Industrial user response potential quantitative evaluation method and system

Abstract

The invention belongs to the technical field of power system demand response, and particularly discloses an industrial user response potential quantitative evaluation method and system, and the method comprises the steps: constructing a training set according to the load characteristics of a plurality of users and corresponding actual average loads; the load characteristic comprises an adjustable index and a multi-dimensional load characteristic index and is obtained according to historical power utilization data and historical response data of the user, the adjustable index is used for measuring the physical adjustment capability of the user, and the multi-dimensional load characteristic index is used for measuring the demand response intention of the user; on the basis of the training set, describing a mapping relation between load characteristics and actual average load through a Gaussian process to obtain an evaluation model; based on the evaluation model, according to the to-be-evaluated user load characteristics, evaluating an actual average load; and quantitatively evaluating the demand response potential by taking the difference value between the baseline average load and the actual average load in the user declaration execution time period as an effective response load. According to the invention, non-intrusive and high-precision evaluation of the demand response potential of the industrial user can be realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system demand response, and more specifically, relates to a method and system for quantitatively evaluating the response potential of industrial users. Background Art

[0002] Against the backdrop of continued global energy demand expansion, the international community is committed to promoting a green transformation of the energy system, accelerating the development of renewable energy sources such as solar and wind power, and striving to reduce negative environmental impacts. However, the power supply characteristics of these clean energy sources are highly influenced by natural conditions, with significant intermittency and uncertainty, posing challenges to the stable operation of power systems.

[0003] To ensure a real-time balance between electricity supply and demand, especially with the growing share of renewable energy, grid management faces a more complex regulatory task. One traditional solution is to enhance generation flexibility, increasing reserve capacity, but this often comes at a high cost. By contrast, addressing demand and adjusting consumer electricity usage patterns through demand response strategies has become a more cost-effective means of balancing power supply.

[0004] In the research and application of demand response technology, existing analysis methods are mainly divided into two schools: one focuses on qualitative evaluation, evaluating user response potential through relative numerical values. Although this method has a certain degree of popularity, it is not convenient for the power system to achieve the established demand response goals. The other type focuses on in-depth analysis of specific types of loads, such as specific modeling for air conditioners, electric vehicles, etc. Although this approach can provide highly accurate response potential assessments, it is not universally applicable to industrial users with complex processes.

[0005] Therefore, an efficient demand response potential assessment method is urgently needed for industrial users, a key load group. Summary of the Invention

[0006] In response to the above-mentioned defects or improvement needs of the prior art, the present invention provides a method and system for quantitatively evaluating the response potential of industrial users, the purpose of which is to achieve non-invasive and high-precision evaluation of the demand response potential of industrial power users.

[0007] To achieve the above objectives, according to one aspect of the present invention, a method for quantitatively evaluating the response potential of industrial users is proposed, comprising:

[0008] Model building phase:

[0009] A training set is constructed based on the load characteristics of multiple users and the corresponding actual average loads; the load characteristics include an adjustability index and a multidimensional load characteristic index. The adjustability index is used to measure the user's physical adjustment ability and is obtained based on the user's historical electricity usage data; the multidimensional load characteristic index is used to measure the user's demand response willingness and is obtained based on the user's historical electricity usage data and historical response data;

[0010] Based on the training set data, the mapping relationship between load characteristics and actual average load is described by Gaussian process to obtain the evaluation model;

[0011] Model application phase:

[0012] Based on the evaluation model, the actual average load is evaluated according to the load characteristics of the user to be evaluated;

[0013] The difference between the baseline average load and the actual average load during the user's declared execution period is taken as the effective response load, and the demand response potential is quantitatively evaluated based on the effective response load.

[0014] As a further preferred embodiment, the adjustability index includes the production trend factor and the load step matrix of the user response day, and the acquisition method thereof is:

[0015] The STL algorithm is used to split the historical electricity consumption data into trend load component, periodic load component and residual component;

[0016] The production trend factor is determined according to the trend load component; the load step matrix is ​​determined according to the periodic load component.

[0017] As a further preferred method, the production trend factor is determined according to the trend load component, and the calculation formula is:

[0018]

[0019] Among them, θ is the production trend factor of the response day, To respond to the nth day r Daily trend load component series, e k Indicates the first n r The mean value of the trend load component on the kth day in the world.

