Industrial load zoning and layering interactive quota dynamic pricing method and system

Abstract

The invention provides an industrial load zoning and layering interactive quota dynamic pricing method and system. The method comprises the following steps: constructing a zoning and layering management model; defining spatio-temporal features of the nodes, extracting the spatio-temporal features from the partition and layering management model, and enhancing the spatio-temporal features; generating a refined load elastic portrait used for quantifying the load response capability according to the enhanced spatial-temporal characteristics, checking the tradable interaction quota of each level node according to the load elastic portrait and a partition and layering management model, and establishing a dynamic pricing model linked with market conditions and system values; and according to the checked quota and the dynamic price, performing transfer transaction matching and settlement of the interactive quota on the transaction platform, performing settlement and excitation according to an actual response effect, and performing closed-loop rolling optimization on parameters of the dynamic pricing model. According to the invention, the ability of industrial users to participate in demand response and electric power spot market can be significantly improved.

Description

Technical Field

[0001] This invention relates to the field of load pricing technology, and in particular to a dynamic pricing method and system for industrial load zoning and hierarchical interactive quotas. Background Technology

[0002] As the nation accelerates the construction of a new energy system, electricity load management, as a crucial support for ensuring energy security, is becoming increasingly important. Particularly in the industrial sector, fully exploring and meticulously managing adjustable load resources on the customer side plays a key role in enhancing the grid's supply capacity and promoting the consumption of renewable energy. Against the backdrop of deepening electricity market reforms, especially the gradual establishment of the electricity spot market, the possibility of load resources participating in market transactions has been made available.

[0003] However, there is a disconnect between existing load response models and market mechanisms. On the one hand, traditional load characteristic analysis methods struggle to accurately depict the spatiotemporal correlation characteristics and dynamic response capabilities of loads in complex industrial processes, failing to support the refined management requirements of "zoning and stratification." On the other hand, existing market trading mechanisms are mostly one-way incentives, lacking effective means to accurately quantify, value-assess, and dynamically price adjustable load resources within a multi-level architecture of "region-bus-enterprise-production line-equipment," resulting in insufficient release of the potential for industrial users to participate in demand response and inadequate interaction capabilities. Technically, while ordinary deep learning models can process time-series data, they struggle to effectively depict the spatial topological dependencies of loads and cannot adaptively focus on the critical loads and critical periods that have the greatest impact on system balance and market value, leading to insufficient model accuracy and interpretability, failing to meet the refined load profiling requirements of zoned and stratified interactive quota trading. Therefore, there is an urgent need for a new method that deeply integrates grid characteristics and market mechanisms and employs advanced artificial intelligence technology to achieve refined zoned and stratified management of industrial loads and efficient market-based interaction. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic pricing method and system for industrial load zoning and hierarchical interactive quotas, aiming to solve the shortcomings of existing industrial load management in terms of refined zoning and hierarchical control, in-depth analysis of load characteristics, and coordinated interaction with the electricity spot market.

[0005] In a first aspect, the present invention provides a dynamic pricing method for industrial load zoning and hierarchical interactive quotas, the method comprising:

[0006] The load resources of the target industrial area are divided into five core levels: “region-busbar-enterprise-production line-electrical equipment”, and the affiliation and relationship between each level are defined to construct a zoned and hierarchical management model.

[0007] Define the spatiotemporal characteristics of nodes, extract spatiotemporal characteristics from the partitioned hierarchical management model, and enhance the spatiotemporal characteristics;

[0008] Based on the enhanced spatiotemporal characteristics, a refined load elasticity profile is generated to quantify load response capability. Based on the load elasticity profile and the partitioned and hierarchical management model, the tradable interactive quotas of each level of node are determined, and a dynamic pricing model that is linked to market conditions and system value is established.

[0009] Based on the approved quotas and dynamic prices, the trading platform executes the matching and settlement of interactive quota transfer transactions. Settlement and incentives are based on the actual response results, and the parameters of the dynamic pricing model are optimized in a closed-loop rolling manner.

