat network load prediction sub-model, and obtaining global federated learning model parameters through an attention weighted aggregation algorithm to generate a rolling updated heat network load prediction curve; inputting the heat network load prediction curve and the real-time carbon quota data stream into a fractal dynamics framework to output a high-dimensional spatio-temporal evolution tensor, and generating a dynamic weight coefficient through an attention weighted mechanism; according to the dynamic weight coefficient, the node marginal electricity price and the real-time power generation instructions, establishing an economic objective function and a carbon emission objective function and performing multi-objective cooperative optimization to generate a power generation equipment power setting value; based on the power generation equipment power setting value, performing power generation operation to generate hierarchical control instructions of a heat pump and a heat exchanger, collecting equipment operation state parameters to construct a digital twin, and generating an actual heating load value through a compensation algorithm; and updating the global federated learning model parameters by using the deviation of the actual heating load value and the heat network load prediction curve.

[0008] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid-connection optimization method based on electric-thermal cooperative scheduling, wherein: the steps of obtaining real-time power generation instructions of a superior power grid, and extracting node marginal electricity price and real-time carbon quota data stream, are as follows,

[0009] The encrypted data packet of the real-time power generation instruction issued by the superior power grid is received through a power communication protocol;

[0010] The identity authentication and integrity check are performed on the encrypted data packet, and after decryption, the power setting feature, the node marginal price feature and the timestamp validity feature in the real-time power generation instruction are extracted to obtain the node marginal electricity price;

[0011] The carbon quota data stream is obtained through a real-time data interface of a carbon trading market, and a carbon price fluctuation prediction model is called synchronously to generate the real-time carbon quota data stream.

[0012] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid-connection optimization method based on electric-thermal cooperative scheduling, wherein: the steps of calculating the gradient vector of the heat network load prediction sub-model based on the real-time power generation instructions, combining historical heat network data are as follows,

[0013] According to the real-time power generation instructions, the historical heat network data of each factory area is aligned in space and time to generate a standardized training data set;

[0014] Based on the standardized training data set, the heat network load prediction sub-model is trained locally in each factory area, and the gradient vector of the heat network load prediction sub-model is calculated.

[0015] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid-connected optimization method based on electric-thermal cooperative scheduling, the specific steps of generating the rolling updated heat network load prediction curve are as follows,

[0016] The gradient vector of the heat network load prediction sub-model is added with differential privacy noise, and the global federated learning model parameters are generated by an attention weighted aggregation algorithm,

[0017] The global federated learning model parameters are distributed to each factory area to generate a rolling updated heat network load prediction curve.

[0018] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid-connected optimization method based on electric-thermal cooperative scheduling, the specific steps of outputting the high-dimensional spatio-temporal evolution tensor are as follows,

[0019] According to the heat network load prediction curve and the real-time carbon quota data stream, a state feature vector is generated by performing embedding operation through a variational circuit,

[0020] The heat load change characteristics of the heat network load prediction curve are extracted through Fourier spectrum analysis,

[0021] The state feature vector is input into a fractal dynamics framework, and spatio-temporal evolution is performed in combination with the heat load change characteristics to generate a high-dimensional spatio-temporal evolution tensor.

[0022] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid-connected optimization method based on electric-thermal cooperative scheduling, the specific steps of generating the dynamic weight coefficient are as follows,

[0023] Based on the high-dimensional spatio-temporal evolution tensor, chaotic intrinsic feedback quantities of each factory area are collected and a global feedback matrix is generated through an attention weighted mechanism,

[0024] The state feature vector and the global feedback matrix are fused to generate an original weight distribution field,

[0025] The dynamic weight field is spatially integrated and nonlinearly converted to generate a dynamic weight coefficient.

[0026] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid-connected optimization method based on electric-thermal cooperative scheduling, the specific steps of generating the power setting value of the power generation equipment are as follows,

[0027] According to the dynamic weight coefficient, the node marginal electricity price and the real-time power generation instruction, an economic objective function and a carbon emission objective function are established,

[0028] According to the physical characteristics of the generator set, the thermodynamic limit of the waste heat recovery device and the safety requirements of the power grid, multi-dimensional constraint conditions are set,

[0029] The economic objective function and the carbon emission objective function are multi-objective collaborative optimized under multi-dimensional constraints to obtain a non-dominated solution set;

[0030] The non-dominated solution set is weighted and sorted using a dynamic weight coefficient, and the highest priority solution is selected as the power setting value of the power generation equipment.

[0031] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid optimization method based on the electric-thermal collaborative scheduling, the specific steps of generating the hierarchical control instructions of the heat pump and the heat exchanger are as follows,

[0032] According to the power setting value of the power generation equipment, the power generation operation is performed, and the micro-energy level signal and the macro-temperature field distribution of the waste heat medium are collected;

[0033] The micro-energy level signal and the macro-temperature field distribution are fused into a waste heat medium multi-source heterogeneous data set, and a waste heat grade classification strategy vector is output;

[0034] Based on the waste heat grade classification strategy vector, the hierarchical control strategy is used to generate the hierarchical control instructions of the heat pump and the heat exchanger.