[0020] As a further preferred method, determining a load step matrix according to periodic load components includes:

[0021] A load step is a steady-state load segment in a periodic load component whose duration exceeds a preset time threshold and whose local change rate is lower than a preset change rate threshold. The steady-state load segments in the periodic load component are combined to obtain a load step matrix.

[0022] As a further preferred embodiment, the preset time threshold is 1 hour, the preset change rate threshold is 3; and the calculation formula of the local change rate α is:

[0023]

[0024] Where: p max 、p min are the maximum and minimum load values ​​in the local load data window respectively; Δp max ,Δp mean ,Δp std are the maximum value, mean value and standard deviation of the load difference between adjacent sampling points, respectively.

[0025] As a further preferred method, the baseline average load is determined by:

[0026] The baseline is determined based on the trend load component, and the arithmetic mean load calculated based on the baseline is used as the baseline average load.

[0027] As a further preference, the multi-dimensional load characteristic indicators include historical declaration participation rate, historical effective response rate, industry relative added value, relative response volume and invitation price ratio.

[0028] As a further preferred method, the formula for calculating the relative added value of an industry is:

[0029]

[0030] Among them, D p,t,his is the relative added value of the industry p to which the user belongs at the demand response time t, y p,t,q Y is the total demand response invitation quantity of industry p at time t in the qth demand response invitation, t,q is the total demand response invitation quantity at demand response time t in the qth demand response invitation, and M is the number of demand response invitations.

[0031] As a further preferred embodiment, based on the user's load characteristics, the mean vector and variance vector of actual electricity consumption are obtained through the evaluation model, thereby obtaining a confidence interval and realizing the quantification of evaluation uncertainty.

[0032] According to another aspect of the present invention, a system for quantitatively evaluating the response potential of industrial users is provided, comprising a processor configured to execute the above-mentioned method for quantitatively evaluating the response potential of industrial users.

[0033] In general, the above technical solutions conceived by the present invention have the following technical advantages compared with the existing technology:

[0034] 1. Targeting industrial users, a key load group, this invention constructs a multi-dimensional feature system based on "physical regulation capability and subjective response willingness" and introduces a Gaussian process regression algorithm. This method enables a non-invasive, highly accurate, and interpretable assessment of industrial users' demand response potential based solely on historical electricity consumption and response data.

[0035] 2. The present invention introduces a decomposition mechanism for trend loads and periodic loads. By analyzing the load step characteristics of historical electricity consumption data and combining it with the response willingness characteristic system constructed based on users' historical response behaviors, a multi-dimensional indicator system is formed. This system can achieve comprehensive quantification of physical regulation capabilities and subjective willingness without deploying dedicated monitoring equipment. It effectively solves the problem of traditional methods' dependence on device-level data, realizes non-invasive and accurate assessment, and reduces implementation costs while protecting user privacy.

[0036] 3. This invention generates probabilistic assessment results with confidence intervals based on Gaussian process regression, breaking through the technical limitations of traditional deterministic assessment models. It quantifies the risk boundaries of demand response potential, enables accurate identification of industrial user load characteristics and prediction of potential response capabilities, and fills a gap in existing technologies for data-based assessment of the overall demand response potential of industrial users. It provides a two-dimensional decision-making basis for power grid dispatching, namely "assessment mean-fluctuation range," thereby improving the stability and reliability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a diagram of a quantitative evaluation method for industrial user response potential according to an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of STL decomposition according to an embodiment of the present invention;

[0039] Figure 3 This is a corresponding diagram of industrial user load components and STL algorithm components in an embodiment of the present invention;

[0040] Figure 4 This is a load step extraction diagram of an embodiment of the present invention;

[0041] Figure 5 The response day load and baseline load diagram of the embodiment of the present invention;

[0042] Figure 6 2 is a comparison chart of the evaluation results of the embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0044] The embodiment of the present invention provides a method for quantitatively evaluating the response potential of industrial users, such as Figure 1 As shown, the following steps are included:

[0045] S1, load data collection, including historical electricity consumption data and historical response data of demand response invitation users.

[0046] Specifically, the data is collected from the historical electricity consumption data and historical response data of multiple industrial users for 15 regular working days before the response date, and the data sampling interval is 15 minutes.

[0047] S2, load data preprocessing and feature extraction, obtain load characteristics and the corresponding actual average load to construct a training set.