[0010] Secondly, the present invention provides an industrial load zoning and hierarchical interactive quota dynamic pricing system, the system comprising:

[0011] The management model construction module is used to divide the load resources of the target industrial area into five core levels: "area-bus-enterprise-production line-electrical equipment", and define the subordinate and related relationships between each level to build a zoned and hierarchical management model.

[0012] The feature extraction module is used to define the spatiotemporal features of nodes, extract spatiotemporal features from the partitioned hierarchical management model, and enhance the spatiotemporal features.

[0013] The pricing model construction module is used to generate a refined load elasticity profile for quantifying load response capability based on the enhanced spatiotemporal characteristics, and to determine the tradable interactive quota of each level node based on the load elasticity profile and the partitioned hierarchical management model, and to establish a dynamic pricing model that is linked to market conditions and system value.

[0014] The optimization module is used to perform interactive quota transfer transaction matching and settlement on the trading platform based on the approved quota and dynamic price, to conduct settlement and incentive based on the actual response effect, and to perform closed-loop rolling optimization of the parameters of the dynamic pricing model.

[0015] Thirdly, the present invention provides a storage medium that stores one or more programs that, when executed by a processor, implement the above-described industrial load zoning and hierarchical interactive quota dynamic pricing method.

[0016] Fourthly, the present invention provides an electronic device, the electronic device comprising a memory and a processor, wherein:

[0017] The memory is used to store computer programs;

[0018] When the processor executes the computer program stored in the memory, it implements the above-mentioned dynamic pricing method for industrial load zoning and hierarchical interactive quotas.

[0019] Compared with the prior art, the present invention has the following advantages:

[0020] 1. This invention utilizes a novel spatiotemporal graph convolutional-attention network deep learning framework to extract the spatiotemporal correlation features of load and identify key adjustable resources, thereby establishing a refined load elasticity profile. Based on this, an interactive quota dynamic pricing and transfer trading model is designed to improve the ability and efficiency of industrial load resources to participate in the electricity market.

[0021] 2. This invention constructs an industrial load zoning and hierarchical management model and employs a novel deep learning framework combining spatiotemporal graph convolutional networks and attention mechanisms to accurately characterize the spatiotemporal characteristics and response capabilities of the load, generating a refined load elasticity profile. Based on this, an interactive quota dynamic pricing and hierarchical transfer trading mechanism based on this profile is designed, effectively solving the problems of the traditional extensive load zoning and hierarchical management and difficulty in coordinating with the market. It significantly improves the ability of industrial users to participate in demand response and the electricity spot market, which is of great significance for enhancing the grid's supply guarantee capacity, promoting the consumption of renewable energy, and advancing the construction of a new power system. Attached Figure Description

[0022] Figure 1 This is a flowchart of a dynamic pricing method for industrial load zoning and hierarchical interactive quotas proposed in an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of the structure of an industrial load zoning and hierarchical interactive quota dynamic pricing system proposed in an embodiment of the present invention.

[0024] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0026] like Figure 1 As shown, an embodiment of the present invention proposes a dynamic pricing method for industrial load zoning and hierarchical interactive quotas. This method includes steps S101 to S104, wherein:

[0027] Step S101: Divide the load resources of the target industrial area into five core levels: “region-bus-enterprise-production line-electrical equipment”, and define the affiliation and association between each level to construct a zoned and hierarchical management model;

[0028] First, a multi-dimensional, zoning-based, hierarchical management model for industrial loads is established to provide clear spatial constraints and hierarchical relationships for subsequent algorithms, ensuring a structured foundation for load modeling. Compared to traditional management methods based on single nodes or enterprises, this multi-dimensional modeling approach enables cross-level correlation analysis, improving the controllability and interpretability of zoning and hierarchical management.

[0029] The model divides the load resources of the target industrial area into five core levels: “region-busbar-enterprise-production line-electrical equipment” from three dimensions: power grid physical structure, administrative management unit and production organization process, and defines the affiliation and relationship between each level.

