Power efficiency optimization method based on intelligent load management

Abstract

The invention discloses a power efficiency optimization method based on intelligent load management. The method comprises the following steps: S1, generating an initial load characteristic matrix; s2, decomposing the initial load characteristic matrix into a short-term behavior characteristic matrix and a long-term trend characteristic matrix through pseudo-symmetry decoupling; s3, multi-level load demand prediction is carried out through a fusion structure of the multi-scale convolution and the recurrent neural network; s4, coding a cognitive gene sequence, triggering gene mutation and recombination operation when significant change occurs, and generating a load distribution strategy; s5, a load distribution strategy is transmitted into a causal reasoning and self-supervision feedback control module, and if the deviation between feedback data and a prediction result exceeds a threshold value, a self-learning mechanism is triggered; s6, dynamically distributing the load priority, and generating a decision path of load management; and S7, performing adaptive optimization through a multi-layer feature feedback closed-loop optimization system. The intelligent load management method is adopted, and the power efficiency is optimized through multi-layer feature decoupling and closed loop.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent load management, and in particular to an electric power energy efficiency optimization method based on intelligent load management. Background Art

[0002] In the field of intelligent power management and load regulation, the operation efficiency and energy efficiency optimization of power systems have long been key research topics. With the rapid growth of social electricity demand and the gradual popularization of renewable energy, the load of the power system has shown great volatility and uncertainty. Traditional load management methods have been unable to meet the needs of modern power management. Generally, the main purpose of power system load management is to improve the efficiency of power resource utilization and system stability through load forecasting and priority adjustment on the basis of ensuring the balance of power supply and demand, so as to achieve energy efficiency optimization of the system. However, existing load management technologies still have significant deficiencies in load forecasting accuracy, strategy adaptability, dynamic response speed, etc.

[0003] Traditional power load management methods mostly use control methods based on experience or rules to analyze and predict factors such as electricity prices, load demand, and environmental changes. This type of method mainly relies on historical data, uses simple mathematical models to predict short-term load demand and formulate load distribution strategies. However, due to the lack of flexible adaptability of traditional methods, they appear to be slow to respond to rapid load fluctuations or sudden changes. At the same time, traditional load management systems mostly rely on fixed rules and thresholds, and cannot adjust load strategies in real time according to dynamic changes in actual conditions, which limits the system's sensitivity to load fluctuations and its ability to handle different load priorities. As a result, the power system is prone to reduced energy efficiency during high-load peak periods, and even causes system instability. In addition, traditional methods are difficult to effectively integrate the impact of multi-dimensional factors such as electricity prices, load demand fluctuations, and response time on load management, resulting in inaccurate load forecasting, low utilization of power resources, and hindering the further improvement of power system energy efficiency.

[0004] In recent years, intelligent power management methods have made some progress in improving load forecasting accuracy and strategy optimization. Some load forecasting methods based on machine learning and deep learning have gradually been applied to load management systems. Through deep mining of big data and historical load data, some methods can achieve high-precision load demand forecasting. However, these methods still face several key challenges. First, load data has highly complex and nonlinear characteristics, and is affected by many factors such as environmental conditions, electricity price fluctuations, and user behavior. The existing load forecasting methods have limited effects in feature decoupling and dynamic modeling, and it is difficult to accurately separate short-term fluctuations and long-term trends. Secondly, although some systems have introduced deep learning models, their real-time and adaptability in dealing with complex load scenarios are still insufficient, and model parameters cannot be updated at any time, which makes it difficult for the system to flexibly adjust load strategies in situations where demand changes rapidly.

[0005] On the other hand, in terms of load management strategy generation and adjustment, traditional methods mostly rely on a single optimization algorithm and lack multi-dimensional decision support for strategy changes. For example, under conditions of large load fluctuations or frequent changes in electricity prices, existing technologies lack an effective strategy control mechanism and cannot dynamically optimize resource allocation according to the priority requirements of different load units. In addition, existing methods rarely introduce causal reasoning and feedback mechanisms, resulting in the system being unable to effectively identify external environmental interference during the execution of load strategies, and lacking means to eliminate external influences, which affects the accuracy and robustness of load distribution strategies. Traditional feedback mechanisms mainly rely on fixed thresholds or static feedback, which cannot adaptively adjust strategies and has great limitations on the system's long-term load forecasting and distribution optimization. When the system executes the load management strategy, it is difficult to continuously adaptively adjust and optimize the load response, which in turn affects the overall energy efficiency of the power system.

[0006] In terms of priority allocation of load management, existing technologies mostly adopt simple priority setting based on electricity price or load demand, ignoring the interactive effects of response time, electricity price fluctuations and demand fluctuations, resulting in difficulty in forming a dynamic allocation strategy that adapts to variable load demand in the actual load environment. The usual priority setting method uses load demand or real-time electricity price as the main criterion, and fails to make hierarchical decisions based on the multi-level demands of load characteristics. As a result, high-priority loads may not be fully guaranteed, while low-priority loads may waste resources in resource allocation.

[0007] Therefore, how to provide a power energy efficiency optimization method based on intelligent load management is an urgent problem that technicians in this field need to solve. Summary of the invention

[0008] One purpose of the present invention is to propose a method for optimizing power energy efficiency based on intelligent load management. The present invention adopts an intelligent load management method, and realizes accurate optimization of power energy efficiency through multi-layer feature decoupling and closed-loop optimization technology. The present invention first decomposes the load characteristics into short-term behavior characteristics and long-term trend characteristics through a dynamic feature decoupling module, thereby ensuring accurate prediction of load demand at different time scales; on this basis, a closed-loop feedback optimization mechanism is adopted to adjust the load distribution strategy in real time to improve the sensitivity and response speed to load fluctuations. The method of the present invention has the advantages of rapid response, high energy efficiency, and flexible distribution, and provides an effective solution for intelligent load management in complex power environments.

[0009] A method for optimizing power energy efficiency based on intelligent load management according to an embodiment of the present invention comprises the following steps:

[0010] S1, collect load data and perform data preprocessing to generate an initial load characteristic matrix;

[0011] S2, inputting the initial load feature matrix into a dynamic feature decoupling module driven by adaptive contrast learning, and decomposing the initial load feature matrix into a short-term behavior feature matrix and a long-term trend feature matrix through pseudo-symmetry decoupling;

[0012] S3, input the short-term behavior feature matrix and the long-term trend feature matrix into the multi-scale focused load forecasting model, perform multi-level load demand forecasting through the fusion structure of multi-scale convolution and recursive neural network, and output short-term load demand, medium-term load demand and long-term load demand;

[0013] S4. Through the cognitive gene sequence generation module, load characteristics, load priorities and historical strategies are encoded into cognitive gene sequences. When the environment or load demand changes significantly, gene mutation and recombination operations are triggered to generate adaptive load distribution strategies.

