A Joint Optimization Control Method and Device for Scalable Adjustable Resources

Abstract

The present invention discloses a method and device for jointly optimizing and controlling scalable adjustable resources, including: considering the dynamic process of cluster response, constructing multi-dimensional response performance parameters of adjustable resources, and solving the comprehensive response performance parameters of adjustable resources; considering the impact of response uncertainty on the response performance of adjustable resources, constructing a response performance analysis model; based on the response performance analysis model, decomposing and extracting features from the historical time series data of relevant parameters in the model through data analysis; according to the extracted feature set, obtaining the parameters of the adjustable resource response performance analysis model at multiple time nodes; considering the difference in adjustable resource response performance, determining the objective function of the joint optimization control of adjustable resources for VPP to participate in peak shaving and frequency modulation; determining the constraint conditions for the joint optimization control of adjustable resources for VPP to participate in peak shaving and frequency modulation; substituting the relevant parameters of the model, quickly solving the above linear optimization problem through Cplex software, and adjusting the electricity price structure and electricity load of users based on the solution results to reduce the transmission risk of transmission lines.

Description

Technical Field

[0001] The present invention relates to the field of adjustable resource response control, and in particular, to a joint optimization control method and device for large-scale adjustable resources. Background Art

[0002] With the substantial growth in the proportion of renewable energy, electric vehicles, and end-user re-electrification, problems such as new energy accommodation and grid balance regulation have become more prominent, and the challenges faced by the safe and stable operation of the power grid are increasing. Through advanced digital technologies, the management and regulation of adjustable resource clusters in the form of virtual power plants, load aggregators, etc. are used to achieve peak shaving, frequency modulation, and mitigation of the volatility and accommodation of new energy output in the new power system, which has become a current research hotspot.

[0003] Different grid balance regulations have different requirements for the response performance of adjustable resources. A virtual power plant is an advanced control method for large-scale controllable resources. Optimizing the control of the aggregated adjustable resources requires not only considering the power characteristics of the adjustable resources themselves but also the dynamic characteristics when the adjustable resources participate in the response. Current research on the optimization control strategy of adjustable resources in virtual power plants mainly focuses on three aspects: operation management architecture, optimization model, and optimization algorithm. These studies mainly analyze how to improve the benefits through their own optimization control when virtual power plants participate in grid balance regulation, and usually use the rated power or current power consumption of adjustable resources as the boundary for optimizing the solution of response models and algorithms. However, the adjustable resources in virtual power plants not only include directly controllable adjustable resources such as energy storage and gas-fired power generation sets but also include adjustable resources such as industrial and commercial users, air-conditioning equipment, and electric vehicles that are regulated through incentive prices. The response performance of this part of the resources usually does not reach the set boundary of optimization, thus affecting the accuracy of the optimization model and algorithm solution. Therefore, the research on the optimization control of adjustable resources in virtual power plants should fully consider the response performance boundary of the resources themselves when participating in the response. Summary of the Invention

[0004] The present invention provides a joint optimization control method and device for large-scale adjustable resources. Combining the differences in performance during the response process of adjustable resources, it analyzes the response performance of adjustable resource clusters from multiple dimensions, improves the accuracy of response performance analysis, and improves the resource management efficiency of virtual power plants through joint optimization control. See the following description for details:

[0005] A joint optimization control method for large-scale adjustable resources, the method includes:

[0006] Considering the dynamic process of cluster response, constructing multi-dimensional response performance parameters of adjustable resources, and solving the comprehensive response performance parameters of adjustable resources;

[0007] Considering the impact of response uncertainty on the response performance of adjustable resources, a response performance analysis model is constructed; based on the response performance analysis model, the historical time series data of relevant parameters in the model are decomposed and features are extracted through data analysis;

[0008] According to the extracted feature set, the parameters of the adjustable resource response performance analysis model at multiple time nodes are obtained;

[0009] Considering the difference in the response performance of adjustable resources, the objective function of the joint optimal control of adjustable resources for VPP to participate in peak shaving and frequency modulation is determined; the constraint conditions for the joint optimal control of adjustable resources for VPP to participate in peak shaving and frequency modulation are determined;

[0010] Substitute the relevant parameters of the model, and quickly solve the above linear optimization problem through Cplex software. Based on the solution results, adjust the electricity price structure and electricity load of users to reduce the transmission risk of transmission lines.