[0030] Based on this management model, a weighted directed graph G = (V, E, A, W) is constructed:

[0031] Where V = {v1, v2, ..., v} N} represents a collection of nodes, with each node having additional hierarchical and regional attributes. The set of edges represents not only electrical connections, but also logical relationships such as production process sequence and management hierarchy. Represents the adjacency matrix, defining the basic connection states between nodes. This is a comprehensive weight matrix, whose elements are jointly determined by the management weight coefficient and the production correlation degree. The element W in the i-th row and j-th column of the comprehensive weight matrix... ij for:

[0032] W ij =α h(i，j) ·ρ ij (1)

[0033] Where, α h(i，j) ρ represents the hierarchical management weight coefficient. ij It is a correlation based on the tight coupling of the production process.

[0034] This step outputs a partitioned and hierarchical management topology matrix, which serves as the spatial constraint input for subsequent feature extraction.

[0035] Step S102: Define the spatiotemporal characteristics of the node, extract the spatiotemporal characteristics from the partitioned hierarchical management model, and enhance the spatiotemporal characteristics;

[0036] In this step, multi-source load, electricity price, and production information are uniformly encoded into high-dimensional spatiotemporal features, providing a data foundation for spatiotemporal deep learning models. Compared to traditional modeling methods that use power or electricity price as a single dimension input, this method comprehensively considers information such as operation, planning, and status, enhancing the model's ability to identify dynamic load behavior.

[0037] For each node v i At time step t, its high-dimensional feature vector is defined.

[0038]

[0039] in, P is a high-dimensional feature vector. i For active power, Q i Reactive power I represents the voltage amplitude. i The current amplitude, For power factor, λ DA and λ RT These are the day-ahead and real-time electricity prices, respectively. t For time period type encoding, S oc δ serves as a status indicator for the equipment's operation. plan The binary encoding of the production plan status.

[0040] Furthermore, in some embodiments, the STGCN is used to capture the complex spatiotemporal dynamic characteristics of load clusters under the management model. The spatiotemporal features X and the topology weight matrix W are input into the Spatiotemporal Graph Convolutional Network (STGCN) to extract spatial dependencies and temporal dynamic features. STGCN simultaneously captures spatial dependencies and temporal dynamics between nodes, enabling spatiotemporal joint modeling of multi-level loads. Compared to traditional LSTM, CNN, and other models, this method can explicitly utilize the network topology to identify the spatial coupling between the electrical and production layers, significantly improving the accuracy of extracting complex load variation patterns.

[0041] By performing a self-loop on the topological weight matrix and normalizing it, we obtain the normalized adjacency matrix used for spatial convolution:

[0042]

[0043] Where I is the identity matrix, Let W be the self-looped and normalized comprehensive weight matrix. It is a degree matrix (diagonal matrix). The element in the i-th row and j-th column of the normalized composite weight matrix. The normalized adjacency matrix, This is the normalized degree matrix;

[0044] For each time step, perform graph convolution between the node features and the normalized adjacency matrix:

[0045]

[0046] in, Let S represent the spatial convolution parameters of the l-th layer, σ be the activation function, and S be the spatial parameters of the l-th layer. (t,l) The output feature map of the l-th spatial convolution at time step t describes the spatial association features of nodes in that layer.

[0047] Gated causal dilation temporal convolution is performed on the spatial convolution output sequence of the most recent K time steps to model temporal dynamics:

[0048]

[0049] Where * denotes one-dimensional causal convolution along the time direction, ⊙ denotes element-wise multiplication, and W f W g All are weights, b f b g All are biased. Z is the intermediate feature map (corresponding to time step t) after residual connection and layer normalization of the output feature map S(l,t) of the l-th spatial convolution. It is used to eliminate feature distribution differences and provide standardized input features for subsequent gated causal dilation temporal convolutions. (t) Z represents the final output of the gated causal dilation temporal convolution, and the temporal features filtered through the forgetting gate. (t) for;

[0050] Introducing residual connections and inter-layer normalization ensures training stability and information transfer:

[0051]

[0052] in, Denotes the necessary linear transformations to align dimensions, LayerNorm is for layer normalization, and H (t,l+1) The feature tensor of the (l+1)th layer after residual connection and layer normalization is fused with the convolutional features of the current layer and the output of the previous layer to ensure stable gradient propagation.

[0053] (4) Overall output format: All time steps H (t,l+1) The superposition of these elements yields the output representation tensor of STGCN:

[0054]

[0055] Where L is the last layer number of STGCN, d is the final feature dimension of each node, ⊙ is the Hadamard product, and H out This is the final output representation tensor of the spatiotemporal graph convolutional network.