[0014] S5, the load distribution strategy is passed to the causal reasoning and self-supervisory feedback control module, the real characteristics of the load demand are identified based on causal reasoning, external environmental interference factors are eliminated, and the feedback data is compared with the prediction results. If the deviation exceeds the threshold, the self-learning mechanism is triggered;

[0015] S6. Through the hierarchical load decision network, load priorities are dynamically allocated according to real-time electricity prices, load demand fluctuations and response time, and a decision path for current load management is generated;

[0016] S7. Through a multi-layer feature feedback closed-loop optimization system, load execution data and environmental change data are collected in real time, the actual execution effect is compared with the predicted results, and adaptive optimization is performed based on the feedback results.

[0017] Optionally, the S2 specifically includes:

[0018] S21, inputting the preprocessed initial load feature matrix into a dynamic feature decoupling module driven by adaptive contrastive learning, wherein the dynamic feature decoupling module driven by adaptive contrastive learning performs normalization processing on the initial load feature matrix;

[0019] S22, use the pseudo-symmetric decoupling method to perform characteristic decomposition on the initial load characteristic matrix and establish the short-term behavior characteristic matrix M short and the long-term trend characteristic matrix M long Joint representation model of

[0020] S23. Decompose the initial load characteristic matrix:

[0021]

[0022] Among them, α1, α2 and α3 represent coefficient parameters, β1, β2 and β3 represent exponential parameters, λ represents time gain or attenuation coefficient, T represents the upper limit of time integration, ω represents the frequency parameter of short-term behavior characteristics, The frequency parameter representing the long-term trend characteristics, γ represents the positive offset;

[0023] S24, through the adaptive contrast learning mechanism, the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long The coupling effect is optimized and the loss function of contrastive learning is used to make the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long A balance between similarities and differences;

[0024] S25, in the contrast learning process, the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long After refinement, the short-term behavior feature matrix M short Including the high-frequency characteristics and periodic fluctuations of short-term load changes, the long-term trend characteristic matrix M long Includes long-term trends and low-frequency features;

[0025] S26. Output the final optimized short-term behavior feature matrix M short and the long-term trend characteristic matrix M long .

[0026] Optionally, the S3 specifically includes:

[0027] S31, the short-term behavior feature matrix M shortThe multi-scale convolution layer in the multi-scale focused load forecasting model is input to extract short-term features of different time scales through multi-scale convolution kernels, and small-scale convolution kernels are used to capture high-frequency components of rapid load fluctuations to generate a preliminary short-term load demand forecast matrix.

[0028] S32, long-term trend characteristic matrix M long Perform multi-scale convolution processing to extract the low-frequency features of long-term trends with large-scale convolution kernels in a multi-layer structure. By expanding the receptive field of the convolution kernel, the overall trend of long-term load changes is captured to generate a preliminary long-term load demand forecast matrix.

[0029] S33, passing the generated preliminary short-term load demand forecast matrix to the recursive neural network layer, capturing the time dependency of short-term characteristics, and recursively analyzing and gradually refining the time series characteristics of short-term load demand. Each layer of recursion is based on the short-term changes of the previous time step, and gradually optimizes the preliminary short-term load demand forecast matrix to generate a final short-term load demand matrix that adapts to demand fluctuations in a short period of time;

[0030] S34, performing recursive neural network processing on the generated preliminary long-term load demand forecast matrix, based on the periodic trend of the long-term load, extracting the steady change pattern of the load over a long time span through deep recursion, and generating the final long-term load demand matrix;

[0031] S35, generating a preliminary medium-term load demand forecast matrix, combining the final short-term load demand forecast matrix and the final long-term load demand matrix into a medium-term input matrix, wherein the preliminary medium-term load demand forecast matrix includes the dual characteristics of short-term fluctuations and long-term trends, applying a multi-scale convolution operation to the fused preliminary medium-term load demand forecast matrix, extracting the characteristics of medium-frequency load demand changes, and outputting a final medium-term load demand forecast matrix that adapts to periodic fluctuations and longer-term changes;

[0032] S36. Output the prediction results of the final short-term load demand matrix, the final long-term load demand matrix and the final medium-term load demand prediction matrix, and make comprehensive adjustments through a multi-level feature focusing mechanism. The multi-level feature focusing mechanism adaptively optimizes the prediction results according to the real-time volatility and historical trend data of the load demand.

[0033] Optionally, the S4 specifically includes:

[0034] S41, encoding the load characteristics, load priority and historical strategy data into multi-level cognitive gene sequences, each gene sequence includes a characteristic tag, a priority parameter, a historical adjustment strategy and a context dependency, and each level of the gene sequence is used for short-term load demand, medium-term load demand and long-term load demand respectively;

[0035] S42, constructing an intelligent load management gene library, storing multi-level cognitive gene sequences under different load scenarios in the intelligent load management gene library, dynamically adjusting the sorting rules and storage priorities of gene units according to environmental conditions, load demand change amplitude and load priority, and clustering the multi-level cognitive gene sequences to automatically identify high-frequency load change patterns;

[0036] S43, when a significant change in load demand or external environmental conditions is detected, the change type is identified and the conditional adaptive gene mutation operation is activated, and based on the load priority, mutation probability and context dependency, the strategy weight, feature tag and context dependency in the gene unit are adjusted to form a new mutation gene unit;

[0037] S44, performing a multi-level gene recombination operation, forming an adaptive load distribution strategy by fusing the new mutation gene unit with the different levels of gene units in the intelligent load management gene library, and automatically selecting different levels of strategy features from the gene units at each level during the gene recombination process to generate an adaptive load distribution strategy;

[0038] S45, applying the load distribution strategy to the current load management, based on the comparison between the real-time load response data and the target energy efficiency, if the strategy effect is lower than the preset threshold, re-triggering the gene mutation and recombination, and retaining the traces of the strategy effect in the intelligent load management gene library;

[0039] S46. Add the adaptively optimized efficient gene units to the core layer of the intelligent load management gene library, and update the structure of the intelligent load management gene library through the reinforcement learning mechanism. Weight the gene units according to the frequency of occurrence and adaptive effect of different load scenarios, eliminate low-priority or unused gene units, and dynamically optimize the content of the intelligent load management gene library.