[0035] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid optimization method based on the electric-thermal collaborative scheduling, the specific steps of generating the actual heating load value are as follows,

[0036] The hierarchical control instructions are executed and the waste heat recovery equipment is driven, and the device operating state parameters are collected to construct a digital twin, and a dynamic state mapping matrix is generated;

[0037] Based on the dynamic state mapping matrix, combined with the heat network load prediction curve, a stable heating load is generated through a compensation algorithm;

[0038] The stable heating load is matched with the heat network load prediction curve to generate the actual heating load value.

[0039] As a preferred scheme of the multi-source low-temperature waste heat adaptive grid optimization method based on the electric-thermal collaborative scheduling, the specific steps of updating the global federated learning model parameters using the deviation of the actual heating load value and the heat network load prediction curve are as follows,

[0040] According to the deviation of the actual heating load value and the heat network load prediction curve, a synchronization deviation sequence is generated through a time alignment engine;

[0041] Based on the spatio-temporal fluctuation characteristics of the synchronization deviation sequence, the fractal feature weight factor of each plant area is calculated through a double integral formula;

[0042] According to the fractal feature weight factor, the global federated learning model parameters are updated.

[0043] The application has the beneficial effects that: through the attention-weighted aggregation algorithm, the global federated learning model parameters are generated, collaborative modeling and privacy protection under distributed data are realized, meanwhile, the security risks brought by the original data set are avoided, and then the accuracy and robustness of cross-factory heat load prediction are improved; through the attention-weighted mechanism, dynamic weight coefficients are generated, high-dimensional spatiotemporal modeling of the nonlinear coupling relationship between economy and carbon emission targets is realized, and the technical limitation of being difficult to respond to real-time fluctuations of electricity price and carbon price is broken through. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0045] Fig. 1 Flowchart of the multi-source low-temperature waste heat adaptive grid-connection optimization method based on electric-thermal collaborative scheduling.

[0046] Fig. 2 Flowchart of obtaining real-time power generation instructions and carbon quota data.

[0047] Fig. 3 Flowchart of generating heat network load prediction curve.

[0048] Fig. 4 Flowchart of generating power setting value of power generation equipment. DETAILED DESCRIPTION

[0049] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings in the specification.

[0050] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0051] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0052] REFERENCE Figs. 1-4For an embodiment of the present application, the embodiment provides a multi-source low-temperature waste heat self-adaptive grid-connection optimization method based on electric-thermal cooperative scheduling, including the following steps:

[0053] S1, obtain the real-time power generation instruction of the upper-level power grid, and extract the node marginal price and real-time carbon quota data stream.

[0054] The encrypted data packet of the real-time power generation instruction issued by the upper-level power grid is received through the power communication protocol.

[0055] The specific process includes that the power communication protocol receives the encrypted data packet of the real-time power generation instruction issued by the upper-level power grid, the encrypted data packet is extracted by analyzing the power communication protocol format, the encrypted data packet is decrypted using a pre-shared key, the content of the real-time power generation instruction is restored, and an acknowledgement response is returned to the upper-level power grid after the instruction analysis is completed.

[0056] The real-time power generation instruction includes target power generation power, execution time, instruction number, instruction type, priority, validity period, regulation rate limit, and signature information.

[0057] The pre-shared key is a symmetric encryption key that is negotiated and distributed in advance through a secure channel and is used for subsequent encryption and decryption operations of communication data.

[0058] The encrypted data packet is authenticated and integrity checked, and the power setting feature, node marginal price feature, and timestamp validity feature in the real-time power generation instruction are extracted after decryption to obtain the node marginal price.

[0059] The specific process includes that the identity of the sender is verified through a digital certificate, and the digital signature of the encrypted data packet is compared using a hash function to confirm the integrity. The legality of the data source is verified and it is confirmed that no tampering has occurred in the transmission process. After verification, the encrypted data packet is decrypted to restore the real-time power generation instruction in plaintext form. Then the power setting feature, node marginal price feature, and timestamp validity feature are accurately extracted from the real-time power generation instruction. The power setting feature is used to specify the target value of power control, the node marginal price feature contains the electricity price information, and the timestamp validity feature is used to judge the timeliness and applicability of the instruction. Finally, based on the accurate and effective node marginal price feature, the current node marginal price can be obtained.

[0060] Through the real-time data interface of the carbon trading market, the carbon quota data stream is obtained, the carbon price fluctuation prediction model is called synchronously, and the real-time carbon quota data stream is generated.

[0061] The specific process includes obtaining a continuously updated carbon quota data stream through a real-time data interface of a carbon trading market, ensuring the real-time and continuity of the carbon quota data stream. At the same time of obtaining the carbon quota data stream, a carbon price fluctuation prediction model is immediately called synchronously, which analyzes the real-time incoming carbon quota data stream in depth and obtains future price trends. The prediction results output by the carbon price fluctuation prediction model are fused and superimposed with the real-time carbon quota data stream, and finally an enhanced real-time carbon quota data stream containing real-time data and forward-looking prediction information is generated.