[0048] Load characteristics include adjustability index and multi-dimensional load characteristic index, and the acquisition includes the following steps:

[0049] S21, uses the STL algorithm to perform multi-scale decomposition of users' historical electricity consumption data, and then constructs an adjustability index based on the mathematical definition of load steps. The adjustability index includes the production trend factor and the load step matrix.

[0050] (1) Use STL algorithm to convert user historical electricity consumption data Y t Split into three parts:

[0051] Y t =T v +S v +R v

[0052] Where T v is the trend load component, S v is the periodic load component, R v is the residual component.

[0053] The load characteristics of industrial users' electrical equipment can be analyzed and divided into four categories:

[0054] 1) Trend load component, which represents the baseline load driven by production scale and is composed of continuously adjustable equipment. Its amplitude changes quasi-linearly with production capacity regulation;

[0055] 2) Periodic load component, reflecting the load fluctuations caused by periodic switching of equipment groups within the production process, with typical characteristics of step-like transitions at multiple time scales within a day;

[0056] 3) Uncertain load component (i.e., residual component), which originates from random load fluctuations caused by unplanned disturbances (such as equipment failure and order changes);

[0057] 4) Non-interruptible load component: the basic load required to maintain safe production, with the characteristic of being non-interruptible during the day.

[0058] Because the demand response capacity is derived by subtracting the baseline from the actual average load during the invitation period, the impact of the non-interruptible load component on the industrial user's demand response potential can be eliminated. Furthermore, the residual component can be considered as fluctuations outside the production plan and is therefore not considered. A feature analysis is then performed on the trend and cycle components. The extracted time series features include the production trend factor and load step matrix for the industrial user's response day, resulting in an adjustability index.

[0059] (2) Based on the trend load component T of the user k days before the response date v Predict the user's production trend factor θ on the response day:

[0060] Decompose the user's response date n by STL algorithm r The load of the day, the trend load component is obtained as (sampling frequency is 15min / time), T 1,1 It represents the value of the first load point on the first day. The average value of the 96 load points of the trend load component of each day is calculated and expressed as e k Indicates the first n r The average trend load of the kth day in the day can be obtained r Daily trend load component series After processing, it becomes the production trend factor θ of the response day.

[0061]

[0062] The larger the production scale factor θ is, the larger the production scale is. Taking 1 as the dividing line, a value greater than 1 represents an increasing trend, and a value less than 1 represents a decreasing trend.

[0063] (3) Based on the periodic load component S of the user k days before the response date v Extract the user's load step matrix T:

[0064] A load step refers to a steady-state load segment in the periodic component of the daily load curve of an industrial user that lasts for more than one hour and whose local change rate is lower than the threshold.

[0065] The local rate of change α is:

[0066]

[0067] where p is the load value in the local load data window; Δp is the load difference value between adjacent sampling points; and the subscripts min, max, mean, std represent the minimum value, maximum value, mean value, and standard deviation, respectively.

[0068] Generally, the local change rate threshold α < 3 is selected as the criterion for determining the load step. Based on the historical load data of the previous day with a sampling frequency of 15 min / time, the load difference value sequence {Δp} approximately obeys a Gaussian distribution. Under this condition, the load step represents that the user is in a steady-state operating condition, and the load fluctuation range Δp max is within the normal fluctuation range [0, p mean + 3Δp std ] of the overall load difference sequence. When α ≥ 3, it indicates that the load fluctuation exceeds the normal fluctuation range, and the user deviates from the steady state and enters a transient process.

[0069] From the physical mechanism analysis, the transition of the load step is essentially caused by the load level step change due to the switching of the operating state of the equipment cluster in industrial production, which specifically manifests in typical scenarios such as the start-stop of high-energy-consumption equipment or the conversion of production line operating modes. From the mathematical analysis, the selection process of the load step discards the transient information between the load steps and retains the steady-state information such as the average value, start time, and end time in the selected time period.

[0070] Specifically, taking the historical data of the previous day with a 15-minute sampling frequency and a sliding time T = 1 h as an example, the load step extraction process is as follows:

[0071] 1) The historical data is decomposed by the STL algorithm to obtain the periodic load component:

[0072]

[0073] 2) The candidate step a is initialized to store S v , and α < 3 is taken as the initial judgment condition for the load step:

[0074] α = S v [1, 2, 3, 4]

[0075] 3) The periodic load component S v is traversed to identify the load step, and the traversal method is as follows:

[0076]

[0077] a is updated. The initial value of t is 5, which represents the latest position of traversal. The updating method of t is as follows: if S v(t) belongs to a load step, then t=t+1; if S v (t) does not belong to a load step but S v (t-1) belongs to a load step, then t=t+3; if S v (t) and S v (t-1) does not belong to the same load step, then t=t+1.