[0040] Optionally, the S5 specifically includes:

[0041] S51, the load distribution strategy is passed to the causal reasoning module, a load demand causal graph is constructed, the main influencing characteristics of the load demand are identified, and interference factors of the external environment are eliminated. The causal reasoning module is based on the causal relationship between load characteristics, environmental parameters and load response, and extracts core features by analyzing causal effects;

[0042] S52. Calculate the causal contribution value of each influencing feature through causal effect analysis. The causal contribution value is used to quantify the actual impact of each feature on load demand:

[0043]

[0044] Among them, C i Represents feature Xi The causal contribution value, w i Represents feature X i The weight coefficient, w j Represents feature X j The weight coefficient, D represents the output variable of load demand, κ i Represents the control feature X i Modulation coefficient of demand impact, κ j Represents the control feature X j Modulation coefficient affecting demand, ω i Represents feature X i The frequency parameter, θ i Represents feature X i The phase shift, Represents feature X j The frequency parameter, φ j Represents feature X j The phase offset, T represents the upper limit of the time window, N represents the total number of features, and η represents the small value offset;

[0045] S53, the load distribution strategy is transmitted to the self-supervisory feedback control module, and the deviation between the load execution effect and the predicted result is monitored in real time. The self-supervisory feedback control module uses a self-supervised learning mechanism to analyze the potential patterns in the load response data and adaptively adjust the load distribution strategy according to the actual load execution effect;

[0046] S54, setting a deviation threshold, when the deviation between the load execution result and the prediction result exceeds the threshold, triggering a self-learning mechanism, the self-learning mechanism adjusting the parameters and causal structure of the load distribution strategy based on the deviation;

[0047] S55. Construct an adaptive deviation correction model, dynamically adjust the control parameters according to the deviation size and change trend, and calculate the deviation correction signal:

[0048]

[0049] Among them, U represents the deviation correction signal, E(t) represents the deviation signal between the load execution result and the prediction result, α k represents the exponent used for deviation amplitude adjustment, ω k and γ l represents the frequency parameter, δ k and θ l represents the phase shift, β k represents the adjustment coefficient, E avg represents the average value of the deviation signal, λ l represents the regulation index, μ lrepresents the adjustment coefficient, ζ represents the denominator offset, M represents the number of modulation items used to process the deviation signal, and L represents the number of adjustment items used to process the difference between the average value of the deviation signal and the current deviation signal;

[0050] S56. The load distribution strategy adjusted by feedback is re-transmitted to the causal reasoning module, and the contribution of each feature is evaluated based on the updated load demand causal graph. The strategy parameters that are not suitable for the current environment are eliminated or optimized, and the parameters of the causal reasoning and self-supervisory feedback control modules are continuously updated through a double-loop mechanism to achieve dynamic adaptive optimization of the load management strategy.

[0051] Optionally, the S6 specifically includes:

[0052] S61, constructing a hierarchical load decision network, passing the real-time electricity price, load demand fluctuation and response time as input features into the dynamic allocation layer of the network, preliminarily dividing the priority factors of the load units, and the hierarchical load decision network adjusts the response strategy of each load unit in real time according to the load demand change and electricity price fluctuation;

[0053] S62. In the dynamic allocation layer, the current values ​​of real-time electricity prices, load demand fluctuations, and response time are analyzed to generate load priority factors:

[0054]

[0055] Among them, P i represents the priority factor of load unit i, α demand represents the priority weight when adjusting for high demand fluctuations, β price Indicates the rate of change of electricity price response, C threshold represents the high priority threshold of electricity price, γ d represents the demand fluctuation weight under medium demand fluctuation conditions, δ price represents the electricity price smoothing coefficient, C median represents the median value of electricity price, θ e represents the response rate of medium electricity price fluctuation, ∈ time represents the weight of adjusting low priority load, η time represents the reduction factor of the response time of controlling low priority loads, C low represents the low priority value of the electricity price, represents the response intensity of low priority load electricity price fluctuation, T thr Represents the high priority threshold of response time, T max Indicates the maximum priority threshold of response time, D high Indicates the high fluctuation threshold of load demand fluctuation, D low Indicates the low fluctuation threshold of load demand fluctuation, C irepresents the real-time electricity price of load unit i, ΔD i represents the load demand fluctuation of load unit i, T i represents the response time of load unit i;

[0056] S63, classify the load units according to the priority factors, and divide the loads into high priority, medium priority and low priority; the high priority load units are loads that are sensitive to electricity price fluctuations or are given priority response when demand fluctuations are large; the medium priority load units are relatively important and flexibly adjusted loads; the low priority load units are non-critical loads with long response times and small load fluctuations;

[0057] S64, generating a dynamic load management path in the hierarchical load decision network, giving priority to responding to high-priority load units, processing medium-priority load units when the electricity price is medium or the demand is stable, and arranging low-priority load units to be executed during low-price periods;

[0058] S65. Monitor the execution of the load management path. If it is detected during the execution of the path that the real-time electricity price, load demand fluctuation or response time change exceeds a preset threshold, recalculate the priority factor of each load unit and dynamically adjust the load management path.

[0059] The beneficial effects of the present invention are:

[0060] First, the present invention decomposes the load data into short-term behavior characteristics and long-term trend characteristics through a dynamic feature decoupling module driven by adaptive contrast learning, and can accurately extract short-term fluctuations and long-term changes in load demand. This feature decoupling method enables the load management system to effectively separate the different time characteristics of the load in a complex load fluctuation environment, thereby providing a more accurate data basis for multi-scale load forecasting. Combined with a multi-scale focused load forecasting model, the present invention can perform hierarchical forecasts on the high-frequency changes in short-term load demand, the medium-frequency fluctuations in medium-term demand, and the long-term load trends, effectively avoiding the limitation that traditional models are difficult to accurately capture load fluctuations on a single time scale.

[0061] The present invention enables the load management system to have efficient adaptability through the cognitive gene sequence generation module and the causal reasoning and self-supervisory feedback control module. When the load demand changes or the environmental conditions fluctuate significantly, the system can quickly generate a load distribution strategy that adapts to the current situation through the cognitive gene mutation and recombination mechanism, which significantly improves the response speed of the system. The causal reasoning module can identify the main influencing features in the load demand, eliminate environmental interference factors, and ensure the accuracy and robustness of the load strategy. At the same time, the self-supervisory feedback control module realizes the closed-loop optimization of the system by real-time monitoring of the deviation between the execution effect and the prediction result, so that the load management strategy can continuously self-adjust and improve in actual execution, thereby ensuring the efficiency and dynamic adaptability of load distribution.