[0062] Further, the construction process of the carbon price fluctuation prediction model includes the collection and cleaning of historical carbon trading market data, including carbon quota price, trading volume, and market supply and demand information, and the like. Then, an algorithm architecture suitable for time series prediction (such as a long short-term memory network) is selected and trained to capture the regularity and characteristics of carbon price fluctuations. After training, the carbon price fluctuation prediction model is verified and optimized to ensure its accuracy in actual application, and finally a usable carbon price fluctuation prediction model is formed.

[0063] Further, the pre-training process of the carbon price fluctuation prediction model needs to collect and clean a large amount of historical carbon market data, including carbon quota price, trading volume, macroeconomic indicators, and related policy texts, and then use these data to preliminarily train the model under a deep learning framework, allowing the carbon price fluctuation prediction model to learn the basic patterns and potential laws in the data through unsupervised or self-supervised learning. This stage focuses on enabling the carbon price fluctuation prediction model to master the basic characteristics and fluctuation characteristics of the carbon market, laying a solid foundation for subsequent fine tuning.

[0064] S2, based on the real-time power generation instruction, a gradient vector of a heat network load prediction sub-model is calculated in combination with historical heat network data, and a global federated learning model parameter is obtained through an attention weighted aggregation algorithm to generate a rolling updated heat network load prediction curve.

[0065] According to the real-time power generation instruction, historical heat network data of each plant area is spatio-temporally aligned to generate a standardized training data set.

[0066] The specific process includes, according to the specific content and requirements in the real-time power generation instruction, synchronously associated extraction of historical heat network data of each plant area, accurate matching and alignment of the time stamp of the real-time power generation instruction and the time series of the historical heat network data, and uniform regularization of the historical heat network data of each plant area in the spatial dimension, eliminating the heterogeneity and scale difference among the historical heat network data, and finally fusing to generate a standardized training data set with consistent spatio-temporal characteristics and directly usable for subsequent analysis.

[0067] It should be noted that the specific content and requirements in the real-time power generation instruction refer to the power set value, execution time window, node marginal price parameter and power grid scheduling priority index contained in the real-time power generation instruction.

[0068] Based on the standardized training data set, the heat network load prediction sub-model is trained locally in each factory area, and the gradient vector of the heat network load prediction sub-model is calculated, and the expression is:

[0069]

[0070] Among them, Loss function of heat network load prediction sub-model Gradient vector of model parameter θ, Indicates the loss function of the heat network load prediction sub-model, Indicates the model parameter, Indicates the number of samples in a batch in the training of the heat network load prediction sub-model, Indicates the index number of the sample in the training data set, The input feature vector of the first training sample, Indicates the output prediction value of the heat network load prediction sub-model to the input feature vector , Indicates the real heat network load value of the first training sample, Indicates the gradient vector of the model parameter of the heat network load prediction sub-model when the input feature vector .

[0071] The specific process includes that based on the standardized training data set, the heat network load prediction sub-model is trained locally in each factory area, and the gradient vector of the heat network load prediction sub-model is calculated. The gradient vector reflects the sensitivity of the heat network load prediction sub-model loss function to the model parameter. By aggregating the differences between the prediction values and the real heat network load values of all samples in the batch, and multiplying these differences with the corresponding sample heat network load prediction sub-model output gradient, a mathematical correlation between the prediction bias and the heat network load prediction sub-model parameter update direction is established, and the optimization direction of the heat network load prediction sub-model under the current parameter is comprehensively evaluated.

[0072] ​Further, the construction process of the heat network load prediction sub-model first collects historical meteorological data, user heat consumption behavior records and pipe network operation parameters as input features, and after cleaning and standardizing these data, adopts a long short-term memory neural network architecture for training. The network input layer receives a multi-source feature sequence within a 24-hour sliding window, extracts spatio-temporal correlation features through three hidden layers, and generates heat network load prediction values for the next 1 to 6 hours. During the training process, the mean square error loss function and the Adam optimizer are used for parameter updating, and finally a heat network load prediction sub-model capable of dynamically predicting load changes according to real-time input is obtained.

[0073] The gradient vector of the heat network load prediction sub-model is added with differential privacy noise, and is fused through an attention weighted aggregation algorithm to generate global federated learning model parameters.

[0074] The specific process includes adding differential privacy noise to the gradient vector of the heat network load prediction sub-model to mask the individual characteristics of the gradient vector of the single heat network load prediction sub-model. Then, the attention weighted aggregation algorithm is used to fuse all the noise processed gradient vectors. The attention weighted aggregation algorithm dynamically determines the contribution of each heat network load prediction sub-model gradient vector and assigns a corresponding weight. Through weighted averaging, the multi-source gradient vectors are effectively integrated, and finally the optimized global federated learning model parameters are generated.