[0078] 4) Record the identified j-th load step information P j (k):

[0079] P j (k)=[p j,1 (k),p j,2 (k),p j,3 (k)],j∈[1,n k ]

[0080] Where p j,1 (k),p j,2 (k),p j,3 (k) storing the start time, end time and average value of the original periodic load component of the load step respectively; n i is the number of steps identified in the periodic load component.

[0081] The step matrix T of the daily load curve of the kth day before the current day k for:

[0082]

[0083] The load step matrix T for the 15 days prior to the present is:

[0084] T=[T1;T2;…;T 15 ]

[0085] S22, obtain multi-dimensional load characteristic indicators based on the user's historical electricity consumption data and historical response data. The user's own willingness to participate in the demand response invitation is also an important factor affecting the user's demand response potential. Regarding the user's demand response willingness characteristics, the present invention combines the user's historical electricity consumption data and historical response data to propose 5 indicators to measure the user's demand response willingness: historical declaration participation rate, historical effective response rate, industry relative added value, relative response volume, and invitation price ratio. Among them, the historical declaration participation rate and the historical effective response rate are mainly used to measure the user's historical demand response participation; the industry relative added value is to assess the impact of the user's industry itself on the demand response willingness, such as the high-value-added computer and communications industries, which have low price sensitivity to demand response and basically do not participate in demand response; the relative response volume and invitation price ratio are indicators declared by the user in the system after receiving the invitation, reflecting the user's awareness of his or her own demand response potential. The specific calculation method of the indicators is as follows:

[0086] (1) Historical application participation rate:

[0087]

[0088] Where: R his is the participation rate of users after receiving invitations in the historical demand response invitation records; M is the number of historical demand response invitations; j q Indicates the user's declared participation in the qth demand response invitation. If the user applies to participate, the value is 1, otherwise the value is 0.

[0089] (2) Historical effective response rate:

[0090]

[0091] Where: E his is the effective response rate of users in the historical demand response invitation records; q Indicates whether the user is considered to have effectively participated in the qth demand response invitation. If so, the value is 1, otherwise the value is 0.

[0092] If the user meets the following two conditions at the same time during the demand response invitation period, it will be deemed as a valid response: 1) the maximum load during the response period is less than the baseline maximum load; 2) the average load during the response period is lower than the baseline average load, and the difference is not less than 50% of the declared response amount.

[0093] (3) Relative added value of the industry:

[0094]

[0095] Among them, D p,t,his is the relative added value of the industry p to which the user belongs at the demand response time t, yp,t,q Y is the total demand response invitation quantity of industry p at time t in the qth demand response invitation, t,q is the total demand response invitation quantity at demand response time t in the qth demand response invitation.

[0096] (4) Relative number of responses to the invitation recently:

[0097] C rec =c r / c

[0098] Where: C rec is the ratio of the response amount declared by the user after receiving the invitation to respond to the previous day to its own power consumption capacity; c is the user's power consumption capacity, which is the maximum load power value that may occur under the user's normal power demand; c r The amount of responses declared for the user.

[0099] (5) Price ratio of the response invitation recently:

[0100] P rec =P r / P max

[0101] Where: P rec P is the ratio of the price reported by the user in response to the invitation to the price ceiling provided by the power company; r The price of the subsidy is the price that the user declares after considering the economic benefits that may be caused by changing the daily working mode; max It is the upper limit of the subsidy price in the demand response policy issued by the power company on the response day.

[0102] S23, combining the adjustability index and the multi-dimensional load characteristic index to obtain the load characteristic x of user i i :

[0103] x i =[θ i ,T i ,R i,his ,E i,his ,D i,p,t,his ,C i,rec ,P i,rec ]

[0104] S3, Gaussian process regression is used to establish a nonlinear mapping between load characteristics and potential to obtain an evaluation model.