[0062] In addition, the hierarchical load decision network proposed in the present invention dynamically allocates load priorities through factors such as real-time electricity prices, load demand fluctuations and response time, thereby effectively solving the problem of single priority setting in the prior art. The system dynamically adjusts the priority factor according to different load characteristics through a piecewise function formula to form a hierarchical response mechanism with three levels of high, medium and low priorities. High-priority load units are given priority response in emergency situations, while medium and low priority loads are arranged for execution according to real-time electricity prices and demand fluctuations, respectively, which reduces the system operating costs while ensuring the reasonable allocation of load resources. Through this multi-dimensional decision network, the present invention achieves a flexible response to load demand. Even under conditions of complex load fluctuations, the system can still maintain stable operation and achieve energy efficiency optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0064] Figure 1 A flow chart of a method for optimizing power energy efficiency based on intelligent load management proposed by the present invention;

[0065] Figure 2 This is a schematic diagram of the structure of a multi-scale focused load prediction model for an electric power energy efficiency optimization method based on intelligent load management proposed in the present invention. DETAILED DESCRIPTION

[0066] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0067] refer to Figure 1 and Figure 2 , a method for optimizing power energy efficiency based on intelligent load management, comprising the following steps:

[0068] S1, collect load data and perform data preprocessing to generate an initial load characteristic matrix;

[0069] S2, inputting the initial load feature matrix into a dynamic feature decoupling module driven by adaptive contrast learning, and decomposing the initial load feature matrix into a short-term behavior feature matrix and a long-term trend feature matrix through pseudo-symmetry decoupling;

[0070] S3, input the short-term behavior feature matrix and the long-term trend feature matrix into the multi-scale focused load forecasting model, perform multi-level load demand forecasting through the fusion structure of multi-scale convolution and recursive neural network, and output short-term load demand, medium-term load demand and long-term load demand;

[0071] S4. Through the cognitive gene sequence generation module, load characteristics, load priorities and historical strategies are encoded into cognitive gene sequences. When the environment or load demand changes significantly, gene mutation and recombination operations are triggered to generate adaptive load distribution strategies.

[0072] S5, the load distribution strategy is passed to the causal reasoning and self-supervisory feedback control module, the real characteristics of the load demand are identified based on causal reasoning, external environmental interference factors are eliminated, and the feedback data is compared with the prediction results. If the deviation exceeds the threshold, the self-learning mechanism is triggered;

[0073] S6. Through the hierarchical load decision network, load priorities are dynamically allocated according to real-time electricity prices, load demand fluctuations and response time, and a decision path for current load management is generated;

[0074] S7. Through a multi-layer feature feedback closed-loop optimization system, load execution data and environmental change data are collected in real time, the actual execution effect is compared with the predicted results, and adaptive optimization is performed based on the feedback results.

[0075] In this implementation, S2 specifically includes:

[0076] S21, inputting the preprocessed initial load feature matrix into a dynamic feature decoupling module driven by adaptive contrastive learning, wherein the dynamic feature decoupling module driven by adaptive contrastive learning performs normalization processing on the initial load feature matrix;

[0077] S22, use the pseudo-symmetric decoupling method to perform characteristic decomposition on the initial load characteristic matrix and establish the short-term behavior characteristic matrix M short and the long-term trend characteristic matrix M long Joint representation model of

[0078] S23. Decompose the initial load characteristic matrix:

[0079]

[0080] Among them, α1, α2 and α3 represent coefficient parameters, β1, β2 and β3 represent exponential parameters, λ represents time gain or attenuation coefficient, T represents the upper limit of time integration, ω represents the frequency parameter of short-term behavior characteristics, The frequency parameter representing the long-term trend characteristics, γ represents the positive offset;

[0081] S24, through the adaptive contrast learning mechanism, the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long The coupling effect is optimized and the loss function of contrastive learning is used to make the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long A balance between similarities and differences;

[0082] S25, in the contrast learning process, the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long After refinement, the short-term behavior feature matrix M short Including the high-frequency characteristics and periodic fluctuations of short-term load changes, the long-term trend characteristic matrix M long Includes long-term trends and low-frequency features;

[0083] S26. Output the final optimized short-term behavior feature matrix M short and the long-term trend characteristic matrix M long .

[0084] In this implementation, S3 specifically includes:

[0085] S31, the short-term behavior feature matrix M short The multi-scale convolution layer in the multi-scale focused load forecasting model is input to extract short-term features of different time scales through multi-scale convolution kernels, and small-scale convolution kernels are used to capture high-frequency components of rapid load fluctuations to generate a preliminary short-term load demand forecast matrix.

[0086] S32, long-term trend characteristic matrix M long Perform multi-scale convolution processing to extract the low-frequency features of long-term trends with large-scale convolution kernels in a multi-layer structure. By expanding the receptive field of the convolution kernel, the overall trend of long-term load changes is captured to generate a preliminary long-term load demand forecast matrix.

[0087] S33, passing the generated preliminary short-term load demand forecast matrix to the recursive neural network layer, capturing the time dependency of short-term characteristics, and recursively analyzing and gradually refining the time series characteristics of short-term load demand. Each layer of recursion is based on the short-term changes of the previous time step, and gradually optimizes the preliminary short-term load demand forecast matrix to generate a final short-term load demand matrix that adapts to demand fluctuations in a short period of time;

[0088] S34, performing recursive neural network processing on the generated preliminary long-term load demand forecast matrix, based on the periodic trend of the long-term load, extracting the steady change pattern of the load over a long time span through deep recursion, and generating the final long-term load demand matrix;

[0089] S35, generating a preliminary medium-term load demand forecast matrix, combining the final short-term load demand forecast matrix and the final long-term load demand matrix into a medium-term input matrix, wherein the preliminary medium-term load demand forecast matrix includes the dual characteristics of short-term fluctuations and long-term trends, applying a multi-scale convolution operation to the fused preliminary medium-term load demand forecast matrix, extracting the characteristics of medium-frequency load demand changes, and outputting a final medium-term load demand forecast matrix that adapts to periodic fluctuations and longer-term changes;

[0090] S36. Output the prediction results of the final short-term load demand matrix, the final long-term load demand matrix and the final medium-term load demand prediction matrix, and make comprehensive adjustments through a multi-level feature focusing mechanism. The multi-level feature focusing mechanism adaptively optimizes the prediction results according to the real-time volatility and historical trend data of the load demand.