[0011] Among them, the response performance analysis model describes the relationship between the multi-dimensional performance parameters of users' participation in demand response and the incentive intensity as a piecewise function,

[0012] The adjustable resource response performance parameters are:

[0013]

[0014] In the formula: η represents the multi-dimensional performance parameters of the adjustable resource response; r 4 The parameter is a random parameter considering the impact of uncertainty. The random characteristics of the adjustable resource participating in the response process can be characterized by the random change of r 4 ; δ is the incentive intensity;

[0015] The uncertainty parameter model of the adjustable resource response performance parameter is:

[0016]

[0017] Among them, r 1 , r 2 , r 3 are known deterministic model parameters; r 4 follows a normal distribution that satisfies a certain law, where the mean of this distribution is and the standard deviation is The parameters μ 0 and σ 0 are estimated values obtained by point estimation based on the historical response data set of r 4 , and these estimated values are regarded as the normal distribution parameters that r 4 satisfies.

[0018] Furthermore, the constraint conditions for determining the joint optimal control of adjustable resources for VPP to participate in peak shaving and frequency modulation are:

[0019] 1) VPP internal power balance constraint;

[0020] 1. Frequency modulation capacity power balance constraint:

[0021] 2. Peak shaving capacity power balance constraint:

[0022] 3. Power balance constraint of the battery:

[0023] Among them, is the bidding power of the z-th battery participating in frequency modulation and peak shaving auxiliary services at time node t;

[0024] 2) Adjustable resource cluster regulation constraint;

[0025]

[0026]

[0027] 3) Energy storage charge and discharge constraint;

[0028]

[0029]

[0030] The state of charge SOC of the battery at each moment is represented as S t , and is calculated by the following formula:

[0031] S t =(1 - ε)S t-1 +P ch Δt / μ c -P dis Δt / μ d (7)

[0032] In the formula: ε is the self-discharge rate of the battery; μ c is the charging efficiency; μ d is the discharging efficiency.

[0033] A joint optimization control device for adjustable resources of a virtual power plant, the device includes: a data processing unit, a data decomposition unit, a data processing unit, a feature extraction unit based on SAE, and a resource joint optimization control unit,

[0034] The data processing unit is used for collecting historical power consumption data and response data, identifying and correcting abnormal points;

[0035] The data decomposition unit is used for decomposing historical power consumption data to obtain each IMF component;

[0036] The SAE-based feature extraction unit is used to obtain the feature relationship data h of the IMF components and the adjustable resource response data i ;

[0037] The resource joint optimization control unit is used to calculate the response performance parameters of the adjustable resources at each time node and solve the optimization control strategy

[0038] The beneficial effects of the technical solution provided by the present invention are as follows

[0039] 1. The present invention combines the differences in the electricity consumption characteristics of adjustable resources and the dynamic complementary characteristics of cluster responses, analyzes the response performance of the adjustable resource cluster from multiple dimensions, improves the accuracy of potential analysis, and fully considers the impact of uncertain factors in the response process of adjustable resources, improving the practicality of response performance analysis; based on the solution results, the electricity price structure and electricity load of users are adjusted to reduce the transmission risk of transmission lines

[0040] 2. The present invention processes and extracts key features from the historical electricity consumption power time series data and historical response data of adjustable resources through the improved EEMD algorithm and SAE algorithm, and then uses these features to calculate the key parameters in the response performance analysis model to form the probability distribution of the response performance parameters of the adjustable resource cluster at each time node

[0041] 3. The present invention fully considers the influence of the response performance of adjustable resources themselves in the joint optimization control strategy of adjustable resources in virtual power plants, making the boundary conditions of the optimization algorithm more in line with the actual situation, reducing the difficulties caused by the insufficient response capacity when virtual power plants participate in ancillary services, and the present invention combines the differences in performance during the response process of adjustable resources, analyzes the response performance of the adjustable resource cluster from multiple dimensions, and improves the accuracy of response performance analysis Description of the Drawings

