An industry typical user power demand response potential assessment method

Abstract

The application provides a kind of industry typical user electric power demand response potential evaluation method, this method includes: S1, based on integrated empirical mode load decomposition and demand response willingness extraction industry user demand response characteristic;S2, based on industry user demand response characteristic and convolution self-attention mechanism realizes the evaluation of industry typical user demand response potential. The effectiveness of the method is verified by analyzing the demand response case of metal processing machinery manufacturing industry in a city of Zhejiang Province, China. The results show that the demand response potential evaluation results obtained based on the method are consistent with the actual response amount of the user. The application overcomes the modal aliasing problem, can capture the long-term and short-term dependence in time series, and considers the local context information of each period feature, which can ensure high accuracy and provide support for power supply companies or load aggregators to fully utilize user-side adjustable resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power demand response, in particular to a method for evaluating power demand response potential of industry typical users. BACKGROUND

[0002] Developing clean new energy such as wind power and photovoltaic power has become the only way for China's energy transformation. However, due to the randomness and uncontrollability of new energy, it brings great risk to the safe and stable operation of the power grid and puts forward higher requirements for the regulation capacity of the power grid. Demand response can overcome the uncertainty brought by high penetration of new energy by encouraging users to actively change their power consumption behavior, and is an effective means to improve the regulation capacity of the power grid. The power selling company clusters industry users, extracts typical power consumption types, and matches the demand response indicators to be completed according to the demand response potential of each typical power consumption type, and uses various incentive strategies to launch demand response invitations to industry users of the corresponding power consumption type. Users can decide whether to participate in demand response according to their load characteristics and demand response willingness. However, the existing method cannot extract both the objective demand response characteristics and the subjective demand response willingness characteristics from the load data of industry users, so the accuracy of the existing method for evaluating the typical demand response potential of industry users is not high. SUMMARY

[0003] The purpose of the present application is to provide a method for evaluating the power demand response potential of industry typical users. The method first uses the integrated empirical mode decomposition method to decompose the user load sequence, overcoming the modal aliasing problem. The resulting intrinsic mode function component can accurately represent the characteristics of the load sequence, effectively improving the accuracy of the objective demand response characteristic sequence extraction. Then, based on the demand response characteristics of industry users, the convolution self-attention is used to establish the mapping relationship between the demand response characteristics and the demand response potential of the users. Not only can it capture the long-term and short-term dependencies in the time series, but also consider the local context information of each period of characteristics, which can better be incorporated into the multi-head self-attention mechanism, more effectively extract the demand response potential characteristics of the users, and improve the accuracy of the evaluation results, providing a new idea for the precise scheduling of power selling companies, and having good practical application value.

[0004] The present application specifically adopts the following technical solutions to achieve the above purposes:

[0005] A method for evaluating the power demand response potential of industry typical users, comprising the following steps:

[0006] S1, extracting demand response characteristics of industry users based on load decomposition results and demand response willingness;

[0007] S2, evaluating the demand response potential of industry typical users based on the demand response characteristics of industry users and the convolution self-attention mechanism.

[0008] S1, comprises:

[0009] S11, based on the integrated empirical mode decomposition method, the load sequence of the industry typical user is decomposed, and the index vector reflecting the objective demand response characteristics of the industry typical user is extracted based on the load decomposition result;

[0010] The power consumption types of each industry and the power consumption characteristics of each user in the industry are quite different, and in-depth analysis of the load change characteristics of industry users and accurate grasp of the change law of typical load are the premise of accurate evaluation of the demand response potential of typical users in the industry. The demand response potential of the typical user in the industry refers to the demand response potential corresponding to the load of the typical user in the industry, which represents the average demand response ability of the corresponding power consumption type. Therefore, in order to accurately evaluate the demand response potential of the typical user in the industry, it is necessary to first divide the power consumption types of the industry and identify the typical load curve representing different power consumption types. Considering that the normal cloud model can fully mine the local dynamic characteristics of the load curve, and the improved density peak value fast clustering algorithm has the characteristics of less preset parameters, high calculation efficiency, good adaptability to data distribution, but it is difficult to accurately determine the clustering center, the present application intends to divide the power consumption types of the industry based on the normal cloud model and the improved density peak value fast clustering algorithm, and identify the typical load curve. Since the focus of the present application is the evaluation of the typical demand response potential of the industry user, the steps of dividing the power consumption types of the industry and identifying the typical load curve will not be repeated here.

[0011] In practice, the typical load of the industry user can be regarded as the superposition of different components, and the influence of different components on the load change characteristics is different, so it is necessary to decompose the typical load sequence of the industry user and analyze the component representing the typical demand response characteristics of the industry. The use of empirical mode decomposition method to decompose the load sequence of the industry user can obtain intrinsic mode function and residual term which can represent different local characteristics of the original load sequence, but the empirical mode decomposition method has the problem of mode aliasing, which makes the decomposition result produce false intrinsic mode function component, so that the intrinsic mode function cannot accurately represent the characteristics of the load sequence. Considering that the integrated empirical mode decomposition method can overcome the shortcomings of the traditional empirical mode decomposition method by adding white noise, the present application intends to propose an industry user load sequence decomposition method based on the integrated empirical mode decomposition method, and extract the sequence representing the objective demand response characteristics of the user.

[0012] Based on the normal cloud model and the improved density peak value fast clustering algorithm, an industry is divided into M power consumption types, represented as a set C T ={1, 2, …, m, …, M}, wherein the identification result of the typical load curve of the mth power consumption type is represented as P m ={P m,1 ,P m,2 ,…,Pm,j ,…,P m,J}, P m,j Let J represent the load at the j-th moment of the typical load curve m, where J is the total number of sampling points. Typically, J = 96, meaning the sampling frequency is 15 min / time. The set of users in the m-th electricity consumption type within the industry is... This represents the b-th user in the m-th electricity consumption type. The load sequence is in, Indicates user The load at time j. The typical load sequence P is analyzed using the integrated empirical mode decomposition method. m The decomposition yields the intrinsic mode function components and remainder terms as follows:

[0013]

[0014] Among them, G IMF This represents the total number of intrinsic mode functions; The remainder is called the trend component of the typical load sequence m, which is the deterministic part and characterizes the overall trend of change in the typical load of industry users; For typical load sequence P m The basic idea behind obtaining the g-th eigenmode function through integrated empirical mode decomposition is as follows: 1) Initialize the count variables g = 1 and k = 1; 2) In the original typical load sequence P m Adding the kth type of Gaussian white noise to obtain 3) Calculation All local maxima and minima are obtained. upper envelope and lower envelope 4) According to Sure Let the g-th eigenmode function be... And g = g + 1; 5) If If it is not a monotonic function, repeat steps 3) and 4); otherwise, calculate the residue. Simultaneously, let k = k + 1; 6) If k is less than the given value of K, then use different Gaussian white noise ξ k Repeat steps 2)-5); 7) Based on the above steps, calculate the typical load sequence P. m Decomposed into The final eigenmode functions and corresponding residues are obtained as follows: and Corresponding to typical load sequences P m The periodic component and trend component are given. The periodic component consists of eigenmode function components with different fluctuation frequencies, namely high-frequency component, mid-frequency component and low-frequency component.