[0091] In this implementation, S4 specifically includes:

[0092] S41, encoding the load characteristics, load priority and historical strategy data into multi-level cognitive gene sequences, each gene sequence includes a characteristic tag, a priority parameter, a historical adjustment strategy and a context dependency, and each level of the gene sequence is used for short-term load demand, medium-term load demand and long-term load demand respectively;

[0093] S42, constructing an intelligent load management gene library, storing multi-level cognitive gene sequences under different load scenarios in the intelligent load management gene library, dynamically adjusting the sorting rules and storage priorities of gene units according to environmental conditions, load demand change amplitude and load priority, and clustering the multi-level cognitive gene sequences to automatically identify high-frequency load change patterns;

[0094] S43, when a significant change in load demand or external environmental conditions is detected, the change type is identified and the conditional adaptive gene mutation operation is activated, and based on the load priority, mutation probability and context dependency, the strategy weight, feature tag and context dependency in the gene unit are adjusted to form a new mutation gene unit;

[0095] S44, performing a multi-level gene recombination operation, forming an adaptive load distribution strategy by fusing the new mutation gene unit with the different levels of gene units in the intelligent load management gene library, and automatically selecting different levels of strategy features from the gene units at each level during the gene recombination process to generate an adaptive load distribution strategy;

[0096] S45, applying the load distribution strategy to the current load management, based on the comparison between the real-time load response data and the target energy efficiency, if the strategy effect is lower than the preset threshold, re-triggering the gene mutation and recombination, and retaining the traces of the strategy effect in the intelligent load management gene library;

[0097] S46. Add the adaptively optimized efficient gene units to the core layer of the intelligent load management gene library, and update the structure of the intelligent load management gene library through the reinforcement learning mechanism. Weight the gene units according to the frequency of occurrence and adaptive effect of different load scenarios, eliminate low-priority or unused gene units, and dynamically optimize the content of the intelligent load management gene library.

[0098] In this implementation manner, S5 specifically includes:

[0099] S51, the load distribution strategy is passed to the causal reasoning module, a load demand causal graph is constructed, the main influencing characteristics of the load demand are identified, and interference factors of the external environment are eliminated. The causal reasoning module is based on the causal relationship between load characteristics, environmental parameters and load response, and extracts core features by analyzing causal effects;

[0100] S52. Calculate the causal contribution value of each influencing feature through causal effect analysis. The causal contribution value is used to quantify the actual impact of each feature on load demand:

[0101]

[0102] Among them, C i Represents feature X i The causal contribution value, w i Represents feature X i The weight coefficient, w j Represents feature X j The weight coefficient, D represents the output variable of load demand, κ i Represents the control feature X i Modulation coefficient of demand impact, κ jRepresents the control feature X j Modulation coefficient affecting demand, ω i Represents feature X i The frequency parameter, θ i Represents feature X i The phase shift, Represents feature X j The frequency parameter, φ j Represents feature X j The phase offset, T represents the upper limit of the time window, N represents the total number of features, and η represents the small value offset;

[0103] S53, the load distribution strategy is transmitted to the self-supervisory feedback control module, and the deviation between the load execution effect and the predicted result is monitored in real time. The self-supervisory feedback control module uses a self-supervised learning mechanism to analyze the potential patterns in the load response data and adaptively adjust the load distribution strategy according to the actual load execution effect;

[0104] S54, setting a deviation threshold, when the deviation between the load execution result and the prediction result exceeds the threshold, triggering a self-learning mechanism, the self-learning mechanism adjusting the parameters and causal structure of the load distribution strategy based on the deviation;

[0105] S55. Construct an adaptive deviation correction model, dynamically adjust the control parameters according to the deviation size and change trend, and calculate the deviation correction signal:

[0106]

[0107] Among them, U represents the deviation correction signal, E(t) represents the deviation signal between the load execution result and the prediction result, α k represents the exponent used for deviation amplitude adjustment, ω k and γ l represents the frequency parameter, δ k and θ l represents the phase shift, β k represents the adjustment coefficient, E avg represents the average value of the deviation signal, λ l represents the regulation index, μ l represents the adjustment coefficient, ζ represents the denominator offset, M represents the number of modulation items used to process the deviation signal, and L represents the number of adjustment items used to process the difference between the average value of the deviation signal and the current deviation signal;

[0108] S56. The load distribution strategy adjusted by feedback is re-transmitted to the causal reasoning module, and the contribution of each feature is evaluated based on the updated load demand causal graph. The strategy parameters that are not suitable for the current environment are eliminated or optimized, and the parameters of the causal reasoning and self-supervisory feedback control modules are continuously updated through a double-loop mechanism to achieve dynamic adaptive optimization of the load management strategy.

[0109] In this implementation manner, S6 specifically includes:

[0110] S61, constructing a hierarchical load decision network, passing the real-time electricity price, load demand fluctuation and response time as input features into the dynamic allocation layer of the network, preliminarily dividing the priority factors of the load units, and the hierarchical load decision network adjusts the response strategy of each load unit in real time according to the load demand change and electricity price fluctuation;

[0111] S62. In the dynamic allocation layer, the current values ​​of real-time electricity prices, load demand fluctuations, and response time are analyzed to generate load priority factors:

[0112]

[0113] Among them, P i represents the priority factor of load unit i, α demand represents the priority weight when adjusting for high demand fluctuations, β price Indicates the rate of change of electricity price response, C threshold represents the high priority threshold of electricity price, γ d represents the demand fluctuation weight under medium demand fluctuation conditions, δ price represents the electricity price smoothing coefficient, C median represents the median value of electricity price, θ e represents the response rate of medium electricity price fluctuation, ∈ time represents the weight of adjusting low priority load, η time represents the reduction factor of the response time of controlling low priority loads, C low represents the low priority value of the electricity price, represents the response intensity of low priority load electricity price fluctuation, T thr Represents the high priority threshold of response time, T max Indicates the maximum priority threshold of response time, D high Indicates the high fluctuation threshold of load demand fluctuation, D low Indicates the low fluctuation threshold of load demand fluctuation, C i represents the real-time electricity price of load unit i, ΔD i represents the load demand fluctuation of load unit i, T i represents the response time of load unit i;

[0114] S63, classify the load units according to the priority factors, and divide the loads into high priority, medium priority and low priority; the high priority load units are loads that are sensitive to electricity price fluctuations or are given priority response when demand fluctuations are large; the medium priority load units are relatively important and flexibly adjusted loads; the low priority load units are non-critical loads with long response times and small load fluctuations;

[0115] S64, generating a dynamic load management path in the hierarchical load decision network, giving priority to responding to high-priority load units, processing medium-priority load units when the electricity price is medium or the demand is stable, and arranging low-priority load units to be executed during low-price periods;

[0116] S65. Monitor the execution of the load management path. If it is detected during the execution of the path that the real-time electricity price, load demand fluctuation or response time change exceeds a preset threshold, recalculate the priority factor of each load unit and dynamically adjust the load management path.