[0105] The mapping relationship between load characteristics and the actual average load during the demand response invitation period is described by a Gaussian process f(X). The prior distribution of f(X) is a Gaussian distribution, and the parameters of f(X) are random variables. The user data is used to construct the actual average load evaluation training set during the demand response invitation period:

[0106] D={(x i ,y i )|i=1,2,…,n}=(X,Y),

[0107] Where n is the number of demand response users included in the training set samples, X=[x1,x2,…,x n ] T is the input matrix of the actual average load evaluation training set during the demand response invitation period, Y=[y1,y2,…,y n ] T is the output label of the actual average load evaluation training set during the demand response invitation period, y i is the actual average load of user i during the demand response invitation period. The training set D can form a Gaussian joint distribution, and its corresponding Gaussian process is composed of the actual average load mean function μ(X) during the demand response period and the covariance matrix K(x i ,x j ) is uniquely defined, denoted as GP[.]:

[0108] f(X)~GP[μ(X),K(x i ,x j )]

[0109] The actual average load collected from users during the demand response period can generally be considered to be superimposed with noise that follows a Gaussian distribution, namely:

[0110]

[0111] Where N(.) represents Gaussian distribution, σ N is the standard deviation of the Gaussian distribution, so the Gaussian process can be further expressed as:

[0112]

[0113] Where, δ ij is the Kronecker delta function, which is used to characterize the equality of two variable algorithms. When i=j, δ ij =1; when i≠j, δ ij =0; I is the unit matrix.

[0114] Furthermore, considering the nonlinear relationship between the demand response potential characteristics and the actual average load during the user demand response invitation period, the kernel function is selected as:

[0115]

[0116] Where x, x′ are input samples (load characteristics), ||·|| represents the Euclidean distance, and σ represents the decay rate of feature space similarity. A too small σ value will lead to excessive complexity in the hypothesis space, which is manifested as overfitting, where the training error approaches zero while the test error surges. Conversely, a too large σ value will compress the effective dimension of the hypothesis space, leading to underfitting.

[0117] In order to evaluate the prediction effect of the model on the actual average load during the user demand response invitation period, another part of the user electricity consumption data is taken as the test set D * :

[0118] D * ={(x i ,f(x i ))|i=n+1,n+2,…,n+n *}

[0119] =(X * ,y * )

[0120] Where: X * ,y * are the input matrix and output label of the actual average load evaluation training set during the demand response invitation period; n * The number of users included in the training set is used to estimate the actual average load during the demand response invitation period.

[0121] From this, we can obtain the prior matrix of the joint Gaussian distribution composed of the actual average load evaluation training set and test set during the demand response invitation period:

[0122]

[0123] Where: is the covariance matrix between the training set data and the test set data; K(X,X) is the autocovariance matrix of the training set data; K(X * ,X * ) is the autocovariance matrix of the test set data.

[0124] This model uses K-Fold cross-validation, which divides the dataset into K equal-sized subsets, with K-1 subsets used for model training and the remaining subset used for model validation. K-Fold cross-validation can provide a more robust and reliable evaluation of model performance than single-shot validation.

[0125] According to Bayesian theory, the posterior distribution of the test set output is obtained, that is, the evaluation result of the actual electricity consumption during the user demand response period:

[0126]

[0127] Where: is the mean vector of the evaluation results; is the variance vector of the evaluation result. If the confidence interval is taken as 95%, the evaluation result of the actual electricity consumption during the user demand response period is:

[0128]

[0129] S4, based on the evaluation model, realizes the demand response potential evaluation and its uncertainty quantification.

[0130] Based on the load characteristics of the user being evaluated, the assessment model is used to determine the actual average load. The difference between the baseline average load and the actual average load during the user's declared execution period is used as the effective response load. This effective response load is then used to quantify the demand response potential. Furthermore, the assessment model can be used to derive confidence intervals to quantify risk.

[0131] Specifically, see S3. According to the load characteristics of the user, the evaluation model can be used to obtain the mean vector and variance vector of the actual power consumption, thereby obtaining the 95% confidence interval of the Gaussian distribution of the user's actual average load. The difference can be further used to obtain the 95% confidence interval of the Gaussian distribution of the demand response potential, thus achieving uncertainty quantification. The mean vector of actual power consumption obtained by the evaluation model is used as the actual average load in the subsequent calculation of the effective response load.