[0056] Furthermore, in some embodiments, a hierarchical attention mechanism is introduced to adaptively focus on key loads and key time periods at different levels in the management model. Based on the STGCN output, a multi-head attention mechanism is used to achieve adaptive focusing on key time periods and key loads. Unlike traditional weighted averaging or fixed-weight aggregation, the Transformer structure can dynamically allocate feature attention, enabling the model to better identify high-value adjustable points and enhancing interpretability and real-time performance.

[0057] (1) Sequence Reshaping and Position Encoding: The features H output by the spatiotemporal graph convolutional network are reshaped and encoded. out Remodeling into a sequence And inject hierarchical position encoding:

[0058] Z in =H out +PE hier (pos) (8)

[0059] Among them, PE hier (pos) is the hierarchical position encoding function, Z in The sequence features are encoded after the injection level position.

[0060] (2) Hierarchical-aware multi-head self-attention: When calculating attention, a hierarchy-based bias matrix B is introduced. layer :

[0061]

[0062] Where, d k B represents the attention vector dimension. layer The settings are configured according to the required level of the node to guide the model to focus on more management value. This is the output feature of the multi-head attention mechanism, which integrates the association results of multiple attention groups. M is the "number of heads" in the multi-head attention mechanism. Through parallel computation of multiple attention groups, it captures association features of different dimensions. Q is the "query vector matrix" in the attention mechanism, which is used to match key features. K is the "key vector matrix" in the attention mechanism, which calculates the similarity with the query vector. T0 is the transpose. V is the "value vector matrix" in the attention mechanism, which outputs features based on similarity weighting.

[0063] (3) Feedforward network and coding layer:

[0064] Z ffn =ReLU(Z) attW1+b1)W2+b2 (10)

[0065] H att =LayerNorm(Z) ffn +Z att (11)

[0066] Where W1, W2, b1, and b2 are all feedforward network parameters, and H att Z represents the features after attention enhancement. ffn The output features of the feedforward network are enhanced by ReLU activation and linear transformation to strengthen the expression of key features.

[0067] Step S103: Generate a refined load elasticity profile for quantifying load response capability based on the enhanced spatiotemporal characteristics, and determine the tradable interactive quota of each level node based on the load elasticity profile and the partitioned hierarchical management model, and establish a dynamic pricing model that is linked to market conditions and system value.

[0068] It should be noted that in this step, the enhanced feature H... att This is mapped to refined load resilience profile parameters. Deep features are transformed into quantifiable "load resilience profiles," enabling a detailed description of node adjustment capabilities, costs, and response characteristics. Compared to traditional load resilience estimation based on experience or linear models, this method simultaneously outputs multi-dimensional parameters through multi-task learning, resulting in more comprehensive results and higher prediction accuracy.

[0069] (1) Multi-scale feature pooling: Perform multi-scale temporal pooling for each node:

[0070]

[0071] The global features are obtained after concatenation:

[0072] in, H is the average temporal feature vector of the i-th node, obtained by average pooling of the attention-enhanced features of node i at all time steps, which characterizes the long-term average trend of the node's power consumption status. att (i,t,:) represents the attention enhancement feature vector of the i-th node at time step t. Let be the maximum temporal feature vector of the i-th node. Let be the global fusion feature vector of the i-th node;

[0073] (2) Multi-task regression: Using a multi-task learning head, various parameters of the load elasticity profile are regressed simultaneously.

[0074]

[0075] Where, φ i Profiling the load resilience of nodes. For interruptible capacity, For the capacity to be moved, To respond to delay, T i duration For the duration, To adjust costs in multiple dimensions, For load elasticity margin, W multi and b multi These are the parameters for the multi-task regression layer.

[0076] Furthermore, in some embodiments, a zoned and tiered quota allocation and dynamic pricing mechanism is established based on management models and load resilience profiles to achieve quantitative coupling between load value and market signals. Compared to static pricing or single-tiered quota mechanisms, this method introduces tiered dynamic coefficients and market feedback to achieve a closed loop of "load characteristics - price response," thereby improving market flexibility and fairness.