[0015] The high-frequency component IMF1 of a typical load sequence, i.e. The fluctuation frequency is large, but the amplitude is small, reflecting the impact of random factors on the electricity consumption of industry users, and having a relatively small impact on the typical demand response potential; the low- and medium-frequency components of the typical load sequence Right now The fluctuation frequency is lower than that of the high-frequency component, and the fluctuation amplitude is larger, reflecting the electricity consumption habits of industry users and having a greater impact on the typical demand response potential. Therefore, this invention intends to extract the objective demand response characteristics of industry users based on ignoring the high-frequency component of the load sequence, as shown in Equation (2).

[0016]

[0017] in, The remaining periodic component is the original typical load sequence after removing trend and high-frequency components. To further reduce the impact of sampling noise on the load sequence, considering that SG filtering can ensure that the signal shape and width remain unchanged while filtering out noise, this invention proposes to use SG filtering for the remaining periodic component. Smoothing filters are applied to improve the smoothness of the remaining periodic sequences and reduce noise interference.

[0018] In actual production, depending on the differences in production processes and procedures, the remaining cycle components obtained above can be divided into time periods. Divided into S segments, as shown in the appendix. Figure 1 As shown, where t1+t2+…t s +…+t S =24. The switching start-stop of combined power equipment for industry users corresponds to the rise and fall of the remaining cycle component in each time period, which is called a step. It represents the demand response potential of industry users in that time period, which is either valley filling or peak shaving, and is expressed as:

[0019]

[0020] in, This represents the average value of the remaining periodic component s segment of a typical load curve m. The steps representing the remaining periodic component of the typical industry load curve at hour θ are called the step sequence. When... When, it indicates that users in this industry have the potential for peak-shaving demand response in the θth hour; when The value at time θ indicates that users in this industry have the potential to fill the demand gap in the θth hour.

[0021] In summary, the objective demand response characteristic vector of a typical load m It can be represented as:

[0022]

[0023] in, The demand response characteristic sequence for industry users, i.e.

[0024] S12, based on survey statistics such as users' historical response electricity consumption, number of effective responses, and contracted capacity, establish an indicator system that reflects users' willingness to respond to their needs and extract users' subjective needs response characteristics.

[0025] The demand response potential of typical users in an industry is not only related to the changing trends of typical load sequences, remaining cycle sequences, and step sequences, but also requires comprehensive consideration of the demand response willingness of each user in each time period within the industry's electricity consumption type. Based on historical actual response electricity volume, effective response frequency, contracted capacity, and user quotations for each user in the industry's electricity consumption type, subjective demand response characteristics that characterize the industry users' demand response willingness are extracted, namely, the ratio of actual response electricity volume, the ratio of effective response frequency, the ratio of invited response capacity, and the ratio of demand response subsidy electricity price.

[0026] (1) Actual response power ratio

[0027]

[0028] in, The ratio of actual response power consumption in the θth hour of typical load m represents the average ratio of the actual response power consumption to the reported response power consumption of each user in industry power consumption type m during that period. and users respectively Actual response power (kW) and declared response power (kW) for hour θ. Actual response power ratio for industry users. The larger the value, the stronger the user's willingness to participate in demand response, and the greater the demand response potential.

[0029] (2) Effective response rate

[0030]

[0031] in, N represents the ratio of effective response counts for a typical load m in hour θ, indicating the average ratio of the historical effective response counts to the total number of invitations for each user in industry electricity consumption type m during that period; iv (b, θ) represents the electricity sales company's transaction with the user in the θth hour. Total number of invitations; n i These are 0-1 variables, representing users. Is the response valid in the i-th iteration? If it is valid, then n i The value is 1; otherwise, ni The value is 0. The ratio of valid responses from industry users. The larger the value, the stronger the user's willingness to participate in demand response, and the greater the demand response potential.

[0032] (3) Contract response capacity ratio

[0033]

[0034] in, Let Q be the contracted response capacity ratio for a typical load m at hour θ, representing the average ratio of the contracted demand response capacity to the user's receiving capacity for each user in industry electricity consumption type m at hour θ; contr (b,θ) and Q cap (b) respectively users Contracted capacity (kWh) and total received capacity (kWh) for hour θ. Contracted capacity ratio for industry users. The larger the value, the stronger its willingness to participate in demand response, and the greater its demand response potential.

[0035] (4) DR subsidy unit price ratio

[0036]

[0037] in, p represents the ratio of the demand response subsidy unit price to the average ratio of the subsidy unit price submitted by each user in electricity consumption type m in the demand response declaration at hour θ to the upper limit of the declared price for subsidy unit price in the electricity sales company's demand response policy; cus (b,θ) and p hat (θ) represents the user at the θth hour. Price quotes (RMB / kWh) and the upper limit of electricity sales company subsidies (RMB / kWh). The higher the demand response subsidy per unit price ratio, the weaker the willingness of users to participate in demand response, and the smaller the demand response potential.

[0038] In summary, the subjective demand response characteristic vector of a typical load m is:

[0039]

[0040] Based on equations (4) and (9), the characteristic vector of subjective and objective demand response of typical load m is obtained as follows:

[0041] Specifically, S2 is:

[0042] Based on the extracted demand response characteristics of typical industry users, the demand response potential of typical industry users is assessed using multi-head self-attention.

[0043] To obtain accurate assessment results of the demand response potential of typical industry users, this invention establishes a mapping relationship between demand response features and demand response potential based on convolutional self-attention. The demand response feature vectors of industry users exhibit significant temporal characteristics, and their demand response potential also varies across different time periods. Considering that the self-attention mechanism can learn the dependencies between features in each time period of the demand response feature sequence and encode them into a vector, and that demand response features from different time periods can be weighted using different attention heads to achieve the fusion of demand response features from different time periods, exhibiting good generalization ability, assessing the demand response potential of typical industry users based on a multi-head self-attention mechanism can improve the accuracy of the assessment results. Taking a multi-head self-attention model consisting of two encoder layers and two decoder layers as an example, the basic principle is as follows: Figure 2 As shown. Here, B0 is the start symbol for decoder #1, and the input is R. m,θ The demand response feature vector representing typical load m in time period θ is used to obtain the demand response potential of typical load m in time period θ through iterative encoder-decoder operation. Appendix Figure 2 The encoder in the model includes position encoding, multi-head self-attention, residual connections and layer normalization, and a fully connected feedforward network; the decoder, in addition to the above modules of the encoder, also has an encoder-decoder attention module, whose input comes not only from the output of the residual connection and layer normalization modules of its previous layer, but also from the output of the last layer encoder.