[0132] Specifically, considering the day-ahead invitation-based demand response, the difference between the baseline average load and the actual average load during the user's declared execution period is used as the effective response load. The demand response potential is determined based on the effective response load. The calculation method is as follows:

[0133]

[0134] Where, P DR potential for user demand response, is the baseline average load during the user demand response period, It is the actual average load during the user demand response period.

[0135] Furthermore, the baseline is calculated from the trend load components derived from the STL algorithm decomposition of sample daily loads selected before the response date. The maximum load occurring in the baseline is called the baseline maximum load, and the arithmetic mean load calculated from the baseline is called the baseline average load. Baselines for the same user participating in the response are divided into workday baselines and non-workday baselines.

[0136] The calculation of the baseline load on working days must follow the following selection criteria: Taking the date on which the demand response event is initiated as the base date, first select the five normal working days before that date (excluding non-working days, statutory holidays and historical demand response implementation days) to form the initial sample set. Then, use the statistical outlier removal method to calculate the average daily load μ of the sample set, and exclude abnormal day data that meet μ<0.25μ or μ>2μ. When there are fewer than five valid samples, candidate days are supplemented in reverse order according to the time series, and the supplement range is limited to 30 natural days (inclusive) before the base date. The termination condition for supplementation under the constraints of load periodicity is the first of the following two: a) the cumulative number of qualified samples reaches 5; b) the supplementation period exceeds 30 days.

[0137] After n sample days are selected, the STL algorithm is used to decompose the selected sample day data to obtain the trend load component T v ={T 1,1 ,…,T 1,96 ,T 2,1 ,…,T n,96 The average value of the 96 load points of the trend load component at the same time every day is calculated as follows:

[0138]

[0139] in T i The average load value at the time is the calculated baseline, and n is the number of sample days selected.

[0140] The calculation of the baseline load on non-working days must follow the following selection criteria: Taking the date of initiation of the demand response event as the base date, first select the three non-working days before that date (excluding working days and historical demand response implementation days) to form the initial sample set. Then, use the statistical outlier removal method to calculate the average daily load μ of the sample set, and exclude abnormal day data that meet μ<0.25μ or μ>2μ. When there are less than 3 valid samples, the candidate days are supplemented in reverse order according to the time series, and the supplement range is limited to 30 natural days (inclusive) before the base date. The termination condition for supplementation under the constraints of the load periodic characteristics is the first of the following two: a) the cumulative number of qualified samples reaches 3; b) the supplementation period exceeds 30 days. When n sample days are selected, the calculation formula is the same as the working day baseline.

[0141] The following are specific embodiments:

[0142] The data set of this embodiment comes from actual data of industrial users participating in peak-shaving demand response in a demand response event in a certain province of China in 2022. The demand response invitation period was 17:00-22:00. The number of invited users was 198. The data for the case analysis included the historical electricity consumption data, historical response data, and demand response declaration data of the users for the 15 regular working days before the response date. The data sampling interval was 15 minutes. The load base was large enough and showed good evaluation results at the overall level, providing a data basis for verifying the technical solution proposed in this application. An evaluation model based on Gaussian process regression was constructed based on this data set.

[0143] Perform STL decomposition on the historical electricity consumption data of the user to be evaluated for 15 regular working days before the response date. Figure 2 To decompose the results, Figure 3 This is the correspondence diagram between industrial user load components and STL algorithm components.

[0144] The production trend factor of the user on the response day is predicted based on the trend load component of the user k days before the response date: the production trend factor obtained from the user's load trend component is 1.59>1, that is, the production trend has an increasing trend compared to the average level of 15 days. The load step matrix of the user is extracted based on the periodic load component of the user k days before the response date: the local change rate threshold α<3 is selected as the load step judgment criterion, the historical data of the 15 days before the response date with a sampling frequency of 15 minutes, the slider time T=1h, and the load step extracted from the load periodic component of the last day before the response date is as follows Figure 4 The multi-dimensional load characteristic index of the user to be evaluated is calculated accordingly, and the load characteristics are obtained by combining the adjustability index.

[0145] Based on the evaluation model, the user's demand response potential is evaluated, and the probability distribution characteristics of the response potential are obtained: mean 11741.39kW, standard deviation 851.23kW. Under the condition of 95% confidence level, the confidence interval of its demand response potential is [10072.98kW, 13409.80kW]. Figure 5 As shown in the actual response data, the user achieved a load adjustment of 13070.80kW during the test period, and its value is located in the upper area of ​​the confidence interval. The empirical results verify the reliability of the evaluation model in evaluating users with high response willingness.