[0077] (1) Tiered Interactive Quota Determination: Considering the node level, calculate its tradable quota. i :

[0078]

[0079] Among them, ψ(L(v) i )) represents the hierarchical quota allocation coefficient, τ ref ΔT is the reference delay time, and ΔT is the difference between the actual response delay and the reference delay.

[0080] (2) Basic value assessment: The basic value of the quota V i base Determined by its adjustment costs and system value:

[0081]

[0082] Where Priority(·) is a priority function based on regional importance, and α, β, and ζ are all weight coefficients. For the multidimensional adjustment cost of nodes, Let i be the load elasticity feature vector of the i-th node. For the average electricity price, Z(v) i ) is node v i The regional attribute feature vector.

[0083] (3) Real-time dynamic pricing: Quota i Real-time transaction price i for:

[0084]

[0085] Where ε and v are both price elasticity indices. This refers to the real-time electricity price.

[0086] (4) Final Real-Time Dynamic Pricing: The final pricing takes into account both zone and tier bonuses.

[0087]

[0088] Where ξ(·) and η(·) are the addition functions for partitioning and hierarchy, respectively. This is the final real-time dynamic price.

[0089] Step S104: Based on the approved quota and dynamic price, perform interactive quota transfer transaction matching and settlement on the trading platform, conduct settlement and incentives based on the actual response effect, and perform closed-loop rolling optimization of the parameters of the dynamic pricing model.

[0090] The trading platform executes quota transfers and completes clearing and settlement. This transforms load resources into tradable quotas, enabling flexible complementarity and economic optimization across multiple market levels. Compared to traditional centralized control methods, this mechanism supports distributed, autonomous trading and revenue settlement, incentivizing enterprises to actively participate in demand response and improving market efficiency.

[0091] (1) Transaction matching: For the seller set S and the buyer set B, perform mixed-integer linear programming matching with the goal of maximizing overall welfare:

[0092]

[0093] Constraints:

[0094]

[0095] Where, π ij Quota is the decision variable for matching transactions between nodes i and j. ij Quota represents the transaction volume from node i to node j. i Quota is the hierarchical interaction quota for node i. j The hierarchical interaction quota for node j.

[0096] (2) Settlement and reward / penalty mechanism:

[0097]

[0098] in, For the actual transaction volume, η1 and η2 are the reward and penalty coefficients, respectively, and R... i For transaction and settlement proceeds, Let be the total quota requirement for node j.

[0099] Finally, based on the latest operational data, a closed-loop rolling optimization is performed on the deep learning model and market parameters. By continuously optimizing model parameters and pricing strategies, a data-driven dynamic closed-loop system is formed. Unlike one-time static models, this method achieves self-learning and self-correction, possesses long-term adaptability and real-time optimization capabilities, and can sustainably improve the system's predictive and economic benefits.

[0100] Define a comprehensive loss function to simultaneously optimize prediction accuracy and market utility:

[0101]

[0102] in, This is the load forecasting error term. For market utility loss, μ1 and μ2 are trade-off coefficients, and θ is the rolling optimization implementation model parameter.

[0103] The final output of this system includes:

[0104]

[0105] In summary, by deeply integrating the spatiotemporal graph convolutional network and the Transformer, a cutting-edge deep learning architecture, and innovatively applying it to load characteristic modeling and market mechanism design, the challenges of industrial load zoning and hierarchical management and market coordination have been effectively solved.

[0106] For example, the method of the present invention establishes a zoned and hierarchical management architecture to achieve refined zoned and hierarchical management of industrial load and market collaboration. The specific implementation steps are as follows:

[0107] Step 1: Construct an industrial load zoning and hierarchical management model. This model divides the load resources of the target industrial area into five core levels: “region-bus-enterprise-production line-electrical equipment” from three dimensions: power grid physical structure, administrative management unit and production organization process. For details, please refer to formula (1).

[0108] Step 2: Construct the definition and input of the spatiotemporal features of the nodes, which can be referred to in Equation (2).

[0109] Step 3: Extract the spatiotemporal features of the load based on the spatiotemporal graph convolutional network, and refer to equations (3) to (5) for details.

