Factory load clean power consumption optimization method and system based on DCRNN and CATS

Abstract

The invention provides a factory load clean power utilization optimization method and system based on DCRNN and CATS. The method comprises the following steps: predicting a fossil fuel power generation proportion in unit electric energy of a factory in target time by adopting a CATS model; determining the total interruptible load capacity of the factory in the target time by adopting a DCRNN model; determining the actual electrical load of the factory at the target time; under the condition that the total interruptible load capacity of the factory in the target time meets a constraint condition, determining a clean power utilization responsiveness index and an actual weighted average fossil power ratio according to a fossil fuel power generation ratio in unit electric energy of the factory in the target time, and determining a weighted load of unit electric quantity carbon emission intensity in combination with an actual power utilization load; and the clean power utilization of the factory load is optimized. According to the invention, precise cooperation of factory load and clean power is realized, preferential utilization of green power is promoted, power consumption in a high-carbon period is avoided, and green low-carbon transformation and carbon emission optimization of a factory are facilitated.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system regulation and control, and in particular relates to a method and system for optimizing clean electricity consumption of factory loads based on DCRNN and CATS. Background Art

[0002] It is of great significance to reduce the proportion of fossil fuel power generation and increase the use of clean energy such as wind power and hydropower. It can effectively reduce carbon emission intensity, promote green production transformation, and improve green electricity consumption efficiency, which will not only help achieve corporate carbon peak and carbon neutrality goals.

[0003] Existing research and application methods for clean electricity consumption for factory loads generally have the following key deficiencies: 1) The granularity of fossil fuel share prediction is coarse and the dynamic response is poor. Existing methods are mostly based on daily averages or time period static statistics, which makes it difficult to accurately reflect the changes in the power generation structure of the power grid at different time points. Especially in the context of fluctuations in renewable energy output and / or frequent changes in grid scheduling, there is a lack of sensitive capture and accurate characterization of the dynamic fluctuations of "carbon emission composition per kilowatt-hour" over time, which affects the timeliness and matching degree of clean electricity consumption strategies. 2) There is a lack of intelligent screening mechanism for interruptible equipment on the load side. The types of loads in factories, such as refrigeration systems, air compressors, and electric drying equipment, are diverse and have complex operating logic. Traditional methods that rely on experience or static rules to identify interruptible equipment cannot accurately determine whether they have the conditions for safe interruption at a specific time, and are prone to underestimate or overestimate the response potential, affecting the feasibility of the regulation plan. 3) The clean electricity regulation model fails to achieve the unity of source-load linkage and response optimization. Current methods generally model power supply cleanliness and load responsiveness in steps, lacking a linkage mechanism between "power supply cleanliness ratio - equipment interruptibility - load regulation strategy". This results in the inability to actively increase load and utilize green electricity when clean power is abundant, and the inability to timely reduce non-critical loads during high-carbon periods, failing to form a complete and effective clean electricity regulation closed loop. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method and system for optimizing clean electricity consumption of factory loads based on DCRNN and CATS.

[0005] In a first aspect, the present invention provides a method for optimizing clean electricity consumption of factory loads based on DCRNN and CATS, comprising:

[0006] Use the CATS model to predict the proportion of fossil fuel generation per unit of electricity produced by the plant at the target time;

[0007] The DCRNN model is used to determine the total interruptible load capacity of the plant at the target time;

[0008] Determine the actual power load of the factory at the target time;

[0009] Under the condition that the total interruptible load capacity of the factory at the target time meets the constraints, the clean electricity responsiveness index and the actual weighted average fossil electricity ratio are determined according to the proportion of fossil fuel power generation in the unit electricity of the factory at the target time, and the weighted load of carbon emission intensity per unit electricity is determined in combination with the actual electricity load;

[0010] The clean electricity consumption of factory load is optimized according to the weighted load of clean electricity responsiveness index, actual weighted average proportion of fossil electricity and carbon emission intensity per unit electricity.

[0011] Optionally, the using the CATS model to predict the proportion of fossil fuel power generation in the unit electricity of the factory at the target time includes:

[0012] The percentage of fossil fuel electricity generated per unit of electricity consumed by the plant at the target time is calculated using the following formula:

[0013]

[0014] E=[e(tk),…,e(t-1)];

[0015] e(tk)=Embed(x(tk));

[0016] x(t′)=[P lood (t′),G wind (t′),G solar (t′),G cood (t′),G gos (t′),L cood (t′),T(t′),H(t′),P price (t′),D dow (t′),H hour (t′)];

[0017] t′=tk;

[0018] Among them, φ fossil(t) is the proportion of fossil fuel power generation in the unit electricity of the factory at time t'; σ(·) is the Sigmoid function; w2 is the weight matrix of the first multi-layer perceptron; ReLU(·) is the activation function; w1 is the weight matrix of the first multi-layer perceptron; MeanPool(·) is the average pooling of the time series dimension; CATS(E) is the context feature sequence after attention fusion; b1 is the first bias term; b2 is the second bias term; T represents matrix transpose; Softmax(·) is the attention weight normalization function; Q is the query vector matrix in the attention mechanism; K is the key vector matrix in the attention mechanism; d is the scaling factor used to prevent gradient explosion; B is the bias matrix below, used to fuse global background information; V is the median vector matrix of the attention mechanism; E is the input sequence matrix composed of the embedding vectors at all times; e(tk) is the multi-dimensional feature vector at time tk mapped to the d-dimensional embedding vector; Embed(·) is the embedding function; x(tk) is the multi-dimensional feature vector at time tk; P lood (t′) is the factory load at time t'; G wind (t′) is the wind power output at time t'; G solar (t′) is the photovoltaic power generation output at time t'; G cood (t′) is the coal-fired power generation output at time t'; G gos (t′) is the gas-fired power generation output at time t'; L cood (t') is the total load of the entire area at time t'; T(t') is the temperature at time t'; H(t') is the humidity at time t'; P price (t′) is the electricity price at time t'; D dow (t′) is the number of the week in which time t' occurs; H hour (t′) The hour in the day at time t'.