[0146] Notably, the difference between the maximum and minimum load steps in this user's load monitoring data reached 12,773.83 kW. This load step difference not only reflects the fluctuations in equipment operating conditions but also serves as an a priori indicator to help determine the scale of a user's adjustable capacity. In this case study, the actual response reached 102.32% of the maximum load step difference, further demonstrating the user's strong load adjustment flexibility and response execution rate during demand response.

[0147] In order to verify the accuracy of the evaluation method proposed in the present invention, the accuracy rate Z of the evaluation result is defined, which means the ratio of the number of users whose potential is accurately evaluated to the total number of users n who participate in the demand response invitation.

[0148]

[0149] Where z i =1 means there are two situations: 1) the actual demand response amount of user i is less than 0 and the mean of the Gaussian process regression model evaluation is also less than 0, that is, the model predicts that the user will not participate in demand response; 2) the actual demand response amount of user i falls within the 95% confidence interval of the evaluation result of the Gaussian process regression model. Otherwise, z i =0.

[0150] A control group experiment validated the necessity of the demand response willingness indicator in the evaluation model. As shown in Table 1, using a Gaussian process regression evaluation model that incorporates demand response willingness, the evaluation accuracy Z reached 91.4% (the actual response behavior of 181 households out of 198 samples was correctly predicted, including confidence interval coverage and accurate identification of non-participation). When the influencing factor of response willingness was removed, the evaluation accuracy Z dropped significantly to only 57.1% (the actual response behavior of 113 households out of 198 samples was correctly predicted, including confidence interval coverage and accurate identification of non-participation). This precipitous drop in accuracy reveals the strong explanatory power of users' subjective response willingness in potential assessment: in industrial user scenarios, the physical regulation capability of equipment, as represented by load time series characteristics, is only a necessary condition for response potential. However, constrained by factors such as economic incentives and production plans, user participation intention is a sufficient condition for determining the actual conversion rate of potential.

[0151] Table 1 Results of industrial user response potential assessment

[0152]

[0153]

[0154] In order to verify the superiority of the evaluation method proposed in the present invention over other demand response evaluation methods, under the condition of considering the same demand response data granularity, the mean absolute percentage error (MAPE) and the root mean square error (RSME) are used as indicators to measure the evaluation results of the method proposed in the present invention and the other two demand response potential evaluation methods.

[0155] Among them, the evaluation method based on two-stage clustering extracts typical user patterns by constructing user features. The user's demand response potential is expressed as the difference between the maximum load on the user's typical day and the average load during the response period. The evaluation method based on deep sub-domain adaptation constructs a quadratic regression parameter library of the user's typical electricity consumption pattern. On this basis, a demand response potential evaluation method based on parameter feature similarity is used to obtain an estimated value of the user's participation in the response. The comparison results are as follows. Figure 6 shown.

[0156] Depend on Figure 6 It can be seen that the evaluation method (GPR) proposed in this invention has significant advantages. Specifically: Although the traditional two-stage clustering method has the advantage of simple calculation and is suitable for preliminary estimation in engineering scenarios, its evaluation error is large and it is difficult to meet the needs of power companies for precise scheduling; the adaptive learning method improves the evaluation accuracy compared to the traditional clustering method by introducing a user similarity analysis mechanism. However, because it does not fully consider the dynamic characteristics of user response willingness and the differences in load conditions during the invitation period, even for industrial user groups with the same electricity consumption mode, its average absolute percentage error is still 16.4 percentage points higher than the GPR method.

[0157] It's worth noting that the demand response potential assessment results of the two aforementioned methods are numerical outputs, rather than the probabilistic distributions generated by the method of this invention. When power companies or load aggregators make actual dispatch decisions, confidence intervals for the response volume can help them understand the potential response volume and the corresponding uncertainty, providing critical information support for developing risk-controlled dispatch strategies.

[0158] The above examples demonstrate that the proposed assessment method effectively improves the refinement of demand response potential assessments. This technical advantage can fundamentally avoid scheduling mismatches caused by assessment bias, and has significant application value in ensuring the proper allocation of winning capacity for demand response projects and the reliable execution of dispatch instructions.