[0117] Embodiment 1:

[0118] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the power management of a large commercial park. Since the power demand fluctuates greatly with time and weather changes, the instability of the power load becomes a major problem. Especially during peak power consumption periods and extreme climate conditions, the power load fluctuates greatly, posing a challenge to the stability and energy efficiency management of the power system. Traditional load management methods cannot flexibly respond to complex load fluctuations, and usually only use historical data and static rules for simple prediction and allocation, resulting in unreasonable allocation of power resources and low energy efficiency under peak load. In order to solve this problem, this embodiment introduces a power energy efficiency optimization method based on intelligent load management. By applying multi-layer feature decoupling, multi-scale focused load prediction models, cognitive gene sequence generation, and hierarchical load decision networks, accurate prediction and intelligent allocation optimization of load demand in the large commercial park are achieved.

[0119] In this embodiment, we decompose the power load demand in the park into short-term behavior characteristics and long-term trend characteristics, and input these characteristics into the multi-scale focused load forecasting model. The load characteristic decoupling module generates a short-term behavior characteristic matrix and a long-term trend characteristic matrix according to the short-term behavior and long-term changes of load fluctuations, respectively, to ensure that the load management system can accurately predict the load demand under the conditions of intense short-term load fluctuations and slow long-term trend changes. After the data preprocessing is completed, the short-term and long-term characteristic matrices are respectively passed into the multi-scale focused load forecasting model, and the high-frequency and low-frequency load characteristics are captured by the multi-scale convolution layer, and the time-dependent features are gradually extracted using the recursive neural network layer to output short-term, medium-term and long-term load demand forecasts.

[0120] On this basis, the intelligent load management system adopts the cognitive gene sequence generation module, which encodes the priority, feature mark and historical strategy of the load unit through cognitive genes to adapt to the real-time changes in load demand in the park. In the actual operation process, when the load demand or environmental conditions change significantly, the system triggers the cognitive gene mutation and recombination mechanism, and quickly generates an adaptive load distribution strategy according to the current environment to ensure that the system load is still distributed smoothly when the demand changes suddenly.

[0121] To verify the effectiveness of the system, we conducted a field test of load management during a four-month winter and spring transition period, and collected and compared the load distribution and energy efficiency performance of the system and the traditional load management system in the park. The experimental results show that the response speed of the intelligent load management system during peak hours is significantly improved, and the energy efficiency is improved by about 13.8%. For example, during the typical peak load period in December, the total load demand of the park once reached 1,650 kilowatts. Under the precise prediction and optimized distribution of the intelligent load management system, the distribution deviation of the peak load was kept within 3%, while the distribution deviation of the traditional system reached 12%, which is significantly lower than the system of the present invention.

[0122] Through the hierarchical load decision network, we further dynamically prioritize the loads. Especially in the case of real-time electricity price fluctuations and surges in load demand, the system responds hierarchically according to the priorities of different load units. On the premise of ensuring the stable supply of high-priority loads, the execution period of low-priority loads is adjusted to execute non-critical loads during the low electricity price period, thereby saving electricity costs. For example, on a certain day in mid-January, the load demand of the heating equipment in the park surged to 1,200 kilowatts due to low temperature weather, of which the high-priority heating load reached 700 kilowatts. The system prioritized sufficient electricity according to the high-priority load demand, and postponed some non-critical loads such as landscape lighting and non-emergency water supply to the low electricity price period, achieving a smooth optimization of energy efficiency. Compared with the traditional system, the system of the present invention is more balanced in the distribution of peak loads and valley loads, which improves the overall power utilization efficiency of the park by about 15%.

[0123] During the feedback and adjustment phase, the system monitors the deviation between the load forecast results and the actual load demand in real time through the self-supervision feedback control module. When the deviation is detected to exceed the set threshold, the self-learning mechanism is automatically triggered to update the parameters of the forecast model. For example, on a certain day in March, the medium-term load demand forecast of the park was 850 kilowatts, while the actual load demand reached 910 kilowatts, with a deviation of about 7%. After the system recognized that the load forecast deviation exceeded the threshold, it automatically optimized and adjusted the weight parameters of the short-term and medium-term load forecast models, which increased the subsequent load forecast accuracy by about 8%, greatly reducing the waste of power resources.

[0124] The intelligent load management method of the present invention has shown significant superiority in practical applications in the park. Through the synergy of modules such as multi-layer feature decoupling, multi-scale load prediction and dynamic priority allocation, the system responds quickly during peak loads, ensuring the stable supply of high-priority loads, while reasonably allocating non-critical loads during valley loads, optimizing the power utilization efficiency of the park. Especially for periods of seasonal changes and large changes in load demand, the energy efficiency of the system is particularly improved, providing an effective solution for power management in the park.

[0125] Table 1 Comparison of operating data between traditional system and intelligent load management system

[0126]

[0127] Table 2 Load demand and allocation priority data table

[0128]

[0129] In this experiment, the data in Table 1 above revealed significant differences in load distribution and energy efficiency between the intelligent load management system and the traditional system. The traditional system has a large deviation in peak load management, ranging from 10% to 14%, resulting in unstable energy efficiency performance; in contrast, the load deviation of the intelligent load management system during the same peak period is significantly reduced, remaining between 3% and 4%, with an average deviation of only 3.325%. From the perspective of energy efficiency indicators, the average energy efficiency of the traditional system is 84.25kWh / kilowatt-hour, while the intelligent system reaches 95.25kWh / kilowatt-hour, with an overall energy efficiency improvement of 13.2%. This difference shows that the intelligent load management system performs well in real-time load regulation and resource optimization, effectively reducing the deviation rate of peak load management and improving system energy efficiency.