[0075] The global federated learning model parameters are distributed to each factory area to generate a rolling updated heat network load prediction curve.

[0076] The specific process includes safely distributing the global federated learning model parameters to each factory area. After receiving the global federated learning model parameters, each factory area immediately updates the configuration of the local heat network load prediction sub-model parameters. The updated heat network load prediction sub-model combines real-time collected factory operation data and performs multi-step ahead prediction at fixed time intervals. Through continuous iterative prediction, a heat network load prediction curve that evolves synchronously with the time series is generated. The heat network load prediction curve is continuously updated as new data is continuously incorporated, forming a dynamically adjusted rolling updated heat network load prediction curve.

[0077] S3, input the heat network load prediction curve and real-time carbon quota data stream into the fractal dynamics framework, output a high-dimensional spatio-temporal evolution tensor, and generate a dynamic weight coefficient through an attention weighted mechanism.

[0078] According to the heat network load prediction curve and the real-time carbon quota data stream, a state feature vector is generated by performing embedding operation through a variational circuit.

[0079] The specific process includes feature alignment and data fusion of the time-series load data contained in the heat network load prediction curve and the dynamic carbon price information contained in the real-time carbon quota data stream to form joint input features. The joint input features are embedded by a variational circuit, which encodes high-dimensional classical data into amplitude or phase information. After a series of controlled rotation gates and entanglement gates, a state feature vector expressing complex data features is finally generated in Hilbert space.

[0080] The heat load variation characteristics of the heat network load prediction curve are extracted by Fourier spectrum analysis.

[0081] The specific process includes discrete Fourier transform of the heat network load prediction curve by Fourier spectrum analysis, which converts time-domain load data (referring to a sequence of heat network load values continuously changing over time, directly measured by sensors installed on the heating pipe network at a fixed sampling frequency) into frequency-domain representation, generating complex spectrum containing amplitude and phase information. The power spectral density function is obtained by squaring the spectral amplitude, the characteristic frequencies and harmonic components corresponding to the peak values in the power spectrum are identified, and the energy proportion and distribution pattern of these characteristic frequencies are analyzed. In the key frequency band range of 0.1Hz to 10Hz, the relative energy ratios of the fundamental frequency and each harmonic are quantified, and three main frequency-domain feature parameters reflecting the periodicity, abruptness and persistence of the load are extracted: fundamental frequency energy proportion, harmonic energy dispersion and frequency band energy entropy value. These three parameters together constitute the frequency-domain fingerprint describing the dynamic variation characteristics of the heat load.

[0082] The state feature vector is input into the fractal dynamics framework to combine the heat load variation characteristics for spatio-temporal evolution, generating a high-dimensional spatio-temporal evolution tensor.

[0083] The specific process includes inputting the state feature vector into the fractal dynamics framework for multi-scale analysis. The fractal dynamics framework simulates the spatio-temporal evolution of the state feature vector through iterative functions and fractional differential equations. In the evolution process, the fractal dimension and Hurst index contained in the heat load variation characteristics are integrated, which guides the state feature vector to exhibit long-term memory characteristics in the time dimension and self-similar structure characteristics in the spatial dimension. Through continuous iteration of evolution, a high-dimensional spatio-temporal evolution tensor containing time evolution information and spatial structure information is obtained.

[0084] Based on the high-dimensional spatio-temporal evolution tensor, the chaotic intrinsic feedback quantities of each plant area are collected and a global feedback matrix is generated through an attention weighting mechanism.

[0085] The specific process includes extracting the corresponding chaotic intrinsic feedback quantity of each plant area based on the high-dimensional spatiotemporal evolution tensor, the chaotic intrinsic feedback quantity including characteristic indexes such as Lyapunov exponent and Kolmogorov entropy, obtaining the weight coefficient of the chaotic intrinsic feedback quantity of each plant area through an attention weighting mechanism, the attention weighting mechanism dynamically allocating the weight coefficient according to the amplitude and change trend of the chaotic intrinsic feedback quantity, and finally matrixing and integrating the chaotic intrinsic feedback quantity of each plant area after the weight coefficient is allocated to generate a global feedback matrix containing global information.

[0086] The state feature vector is fused with the global feedback matrix to generate an original weight distribution field.

[0087] The specific process includes performing a tensor fusion operation on the state feature vector and the global feedback matrix to generate a fusion tensor, and performing spatial search optimization on the fusion tensor through an optimization algorithm, the optimization algorithm using parallel computing characteristics to explore the optimal solution space, minimizing the objective function by adjusting the superposition state probability amplitude, and finally outputting the original weight distribution field representing the optimal weight configuration.

[0088] It should be noted that the superposition state probability amplitude is a probability distribution coefficient between the ground state and the excited state, which is obtained by linearly transforming the initial state through the Hadamard gate operation of the circuit.