[0019] Optionally, the determining of the total interruptible load capacity of the plant at the target time by using the DCRNN model includes:

[0020] Construct a process dependency graph for electrical equipment within the factory; nodes represent electrical equipment, and edges represent the order in which each electrical equipment operates or the resource coupling relationship;

[0021] Calculate the total interruptible load capacity of the plant at the target time using the following formula:

[0022]

[0023] Among them, P interruptible (t) is the total interruptible load capacity of the factory at time t; M is the total number of interruptible equipment in the factory; is the actual interruptible load power of equipment i; γ i(t) is the interruption allowable factor under the process constraints of equipment i; The interruptible load potential of device i at time t predicted by the DCRNN model; Slack i (t) is the minimum time window that device i and the downstream device can wait; is the minimum continuous operation limit of device i; ReLU(·) is the activation function; W out is the output mapping weight matrix; T represents the transpose of the matrix; is the hidden state of device i at time t in the last layer of the DCRNN model; b is the third bias term; GRU(·) represents the gated recurrent unit, which is used to capture the evolution of temporal features; N(i) is the set of neighboring nodes directly connected to node i; A ij is the edge weight from node i to node j in the adjacency matrix; W (L) is the learnable weight matrix of the Lth layer of graph convolution; is the hidden state of neighbor node j in the L-1 layer at time t.

[0024] Optionally, determining the actual power load of the factory at the target time includes:

[0025] The sum of the factory's adjustable load and non-adjustable load at time t is taken as the factory's actual power load at time t.

[0026] Optionally, when the total interruptible load capacity of the factory at the target time satisfies the constraint conditions, the clean electricity responsiveness index and the actual weighted average fossil electricity ratio are determined according to the fossil fuel power generation ratio per unit electricity of the factory at the target time, and the weighted load of the carbon emission intensity per unit electricity is determined in combination with the actual electricity load, including:

[0027] Construct the constraints that the total interruptible load capacity of the plant at the target time satisfies:

[0028]

[0029] Among them, P interruptible (t) is the total interruptible load capacity of the plant at time t; P shift (t) is the adjustable load of the plant at time t; is the load capacity that the plant can operate in advance at time t; T' is the total number of preset times; P load (t) is the non-adjustable load of the factory at time t; P base (t) is the non-adjustable load of the factory at time t; E total is the total power consumption demand of the factory; P shift (t-1) is the adjustable load of the factory at time t-1; Δ max is the maximum allowable total amount of adjustment change;

[0030] The clean electricity responsiveness index R is calculated according to the following formula clean :

[0031] R clean =-corr(φ fossil (t),P shift (t));

[0032] Where corr(·) is the correlation coefficient function; φ fossil (t) is the proportion of fossil fuel electricity generation in the unit electricity of the factory at time t'; the actual weighted average fossil fuel electricity generation is calculated according to the following formula

[0033]

[0034] The weighted load of carbon emission intensity per unit of electricity is calculated according to the following formula:

[0035]

[0036] Among them, P′ QC (t) is the weighted load of carbon emission intensity per unit electricity at time t.

[0037] In a second aspect, the present invention provides a factory load clean electricity optimization system based on DCRNN and CATS, comprising:

[0038] A forecasting module, which uses the CATS model to predict the percentage of fossil fuel-based electricity generated per unit of power at the plant at a target time.

[0039] A first determination module is used to determine the total interruptible load capacity of the plant at the target time using a DCRNN model;

[0040] The second determination module is used to determine the actual power load of the factory at the target time;

[0041] The third determination module is used to determine the clean electricity responsiveness index and the actual weighted average fossil electricity ratio based on the fossil fuel power generation ratio per unit of electricity consumed by the plant at the target time, provided that the total interruptible load capacity of the plant at the target time meets the constraint conditions, and to determine the weighted load of carbon emission intensity per unit of electricity consumed based on the actual electricity load;

[0042] The optimization module is used to optimize the clean electricity consumption of factory loads based on the clean electricity responsiveness index, the actual weighted average fossil electricity share, and the weighted load of carbon emission intensity per unit of electricity.

[0043] Optionally, the prediction module includes:

[0044] The first calculation unit is used to calculate the proportion of fossil fuel power generation in the unit electricity of the factory at the target time according to the following formula:

[0045]

[0046] E=[e(tk),…,e(t-1)];

[0047] e(tk)=Embed(x(tk));

[0048] x(t′)=[P load (t′), G wind (t′), G solar (t′), G cood (t′), G gos (t′), L coos (t′), T(t′), H(t′), P price (t′), D dow (t′), H hour (t′)];

[0049] t′=tk;

[0050] Among them, φ fossil (t) is the proportion of fossil fuel power generation in the unit electricity of the factory at time t'; σ(·) is the Sigmoid function; w2 is the weight matrix of the first multi-layer perceptron; ReLU(·) is the activation function; w1 is the weight matrix of the first multi-layer perceptron; MeanPool(·) is the average pooling of the time series dimension; CATS(E) is the context feature sequence after attention fusion; b1 is the first bias term; b2 is the second bias term; T represents matrix transpose; Softmax(·) is the attention weight normalization function; Q is the query vector matrix in the attention mechanism; K is the key vector matrix in the attention mechanism; d is the scaling factor used to prevent gradient explosion; B is the bias matrix below, used to fuse global background information; V is the median vector matrix of the attention mechanism; E is the input sequence matrix composed of the embedding vectors at all times; e(tk) is the multi-dimensional feature vector at time tk mapped to the d-dimensional embedding vector; Embed(·) is the embedding function; x(tk) is the multi-dimensional feature vector at time tk; P lood (t′) is the factory load at time t'; G wind (t′) is the wind power output at time t'; G solar (t′) is the photovoltaic power generation output at time t'; G cood (t′) is the coal-fired power generation output at time t'; G gos (t′) is the gas-fired power generation output at time t'; L cood(t') is the total load of the entire area at time t'; T(t') is the temperature at time t'; H(t') is the humidity at time t'; P price (t′) is the electricity price at time t'; D dow (t′) is the number of the week in which time t' occurs; H hour (t′) The hour in the day at time t'.

[0051] Optionally, the first determining module includes:

[0052] The first construction unit is used to construct a process dependency graph of electrical equipment within the factory, wherein nodes represent electrical equipment and edges represent the operation sequence or resource coupling relationship of each electrical equipment;

[0053] The second calculation unit is used to calculate the total interruptible load capacity of the plant at the target time according to the following formula:

[0054]

[0055] Among them, P interruptible (t) is the total interruptible load capacity of the factory at time t; M is the total number of interruptible equipment in the factory; is the actual interruptible load power of equipment i; γ i (t) is the interruption allowable factor under the process constraints of equipment i; The interruptible load potential of device i at time t predicted by the DCRNN model; Slack i (t) is the minimum time window that device i and the downstream device can wait; is the minimum continuous operation limit of device i; ReLU(·) is the activation function; W out is the output mapping weight matrix; T represents the transpose of the matrix; is the hidden state of device i at time t in the last layer of the DCRNN model; b is the third bias term; GRU(·) represents the gated recurrent unit, which is used to capture the evolution of temporal features; N(i) is the set of neighboring nodes directly connected to node i; A ij is the edge weight from node i to node j in the adjacency matrix; W (L) is the learnable weight matrix of the Lth layer of graph convolution; is the hidden state of neighbor node j in the L-1 layer at time t.