[0110] Step 4: Enhance key features of the load based on the Transformer attention mechanism, specifically refer to equations (6) to (10).

[0111] Step 5: Generate a load resilience profile and map the enhanced features to the refined load resilience profile parameters. For details, please refer to equations (11) to (12).

[0112] Step 7: Interactive quota transfer transaction settlement. Execute the quota transfer on the trading platform and complete the final transaction settlement. For details, please refer to formulas (17) to (21).

[0113] Step 8: Rolling optimization of the model and trading strategy, and finally output the optimal operating strategy and market trading utility in the region. For details, please refer to formula (22).

[0114] In summary, this invention constructs a refined industrial load management model, uses a spatiotemporal graph convolutional network to extract load spatiotemporal features, enhances key load features based on the Transformer attention mechanism, and generates a load elasticity profile. In addition, it establishes a dynamic electricity market pricing model under load zoning and hierarchical management, and performs closed-loop rolling optimization of the deep learning model and market parameters to achieve optimized allocation and dynamic pricing of load resources under zoning and hierarchical management.

[0115] like Figure 2 As shown, this invention proposes an industrial load zoning and hierarchical interactive quota dynamic pricing system, the system comprising:

[0116] The management model construction module 10 is used to divide the load resources of the target industrial area into five core levels: “region-bus-enterprise-production line-electrical equipment”, and define the subordinate and related relationships between each level to construct a zoned and hierarchical management model.

[0117] The feature extraction module 20 is used to define the spatiotemporal features of nodes, extract spatiotemporal features from the partitioned hierarchical management model, and enhance the spatiotemporal features;

[0118] The pricing model construction module 30 is used to generate a refined load elasticity profile for quantifying load response capability based on the enhanced spatiotemporal characteristics, and to determine the tradable interactive quota of each level node based on the load elasticity profile and the partitioned hierarchical management model, and to establish a dynamic pricing model that is linked to market conditions and system value.

[0119] The optimization module 40 is used to perform interactive quota transfer transaction matching and settlement on the trading platform according to the approved quota and dynamic price, to conduct settlement and incentive based on the actual response effect, and to perform closed-loop rolling optimization of the parameters of the dynamic pricing model.

[0120] In another aspect, the present invention also proposes a storage medium on which one or more programs are stored, which, when executed by a processor, implement the above-described dynamic pricing method for industrial load zoning and hierarchical interactive quotas.

[0121] In another aspect, the present invention also proposes an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to realize the above-mentioned dynamic pricing method for industrial load zoning and hierarchical interactive quotas.

[0122] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain stored, communicated, propagated, or transmitted programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0123] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0124] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0125] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A dynamic pricing method for industrial load zoning and hierarchical interactive quotas, characterized in that, The method includes: The load resources of the target industrial area are divided into five core levels: "area-busbar-enterprise-production line-electrical equipment", and the affiliation and relationship between each level are defined to construct a zoned and hierarchical management model. Define the spatiotemporal characteristics of nodes, extract spatiotemporal characteristics from the partitioned hierarchical management model, and enhance the spatiotemporal characteristics; Based on the enhanced spatiotemporal characteristics, a refined load elasticity profile is generated to quantify load response capability. Based on the load elasticity profile and the partitioned hierarchical management model, the tradable interactive quotas of each level of node are determined, and a dynamic pricing model linked to market conditions and system value is established. Based on the approved quotas and dynamic prices, the trading platform executes the matching and settlement of interactive quota transfer transactions. Settlement and incentives are based on the actual response results, and the parameters of the dynamic pricing model are optimized in a closed-loop rolling manner.

2. The industrial load zoning and hierarchical interactive quota dynamic pricing method according to claim 1, characterized in that, The spatiotemporal characteristics of a node are defined according to the following formula: in, P is a high-dimensional feature vector. i For active power, Q i Reactive power I represents the voltage amplitude. i The current amplitude, For power factor, λ DA and λ RT These are the day-ahead and real-time electricity prices, respectively. t For time period type encoding, S oc δ serves as a status indicator for the equipment's operation. plan The binary encoding of the production plan status.