[0044] (1) Location coding of demand response characteristics

[0045] The demand response potential of typical users in an industry varies across different time periods, depending on the demand response characteristics of the input at different time periods. Therefore, assessing the demand response potential of typical users in an industry requires considering the relative order of demand response characteristics across different time periods. Thus, each time period can be viewed as a position in a demand response characteristic sequence, and the input demand response characteristics can be positionally encoded, as shown in Equation (10). This allows the multi-head self-attention model to better identify the relative positional relationships of different time periods within the demand response characteristic sequence, improving the model's understanding of the demand response characteristic sequence and the accuracy of the demand response potential assessment results.

[0046] The basic idea of ​​location encoding for demand response features is as follows: ① Let the number of time periods contained in the demand response feature sequence be N, that is, there are N locations that need to be encoded, represented as p. os =1,2,…,n,…,N, representing the position of the demand response feature in the demand response feature sequence; ② Let the dimension of the position embedding vector of the demand response feature be d, and calculate ③ Determine p os = The positional encoding corresponding to n:

[0047] P n E =[sin(nω1), cos(nω2), sin(nω3),…] (10)

[0048] When the vector dimension d is odd, the last term of equation (10) is sin(nω) d When the vector dimension d is even, the last term of equation (10) is cos(nω). d ).

[0049] The aforementioned location encoding method ensures that the distance between the encoding vectors at different locations in the demand response feature sequence is equal, thereby enabling the demand response potential assessment model to identify the relative positional relationships between different locations and thus better process the demand response feature sequence data.

[0050] (2) Multi-headed self-attention layer for demand response potential assessment

[0051] The assessment of demand response potential depends not only on the demand response characteristics of the corresponding time period, but also on the demand response characteristics of adjacent time periods. (Appendix) Figure 2 The multi-head self-attention layer in the demand response potential assessment model shown in the figure obtains the relationship between demand response features in different time periods in the input demand response feature sequence by performing self-attention calculation on the demand response feature sequence. That is, it not only focuses on the demand response features of adjacent time periods, but also on the relationship between demand response features of any two time periods, thereby enhancing the representation ability of the demand response feature sequence.

[0052] In multi-head self-attention, "multi-head" refers to learning different features and relationships from different "perspectives" within the input demand response feature sequence, such as time-period features, peak-shaving or valley-filling categories, and the confidence level of demand response potential assessment results. The multi-head self-attention mechanism linearly maps the input demand response feature sequence, then divides it into multiple "heads." Each head performs attention calculations on the mapped input demand response feature sequence and learns different weight assignments. The outputs of the multiple attention heads are then concatenated to form the final multi-head attention representation.

[0053] The basic principles of multi-head self-attention are as follows: Figure 3 First, the user's demand response feature matrix R is mapped to h (h = 1, 2, 3) different linear mappings. and in, and The weight matrix to be learned is given; then, the attention of the mapped input demand response features is calculated, as shown in equation (11), where d k For matrix K hThe dimension is M, where M is the mask matrix; then, h attention matrices are concatenated and multiplied by the weight matrix W. 0 The multi-head attention evaluation results are shown in Equation (12).

[0054]

[0055]

[0056] in, and M head These represent the attention of the h-th head and multi-head attention, respectively; f Atte (·) is the attention calculation function; f sm (·) is the softmax function, used to convert Q... h and K h Similarity scores are normalized; f ct (·) is the concatenation function.

[0057] (3) Residual connectivity and hierarchical normalization for demand response potential assessment

[0058] To accelerate the training of multi-head self-attention models and improve their generalization ability, cross-layer connections, i.e., additional connections, need to be introduced. Figure 2 Each module in the encoder and decoder is followed by a residual connection and layer normalization. Specifically, the residual connection adds the input and output of each sub-layer to obtain a new output, and the output vector is normalized to a normal distribution through layer normalization, as shown in Equation (13). This creates a direct path for gradient backpropagation, prevents gradient vanishing, and thus improves the training efficiency and generalization ability of the model.

[0059]

[0060] in, This represents the result of residual connectivity and layer normalization; f LN (·) represents the layer normalization function; X is the input of each sublayer; f SL (X) represents the output of each sub-layer; and The result of the residual connection is X+f SL The mean and variance of (X), λ and γ are learning parameters to compensate for information lost during normalization; ε is set to a small number to prevent the denominator from being 0; ⊙ represents the Hadamard product.

[0061] (4) Fully connected feedforward network for demand response potential assessment

[0062] To address the nonlinear relationships in the input demand response feature sequence and further improve the expressive and generalization capabilities of the multi-head self-attention demand response potential assessment model, a fully connected feedforward network is used to perform one or more linear transformations and one nonlinear transformation on the output of each residual connection and layer normalization, as shown in Equation (14). The linear transformation converts the output of the residual connection and layer normalization into a new vector representation, while the nonlinear transformation applies a nonlinear transformation to the result of the linear transformation using an activation function.

[0063]

[0064] Among them, f FFN (·) represents the fully connected feedforward network function in the model, f ReLU (·) is the ReLU activation function. Let be the T-th linear transformation function.

[0065] Building upon multi-head self-attention, this paper proposes a method for assessing the demand response potential of typical industry users based on convolutional self-attention, so as to better incorporate local contextual information of features at different time periods into the multi-head self-attention mechanism.

[0066] Although load or demand response characteristics may be the same across different time periods, their corresponding demand response potentials may still differ, such as peak shaving or valley filling attributes. Therefore, point-by-point dot product self-attention, which only considers the demand response characteristics of a single time period, will reduce the accuracy of the demand response potential assessment results. To improve the accuracy of the assessment results, it is necessary to consider the information of adjacent time periods of the input time period characteristics. Therefore, this invention proposes convolutional self-attention based on multi-head self-attention, generating Q-factors of demand response characteristics across multiple time periods through causal convolution. h and K h To better incorporate the local contextual information of features from different time periods into the multi-head self-attention mechanism, the corresponding architecture is shown in the attached figure. Figure 4 As shown.

[0067] From the appendix Figure 4 It can be seen that if the traditional pointwise dot product self-attention mechanism is directly used, time periods with the same value in the input demand response feature sequence may correspond to the same attention score, but the demand response potential corresponding to each time period is different; if the convolutional self-attention mechanism is used, the Q generated by causal convolution can be used to achieve the same attention score. h and K h This approach aims to understand the local contextual information corresponding to each time period of the demand response feature sequence, and to calculate the corresponding attention score using this local contextual information. This overcomes the limitations of the Q-sense mechanism in traditional self-attention mechanisms. h and K h This approach avoids the drawback of only considering the demand response characteristics of the current period and improves the accuracy of the assessment results.