[0056] Optionally, the second determining module includes:

[0057] The determination unit is used to take the sum of the adjustable load and the non-adjustable load of the factory at time t as the actual power load of the factory at time t.

[0058] Optionally, the third determining module includes:

[0059] The second building block is used to build the constraints that the total interruptible load capacity of the plant at the target time satisfies:

[0060]

[0061] Among them, P interruptible (t) is the total interruptible load capacity of the plant at time t; P shift (t) is the adjustable load of the plant at time t; is the load capacity that the plant can operate in advance at time t; T' is the total number of preset times; P load (t) is the non-adjustable load of the factory at time t; P base (t) is the non-adjustable load of the factory at time t; E total is the total power consumption demand of the factory; P shift (t-1) is the adjustable load of the factory at time t-1; Δ max is the maximum allowable total amount of adjustment change;

[0062] The third calculation unit is used to calculate the clean electricity responsiveness index R according to the following formula clean :

[0063] R clean =-corr(φ fossil (t),P shift (t));

[0064] Where corr(·) is the correlation coefficient function; φ fossil (t) is the proportion of fossil fuel electricity generation in the unit electricity of the factory at time t';

[0065] The fourth calculation unit is used to calculate the actual weighted average fossil electricity ratio according to the following formula:

[0066]

[0067] The fifth calculation unit is used to calculate the weighted load of carbon emission intensity per unit electricity according to the following formula:

[0068]

[0069] Among them, P′ QC (t) is the weighted load of carbon emission intensity per unit electricity at time t.

[0070] The present invention provides a method and system for optimizing clean electricity consumption for factory loads based on a distributed control neural network (DCRNN) and a computerized, automated, and time-sensitive (CATS) (CATS)-based system. The method utilizes the global-local hybrid attention mechanism of the CATS model to accurately model the evolution of power generation structures and output a high-resolution forecast sequence for the proportion of fossil power generation, overcoming the limitations of traditional forecasting models in capturing temporal patterns and providing an accurate dynamic benchmark for carbon emission intensity for load regulation decisions. Subsequently, based on the DCRNN spatiotemporal graph convolution, interruptible load equipment is identified, a process dependency topology network for each device is constructed, the devices are abstracted as graph nodes, and upstream and downstream associated edges based on the production process are established. Leveraging the spatiotemporal joint modeling capabilities of the DCRNN model, deep feature extraction is performed on multi-source data, including equipment load sequences, control state signals, and process parameters. A diffuse convolution operator is introduced to capture the energy transfer characteristics in the process dependency graph. Combined with a knowledge base of equipment operation constraints, a dynamic screening mechanism is established, encompassing boundary conditions such as minimum continuous operating time and process buffer margin. This outputs real-time interruptible load capacity that meets process safety requirements, significantly improving the reliability and refinement of flexible resource identification for industrial loads. The present invention constructs a load regulation model with the ability to dynamically perceive carbon emission intensity through the predicted results of the proportion of fossil fuel power generation and the boundary of interruptible load capacity. A load elasticity control strategy based on the time-varying characteristics of the power generation structure is adopted: flexible load reduction is implemented during periods when the proportion of fossil fuel power generation is high, and the carbon footprint is reduced through the precise allocation of interruptible loads; load utilization is dynamically improved during periods when clean energy output is sufficient, achieving spatiotemporal matching of electricity consumption and low-carbon power generation. The present invention breaks through the limitation of traditional demand response that only considers economic indicators, forms a new load regulation paradigm guided by carbon emission intensity, and significantly improves the cleanliness level of the factory's energy structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0072] Figure 1 A flow chart of a method for optimizing clean electricity consumption of factory loads based on DCRNN and CATS provided in an embodiment of the present invention;

[0073] Figure 2 A schematic structural diagram of a factory load clean electricity optimization system based on DCRNN and CATS provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0075] Example 1

[0076] like Figure 1 As shown, an embodiment of the present invention provides a method for optimizing clean electricity consumption of factory loads based on DCRNN and CATS, including:

[0077] Step 101: Use the CATS model to predict the proportion of fossil fuel power generation in the unit electricity of the factory at the target time.

[0078] This step aims to use the CATS (Context-Aware Transformer Structure) model, combining multi-dimensional contextual information such as the plant's historical load, grid operating status, renewable power output, electricity prices, weather conditions, and time, to predict the fossil fuel generation share of the unit energy of the plant's connected nodes at each moment. Specifically, a context-aware time series input sequence is constructed and embedded into the CATS model, using its global-local hybrid attention mechanism to model long-term trends and sudden changes. The CATS model outputs a continuous sequence of fossil fuel generation share predictions, which is used to guide subsequent load regulation decisions.

[0079] Collect multi-dimensional time series data, including plant load, clean / fossil power output, electricity prices, weather conditions, and time, as input to the CATS model. This data comprehensively reflects the current power system power supply structure and operating environment, providing a data foundation for subsequent fossil fuel share predictions. A multimodal input sequence tensor is constructed, as shown in Formula (1); each x(t') contains multi-dimensional features, as shown in Formula (2).

[0080] X CATS ={x(tk),…,x(t-1)}(1)

[0081] x(t′)=[P lood (t′),G wind (t′),G solar (t′),G cood (t′),G gos (t′),L cood (t′),T(t′),H(t′),P price (t′),D dow (t′),H hour (t′)](2)

[0082] Among them, X CATS is an input feature sequence of length k, used to predict the carbon intensity at time t; x(tk) is the multidimensional feature vector at time tk; P lood (t′) is the factory load at time t'; G wind (t′) is the wind power output at time t'; G solar (t′) is the photovoltaic power generation output at time t'; G cood (t′) is the coal-fired power generation output at time t'; G gos (t′) is the gas-fired power generation output at time t'; L cood (t') is the total load of the entire area at time t'; T(t') is the temperature at time t'; H(t') is the humidity at time t'; P price (t′) is the electricity price at time t'; D dow (t′) is the number of the week in which time t' occurs; H hour (t′) The hour in the day at time t'; t'=tk.

[0083] Different types of input features are uniformly mapped into fixed-dimensional vector representations, preserving temporal order information to form a temporal input tensor suitable for Transformer encoder processing, laying the foundation for context modeling.