[0089] The dynamic weight field is spatially integrated and nonlinearly converted to generate a dynamic weight coefficient.

[0090] The specific process includes spatial integration operation on the dynamic weight field, the spatial integration operation performing sliding analysis in the dynamic weight field through a convolution kernel and aggregating the weight values of adjacent regions in the dynamic weight field, and then applying nonlinear conversion to the result of the spatial integration operation, the nonlinear conversion process using an activation function to nonlinearly map the weight distribution after the spatial integration operation. Finally, a dynamic weight coefficient suitable for multi-scale calculation is generated.

[0091] S4, according to the dynamic weight coefficient, the node marginal electricity price and the real-time power generation instruction, establishing an economic objective function and a carbon emission objective function and performing multi-objective collaborative optimization to generate a power setting value of the power generation equipment.

[0092] According to the dynamic weight coefficient, the node marginal electricity price and the real-time power generation instruction, an economic objective function and a carbon emission objective function are established.

[0093] The specific process includes constructing an economic objective function according to the dynamic weight coefficient, the node marginal price and the real-time power generation instruction, the economic objective function taking minimization of power generation operation cost and maximization of power generation benefit as optimization objectives, wherein the power generation operation cost includes fuel cost and maintenance cost, and the power generation benefit is calculated according to the product of the node marginal price and the power generation amount. Meanwhile, a carbon emission objective function is established, the carbon emission objective function taking the real-time carbon quota as a benchmark to constrain the total carbon emission in the power generation process, realizing low-carbon optimization by minimizing the deviation between the actual carbon emission and the carbon quota, and the two objective functions are weighted and integrated through the dynamic weight coefficient to form a multi-objective optimization problem.

[0094] According to the physical characteristics of the generator set, the thermodynamic limit of the waste heat recovery device and the safety requirements of the power grid, multi-dimensional constraint conditions are set.

[0095] The specific process includes setting the power ramping rate and the upper and lower limit constraints of the output according to the physical characteristics of the generator set, setting the thermal power conversion efficiency and the heat transfer temperature difference boundary constraints according to the thermodynamic limit of the waste heat recovery device, and setting the node voltage safety and line transmission capacity constraints in combination with the safety requirements of the power grid, to jointly constitute the safe operation multi-dimensional constraint conditions covering the electrical and thermal fields.

[0096] The economic objective function and the carbon emission objective function are subjected to multi-objective collaborative optimization under the multi-dimensional constraint conditions to obtain a non-dominated solution set.

[0097] The specific process includes multi-objective collaborative optimization of the economic objective function and the carbon emission objective function under the multi-dimensional constraint conditions, simultaneous optimization of the economic objective function and the carbon emission objective function by using a multi-objective optimization algorithm, obtaining a solution set that simultaneously satisfies the optimization of the economic objective function and the carbon emission objective function by Pareto frontier search, and these solutions do not exist absolute optimal solution but exist mutually non-dominated optimal solutions, finally forming a non-dominated solution set containing multiple Pareto optimal solutions.

[0098] The non-dominated solution set is weighted and sorted using the dynamic weight coefficient, and the highest priority solution is selected as the power setting value of the power generation equipment.

[0099] The specific process includes weighted scoring of each solution in the non-dominated solution set using the dynamic weight coefficient, adjustment of the priority weight of the economic objective function and the carbon emission objective function according to the real-time operation demand, obtaining of the comprehensive evaluation value of each solution by weighted summation, subsequent sorting of the non-dominated solution set in descending order of the comprehensive evaluation value, and finally selection of the highest priority solution with the highest ranking as the power setting value of the power generation equipment.

[0100] S5, based on the power generation equipment power set value to perform power generation operation, generate heat pump and heat exchanger hierarchical control instruction, and collect equipment operation state parameters to construct digital twin, and generate actual heating load value through compensation algorithm.

[0101] According to the power generation equipment power set value to perform power generation operation, and collect the micro energy level signal and the macro temperature field distribution of the waste heat medium.

[0102] The specific process includes: adjusting the power generation unit output according to the power generation equipment power set value to perform power generation operation, synchronously collecting the micro energy level signal of the waste heat medium molecular vibration energy level and electron transition signal, and measuring the macro temperature field distribution of the waste heat recovery device surface through an infrared thermal imager. The micro energy level signal reflects the energy quality characteristics of the waste heat medium (such as the thermal excitation potential corresponding to the molecular bond vibration frequency and the internal energy storage density represented by the electron energy level transition), and the macro temperature field distribution shows the spatial thermodynamic state of the waste heat medium. The micro energy level signal and the macro temperature field distribution of the waste heat medium together constitute the complete energy characteristic description of the waste heat medium.

[0103] The micro energy level signal and the macro temperature field distribution are fused into a waste heat medium multi-source heterogeneous data set, and a waste heat grade classification strategy vector is output.