[0130] Further analysis of the load demand and allocation priority data in Table 2 shows that the intelligent load management system has obvious advantages in ensuring high-priority load demand. Taking mid-December as an example, the high-priority load demand detected by the system was 700kW, and this priority load was fully met in the actual allocation; the medium and low priority loads were also relatively close to the demand, with only slight deviations. This refined load allocation optimization effect was repeatedly verified in the experimental data in January, February and March. The intelligent load management system arranges non-critical loads to execute during periods with lower electricity prices based on real-time electricity prices and load fluctuations, minimizing the load pressure during peak hours. Overall, the load priority dynamic allocation strategy of the intelligent system significantly improves the utilization efficiency of power resources, can flexibly allocate loads of various priorities, and realizes the stable operation of the system under different load demand conditions.

[0131] It can be seen that the data in Tables 1 and 2 further verify the superior performance of the intelligent load management system in a complex power environment. The system can effectively reduce peak load deviation, improve system energy efficiency, and ensure the needs of high-priority loads in load distribution, and meet the execution arrangements of medium and low-priority loads in different electricity price periods. This hierarchical load priority decision-making method enables the system to show excellent flexibility and stability in practical applications, providing strong technical support for efficient power energy efficiency optimization.

[0132] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing power energy efficiency based on intelligent load management, characterized in that: The steps include: S1, collect load data and perform data preprocessing to generate an initial load characteristic matrix; S2, inputting the initial load feature matrix into a dynamic feature decoupling module driven by adaptive contrast learning, and decomposing the initial load feature matrix into a short-term behavior feature matrix and a long-term trend feature matrix through pseudo-symmetry decoupling; S3, input the short-term behavior feature matrix and the long-term trend feature matrix into the multi-scale focused load forecasting model, perform multi-level load demand forecasting through the fusion structure of multi-scale convolution and recursive neural network, and output short-term load demand, medium-term load demand and long-term load demand; S4. Through the cognitive gene sequence generation module, load characteristics, load priorities and historical strategies are encoded into cognitive gene sequences. When the environment or load demand changes significantly, gene mutation and recombination operations are triggered to generate adaptive load distribution strategies. S5, the load distribution strategy is passed to the causal reasoning and self-supervisory feedback control module, the real characteristics of the load demand are identified based on causal reasoning, external environmental interference factors are eliminated, and the feedback data is compared with the prediction results. If the deviation exceeds the threshold, the self-learning mechanism is triggered; S6. Through the hierarchical load decision network, load priorities are dynamically allocated according to real-time electricity prices, load demand fluctuations and response time, and a decision path for current load management is generated; S7. Through a multi-layer feature feedback closed-loop optimization system, load execution data and environmental change data are collected in real time, the actual execution effect is compared with the predicted results, and adaptive optimization is performed based on the feedback results.

2. The method for optimizing power energy efficiency based on intelligent load management according to claim 1, characterized in that: The S2 specifically includes: S21, inputting the preprocessed initial load feature matrix into a dynamic feature decoupling module driven by adaptive contrastive learning, wherein the dynamic feature decoupling module driven by adaptive contrastive learning performs normalization processing on the initial load feature matrix; S22, use the pseudo-symmetric decoupling method to perform characteristic decomposition on the initial load characteristic matrix and establish the short-term behavior characteristic matrix M short and the long-term trend characteristic matrix M long Joint representation model of S23. Decompose the initial load characteristic matrix: Among them, α1, α2 and α3 represent coefficient parameters, β1, β2 and β3 represent exponential parameters, λ represents time gain or attenuation coefficient, T represents the upper limit of time integration, ω represents the frequency parameter of short-term behavior characteristics, φ represents the frequency parameter of long-term trend characteristics, and γ represents the positive offset; S24, through the adaptive contrast learning mechanism, the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long The coupling effect is optimized and the loss function of contrastive learning is used to make the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long A balance between similarities and differences; S25, in the contrast learning process, the short-term behavior feature matrix M short and the long-term trend characteristic matrix M long After refinement, the short-term behavior feature matrix M short Including the high-frequency characteristics and periodic fluctuations of short-term load changes, the long-term trend characteristic matrix M long Includes long-term trends and low-frequency features; S26. Output the final optimized short-term behavior feature matrix M short and the long-term trend characteristic matrix M long .

3. The method for optimizing power energy efficiency based on intelligent load management according to claim 1, characterized in that: The S3 specifically includes: S31, the short-term behavior feature matrix M short The multi-scale convolution layer in the multi-scale focused load forecasting model is input to extract short-term features of different time scales through multi-scale convolution kernels, and small-scale convolution kernels are used to capture high-frequency components of rapid load fluctuations to generate a preliminary short-term load demand forecast matrix. S32, long-term trend characteristic matrix M long Perform multi-scale convolution processing to extract the low-frequency features of long-term trends with large-scale convolution kernels in a multi-layer structure. By expanding the receptive field of the convolution kernel, the overall trend of long-term load changes is captured to generate a preliminary long-term load demand forecast matrix. S33, passing the generated preliminary short-term load demand forecast matrix to the recursive neural network layer, capturing the time dependency of short-term characteristics, and recursively analyzing and gradually refining the time series characteristics of short-term load demand. Each layer of recursion is based on the short-term changes of the previous time step, and gradually optimizes the preliminary short-term load demand forecast matrix to generate a final short-term load demand matrix that adapts to demand fluctuations in a short period of time; S34, performing recursive neural network processing on the generated preliminary long-term load demand forecast matrix, based on the periodic trend of the long-term load, extracting the steady change pattern of the load over a long time span through deep recursion, and generating the final long-term load demand matrix; S35, generating a preliminary medium-term load demand forecast matrix, combining the final short-term load demand forecast matrix and the final long-term load demand matrix into a medium-term input matrix, wherein the preliminary medium-term load demand forecast matrix includes the dual characteristics of short-term fluctuations and long-term trends, applying a multi-scale convolution operation to the fused preliminary medium-term load demand forecast matrix, extracting the characteristics of medium-frequency load demand changes, and outputting a final medium-term load demand forecast matrix that adapts to periodic fluctuations and longer-term changes; S36. Output the prediction results of the final short-term load demand matrix, the final long-term load demand matrix and the final medium-term load demand prediction matrix, and make comprehensive adjustments through a multi-level feature focusing mechanism. The multi-level feature focusing mechanism adaptively optimizes the prediction results according to the real-time volatility and historical trend data of the load demand.