[0084] Each input time slice x(t') is embedded through a multilayer perceptron or linear transformation, as shown in formula (3); further forming a sequence matrix, as shown in formula (4):

[0085] e(tk)=Embed(x(tk))(3)

[0086] E=[e(tk),…,e(t-1)](4)

[0087] Where e(tk) is the mapping of the multi-dimensional feature vector at time tk to a d-dimensional embedding vector; Embed(·) is the embedding function; and E is the input sequence matrix composed of the embedding vectors at all times.

[0088] An improved attention mechanism is used to simultaneously focus on short-term fluctuations and long-term trends. External contextual bias information, such as electricity prices and weather conditions, is introduced to improve the model's ability to perceive changes in the proportion of fossil power generation and its prediction accuracy. A global-local hybrid attention mechanism is introduced to capture long-range dependencies and local mutations. The multi-head attention mechanism is shown in Equation (5).

[0089]

[0090] Among them, CATS(E) is the context feature sequence after attention fusion; T represents the matrix transpose; Softmax(·) is the attention weight normalization function; Q is the query vector matrix in the attention mechanism, Q = EW Q ; K is the key vector matrix in the attention mechanism, K = EW K ; d is the scaling factor used to prevent gradient explosion; B is the bias matrix below, used to integrate global background information such as electricity prices and weather; V is the median vector matrix of the attention mechanism, V = EW V ;W Q 、W K and W V Both are learnable projection matrices, both belong to is a learnable projection matrix of d×d dimensions.

[0091] The attention output features are fused and mapped to prediction values ​​through a regression network. The sigmoid function is used to normalize the output, and the current fossil generation percentage per kilowatt-hour is output, providing a basis for subsequent load optimization. The CATS encoder output sequence is average-pooled and then enters the multi-layer perceptron regression layer to obtain the prediction result, as shown in formula (6).

[0092]

[0093] Among them, φ fossil (t) is the proportion of fossil fuel power generation in the unit electricity of the factory at time t'; σ(·) is the Sigmoid function, which is used to compress the output to the interval [0,1]; w2 is the weight matrix of the first multi-layer perceptron; ReLU(·) is the activation function; w1 is the weight matrix of the first multi-layer perceptron; MeanPool(·) is the average pooling of the time series dimension; b1 is the first bias term; b2 is the second bias term.

[0094] Step 102 : Determine the total interruptible load capacity of the plant at the target time using the DCRNN model.

[0095] This step uses the DCRNN (Diffusion Convolutional Recurrent Neural Network) model to identify the interruptibility of various loads from the equipment load data within the factory. First, a process dependency graph between the equipment is established, with each key equipment (such as air compressors, refrigeration systems, and electric drying equipment) as a node in the graph, and the upstream and downstream relationships between them are represented. Then, a time series input is constructed for each device, including load sequence, control status, process characteristics, etc., and spatiotemporal modeling is performed through DCRNN to identify the interruption potential of each device at the current moment. On this basis, combined with constraints such as minimum continuous operating time and process buffer window, the truly interruptible loads are screened out, and the total interruptible power is output to provide a capacity boundary for regulation.

[0096] Construct a process dependency graph for key electrical equipment within the factory, where nodes represent equipment and edges represent the operation sequence or resource coupling relationship of each electrical equipment, providing a structural basis for subsequent graph neural network modeling.

[0097] Assume that the factory has N types of power equipment (such as refrigeration, drying, air compressor, etc.), and establish a directed graph as shown in formula (7).

[0098]

[0099] in, is a set of nodes, each node represents a load device (such as refrigeration system, air compressor, etc.); ε is the process dependency edge between devices (such as "air compressor → drying equipment"), is the adjacency matrix, which represents the strength of topological relationships; This is the equipment process topology diagram within the factory.

[0100] An input sequence is constructed for each device, which includes load history, control status and process information such as operating time, task priority, etc., to fully describe the device behavior and interruption possibility, as shown in formula (8).

[0101] F i (t)=[P i (tk:t-1),c i (tk:t-1),m i (tk:t-1)](8)

[0102] Among them, F i (t) is the multi-source time series input feature of the i-th device at time t; P i (tk:t-1) is the load power sequence of equipment i at the previous t moments; c i (t) is the equipment control status, such as start / stop flag, running time, etc.; m i(t) is the equipment process characteristics, such as temperature, task status, energy storage upper limit, etc.

[0103] Using DCRNN, we simultaneously model the temporal dynamics and graph structure dependencies of equipment, outputting a representation of each equipment's potential interruption capability and capturing the equipment's adjustment space and fluctuation characteristics in the process chain. DCRNN is a time series modeling network on a graph structure, integrating graph convolution (GCN) and gated recurrent unit (GRU), as shown in Equation (9).

[0104]

[0105] in, is the hidden state of device i at time t in the last layer L of the DCRNN model; b is the third bias term; GRU(·) represents the gated recurrent unit, which is used to capture the evolution of temporal features; N(i) is the set of neighboring nodes directly connected to node i; A ij is the edge weight from node i to node j in the adjacency matrix; W (L) is the learnable weight matrix of the Lth layer of graph convolution; is the hidden state of neighbor node j in the L-1 layer at time t.

[0106] Combined with process constraints (such as minimum operating time and downstream waiting time), it is determined whether each device can be safely interrupted, and the interruption power upper limit at time t is output based on the prediction results.

[0107] Finally, the interruption potential of device i at the current moment is estimated as shown in formula (10). And under the premise of considering process constraints, the truly interruptible devices are screened, as shown in formula (11).

[0108]

[0109] The interruptible load potential of device i at time t predicted by the DCRNN model; Slack i (t) is the minimum time window that device i and the downstream device can wait; is the minimum continuous operation limit of device i; ReLU(·) is the activation function; W out is the output mapping weight matrix; T represents the transpose of the matrix.

[0110] All the adjustable loads of the equipment that meet the interruption conditions are aggregated at the current moment to form the interruptible load capacity of the entire plant, which serves as the input of the subsequent adjustment boundary, as shown in formula (12).

[0111]

[0112] Among them, P interruptible(t) is the total interruptible load capacity of the factory at time t; M is the total number of interruptible equipment in the factory; is the actual interruptible load power of device i, which has passed the process logic screening.

[0113] Step 103: Determine the actual power load of the factory at the target time.

[0114] In this step, the sum of the factory's adjustable load and non-adjustable load at time t is taken as the factory's actual power load at time t.