[0104] The specific process includes: data alignment and feature fusion of the micro energy level signal and the macro temperature field distribution, forming a waste heat medium multi-source heterogeneous data set containing multi-dimensional energy information, parallel processing of the waste heat medium multi-source heterogeneous data set, optimization solving of the waste heat medium multi-source heterogeneous data set, and finally outputting a waste heat grade classification strategy vector representing the waste heat quality grade classification scheme.

[0105] Based on the waste heat grade classification strategy vector, the hierarchical control strategy is used to generate the hierarchical control instruction of the heat pump and the heat exchanger.

[0106] The specific process includes: based on the waste heat quality grade classification scheme defined by the waste heat grade classification strategy vector, the control logic in the waste heat grade classification strategy vector is analyzed through the hierarchical control strategy. The hierarchical control strategy matches the corresponding operation mode according to the quality grade specified by the waste heat grade classification strategy vector (for example: high-temperature waste heat preferentially drives the power generation unit, medium-temperature waste heat drives the absorption heat pump, and low-temperature waste heat directly enters the plate heat exchanger), and generates hierarchical control instructions such as start-stop instructions, power regulation instructions and flow distribution instructions for the heat pump and the heat exchanger.

[0107] It should be pointed out that the waste heat quality grade division scheme divides waste heat into three grades according to temperature interval, energy level characteristics and heat flow density parameters: the first grade is high-grade waste heat with temperature > 200℃ and electronic excitation state proportion > 30%, suitable for steam power generation; the second grade is medium-grade waste heat with 80-200℃ mainly as molecular vibration energy, suitable for heat pumps; the third grade is low-grade waste heat with temperature < 80℃ mainly as lattice vibration energy, suitable for regional heating. The waste heat quality grade division scheme optimizes the dynamic adjustment of the grading threshold value, and optimizes the linkage with the power grid price, while maintaining the monitoring accuracy of 1cm2 temperature field resolution and 0.1eV energy resolution.

[0108] The hierarchical control instructions are executed and the waste heat recovery equipment is driven, and the device operation state parameters are collected to construct a digital twin, and a dynamic state mapping matrix is generated.

[0109] The specific process includes executing hierarchical control instructions and driving waste heat recovery equipment, including heat pumps and heat exchangers, to perform start-stop, power regulation, and flow distribution operations, while collecting device operation state parameters such as inlet and outlet temperatures, working pressure, medium flow, and device vibration frequency of heat pumps and heat exchangers in real time through a sensor network. The device operation state parameters are used to construct a digital twin that is updated synchronously with the physical device. The digital twin obtains the spatial distribution and time evolution relationship of the device operation state parameters through multi-physical field coupling simulation, and finally generates a dynamic state mapping matrix that comprehensively represents the device operation state.

[0110] It should be pointed out that by collecting device operation state parameters such as generator set speed, waste heat recovery device inlet and outlet temperature, pressure and flow in real time, combining the first law of thermodynamics and heat transfer equations to establish a dynamic simulation model, mapping the sensor data stream to the dynamic simulation model with millisecond-level delay, using an adaptive Kalman filter algorithm to correct the parameter deviation of the dynamic simulation model, and simultaneously obtaining microscopic energy level data, a digital twin with multi-scale characteristics is constructed. The digital twin keeps data synchronization with the physical device through OPC UA protocol, and verifies its dynamic response consistency according to Lyapunov index, and finally realizes virtual-real synchronous closed loop updated every 10ms.

[0111] Based on the dynamic state mapping matrix, combined with the heat network load prediction curve, a stable heating load is generated through compensation algorithm.

[0112] The specific process includes providing real-time equipment operation state information based on the dynamic state mapping matrix, combining future load demand data provided by the heat network load prediction curve (for example: the peak and valley distribution of regional heating heat load in the next 15 minutes), and analyzing the difference between the dynamic state mapping matrix and the heat network load prediction curve through a compensation algorithm. The compensation algorithm obtains the compensation amount according to the deviation between the actual operation capacity of the waste heat recovery equipment and the predicted load, and dynamically adjusts the heat supply parameters, and finally generates a stable heat supply load that meets the heating demand and operates stably.

[0113] Match the stable heat supply load with the heat network load prediction curve to generate the actual heat supply load value.

[0114] The specific process includes accurately aligning the stable heat supply load with the heat network load prediction curve in the time dimension, comparing the actual supply value of the stable heat supply load with the predicted demand value of the heat network load prediction curve point by point, obtaining the deviation degree of the stable heat supply load and the heat network load prediction curve through the least squares fitting algorithm and dynamically adjusting the weight distribution, using sliding window mean filtering smoothing processing time sequence fluctuation, and finally generating an actual heat supply load value that meets the real-time heat supply demand and conforms to the long-term trend of the heat network load prediction curve.

[0115] S6, update the global federated learning model parameters using the deviation between the actual heat supply load value and the heat network load prediction curve.

[0116] Based on the actual heat supply load value and the heat network load prediction curve, a synchronization deviation sequence is generated through a time alignment engine.