4. The method for optimizing power energy efficiency based on intelligent load management according to claim 1, characterized in that: The S4 specifically includes: S41, encoding the load characteristics, load priority and historical strategy data into multi-level cognitive gene sequences, each gene sequence includes a characteristic tag, a priority parameter, a historical adjustment strategy and a context dependency, and each level of the gene sequence is used for short-term load demand, medium-term load demand and long-term load demand respectively; S42, constructing an intelligent load management gene library, storing multi-level cognitive gene sequences under different load scenarios in the intelligent load management gene library, dynamically adjusting the sorting rules and storage priorities of gene units according to environmental conditions, load demand change amplitude and load priority, and clustering the multi-level cognitive gene sequences to automatically identify high-frequency load change patterns; S43, when a significant change in load demand or external environmental conditions is detected, the change type is identified and the conditional adaptive gene mutation operation is activated, and based on the load priority, mutation probability and context dependency, the strategy weight, feature tag and context dependency in the gene unit are adjusted to form a new mutation gene unit; S44, performing a multi-level gene recombination operation, forming an adaptive load distribution strategy by fusing the new mutation gene unit with the different levels of gene units in the intelligent load management gene library, and automatically selecting different levels of strategy features from the gene units at each level during the gene recombination process to generate an adaptive load distribution strategy; S45, applying the load distribution strategy to the current load management, based on the comparison between the real-time load response data and the target energy efficiency, if the strategy effect is lower than the preset threshold, re-triggering the gene mutation and recombination, and retaining the traces of the strategy effect in the intelligent load management gene library; S46. Add the adaptively optimized efficient gene units to the core layer of the intelligent load management gene library, and update the structure of the intelligent load management gene library through the reinforcement learning mechanism. Weight the gene units according to the frequency of occurrence and adaptive effect of different load scenarios, eliminate low-priority or unused gene units, and dynamically optimize the content of the intelligent load management gene library.

5. The method for optimizing power energy efficiency based on intelligent load management according to claim 1, characterized in that: The S5 specifically includes: S51, the load distribution strategy is passed to the causal reasoning module, a load demand causal graph is constructed, the main influencing characteristics of the load demand are identified, and interference factors of the external environment are eliminated. The causal reasoning module is based on the causal relationship between load characteristics, environmental parameters and load response, and extracts core features by analyzing causal effects; S52. Calculate the causal contribution value of each influencing feature through causal effect analysis. The causal contribution value is used to quantify the actual impact of each feature on load demand: Among them, C i Represents feature X i The causal contribution value, w i Represents feature X i The weight coefficient, w j Represents feature X j The weight coefficient, D represents the output variable of load demand, κ i Represents the control feature X i Modulation coefficient of demand impact, κ j Represents the control feature X j Modulation coefficient affecting demand, ω i Represents feature X i The frequency parameter, θ i Represents feature X i The phase shift, Represents feature X j The frequency parameter, φ j Represents feature X j The phase offset, T represents the upper limit of the time window, N represents the total number of features, and η represents the small value offset; S53, the load distribution strategy is transmitted to the self-supervisory feedback control module, and the deviation between the load execution effect and the predicted result is monitored in real time. The self-supervisory feedback control module uses a self-supervised learning mechanism to analyze the potential patterns in the load response data and adaptively adjust the load distribution strategy according to the actual load execution effect; S54, setting a deviation threshold, when the deviation between the load execution result and the prediction result exceeds the threshold, triggering a self-learning mechanism, the self-learning mechanism adjusting the parameters and causal structure of the load distribution strategy based on the deviation; S55. Construct an adaptive deviation correction model, dynamically adjust the control parameters according to the deviation size and change trend, and calculate the deviation correction signal: Among them, U represents the deviation correction signal, E(t) represents the deviation signal between the load execution result and the prediction result, α k represents the exponent used for deviation amplitude adjustment, ω k and γ l represents the frequency parameter, δ k and θ l represents the phase shift, β k represents the adjustment coefficient, E avg represents the average value of the deviation signal, λ l represents the regulation index, μ l represents the adjustment coefficient, ζ represents the denominator offset, M represents the number of modulation items used to process the deviation signal, and L represents the number of adjustment items used to process the difference between the average value of the deviation signal and the current deviation signal; S56. The load distribution strategy adjusted by feedback is re-transmitted to the causal reasoning module, and the contribution of each feature is evaluated based on the updated load demand causal graph. The strategy parameters that are not suitable for the current environment are eliminated or optimized, and the parameters of the causal reasoning and self-supervisory feedback control modules are continuously updated through a double-loop mechanism to achieve dynamic adaptive optimization of the load management strategy.

6. The method for optimizing power energy efficiency based on intelligent load management according to claim 1, characterized in that: The S6 specifically includes: S61, constructing a hierarchical load decision network, passing the real-time electricity price, load demand fluctuation and response time as input features into the dynamic allocation layer of the network, preliminarily dividing the priority factors of the load units, and the hierarchical load decision network adjusts the response strategy of each load unit in real time according to the load demand change and electricity price fluctuation; S62. In the dynamic allocation layer, the current values ​​of real-time electricity prices, load demand fluctuations, and response time are analyzed to generate load priority factors: Among them, P i represents the priority factor of load unit i, α demand represents the priority weight when adjusting for high demand fluctuations, β price Indicates the rate of change of electricity price response, C threshold represents the high priority threshold of electricity price, γ d represents the demand fluctuation weight under medium demand fluctuation conditions, δ price represents the electricity price smoothing coefficient, C median represents the median value of electricity price, θ e represents the response rate of medium electricity price fluctuation, ∈ time represents the weight of adjusting low priority load, η time represents the reduction factor of the response time of controlling low priority loads, C low represents the low priority value of the electricity price, represents the response intensity of low priority load electricity price fluctuation, T thr Represents the high priority threshold of response time, T max Indicates the maximum priority threshold of response time, D high Indicates the high fluctuation threshold of load demand fluctuation, D low Indicates the low fluctuation threshold of load demand fluctuation, C i represents the real-time electricity price of load unit i, ΔD i represents the load demand fluctuation of load unit i, T i represents the response time of load unit i; S63, classifying the load units according to the priority factors, and dividing the loads into high priority, medium priority and low priority; S64, generating a dynamic load management path in the hierarchical load decision network, giving priority to responding to high-priority load units, processing medium-priority load units when the electricity price is medium or the demand is stable, and arranging low-priority load units to be executed during low-price periods; S65. Monitor the execution of the load management path. If it is detected during the execution of the path that the real-time electricity price, load demand fluctuation or response time change exceeds a preset threshold, recalculate the priority factor of each load unit and dynamically adjust the load management path.

Images