[0115] Step 104, when the total interruptible load capacity of the factory at the target time meets the constraint conditions, the clean electricity responsiveness index and the actual weighted average fossil electricity ratio are determined according to the ratio of fossil fuel power generation in the unit electricity of the factory at the target time, and the weighted load of carbon emission intensity per unit electricity is determined in combination with the actual electricity load.

[0116] After obtaining the fossil fuel power generation share and interruptible load capacity at each moment, this step constructs an optimization model to adjust the plant's adjustable load operation strategy. The goal is to prioritize the use of more electricity during periods with a lower fossil fuel power share, while ensuring process stability, while ensuring constant total energy consumption and avoiding the operation of interruptible equipment during periods with higher carbon intensity. The optimization variable is the change in adjustable load, whose adjustment range is limited by the output of step 102. By minimizing the weighted carbon emission function, the model outputs the final load regulation sequence and provides evaluation indicators such as clean electricity consumption rate, carbon intensity mean, and response synergy to measure the clean benefits of load regulation.

[0117] With the goal of reducing carbon emissions, a weighted load minimization function based on the proportion of fossil power generation is constructed to guide the use of loads during periods with more clean electricity and avoid their use during periods with more fossil electricity, thereby achieving green load time-shifting response.

[0118] Minimize the weighted load of carbon emission intensity per unit electricity throughout the entire operating cycle, as shown in formula (13).

[0119]

[0120] Among them, P QC (t) is the weighted load of carbon emission intensity per unit electricity at time t; T' is the total number of preset moments, i.e., the total optimization duration; φ fossil (t) is the proportion of fossil fuel power generation in the unit electricity of the factory at time t'; P base (t) is the non-adjustable load of the factory at time t, such as basic lighting, constant water supply, etc.; P shift (t) is the adjustable load of the factory at time t (needs to be optimized).

[0121] The adjustable load boundary (determined by the interruptibility) is shown in formula (14), the total energy consumption is conserved (to ensure constant output) is shown in formula (15), and the adjustment smoothness constraint (to avoid frequent starts and stops) is shown in formula (16), ensuring that the optimization results are feasible and stable under the actual production process. The total interruptible load capacity of the constructed factory at the target time meets the following constraints:

[0122]

[0123] in, is the load capacity that the plant can operate in advance at time t; E total is the total power consumption demand of the factory; P shift (t-1) is the adjustable load Δ of the factory at time t-1 max It is the maximum allowable total amount of adjustment change, used to smooth the load adjustment process.

[0124] The adjustable load P at each moment shift (t) is used as the optimization variable. Under the condition of satisfying all constraints, the adjustment curve is solved by the optimization algorithm to minimize the overall weighted carbon intensity.

[0125] The final goal is shown in formula (17).

[0126]

[0127] Among them, P′ QC (t) is the weighted load of carbon emission intensity per unit electricity at time t.

[0128] The output includes the adjusted average carbon intensity as shown in formula (18) and the clean load regulation response correlation as shown in formula (19), which are used to quantify the effect and response performance of the clean load regulation strategy.

[0129]

[0130] R clean =-corr(φ fossil (t),P shift (t))(19)

[0131] in, is the actual weighted average proportion of fossil electricity, which is used to reflect the degree of cleanliness; R clean is the clean electricity responsiveness index; the closer the value is to 1, the more the load tends to operate during the clean electricity period; corr(·) is the correlation coefficient function.

[0132] Step 105 , optimizing the clean electricity consumption of the factory load according to the clean electricity consumption responsiveness index, the actual weighted average fossil electricity proportion, and the weighted load of the carbon emission intensity per unit electricity.

[0133] In this step, flexible load reduction is implemented during periods when fossil fuel power generation accounts for a high proportion, and the carbon footprint is reduced through precise adjustment of interruptible loads; load utilization is dynamically improved during periods when clean energy output is sufficient, achieving temporal and spatial matching of electricity consumption and low-carbon power generation, breaking through the limitation of traditional demand response that only considers economic indicators, forming a new load regulation paradigm guided by carbon emission intensity, and significantly improving the clean level of the factory's energy structure.

[0134] In summary, the factory load clean electricity optimization method based on DCRNN and CATS provided in this embodiment is based on a constructed context-aware time series input system. By integrating multi-dimensional dynamic data such as the historical load of the food processing plant, the operating status of the power grid, renewable output, electricity prices, weather and time, and adopting the global-local hybrid attention mechanism of the CATS model, it can achieve accurate modeling of the evolution law of the power generation structure. Specifically, through the time series feature embedding coding technology, the multi-dimensional heterogeneous data is converted into a context-aware time series input sequence. The CATS model's ability to collaboratively analyze long-term trend characteristics and short-term fluctuation characteristics is utilized to output a high-resolution fossil power generation share prediction sequence. This embodiment breaks through the limitations of traditional prediction models in capturing time series patterns and provides an accurate dynamic benchmark for carbon emission intensity for load regulation decisions. Then, a process-dependent topological network of the equipment is constructed, and key equipment such as air compressors, refrigeration systems, and electric drying equipment are abstracted as graph nodes, and upstream and downstream association edges based on the production process are established. Leveraging the spatiotemporal joint modeling capabilities of the DCRNN model, deep feature extraction is performed on multi-source data, including equipment load sequences, control state signals, and process parameters. The diffusion convolution operator is introduced to capture the energy transfer characteristics within the process dependency graph. Combined with a knowledge base of equipment operation constraints, a dynamic screening mechanism is established, incorporating boundary conditions such as minimum continuous operating time and process buffer margin. This output outputs real-time interruptible load capacity that meets process safety requirements, significantly improving the reliability and refinement of flexible resource identification for industrial loads. Finally, a load regulation model with dynamic carbon emission intensity awareness is constructed based on the predicted fossil fuel generation share and the interruptible load capacity boundary. A load flexibility control strategy based on the time-varying characteristics of the power generation structure is developed: flexible load reduction is implemented during periods of high fossil fuel generation, reducing the carbon footprint through the precise deployment of interruptible loads. Load utilization is dynamically increased during periods of abundant clean energy output, achieving spatiotemporal matching of electricity consumption with low-carbon generation. This breaks through the limitations of traditional demand response, which only considers economic indicators, and forms a new load regulation paradigm guided by carbon emission intensity, significantly improving the cleanliness of the factory energy structure.

[0135] Example 2

[0136] Based on the same inventive concept as Example 1, this embodiment also provides a factory load clean electricity optimization system based on DCRNN and CATS. Since the principle of solving the problem by this system is similar to the aforementioned factory load clean electricity optimization method based on DCRNN and CATS, the implementation of this system can refer to the implementation of the factory load clean electricity optimization method based on DCRNN and CATS.