[0117] The specific process includes, based on the actual heat supply load value and the heat network load prediction curve, matching the data points of the actual heat supply load value and the heat network load prediction curve according to the unified timestamp through the time alignment engine. The time alignment engine uses an interpolation algorithm to fill in the missing time point data, and point by point obtains the difference between the corresponding values of the actual heat supply load value and the heat network load prediction curve, and generates a synchronization deviation sequence reflecting the deviation between prediction and performance.

[0118] Based on the spatiotemporal fluctuation characteristics of the synchronization deviation sequence, the fractal feature weight factor of each plant area is calculated through a double integral formula, and the expression is:

[0119] ;

[0120] Wherein, represents the fractal feature weight factor of the th plant area, represents the current calculated plant area number, represents the length of the time window, represents the th plant area at time point The deviation between the actual heating load and the predicted load. This represents the first time variable in the integration operation. Indicates the first Each factory area at a specific time point The deviation between the actual heating load and the predicted load. This represents the second time variable in the integration operation. This represents the total number of factory areas participating in federated learning aggregation. This represents the index used to traverse all factory areas. Indicates the first Each factory area at a specific time point The deviation between the actual heating load and the predicted load. Indicates the first Each factory area at a specific time point The deviation between the actual heating load and the predicted load.

[0121] The specific process includes calculating the fractal feature weighting factor for each plant area using a double integral formula based on the spatiotemporal fluctuation characteristics of the synchronized deviation sequence. The numerator of the double integral formula performs a double integral operation on the absolute value of the load deviation difference at all time points within the time window of the plant area, and sums the similar integral results for all plant areas, ultimately generating a fractal feature weighting factor that reflects the spatiotemporal complexity of the load fluctuation of the plant area.

[0122] It should be noted that spatiotemporal fluctuation characteristics refer to the comprehensive characterization of the synchronization deviation sequence in the time dimension (such as the long-range dependence characterized by the Hearst exponent) and the cross-plant correlation in the spatial dimension (such as the fluctuation complexity characterized by the fractal dimension).

[0123] Update the parameters of the global federated learning model based on the fractal feature weight factors.

[0124] The specific process includes adjusting the contribution of each plant area in the federated learning aggregation process according to the fractal feature weight factor. The fractal feature weight factor reflects the spatiotemporal complexity of the plant area load fluctuation. The fractal feature weight factor is used as a weighting coefficient in the global federated learning model parameter update process. The global federated learning model gradients uploaded by each plant area are fused through a weighted average algorithm to finally generate the updated global federated learning model parameters.

[0125] This embodiment also provides a computer device applicable to the multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling proposed in the above embodiment.

[0126] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0127] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the method for realizing multi-source low-temperature waste heat adaptive grid-connection optimization based on electric-thermal cooperative scheduling proposed in the above embodiment; and the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0128] To sum up, the application generates global federated learning model parameters through the attention weighted aggregation algorithm, realizes collaborative modeling and privacy protection under distributed data, avoids the security risks brought by original data sets, and further improves the accuracy and robustness of cross-factory thermal load prediction. Through the attention weighted mechanism, dynamic weight coefficients are generated to realize high-dimensional spatiotemporal modeling of the nonlinear coupling relationship between economy and carbon emission targets, breaking through the technical limitation of being difficult to respond to real-time fluctuations in electricity prices and carbon prices.

[0129] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling, characterized in that: include, Obtain real-time power generation instructions from the upper-level power grid and extract node marginal electricity price and real-time carbon quota data streams; Based on real-time power generation instructions and historical heating network data, the gradient vector of the heating network load prediction sub-model is calculated. Then, the parameters of the global federated learning model are obtained through an attention-weighted aggregation algorithm, generating a continuously updated heating network load prediction curve. The specific steps are as follows: Based on real-time power generation instructions, and combined with historical heating network data from each plant area, a standardized training dataset is generated through spatiotemporal alignment. The standardized training dataset includes historical meteorological data, user heating behavior records, and pipeline operation parameters as input features. Based on a standardized training dataset, a sub-model for predicting the heating network load is trained locally in each plant area, and the gradient vector of the sub-model is calculated. Adding differential privacy noise to the gradient vector of the heating network load prediction sub-model and fusing it through an attention-weighted aggregation algorithm to generate global federated learning model parameters means adding differential privacy noise that meets privacy protection requirements to the gradient vector of the heating network load prediction sub-model to mask the individual characteristics of the gradient vector of a single heating network load prediction sub-model. The parameters of the global federated learning model are distributed to each plant area to generate a rolling updated heating network load prediction curve. Then, the attention weighted aggregation algorithm is used to fuse all the gradient vectors that have been processed by noise. The attention weighted aggregation algorithm dynamically obtains the contribution of the gradient vector of each heating network load prediction sub-model and assigns it a corresponding weight. Through weighted averaging, the multi-source gradient vectors are effectively integrated, and finally the optimized global federated learning model parameters are generated. The heating network load prediction curve and real-time carbon quota data stream are input into the fractal dynamics framework, outputting a high-dimensional spatiotemporal evolution tensor, and generating dynamic weight coefficients through an attention weighting mechanism. Based on dynamic weighting coefficients, nodal marginal electricity prices, and real-time power generation instructions, an economic objective function and a carbon emission objective function are established and multi-objective collaborative optimization is performed to generate a power setpoint for the power generation equipment. The economic objective function and the carbon emission objective function are weighted and integrated through dynamic weighting coefficients to form a multi-objective optimization problem. Power generation is performed based on the power setpoint of the power generation equipment, generating hierarchical control commands for the heat pump and heat exchanger. At the same time, the operating status parameters of the equipment are collected to construct a digital twin, and the actual heating load value is generated through a compensation algorithm. The parameters of the global federated learning model are updated by using the deviation between the actual heating load value and the heating network load prediction curve.