[0159] In summary, the present invention designs a Gaussian process regression model that integrates STL decomposition and load step characteristics to evaluate the response potential of industrial users. By introducing the decomposition mechanism of trend load and periodic load, and the response willingness feature system constructed in combination with historical response behavior, combined with the Gaussian process regression algorithm, a high-precision and explainable evaluation of the demand response potential of industrial users is achieved, and a breakthrough is made in the reliance of traditional evaluation methods on equipment-level monitoring data. Experimental results show that the method of the present invention is superior to existing mainstream methods in both evaluation accuracy and uncertainty quantification, with an evaluation accuracy of 91.4%. It shows good generalization ability and applicability in industrial load scenarios with strong uncertainty and heterogeneity, and can provide flexible resource scheduling decision support with quantifiable risks for high-proportion new energy power grids.

[0160] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A quantitative evaluation method for industrial user response potential, characterized in that: include: Model building phase: A training set is constructed based on the load characteristics of multiple users and the corresponding actual average loads; the load characteristics include an adjustability index and a multidimensional load characteristic index. The adjustability index is used to measure the user's physical adjustment ability and is obtained based on the user's historical electricity usage data; the multidimensional load characteristic index is used to measure the user's demand response willingness and is obtained based on the user's historical electricity usage data and historical response data; Based on the training set data, the mapping relationship between load characteristics and actual average load is described by Gaussian process to obtain the evaluation model; Model application phase: Based on the evaluation model, the actual average load is evaluated according to the load characteristics of the user to be evaluated; The difference between the baseline average load and the actual average load during the user's declared execution period is taken as the effective response load, and the demand response potential is quantitatively evaluated based on the effective response load.

2. The method for quantitatively evaluating the response potential of industrial users according to claim 1, wherein: The adjustability index includes the production trend factor and load step matrix of the user response day, and the acquisition method is: The STL algorithm is used to split the historical electricity consumption data into trend load component, periodic load component and residual component; Determine the production trend factor based on the trend load component; The load step matrix is ​​determined based on the periodic load components.

3. The method for quantitatively evaluating the response potential of industrial users according to claim 2, wherein: The production trend factor is determined based on the trend load component. The calculation formula is: Among them, θ is the production trend factor of the response day, To respond to the nth day r Daily trend load component series, e k Indicates the first n r The mean value of the trend load component on the kth day in the world.

4. The method for quantitatively evaluating the response potential of industrial users according to claim 2, wherein: Determine the load step matrix based on the periodic load components, including: A load step is a steady-state load segment in a periodic load component whose duration exceeds a preset time threshold and whose local change rate is lower than a preset change rate threshold. The steady-state load segments in the periodic load component are combined to obtain a load step matrix.

5. The method for quantitatively evaluating the response potential of industrial users according to claim 4, wherein: The preset time threshold is 1 hour, the preset change rate threshold is 3; the calculation formula of the local change rate α is: Where: p max 、p min are the maximum and minimum load values ​​in the local load data window respectively; Δp max ,Δp mean ,Δp std are the maximum value, mean value and standard deviation of the load difference between adjacent sampling points, respectively.

6. The method for quantitatively evaluating the response potential of industrial users according to claim 2, wherein: The method for determining the baseline average load is: The baseline is determined based on the trend load component, and the arithmetic mean load calculated based on the baseline is used as the baseline average load.

7. The method for quantitatively evaluating the response potential of industrial users according to claim 1, wherein: The multi-dimensional load characteristic indicators include historical declaration participation rate, historical effective response rate, industry relative added value, relative response volume and invitation price ratio.

8. The method for quantitatively evaluating the response potential of industrial users according to claim 7, wherein: The formula for calculating the relative added value of the industry is: Among them, D p,t,his is the relative added value of the industry p to which the user belongs at the demand response time t, y p,t,q Y is the total demand response invitation quantity of industry p at time t in the qth demand response invitation, t,q is the total demand response invitation quantity at demand response time t in the qth demand response invitation, and M is the number of demand response invitations.

9. The method for quantitatively evaluating the response potential of industrial users according to any one of claims 1 to 8, wherein: According to the user's load characteristics, the mean vector and variance vector of actual electricity consumption are obtained through the evaluation model, thereby obtaining the confidence interval and realizing the quantification of evaluation uncertainty.

10. A quantitative evaluation system for industrial user response potential, characterized in that: The method comprises a processor configured to execute the method for quantitatively evaluating the response potential of industrial users according to any one of claims 1 to 9.

Images