3. The industrial load zoning and hierarchical interactive quota dynamic pricing method according to claim 2, characterized in that, The steps for extracting spatiotemporal features from the partitioned hierarchical management model include: By performing a self-loop on the topological weight matrix and normalizing it, we obtain the normalized adjacency matrix used for spatial convolution: Where I is the identity matrix, Let W be the self-looped and normalized comprehensive weight matrix. It is a degree matrix (diagonal matrix). The element in the i-th row and j-th column of the normalized composite weight matrix. The normalized adjacency matrix, This is the normalized degree matrix; For each time step, perform graph convolution between the node features and the normalized adjacency matrix: in, Let S represent the spatial convolution parameters of the l-th layer, σ be the activation function, and S be the spatial parameters of the l-th layer. (t，l) The output feature map of the l-th spatial convolution at time step t describes the spatial association features of nodes in that layer. Gated causal dilation temporal convolution is performed on the spatial convolution output sequence of the most recent K time steps to model temporal dynamics: Where * denotes one-dimensional causal convolution along the time direction, ⊙ denotes element-wise multiplication, and W f W g All are weights, b f b g All are biased. The intermediate feature map S(l,t) output by the l-th spatial convolution layer is obtained after residual connections and layer normalization. This intermediate feature map is used to eliminate feature distribution differences and provide standardized input features for subsequent gated causal dilation temporal convolutions. (t) The final output of the gated causal dilation temporal convolution is the temporal feature filtered by the forget gate; Introducing residual connectivity and interlayer normalization: in, Denotes the necessary linear transformations to align dimensions, LayerNorm is for layer normalization, and H (t，l+1) The feature tensor of the (l+1)th layer after residual connection and layer normalization is fused with the convolutional features of the current layer and the output of the previous layer to ensure stable gradient propagation. All time steps H (t，l+1) The superposition of these elements yields the output representation tensor of STGCN: Where L is the last layer number of STGCN, d is the final feature dimension of each node, ⊙ is the Hadamard product, and H out This is the final output representation tensor of the spatiotemporal graph convolutional network; The element W in the i-th row and j-th column of the comprehensive weight matrix ij for: W ij =a h(i,j) ·r ij W ij =a h(i,j) ·r ij ; Where, α h(i,j) ρ represents the hierarchical management weight coefficient. ij It is a correlation based on the tight coupling of the production process.

4. The industrial load zoning and hierarchical interactive quota dynamic pricing method according to claim 3, characterized in that, The step of enhancing the spatiotemporal features includes: The feature H output by the spatiotemporal graph convolutional network out Remodeling into a sequence And inject hierarchical position encoding: Z in =H out +PE hier (post); Among them, PE hier (pos) is the hierarchical position encoding function, Z in The sequence features are encoded after the injection level position; When calculating attention, a hierarchy-based bias matrix B is introduced. layer : Where, d k B represents the attention vector dimension. layer Configure the nodes according to their required hierarchical levels to guide the model to focus on greater management value. att The output features of the multi-head attention mechanism integrate the association results of multiple attention groups. M is the "number of heads" in the multi-head attention mechanism. Through parallel computation of multiple attention groups, association features of different dimensions are captured. Q is the "query vector matrix" in the attention mechanism, which is used to match key features. K is the "key vector matrix" in the attention mechanism, which calculates the similarity with the query vector. T0 is the transpose. V is the "value vector matrix" in the attention mechanism, which outputs features based on similarity weights. Feedforward networks and coding layers: WITH ffn =ReLU(Z att W1+b1)W2+b2 H att =LayerNorm(Z ffn +Z aat ); Where W1, W2, b1, and b2 are all feedforward network parameters, and H att Z represents the features after attention enhancement. ffn The output features of the feedforward network are enhanced by ReLU activation and linear transformation to strengthen the expression of key features.