[0068] In summary, using the objective and subjective demand response characteristics of industry users at different time periods as input features for demand response potential assessment, a typical demand response potential assessment framework for industry users based on convolutional self-attention is attached. Figure 5 As shown.

[0069] The beneficial effects of this invention are as follows:

[0070] This invention provides a method for assessing the electricity demand response potential of typical users in the industry. It utilizes integrated empirical mode decomposition to decompose user load sequences, overcoming the mode aliasing problem. The resulting intrinsic mode function components accurately characterize the load sequence, effectively improving the accuracy of objective demand response feature extraction. By employing convolutional self-attention to establish a mapping relationship between demand response features and user demand response potential, it not only captures long-term and short-term dependencies in the time series but also considers the local contextual information of features at different time periods. This allows for better integration into the multi-head self-attention mechanism, more effectively extracting user demand response potential features, improving the accuracy of the assessment results, and providing a new approach for precise dispatching by power sales companies. This method has significant practical application value. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] Figure 1 The time period division of the remaining cycle component in the industry typical user electricity demand response potential assessment method provided in the embodiments of the present invention;

[0073] Figure 2 The principle of assessing the demand response potential of typical users in the industry based on multi-head self-attention is provided in the embodiments of the present invention.

[0074] Figure 3 The basic principle of multi-head self-attention in the demand response potential assessment of the industry typical user electricity demand response potential assessment method provided in the embodiments of the present invention;

[0075] Figure 4 The convolutional self-attention basic architecture for demand response potential assessment of the industry typical user electricity demand response potential assessment method provided in the embodiments of the present invention;

[0076] Figure 5The present invention provides a framework for assessing the typical demand response potential of industry users based on convolutional self-attention, which is a method for assessing the typical demand response potential of industry users in an embodiment of the present invention.

[0077] Figure 6 This invention provides an embodiment of a method for assessing the electricity demand response potential of typical industry users, which includes the classification of industry user electricity consumption types and the identification of typical load curves.

[0078] Figure 7 The integrated empirical mode decomposition result of a typical daytime double-peak type I load curve of an embodiment of the industry typical user electricity demand response potential assessment method provided in the present invention;

[0079] Figure 8 The SG-filtered residual cycle component is an embodiment of the industry typical user power demand response potential assessment method provided by the present invention.

[0080] Figure 9 A step sequence of a typical daytime double-peak type I load curve, as provided in an embodiment of the method for assessing the electricity demand response potential of typical users in the industry according to an embodiment of the present invention;

[0081] Figure 10 This invention provides an embodiment of a method for assessing the electricity demand response potential of typical industry users in a daytime bi-peak type I electricity consumption category, illustrating the typical demand response potential of such users.

[0082] Figure 11 A comparison of user demand response potential assessment results obtained by different methods in one embodiment of the industry typical user power demand response potential assessment method provided in this invention. Detailed Implementation

[0083] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0084] In one specific implementation, the method for assessing the electricity demand response potential of typical users in the industry includes:

[0085] S1. Extract industry user demand response characteristics based on load decomposition results and demand response intentions;

[0086] S2. Based on the characteristics of industry user demand response and the convolutional self-attention mechanism, the potential for responding to typical industry user demand is evaluated.

[0087] In one embodiment, step S1 further includes:

[0088] S11, based on the integrated empirical mode decomposition method, decomposes the load sequence of typical users in the industry, and extracts an index vector reflecting the objective demand response characteristics of typical users in the industry based on the load decomposition results.

[0089] The electricity consumption types and characteristics of users within different industries vary significantly. A thorough analysis of the load variation characteristics of industry users and an accurate grasp of the variation patterns of typical loads are prerequisites for accurately assessing the demand response potential of typical users in an industry. The demand response potential of typical users in an industry refers to the demand response potential corresponding to the load of typical users in the industry, representing the average demand response capability of the corresponding industry's electricity consumption type. Therefore, to accurately assess the demand response potential of typical users in an industry, it is necessary to first classify the industry's electricity consumption types and identify the typical load curves representing different electricity consumption types. Considering that the normal cloud model can fully explore the local dynamic characteristics of the load curve, and that the density peak clustering algorithm has the advantages of few preset parameters, high computational efficiency, and good adaptability to data distribution, but it is difficult to accurately determine the cluster center, this invention proposes to classify the industry's electricity consumption types and identify their typical load curves based on the normal cloud model and an improved density peak clustering algorithm. Since the focus of this invention is on assessing the typical demand response potential of industry users, the steps of classifying industry electricity consumption types and identifying typical load curves will not be elaborated here.

[0090] In practice, the typical load of industry users can be considered as a superposition of different components. Different components have different impacts on load variation characteristics. Therefore, it is necessary to decompose the typical load sequence of industry users and analyze the components representing the typical demand response characteristics of the industry. This can be achieved using the empirical mode decomposition method.

[18] Decomposing the load sequences of industry users can yield intrinsic mode functions (IMFs) and remainders that characterize different local features of the original load sequences. However, empirical mode decomposition (EMD) methods suffer from mode aliasing, leading to erroneous IMF components and thus preventing the IMFs from accurately representing the characteristics of the load sequences. Considering that integrated EMF methods can overcome the shortcomings of traditional EMF methods by adding white noise, this invention proposes an industry user load sequence decomposition method based on integrated EMF methods, extracting sequences that characterize user demand response features.

[0091] Based on the normal cloud model and an improved density peak fast clustering algorithm, a certain industry is divided into M electricity consumption types, denoted as set C. T ={1,2,…,m,…,M}, where the typical load curve identification result for the m-th electricity consumption type is represented as P. m ={P m,1,P m,2 ,…,P m,j ,…,P m,J}, P m,j Let J represent the load at the j-th moment of the typical load curve m, where J is the total number of sampling points. Typically, J = 96, meaning the sampling frequency is 15 min / time. The set of users in the m-th electricity consumption type within the industry is... This represents the b-th user in the m-th electricity consumption type. The load sequence is in, Indicates user The load at time j. The typical load sequence P is analyzed using the integrated empirical mode decomposition method. m The decomposition yields the intrinsic mode function components and remainder terms as follows:

[0092]

[0093] Among them, G IMF This represents the total number of intrinsic mode functions; The remainder is called the trend component of the typical load sequence m, which is the deterministic part and characterizes the overall trend of change in the typical load of industry users; For typical load sequence P m The basic idea behind obtaining the g-th eigenmode function through integrated empirical mode decomposition is as follows: 1) Initialize the count variables g = 1 and k = 1; 2) In the original typical load sequence P m Adding the kth type of Gaussian white noise to obtain 3) Calculation All local maxima and minima are obtained. upper envelope and lower envelope 4) According to Sure The g-th IMF, let And g = g + 1; 5) If If it is not a monotonic function, repeat steps 3) and 4); otherwise, calculate the residue. Simultaneously, let k = k + 1; 6) If k is less than the given value of K, then use different Gaussian white noise ξ k Repeat steps 2)-5); 7) Based on the above steps, calculate the typical load sequence P. m Decomposed into The final eigenmode functions and corresponding residues are obtained as follows: and Corresponding to typical load sequences P mThe periodic component and trend component are given. The periodic component consists of eigenmode function components with different fluctuation frequencies, namely high-frequency component, mid-frequency component and low-frequency component.