[0137] like Figure 2 As shown in the figure, the factory load clean electricity optimization system based on DCRNN and CATS includes:

[0138] The prediction module 10 is used to predict the proportion of fossil fuel power generation in the unit electricity of the factory at the target time by using the CATS model.

[0139] The first determination module 20 is configured to determine the total interruptible load capacity of the plant at a target time using a DCRNN model.

[0140] The second determining module 30 is configured to determine the actual power load of the factory at the target time.

[0141] The third determination module 40 is used to determine the clean electricity responsiveness index and the actual weighted average fossil electricity proportion according to the proportion of fossil fuel power generation in the unit electricity of the factory at the target time, when the total interruptible load capacity of the factory at the target time meets the constraint conditions, and determine the weighted load of carbon emission intensity per unit electricity in combination with the actual electricity load.

[0142] The optimization module 50 is used to optimize the clean electricity consumption of the factory load according to the clean electricity consumption responsiveness index, the actual weighted average fossil electricity proportion and the weighted load of the carbon emission intensity per unit electricity.

[0143] Exemplarily, the prediction module includes:

[0144] The first calculation unit is used to calculate the proportion of fossil fuel power generation in the unit electricity of the factory at the target time according to the following formula:

[0145]

[0146] E=[e(tk),…,e(t-1)];

[0147] e(tk)=Embed(x(tk));

[0148] x(t′)=[P load (t′), G wind (t′), G solar (t′), G cood (t′), G gos (t′), L cood(t′), T(t′), H(t′), P price (t′), D dow (t′), H hour (t′)];

[0149] t′=tk;

[0150] Among them, φ fossil (t) is the proportion of fossil fuel power generation in the unit electricity of the factory at time t'; σ(·) is the Sigmoid function; w2 is the weight matrix of the first multi-layer perceptron; ReLU(·) is the activation function; w1 is the weight matrix of the first multi-layer perceptron; MeanPool(·) is the average pooling of the time series dimension; CATS(E) is the context feature sequence after attention fusion; b1 is the first bias term; b2 is the second bias term; T represents matrix transpose; Softmax(·) is the attention weight normalization function; Q is the query vector matrix in the attention mechanism; K is the key vector matrix in the attention mechanism; d is the scaling factor used to prevent gradient explosion; B is the bias matrix below, used to fuse global background information; V is the median vector matrix of the attention mechanism; E is the input sequence matrix composed of the embedding vectors at all times; e(tk) is the multi-dimensional feature vector at time tk mapped to the d-dimensional embedding vector; Embed(·) is the embedding function; x(tk) is the multi-dimensional feature vector at time tk; P lood (t′) is the factory load at time t'; G wind (t′) is the wind power output at time t'; G solar (t′) is the photovoltaic power generation output at time t'; G cood (t′) is the coal-fired power generation output at time t'; G gos (t′) is the gas-fired power generation output at time t'; L cood (t') is the total load of the entire area at time t'; T(t') is the temperature at time t'; H(t') is the humidity at time t'; P price (t′) is the electricity price at time t'; D dow (t′) is the number of the week in which time t' occurs; H hour (t′) The hour in the day at time t'.

[0151] Exemplarily, the first determining module includes:

[0152] The first construction unit is used to construct a process dependency graph of electrical equipment within the factory, wherein nodes represent electrical equipment, and edges represent the operation sequence or resource coupling relationship of each electrical equipment.

[0153] The second calculation unit is used to calculate the total interruptible load capacity of the plant at the target time according to the following formula:

[0154]

[0155] Among them, P interruptible (t) is the total interruptible load capacity of the factory at time t; M is the total number of interruptible equipment in the factory; is the actual interruptible load power of equipment i; γ i (t) is the interruption allowable factor under the process constraints of equipment i; The interruptible load potential of device i at time t predicted by the DCRNN model; Slack i (t) is the minimum time window that device i and the downstream device can wait; is the minimum continuous operation limit of device i; ReLU(·) is the activation function; W out is the output mapping weight matrix; T represents the transpose of the matrix; is the hidden state of device i at time t in the last layer of the DCRNN model; b is the third bias term; GRU(·) represents the gated recurrent unit, which is used to capture the evolution of temporal features; N(i) is the set of neighboring nodes directly connected to node i; A ij is the edge weight from node i to node j in the adjacency matrix; W (L) is the learnable weight matrix of the Lth layer of graph convolution; is the hidden state of neighbor node j in the L-1 layer at time t.

[0156] Exemplarily, the second determining module includes:

[0157] The determination unit is used to take the sum of the adjustable load and the non-adjustable load of the factory at time t as the actual power load of the factory at time t.

[0158] Exemplarily, the third determining module includes:

[0159] The second building block is used to build the constraints that the total interruptible load capacity of the plant at the target time satisfies:

[0160]

[0161] Among them, P interruptible (t) is the total interruptible load capacity of the plant at time t; P shift (t) is the adjustable load of the plant at time t; is the load capacity that the plant can operate in advance at time t; T' is the total number of preset times; P load (t) is the non-adjustable load of the factory at time t; P base (t) is the non-adjustable load of the factory at time t; E total is the total power consumption demand of the factory; P shift (t-1) is the adjustable load of the factory at time t-1; Δmax is the maximum allowable total adjustment change.

[0162] The third calculation unit is used to calculate the clean electricity responsiveness index R according to the following formula clean :

[0163] R clean =-corr(φ fossil (t),P shift (t));

[0164] Where corr(·) is the correlation coefficient function; φ fossil (t) is the proportion of fossil fuel power generation in the unit electricity of the factory at time t'.

[0165] The fourth calculation unit is used to calculate the actual weighted average fossil electricity ratio according to the following formula:

[0166]

[0167] The fifth calculation unit is used to calculate the weighted load of carbon emission intensity per unit electricity according to the following formula:

[0168]

[0169] Among them, P′ QC (t) is the weighted load of carbon emission intensity per unit electricity at time t.

[0170] For more specific working processes of the above modules, please refer to the corresponding content disclosed in Example 1, which will not be repeated here.

[0171] Example 3

[0172] This embodiment provides a computer device, including a processor and a memory; wherein, when the processor executes the computer program stored in the memory, the steps of the factory load clean electricity optimization method based on DCRNN and CATS described in Example 1 are implemented.

[0173] For more specific details about the above method, please refer to the corresponding content disclosed in Example 1, which will not be repeated here.

[0174] Example 4

[0175] This embodiment provides a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, the steps of the factory load clean electricity optimization method based on DCRNN and CATS described in Example 1 are implemented.