2. The multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling as described in claim 1, characterized in that: The specific steps for obtaining real-time power generation instructions from the upper-level power grid and extracting the node marginal electricity price and real-time carbon quota data stream are as follows. Encrypted data packets of real-time power generation instructions issued by the upper-level power grid are received via power communication protocols. The encrypted data packets are authenticated and their integrity is verified. After decryption, the power setting features, node marginal price features, and timestamp validity features in the real-time power generation instructions are extracted to obtain the node marginal price. By using the real-time data interface of the carbon trading market, carbon quota data streams are obtained, and carbon price fluctuation prediction models are called simultaneously to generate real-time carbon quota data streams.

3. The multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling as described in claim 2, characterized in that: The specific steps for outputting the high-dimensional spatiotemporal evolution tensor are as follows. Based on the heating network load forecast curve and the real-time carbon quota data stream, an embedding operation is performed through a variational circuit to generate a state feature vector. The heat load variation characteristics of the heating network load prediction curve were extracted using Fourier spectrum analysis. The state feature vector is input into the fractal dynamics framework and combined with the characteristics of thermal load change to perform spatiotemporal evolution, generating a high-dimensional spatiotemporal evolution tensor.

4. The multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling as described in claim 3, characterized in that: The specific steps for generating dynamic weight coefficients are as follows: Based on the high-dimensional spatiotemporal evolution tensor, the chaotic intrinsic feedback quantities of each plant area are collected and a global feedback matrix is ​​generated through an attention weighting mechanism; The state feature vector and the global feedback matrix are fused to generate the original weight distribution field; The dynamic weight field is spatially integrated and nonlinearly transformed to generate dynamic weight coefficients.

5. The multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling as described in claim 4, characterized in that: The specific steps for generating the power setpoint for the power generation equipment are as follows. Based on dynamic weighting coefficients, nodal marginal electricity prices, and real-time power generation instructions, establish economic objective functions and carbon emission objective functions; Based on the physical characteristics of the generator set, the thermodynamic limits of the waste heat recovery device, and the power grid safety requirements, multidimensional constraints are set. Under multidimensional constraints, multi-objective collaborative optimization is performed on the economic objective function and the carbon emission objective function to obtain a non-dominated solution set; The non-dominated solution set is weighted and sorted using dynamic weighting coefficients, and the highest priority solution is selected as the power setting value for the power generation equipment.

6. The multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling as described in claim 5, characterized in that: The specific steps for generating the graded control commands for the heat pump and heat exchanger are as follows. The power generation operation is performed according to the power setting value of the power generation equipment, while the micro energy level signal and macro temperature field distribution of the waste heat medium are collected. The micro-energy level signal and the macro-temperature field distribution are fused into a multi-source heterogeneous dataset of waste heat medium, and the waste heat grade classification strategy vector is output. Based on the waste heat grade classification strategy vector, graded control commands for heat pumps and heat exchangers are generated through a graded control strategy.

7. The multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling as described in claim 6, characterized in that: The specific steps for generating the actual heating load value are as follows: The system executes hierarchical control commands and drives the waste heat recovery equipment, while simultaneously collecting equipment operating status parameters to construct a digital twin and generate a dynamic status mapping matrix. Based on the dynamic state mapping matrix and combined with the heating network load prediction curve, a stable heating load is generated through a compensation algorithm. The stable heating load is matched with the heating network load prediction curve to generate the actual heating load value.

8. The multi-source low-temperature waste heat adaptive grid-connected optimization method based on electrothermal coordinated scheduling as described in claim 7, characterized in that: The steps for updating the parameters of the global federated learning model by utilizing the deviation between the actual heating load value and the heating network load prediction curve are as follows: Based on the deviation between the actual heating load value and the heating network load prediction curve, a synchronization deviation sequence is generated through a time alignment engine; Based on the spatiotemporal fluctuation characteristics of the synchronization deviation sequence, the fractal characteristic weighting factor of each plant area is calculated using a double integral formula. Update the parameters of the global federated learning model based on the fractal feature weight factors.

Images