[0094] The high-frequency component IMF1 of a typical load sequence, i.e. The fluctuation frequency is large, but the amplitude is small, reflecting the impact of random factors on the electricity consumption of industry users, and having a relatively small impact on the typical demand response potential; the low- and medium-frequency components of the typical load sequence Right now The fluctuation frequency is lower than that of the high-frequency component, and the fluctuation amplitude is larger, reflecting the electricity consumption habits of industry users and having a greater impact on the typical demand response potential. Therefore, this invention intends to extract the objective demand response characteristics of industry users based on ignoring the high-frequency component of the load sequence, as shown in Equation (2).

[0095]

[0096] in, The remaining periodic component is the original typical load sequence after removing trend and high-frequency components. To further reduce the impact of sampling noise on the load sequence, considering that SG filtering can ensure that the signal shape and width remain unchanged while filtering out noise, this invention proposes to use SG filtering for the remaining periodic component. Smoothing filters are applied to improve the smoothness of the remaining periodic sequences and reduce noise interference.

[0097] In actual production, depending on the differences in production processes and procedures, the remaining cycle components obtained above can be divided into time periods. Divided into S segments, as shown in the appendix. Figure 1 As shown, where t1+t2+…t s +…+t S =24. The switching start-stop of combined power equipment for industry users corresponds to the rise and fall of the remaining cycle component in each time period, which is called a step. It represents the demand response potential of industry users in that time period, which is either valley filling or peak shaving, and is expressed as:

[0098]

[0099] in, This represents the average value of the remaining periodic component s segment of a typical load curve m. The steps representing the remaining periodic component of the typical industry load curve at hour θ are called the step sequence. When... When, it indicates that users in this industry have the potential for peak-shaving demand response in the θth hour; when The value at time θ indicates that users in this industry have the potential to fill the demand gap in the θth hour.

[0100] In summary, the objective demand response characteristic vector of a typical load m It can be represented as:

[0101]

[0102] in, The demand response characteristic sequence for industry users, i.e.

[0103] S12, based on survey statistics such as users' historical response electricity consumption, number of effective responses, and contracted capacity, establish an indicator system that reflects users' willingness to respond to their needs and extract users' subjective needs response characteristics.

[0104] The demand response potential of typical users in an industry is not only related to the changing trends of typical load sequences, remaining cycle sequences, and step sequences, but also requires comprehensive consideration of the demand response willingness of each user in each time period within the industry's electricity consumption type. Based on historical actual response electricity volume, effective response frequency, contracted capacity, and user quotations for each user in the industry's electricity consumption type, subjective demand response characteristics that characterize the industry users' demand response willingness are extracted, namely, the ratio of actual response electricity volume, the ratio of effective response frequency, the ratio of invited response capacity, and the ratio of demand response subsidy electricity price.

[0105] (1) Actual response power ratio

[0106]

[0107] in, The ratio of actual response power consumption in the θth hour of typical load m represents the average ratio of the actual response power consumption to the reported response power consumption of each user in industry power consumption type m during that period. and users respectively Actual response power (kW) and declared response power (kW) for hour θ. Actual response power ratio for industry users. The larger the value, the stronger the user's willingness to participate in demand response, and the greater the demand response potential.

[0108] (2) Effective response rate

[0109]

[0110] in, N represents the ratio of effective response counts for a typical load m in hour θ, indicating the average ratio of the historical effective response counts to the total number of invitations for each user in industry electricity consumption type m during that period; iv (b, θ) represents the electricity sales company's transaction with the user in the θth hour. Total number of invitations; n iThese are 0-1 variables, representing users. Is the response valid in the i-th iteration? If it is valid, then n i The value is 1; otherwise, n i The value is 0. The ratio of valid responses from industry users. The larger the value, the stronger the user's willingness to participate in demand response, and the greater the demand response potential.

[0111] (3) Contract response capacity ratio

[0112]

[0113] in, Let Q be the contracted response capacity ratio for a typical load m at hour θ, representing the average ratio of the contracted demand response capacity to the user's receiving capacity for each user in industry electricity consumption type m at hour θ; contr (b,θ) and Q cap (b) respectively users Contracted capacity (kWh) and total received capacity (kWh) for hour θ. Contracted capacity ratio for industry users. The larger the value, the stronger its willingness to participate in demand response, and the greater its demand response potential.

[0114] (4) Demand response subsidy unit price ratio

[0115]

[0116] in, p represents the ratio of the demand response subsidy unit price to the average ratio of the subsidy unit price submitted by each user in electricity consumption type m in the demand response declaration at hour θ to the upper limit of the declared price for subsidy unit price in the electricity sales company's demand response policy; cus (b,θ) and p hat (θ) represents the user at the θth hour. Price quotes (RMB / kWh) and the upper limit of electricity sales company subsidies (RMB / kWh). The higher the demand response subsidy per unit price ratio, the weaker the willingness of users to participate in demand response, and the smaller the demand response potential.

[0117] In summary, the subjective demand response characteristic vector of a typical load m is:

[0118]

[0119] Based on equations (4) and (9), the characteristic vector of subjective and objective demand response of typical load m is obtained as follows:

[0120] In one embodiment, in S2:

[0121] Based on the extracted demand response characteristics of typical industry users, the demand response potential of typical industry users is assessed using multi-head self-attention.

[0122] To obtain accurate assessment results of the demand response potential of typical industry users, this section will establish a mapping relationship between demand response features and demand response potential based on convolutional self-attention. The demand response feature vectors of industry users exhibit significant temporal characteristics, and their demand response potential also varies across different time periods. Considering that the self-attention mechanism can learn the dependencies between features in each time period of the demand response feature sequence and encode them into a vector, and that demand response features from different time periods can be weighted using different attention heads to achieve the fusion of demand response features from different time periods, exhibiting good generalization ability, assessing the demand response potential of typical industry users based on a multi-head self-attention mechanism can improve the accuracy of the assessment results. Taking a multi-head self-attention mechanism consisting of two encoder layers and two decoder layers as an example, the basic principle is as follows: Figure 2 As shown. Here, B0 is the start symbol for decoder #1, and the input is R. m,θ The demand response feature vector representing typical load m in time period θ is used to obtain the demand response potential of typical load m in time period θ through iterative encoder-decoder operation. Appendix Figure 2 The encoder in the model includes position encoding, multi-head self-attention, residual connections and layer normalization, and a fully connected feedforward network; the decoder, in addition to the above modules of the encoder, also has an encoder-decoder attention module, whose input comes not only from the output of the residual connection and layer normalization modules of its previous layer, but also from the output of the last layer encoder.