[0176] For more specific details about the above method, please refer to the corresponding content disclosed in Example 1, which will not be repeated here.

[0177] Example 5

[0178] This embodiment provides a computer program product, including computer-executable instructions or a computer program. When the computer-executable instructions or the computer program are executed by a processor, the steps of the factory load clean electricity optimization method based on DCRNN and CATS described in Example 1 are implemented.

[0179] For more specific details about the above method, please refer to the corresponding content disclosed in Example 1, which will not be repeated here.

[0180] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. References to the same or similar parts between the various embodiments will be sufficient. The systems, devices, storage media, and computer program products disclosed in the embodiments correspond to the methods disclosed in the embodiments, so their descriptions are relatively simplified. For relevant details, refer to the method descriptions.

[0181] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus a necessary general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments of the present invention or certain portions of the embodiments.

[0182] In some embodiments, computer-executable instructions may be in the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0183] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinating files (e.g., files storing one or more modules, subroutines, or code portions).

[0184] By way of example, computer-executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed across multiple sites and interconnected by a communication network.

[0185] The present invention has been described in detail above with reference to specific embodiments and exemplary examples. However, these descriptions should not be construed as limiting the present invention. Those skilled in the art will appreciate that various equivalent substitutions, modifications, or improvements may be made to the technical solutions and implementations of the present invention without departing from the spirit and scope of the present invention, all of which fall within the scope of the present invention. The scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A factory load clean electricity optimization method based on DCRNN and CATS, characterized in that: include: Use the CATS model to predict the proportion of fossil fuel generation per unit of electricity produced by the plant at the target time; The DCRNN model is used to determine the total interruptible load capacity of the plant at the target time; Determine the actual power load of the factory at the target time; Under the condition that the total interruptible load capacity of the factory at the target time meets the constraints, the clean electricity responsiveness index and the actual weighted average fossil electricity ratio are determined according to the proportion of fossil fuel power generation in the unit electricity of the factory at the target time, and the weighted load of carbon emission intensity per unit electricity is determined in combination with the actual electricity load; The clean electricity consumption of factory load is optimized according to the weighted load of clean electricity responsiveness index, actual weighted average proportion of fossil electricity and carbon emission intensity per unit electricity.

2. The factory load clean electricity optimization method according to claim 1, characterized in that: The CATS model is used to predict the proportion of fossil fuel power generation in the unit electricity of the factory at the target time, including: The percentage of fossil fuel electricity generated per unit of electricity consumed by the plant at the target time is calculated using the following formula: E=[e(tk),…,e(t-1)]; e(tk)=Embed(x(tk)); x(t′)=[P lood (t′),G wind (t′),G solar (t′),G cood (t′),G gos (t′),L cood (t′),T(t′),H(t′),P price (t′),D dow (t′),H hour (t′)]; t′=tk; Among them, φ fossil (t) is the proportion of fossil fuel power generation in the unit electricity of the factory at time t'; σ(·) is the Sigmoid function; w2 is the weight matrix of the first multi-layer perceptron; ReLU(·) is the activation function; w1 is the weight matrix of the first multi-layer perceptron; MeanPool(·) is the average pooling of the time series dimension; CATS(E) is the context feature sequence after attention fusion; b1 is the first bias term; b2 is the second bias term; T represents matrix transpose; Softmax(·) is the attention weight normalization function; Q is the query vector matrix in the attention mechanism; K is the key vector matrix in the attention mechanism; d is the scaling factor used to prevent gradient explosion; B is the bias matrix below, used to fuse global background information; V is the median vector matrix of the attention mechanism; E is the input sequence matrix composed of the embedding vectors at all times; e(tk) is the multi-dimensional feature vector at time tk mapped to the d-dimensional embedding vector; Embed(·) is the embedding function; x(tk) is the multi-dimensional feature vector at time tk; P lood (t′) is the factory load at time t'; G wind (t′) is the wind power output at time t'; G solar (t′) is the photovoltaic power generation output at time t'; G cood (t′) is the coal-fired power generation output at time t'; G gos (t′) is the gas-fired power generation output at time t'; L cood (t') is the total load of the entire area at time t'; T(t') is the temperature at time t'; H(t') is the humidity at time t'; P price (t′) is the electricity price at time t'; D dow (t′) is the number of the week in which time t' occurs; H hour (t′) The hour in the day at time t'.

3. The factory load clean electricity optimization method according to claim 1, characterized in that: The DCRNN model is used to determine the total interruptible load capacity of the plant at the target time, including: Construct a process dependency graph for electrical equipment within the factory; nodes represent electrical equipment, and edges represent the order in which each electrical equipment operates or the resource coupling relationship; Calculate the total interruptible load capacity of the plant at the target time using the following formula: Among them, P interruptible (t) is the total interruptible load capacity of the factory at time t; M is the total number of interruptible equipment in the factory; is the actual interruptible load power of equipment i; γ i (t) is the interruption allowable factor under the process constraints of equipment i; The interruptible load potential of device i at time t predicted by the DCRNN model; Slack i (t) is the minimum time window that device i and the downstream device can wait; is the minimum continuous operation limit of device i; ReLU(·) is the activation function; W out is the output mapping weight matrix; T represents the transpose of the matrix; is the hidden state of device i at time t in the last layer of the DCRNN model; b is the third bias term; GRU(·) represents the gated recurrent unit, which is used to capture the evolution of temporal features; N(i) is the set of neighboring nodes directly connected to node i; A ij is the edge weight from node i to node j in the adjacency matrix; W (L) is the learnable weight matrix of the Lth layer of graph convolution; is the hidden state of neighbor node j in the L-1 layer at time t.

4. The method for optimizing clean electricity consumption of factory loads according to claim 1, characterized in that: Determining the actual power load of the factory at the target time includes: The sum of the factory's adjustable load and non-adjustable load at time t is taken as the factory's actual power load at time t.