5. The industrial load zoning and hierarchical interactive quota dynamic pricing method according to claim 4, characterized in that, The step of generating a refined load resilience profile for quantifying load response capability based on the enhanced spatiotemporal features includes: Perform multi-scale temporal pooling for each node: The global features are obtained after concatenation: in, H is the average temporal feature vector of the i-th node, obtained by average pooling of the attention-enhanced features of node i at all time steps, which characterizes the long-term average trend of the node's power consumption status. att (i,t,:) represents the attention enhancement feature vector of the i-th node at time step t. Let be the maximum temporal feature vector of the i-th node. Let be the global fusion feature vector of the i-th node; Using a multi-task learning head, various parameters of the load elasticity profile are simultaneously regressed: Where, φ i Profiling the load resilience of nodes. For interruptible capacity, For the capacity to be moved, To respond to delay, T i duration For the duration, To adjust costs in multiple dimensions, For load elasticity margin, W multi and b multi These are the parameters for the multi-task regression layer.

6. The industrial load zoning and hierarchical interactive quota dynamic pricing method according to claim 5, characterized in that, The steps of determining the tradable interactive quotas for each level of node based on the load elasticity profile and the partitioned hierarchical management model, and establishing a dynamic pricing model that is linked to market conditions and system value, include: Considering the node level, calculate its tradable quota. i : Among them, ψ(L(v) i )) represents the hierarchical quota allocation coefficient, τ ref The reference delay time is ΔT, which is the difference between the actual response delay and the reference delay. The basic value of the quota V i base Determined by its adjustment costs and system value: Where Priority(·) is a priority function based on regional importance, and α, β, and ζ are all weight coefficients. For the multidimensional adjustment cost of nodes, Let i be the load elasticity feature vector of the i-th node. For the average electricity price, Z(v) i ) is node v i The region attribute feature vector; Quota i Real-time transaction price i for: Where ε and ν are both price elasticity indices. For real-time electricity prices; The final pricing takes into account both zone and tier bonuses: Where ξ(·) and η(·) are the addition functions for partitioning and hierarchy, respectively. This is the final real-time dynamic price.

7. The industrial load zoning and hierarchical interactive quota dynamic pricing method according to claim 6, characterized in that, The steps for matching and clearing interactive quota transfer transactions on the trading platform based on the approved quota and dynamic price include: For a seller set S and a buyer set B, perform a mixed-integer linear programming matching with the objective of maximizing overall welfare: Construct constraints based on the following formula: Where, π ij Quota is the decision variable for matching transactions between nodes i and j. ij Quota represents the transaction volume from node i to node j. i Quota is the hierarchical interaction quota for node i. j The hierarchical interaction quota for node j; Establish a clearing and settlement and reward / penalty mechanism based on the following formula: in, For the actual transaction volume, η1 and η2 are the reward and penalty coefficients, respectively, and R... i For transaction and settlement proceeds, Let be the total quota requirement for node j.

8. The industrial load zoning and hierarchical interactive quota dynamic pricing method according to claim 7, characterized in that, The steps of settling and incentivizing based on actual response results, and performing closed-loop rolling optimization of the parameters of the dynamic pricing model include: Define the comprehensive loss function: in, This is the load forecasting error term. For market utility loss, μ1 and μ2 are both trade-off coefficients, and θ is the rolling optimization implementation model parameter; The final output is 9. A dynamic pricing system for industrial load zoning and hierarchical interactive quotas, characterized in that, The system includes: The management model construction module is used to divide the load resources of the target industrial area into five core levels: "area-bus-enterprise-production line-electrical equipment", and define the affiliation and association between each level to build a zoned and hierarchical management model. The feature extraction module is used to define the spatiotemporal features of nodes, extract spatiotemporal features from the partitioned hierarchical management model, and enhance the spatiotemporal features. The pricing model construction module is used to generate a refined load elasticity profile for quantifying load response capability based on the enhanced spatiotemporal characteristics, and to determine the tradable interactive quota of each level node based on the load elasticity profile and the partitioned hierarchical management model, and to establish a dynamic pricing model that is linked to market conditions and system value. The optimization module is used to perform interactive quota transfer transaction matching and settlement on the trading platform based on the approved quota and dynamic price, to conduct settlement and incentive based on the actual response effect, and to perform closed-loop rolling optimization of the parameters of the dynamic pricing model.

10. A storage medium, characterized in that, The storage medium stores one or more programs that, when executed by a processor, implement the industrial load zoning and hierarchical interactive quota dynamic pricing method as described in any one of claims 1-8.

Images