[0123] (1) Location coding of demand response characteristics

[0124] The demand response potential of typical users in an industry varies across different time periods, depending on the demand response characteristics of the input at different time periods. Therefore, assessing the demand response potential of typical users in an industry requires considering the relative order of demand response characteristics across different time periods. Thus, each time period can be viewed as a position in a demand response characteristic sequence, and the input demand response characteristics can be positionally encoded, as shown in Equation (10). This allows the multi-head self-attention model to better identify the relative positional relationships of different time periods within the demand response characteristic sequence, improving the model's understanding of the demand response characteristic sequence and the accuracy of the demand response potential assessment results.

[0125] The basic idea of ​​location encoding for demand response features is as follows: ① Let the number of time periods contained in the demand response feature sequence be N, that is, there are N locations that need to be encoded, represented as p. os=1,2,…,n,…,N, representing the position in the demand response feature sequence; ② Let the dimension of the position embedding vector of the demand response feature be d, and calculate based on equation (10). ③ Determine p os =n's corresponding positional code: P n E =[sin(nω1), cos(nω2), sin(nω3),…] (10)

[0126] When the vector dimension d is odd, the last term of equation (10) is sin(nω) d When the vector dimension d is even, the last term of equation (10) is cos(nω). d ).

[0127] The aforementioned location encoding method ensures that the distance between the encoding vectors at different locations in the demand response feature sequence is equal, thereby enabling the demand response potential assessment model to identify the relative positional relationships between different locations and thus better process the demand response feature sequence data.

[0128] (2) Multi-headed self-attention layer for demand response potential assessment

[0129] The assessment of demand response potential depends not only on the demand response characteristics of the corresponding time period, but also on the demand response characteristics of adjacent time periods. (Appendix) Figure 2 The multi-head self-attention layer in the demand response potential assessment model shown in the figure obtains the relationship between demand response features in different time periods in the input demand response feature sequence by performing self-attention calculation on the demand response feature sequence. That is, it not only focuses on the demand response features of adjacent time periods, but also on the relationship between demand response features of any two time periods, thereby enhancing the representation ability of the demand response feature sequence.

[0130] In multi-head self-attention, "multi-head" refers to learning different features and relationships from different "perspectives" within the input demand response feature sequence, such as time-period features, peak-shaving or valley-filling categories, and the confidence level of demand response potential assessment results. The multi-head self-attention mechanism linearly maps the input demand response feature sequence, then divides it into multiple "heads." Each head performs attention calculations on the mapped input demand response feature sequence and learns different weight assignments. The outputs of the multiple attention heads are then concatenated to form the final multi-head attention representation.

[0131] The basic principles of multi-head self-attention are as follows: Figure 3 First, the user's demand response feature matrix R is mapped to h (h = 1, 2, 3) different linear mappings. and in, and The weight matrix to be learned is given; then, the attention of the mapped input demand response features is calculated, as shown in equation (11), where d k For matrix K h The dimension is M, where M is the mask matrix; then, h attention matrices are concatenated and multiplied by the weight matrix W. 0 The multi-head attention evaluation results are shown in Equation (12).

[0132]

[0133]

[0134] in, and M head These represent the attention of the h-th head and multi-head attention, respectively; f Atte (·) is the attention calculation function; f sm (·) is the softmax function, used to convert Q... h and K h Similarity scores are normalized; f ct (·) is the concatenation function.

[0135] (3) Residual connectivity and hierarchical normalization for demand response potential assessment

[0136] To accelerate the training of multi-head self-attention models and improve their generalization ability, cross-layer connections, i.e., additional connections, need to be introduced. Figure 2 Each module in the encoder and decoder is followed by a residual connection and layer normalization. Specifically, the residual connection adds the input and output of each sub-layer to obtain a new output, and the output vector is normalized to a normal distribution through layer normalization, as shown in Equation (13). This creates a direct path for gradient backpropagation, prevents gradient vanishing, and thus improves the training efficiency and generalization ability of the model.

[0137]

[0138] in, This represents the result of residual connectivity and layer normalization; f LN (·) represents the layer normalization function; X is the input of each sublayer; f SL (X) represents the output of each sub-layer; and The result of the residual connection is X+f SL The mean and variance of (X), λ and γ are learning parameters to compensate for information lost during normalization; ε is set to a small number to prevent the denominator from being 0; ⊙ represents the Hadamard product.

[0139] (4) Fully connected feedforward network for demand response potential assessment

[0140] To address the nonlinear relationships in the input demand response feature sequence and further improve the expressive and generalization capabilities of the multi-head self-attention model, a fully connected feedforward network is used to perform one or more linear transformations and one nonlinear transformation on the output of each residual connection and layer normalization, as shown in Equation (14). The linear transformation converts the output of the residual connection and layer normalization into a new vector representation, while the nonlinear transformation applies a nonlinear transformation to the result of the linear transformation using an activation function.

[0141]

[0142] Among them, f FFN (·) represents the fully connected feedforward network function in the model, f ReLU (·) is the ReLU activation function. Let be the T-th linear transformation function.

[0143] Building upon multi-head self-attention, this paper further proposes a demand response potential assessment method for typical industry users based on convolutional self-attention, so as to better incorporate local contextual information of features at different time periods into the multi-head self-attention mechanism.

[0144] Although load or demand response characteristics may be the same across different time periods, their corresponding demand response potentials may still differ, such as peak shaving or valley filling attributes. Therefore, point-by-point dot product self-attention, which only considers the demand response characteristics of a single time period, will reduce the accuracy of the demand response potential assessment results. To improve the accuracy of the assessment results, it is necessary to consider the information of adjacent time periods of the input time period characteristics. Therefore, this invention proposes convolutional self-attention based on multi-head self-attention, generating Q-factors of demand response characteristics across multiple time periods through causal convolution. h and K h To better incorporate the local contextual information of features from different time periods into the multi-head self-attention mechanism, the corresponding architecture is shown in the attached figure. Figure 4 As shown.