5. The method for optimizing clean electricity consumption of factory load according to claim 4, characterized in that: In the case where the total interruptible load capacity of the factory at the target time meets the constraint conditions, the clean electricity responsiveness index and the actual weighted average fossil electricity ratio are determined according to the fossil fuel power generation ratio in the unit electricity of the factory at the target time, and the weighted load of the carbon emission intensity per unit electricity is determined in combination with the actual electricity load, including: Construct the constraints that the total interruptible load capacity of the plant at the target time satisfies: Among them, P interruptible (t) is the total interruptible load capacity of the plant at time t; P shift (t) is the adjustable load of the plant at time t; is the load capacity that the plant can operate in advance at time t; T' is the total number of preset times; P load (t) is the non-adjustable load of the factory at time t; P base (t) is the non-adjustable load of the factory at time t; E total is the total power consumption demand of the factory; P shift (t-1) is the adjustable load of the factory at time t-1; Δ max is the maximum allowable total amount of adjustment change; The clean electricity responsiveness index R is calculated according to the following formula clean : R clean =-corr(φ fossil (t),P shift (t)); Where corr(·) is the correlation coefficient function; φ fossil (t) is the proportion of fossil fuel electricity generation in the unit electricity of the factory at time t'; The actual weighted average fossil electricity share is calculated using the following formula: The weighted load of carbon emission intensity per unit of electricity is calculated according to the following formula: Among them, P′ QC (t) is the weighted load of carbon emission intensity per unit electricity at time t.

6. A factory load clean electricity optimization system based on DCRNN and CATS, characterized by: include: A forecasting module, which uses the CATS model to predict the percentage of fossil fuel-based electricity generated per unit of power at the plant at a target time. A first determination module is used to determine the total interruptible load capacity of the plant at the target time using a DCRNN model; The second determination module is used to determine the actual power load of the factory at the target time; The third determination module is used to determine the clean electricity responsiveness index and the actual weighted average fossil electricity ratio based on the fossil fuel power generation ratio per unit of electricity consumed by the plant at the target time, provided that the total interruptible load capacity of the plant at the target time meets the constraint conditions, and to determine the weighted load of carbon emission intensity per unit of electricity consumed based on the actual electricity load; The optimization module is used to optimize the clean electricity consumption of factory loads based on the clean electricity responsiveness index, the actual weighted average fossil electricity share, and the weighted load of carbon emission intensity per unit of electricity.

7. The factory load clean electricity optimization system according to claim 6, characterized in that: The prediction module includes: The first calculation unit is used to calculate the proportion of fossil fuel power generation in the unit electricity of the factory at the target time according to the following formula: E=[e(tk),…,e(t-1)]; e(tk)=Embed(x(tk)); x(t′)=[P lood (t′),G wind (t′),G solar (t′),G cood (t′),G gos (t′),L cood (t′),T(t′),H(t'),P price (t'),D dow (t'′),H hour (t′)]; t′=tk; Among them, φ fossil (t) is the proportion of fossil fuel power generation in the unit electricity of the factory at time t'; σ(·) is the Sigmoid function; w2 is the weight matrix of the first multi-layer perceptron; ReLU(·) is the activation function; w1 is the weight matrix of the first multi-layer perceptron; MeanPool(·) is the average pooling of the time series dimension; CATS(E) is the context feature sequence after attention fusion; b1 is the first bias term; b2 is the second bias term; T represents matrix transpose; Softmax(·) is the attention weight normalization function; Q is the query vector matrix in the attention mechanism; K is the key vector matrix in the attention mechanism; d is the scaling factor used to prevent gradient explosion; B is the bias matrix below, used to fuse global background information; V is the median vector matrix of the attention mechanism; E is the input sequence matrix composed of the embedding vectors at all times; e(tk) is the multi-dimensional feature vector at time tk mapped to the d-dimensional embedding vector; Embed(·) is the embedding function; x(tk) is the multi-dimensional feature vector at time tk; P lood (t′) is the factory load at time t'; G wind (t′) is the wind power output at time t'; G solar (t′) is the photovoltaic power generation output at time t'; G cood (t′) is the coal-fired power generation output at time t'; G gos (t′) is the gas-fired power generation output at time t'; L cood (t') is the total load of the entire area at time t'; T(t') is the temperature at time t'; H(t') is the humidity at time t'; P price (t′) is the electricity price at time t'; D dow (t′) is the number of the week in which time t' occurs; H hour (t′) The hour in the day at time t'.

8. The factory load clean electricity optimization system according to claim 6, characterized in that: The first determining module includes: The first construction unit is used to construct a process dependency graph of electrical equipment within the factory, wherein nodes represent electrical equipment and edges represent the operation sequence or resource coupling relationship of each electrical equipment; The second calculation unit is used to calculate the total interruptible load capacity of the plant at the target time according to the following formula: Among them, P interruptible (t) is the total interruptible load capacity of the factory at time t; M is the total number of interruptible equipment in the factory; is the actual interruptible load power of equipment i; γ i (t) is the interruption allowable factor under the process constraints of equipment i; The interruptible load potential of device i at time t predicted by the DCRNN model; Slack i (t) is the minimum time window that device i and the downstream device can wait; is the minimum continuous operation limit of device i; ReLU(·) is the activation function; W out is the output mapping weight matrix; T represents the transpose of the matrix; is the hidden state of device i at time t in the last layer of the DCRNN model; b is the third bias term; GRU(·) represents the gated recurrent unit, which is used to capture the evolution of temporal features; N(i) is the set of neighboring nodes directly connected to node i; A ij is the edge weight from node i to node j in the adjacency matrix; W (L) is the learnable weight matrix of the Lth layer of graph convolution; is the hidden state of neighbor node j in the L-1 layer at time t.

9. The factory load clean electricity optimization system according to claim 6, characterized in that: The second determining module includes: The determination unit is used to take the sum of the adjustable load and the non-adjustable load of the factory at time t as the actual power load of the factory at time t.

10. The factory load clean electricity optimization system according to claim 9, characterized in that: The third determining module includes: The second building block is used to build the constraints that the total interruptible load capacity of the plant at the target time satisfies: Among them, P interruptible (t) is the total interruptible load capacity of the plant at time t; P shift (t) is the adjustable load of the plant at time t; is the load capacity that the plant can operate in advance at time t; T' is the total number of preset times; P load (t) is the non-adjustable load of the factory at time t; P base (t) is the non-adjustable load of the factory at time t; E total is the total power consumption demand of the factory; P shift (t-1) is the adjustable load of the factory at time t-1; Δ max is the maximum allowable total amount of adjustment change; The third calculation unit is used to calculate the clean electricity responsiveness index R according to the following formula clean : R clean =-corr(φ fossil (t),P shift (t)); Where corr(·) is the correlation coefficient function; φ fossil (t) is the proportion of fossil fuel electricity generation in the unit electricity of the factory at time t'; The fourth calculation unit is used to calculate the actual weighted average fossil electricity ratio according to the following formula: The fifth calculation unit is used to calculate the weighted load of carbon emission intensity per unit electricity according to the following formula: Among them, P′ QC (t) is the weighted load of carbon emission intensity per unit electricity at time t.

Images