[0145] From the appendix Figure 4 It can be seen that if the traditional pointwise dot product self-attention mechanism is directly used, time periods with the same value in the input demand response feature sequence may correspond to the same attention score, but the demand response potential corresponding to each time period is different; if the convolutional self-attention mechanism is used, the Q generated by causal convolution can be used to achieve the same attention score. h and K h This approach aims to understand the local contextual information corresponding to each time period of the demand response feature sequence, and to calculate the corresponding attention score using this local contextual information. This overcomes the limitations of the Q-sense mechanism in traditional self-attention mechanisms. h and K hThis approach avoids the drawback of only considering the demand response characteristics of the current period and improves the accuracy of the assessment results.

[0146] In summary, using the objective and subjective demand response characteristics of industry users at different time periods as input features for demand response potential assessment, a typical demand response potential assessment framework for industry users based on convolutional self-attention is attached. Figure 5 As shown.

[0147] To verify the effectiveness of the proposed demand response potential assessment method, a demand response case study of the metal processing machinery manufacturing industry in a city in Zhejiang Province, China, was used as an example. A total of 263 demand response users were invited, and smart meters collected daily load data from each user over a period of 365 days at a sampling frequency of 15 minutes per instance.

[0148] Based on the normal cloud model and an improved density peak fast clustering algorithm, 21,332 typical load curves for all 263 users were identified. Further, their electricity consumption types were categorized, and typical load curves for each type were identified as follows: full-peak type, peak-avoidance type, daytime bi-peak type I, daytime bi-peak type II, and daytime bi-peak type III. The results are attached. Figure 6 As shown, for each electricity consumption type, using the 87% load curve as the training set, ensemble empirical mode decomposition is performed on the daily load data to calculate the objective demand response characteristics of users in each industry, and to extract the subjective demand response characteristics of users. These are used as inputs to the demand response potential assessment method proposed in this invention to train the model and determine the number of encoding layers, decoding layers, weight matrix, and other parameters of the convolutional self-attention for each electricity consumption type. Based on this, using the typical load curves of each electricity consumption type as the object, the method of this invention is used to assess the typical demand response potential of industry users in different time periods.

[0149] Taking daytime double-peak type I electricity consumption as an example, integrated empirical mode decomposition was performed on its typical load curve, and the results are shown in the appendix. Figure 7 As shown, the horizontal axis represents the time series, and the vertical axis represents the original load series and the components after integrated empirical mode decomposition, with units of pu. (See attached diagram) Figure 7 It can be seen that the time-series load of this typical electricity consumption type is decomposed into 5 components. The last component is the residual component, reflecting the electricity consumption trend of users of this typical electricity consumption type. The IMF1 component is the high-frequency component of the load sequence, with the highest frequency and smaller fluctuation amplitude, having a relatively small impact on the demand response potential of industry users. The IMF2-IMF4 components are the medium-to-low-frequency components of the load sequence, with lower frequencies and larger fluctuation amplitudes, reflecting changes in the electricity consumption behavior and habits of industry users over a longer time scale, and having a significant impact on the demand response potential of industry users. Based on this, the residual periodic component of industry users is extracted, resulting in the SG-filtered residual periodic component, as shown in the appendix. Figure 8As shown, the step sequence of the typical daytime bimodal type I load sequence after SG filtering is further obtained, as shown in the attached figure. Figure 9 As shown.

[0150] From the appendix Figure 9 It is evident that the production processes of industry users at different times correspond to different load compositions and different switching and shutdown situations of combined electrical equipment, implying that these industry users have different peak-shaving or valley-filling demand response potentials at different times. Specifically, for the daytime double-peak type I, this typical industry user is in a concentrated electricity consumption phase between 09:00-10:00 and 15:00-16:00, and has significant peak-shaving demand response potential by improving production processes or procedures. Similarly, the industry user's electricity demand is lower between 0:00-6:00 and 21:00-24:00, and demand response strategies can encourage users to consume more electricity during these periods, fully leveraging their valley-filling demand response potential. The above analysis aligns with the actual production situation of this industry user, thus demonstrating the rationality of the load sequence decomposition method proposed in this invention.

[0151] Based on the above load sequence decomposition results, the method of this invention is used to evaluate the typical demand response potential of daytime bimodal type I industry users over 1-24 hours, as shown in the appendix. Figure 10 As shown. (From the appendix) Figure 10 It can be seen that during the period from 9:00 to 11:00, the peak-shaving demand response potential of industry users under this electricity consumption type is relatively large, at 0.3519 pu, 0.4012 pu, and 0.3875 pu respectively; and during 1:00, 4:00, 6:00, and 24:00, there is a relatively large valley-filling demand response potential, at 0.249 pu, 0.3016 pu, and 0.3677 pu respectively. In practice, among the daytime double-peak type I electricity consumption, 17 users participated in demand response between 9:00 and 10:00, with a per-unit response load average of 0.4165 pu. The demand response potential assessment result for typical industry users during this period is 0.4012 pu, indicating that the demand response potential assessment result is consistent with its actual response volume. In addition, industry users under this electricity consumption type have peak-shaving demand response potential from 09:00 to 18:00, and strong valley-filling demand response potential at other times. Electricity sales companies can fully leverage the demand response potential of industry users under this electricity consumption type through certain demand response strategies to achieve the purpose of peak shaving and valley filling.

[0152] To further verify the rationality and effectiveness of the proposed method for assessing the electricity demand response potential of typical users in the industry, this invention compares the proposed method with the objective characteristic method, load factor method, traditional Transformer method, and similarity method. Taking 09:00-10:00 as an example, the accuracy of the demand response potential assessment results for each test user during this time period under each assessment method is calculated, as shown in the appendix. Figure 11 As shown in the figure. Among them, the objective feature method does not consider the user's subjective demand response intention, that is, it only takes the objective demand response feature sequence extracted based on the industry user load curve as input; the load rate method uses the industry user load curve and takes the difference between the maximum load and the load of the corresponding time period as the user's demand response potential; the traditional self-attention method only considers the point-by-point product self-attention of the demand response features of a single time period; the similarity method is based on the user's daily load curve, extracts indicators (daily load rate, daily maximum load utilization hours, daily peak-valley difference rate, peak period load rate, valley period load rate, and normal period load rate) to characterize the electricity consumption characteristics of the sample set industry users and test users, calculates the similarity between the sample set industry users and test users, and evaluates the demand response potential of test users by combining the demand response potential of the sample set users.

[0153] From the appendix Figure 11 It can be seen that the proposed method for assessing the electricity demand response potential of typical users in the industry has the highest accuracy, all exceeding 80%. Traditional self-attention assessment methods generate Q and K based solely on the demand response characteristics of a single time period, ignoring the impact of adjacent time period characteristics on the assessment results, resulting in an accuracy of approximately 60%. Assessment methods that ignore users' subjective demand response intentions yield the lowest accuracy, because even if a user's demand response potential is large, their participation in demand response is limited when their willingness to respond is weak. While load factor-based methods are simple and suitable for rapid, rough estimations in engineering, the assessment results have significant errors and cannot support accurate dispatching by power sales companies. Similarity-based methods consider the similarity of users' electricity consumption habits, but different users have different subjective intentions, and the production processes and technologies corresponding to each user also differ during the demand response invitation period. Therefore, the accuracy of demand response potential assessment results considering only user electricity consumption habits is only about 55%. It is evident that considering users' subjective demand response intentions and the local contextual information of characteristics in each time period can effectively improve the accuracy of user demand response potential assessment results.

[0154] In summary, the method for assessing the electricity demand response potential of typical users in the industry proposed in this invention effectively extracts the objective demand response feature sequence of users, while also considering the subjective willingness of users to respond to demand and the local contextual information of the input time period features, which can effectively improve the accuracy of the assessment results of the demand response potential of users in the industry.

[0155] The above provides a detailed description of the method for assessing the electricity demand response potential of typical users in the industry, as provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this invention.

Claims

1. A method for assessing the electricity demand response potential of typical users in an industry, characterized in that, include: S1. Extract industry user demand response characteristics based on load decomposition results and demand response intentions; S2. Based on the characteristics of industry user demand response and the convolutional self-attention mechanism, the potential for responding to typical industry user demand is evaluated. Step S1 specifically includes: S11, based on the integrated empirical mode decomposition method, decomposes the load sequence of typical users in the industry, and extracts an index vector reflecting the objective demand response characteristics of typical users in the industry based on the load decomposition results. S12. Based on the survey statistics of users' historical response electricity, number of effective responses, contracted capacity and user quotations, establish an indicator system that reflects users' willingness to respond to their needs and extract users' subjective needs response characteristics. The specific steps of step S2 are as follows: based on the extracted demand response characteristics of typical users in the industry, multi-time-period features are obtained based on the local context information of the features of each time period and causal convolution, and the demand response potential of typical users in the industry is evaluated based on the multi-time-period features and the multi-head self-attention mechanism. In step S2, when evaluating the demand response potential of typical industry users based on multi-time period features and multi-head self-attention mechanism, it is necessary to consider the relative order relationship between demand response features in different time periods, treating each time period as a position in the demand response feature sequence, and encoding the position of demand response features of typical industry users. The specific steps are as follows: Suppose the demand response feature sequence contains N time periods, meaning there are N positions that need to be encoded, denoted as p. os =1,2,…,n,…,N, representing the position of the demand response feature in the demand response feature sequence; Let the dimension of the embedding vector of the location of the demand response feature be d, and calculate the result. ; Sure Corresponding position code: When the vector dimension d is odd When the vector dimension d is even 2. The method for assessing the electricity demand response potential of typical users in the industry as described in claim 1, characterized in that, Step S11 specifically involves: First, based on the normal cloud model and the improved density peak fast clustering algorithm, a certain industry will be divided into M electricity consumption types, denoted as set C. T ={1,2,…,m,…,M}, where the typical load curve identification result for the m-th electricity consumption type is represented as P. m ={P m,1 ,P m,2 ,…,P m,j ,…,P m,J }, Let J be the load at the j-th moment of the typical load curve m, where J is the total number of sampling points; the set of users in the m-th electricity consumption type of the industry is... This represents the b-th user in the m-th electricity consumption type; The load sequence is in, Indicates user The load at time j; typical load sequences are analyzed using the integrated empirical mode decomposition method. The decomposition yields the intrinsic mode function components and remainder terms as follows: Among them, G IMF This represents the total number of intrinsic mode functions; The remainder is called the trend component of the typical load sequence m, which is the deterministic part and characterizes the overall trend of change in the typical load of industry users; Typical load sequence The g-th intrinsic mode function obtained through integrated empirical mode decomposition; Next, ignoring high-frequency components of the load sequence, the objective demand response characteristics of industry users are extracted: in, The remaining periodic component after removing the trend component and high-frequency component from the original typical load sequence; Then, SG filtering is used for the remaining periodic components. Perform smoothing filtering: Based on the differences in production processes and procedures, the remaining cycle time is divided into time periods. Divided into S segments; the rise and fall of the remaining cycle component corresponding to the switching start and stop of the combined power consumption equipment of industry users in each time period are called steps, which represent the demand response potential of industry users in that time period, either filling valleys or shaving peaks, and are expressed as: in, This represents the average value of the remaining periodic component s segment of a typical load curve m. The steps representing the remaining periodic components of the typical industry load curve m at the θ-th hour are called the step sequence; when When, it indicates that users in this industry have the potential for peak-shaving demand response in the θth hour; when When, it indicates that users in this industry have the potential to fill the valley in demand response at hour θ; Finally, the objective demand response characteristic vector of typical load m Represented as: in, The demand response characteristic sequence for industry users, i.e.

3. The method for assessing the electricity demand response potential of typical users in the industry as described in claim 1, characterized in that, In step S12, the extracted subjective demand response characteristics of users include the actual response electricity ratio, the effective response frequency ratio, the invited response capacity ratio, and the demand response subsidy electricity price ratio. (1) Actual response power ratio in, The ratio of actual response power consumption to actual response power consumption in the θth hour of typical load m represents the average ratio of actual response power consumption to user-reported response power consumption for each user in industry power consumption type m during that period. and users respectively The actual response power (kW) and the declared response power (kW) for the θth hour; (2) Effective response rate in, N represents the ratio of effective response counts for a typical load m in hour θ, indicating the average ratio of the historical effective response counts to the total number of invitations for each user in industry electricity consumption type m during that period; iv (b, θ) represents the electricity company's transaction with the user in the θth hour. Total number of invitations; n i These are 0-1 variables, representing users. Is the response valid in the i-th iteration? If it is valid, then n i The value is 1; otherwise, n i The value is 0; (3) Contract response capacity ratio in, Let Q be the contracted response capacity ratio for a typical load m at hour θ, representing the average ratio of the contracted demand response capacity to the user's receiving capacity for each user in industry electricity consumption type m at hour θ; contr (b, θ) and Q cap (b) respectively users Contracted capacity (kWh) and total power receiving capacity (kWh) at hour θ; (4) DR subsidy unit price ratio in, p represents the ratio of the demand response subsidy unit price to the average ratio of the subsidy unit price submitted by each user in electricity consumption type m in the demand response declaration at hour θ to the upper limit of the declared price for subsidy unit price in the electricity sales company's demand response policy; cus (b, θ) and p hat (θ) represents the user at the θth hour. Price quote (RMB / kWh) and upper limit of subsidy price for electricity sales companies (RMB / kWh); In summary, the subjective demand response characteristic vector of a typical load m is: (9) The characteristic vector of subjective and objective demand response for a typical load m is represented as follows:

Images