[0042] Figure 1 It is a schematic diagram of the response performance analysis model under the influence of uncertain factors

[0043] Figure 2 It is a structural block diagram of the response performance analysis of an adjustable resource cluster

[0044] Figure 3 It is a flowchart of the response performance analysis method considering the dynamic process of the adjustable resource cluster response

[0045] Figure 4 It is a flowchart for obtaining the original input parameters of feature extraction

[0046] Figure 5 It is a flowchart of the response feature extraction and performance analysis of adjustable resources

[0047] Figure 6 is the real-time electricity purchase price curve in the spot market;

[0048] Figure 7 is the decomposition result diagram of the original load sequence based on EEMD;

[0049] Figure 8 is the calculation result diagram of the parameters of the response performance analysis model for each time node;

[0050] Figure 9 is the probability distribution result diagram of the response performance parameters of the adjustable resource cluster;

[0051] Figure 10 is the calculation result diagram of the effective response load reduction power of the adjustable resource cluster.

[0052] Figure 11 is the intraday call price, frequency regulation, peak shaving clearing price, and response penalty price curve of the adjustable resource;

[0053] Figure 12 is the comparison diagram of the revenue results of the joint optimization control strategy method considering the cluster response performance proposed by the present invention. Detailed implementation manners

[0054] To make the objectives, technical solutions, and advantages of the present invention clearer, the following further describes the embodiments of the present invention in detail.

[0055] Embodiment 1

[0056] A joint optimization control method for large-scale adjustable resources, the method includes the following steps:

[0057] Step 101: Considering the dynamic process of cluster response, construct multi-dimensional response performance parameters of adjustable resources, and then solve the comprehensive response performance parameters of adjustable resources;

[0058] Step 102: Considering the influence of response uncertainty on the response performance of adjustable resources, construct a response performance analysis model;

[0059] Step 103: Based on the response performance analysis model, decompose and extract features from the historical time series data of relevant parameters in the model through data analysis methods;

[0060] Step 104: According to the extracted feature set, obtain the parameters of the response performance analysis model of adjustable resources at multiple time nodes;

[0061] Step 105: Considering the response performance differences of adjustable resources, determine the objective function of the joint optimization control of adjustable resources participating in peak shaving and frequency regulation for the VPP;

[0062] Step 106: Determine the constraint conditions for the joint optimization control of adjustable resources for VPP to participate in peak shaving and frequency modulation;

[0063] Step 107: Substitute the relevant parameters of the model, quickly solve the above linear optimization problem through Cplex software, and adjust the electricity price structure and electricity load of users based on the solution results to reduce the transmission risk of transmission lines.

[0064] Among them, considering the dynamic process of cluster response in Step 101, constructing the multi-dimensional response performance parameters of adjustable resources, and then solving the comprehensive response performance parameters of adjustable resources, specifically including: the maximum adjustment power P adjust , the response reaction time τ r , the average response rate v a , the steady-state response duration T s , the available response frequency f r , specifically:

[0065] 1) The maximum adjustment power P adjust . It represents the difference between the total electricity consumption power when the overall output reaches relative stability after the adjustable resource cluster participates in the response and the initial electricity consumption power.

[0066] 2) The response reaction time τ r . It refers to the time from when the adjustable resource manager issues a response signal to when it starts to participate in the response.

[0067] 3) The average response rate v a . It represents the adjustment power of the adjustable resource cluster per unit time after receiving the response signal.

[0068] 4) The steady-state response duration T s . It represents the duration of the stable response state after the cluster receives the regulation capacity signal. This index reflects the ability of the adjustable resource cluster to participate in refined regulation.

[0069] 5) The available response frequency f r : It represents the difference between the maximum power reached after the cluster response ends and the power before the response.

[0070] 6) The comprehensive response performance parameter of adjustable resources:

[0071]

[0072]

[0073] In the formula, is the normalized result of each cluster participating under VPP, that is, the original value of this parameter is divided by the maximum value of the corresponding parameter in each cluster within VPP. Taking the response reaction time as an example, it can be obtained according to formula (3)

[0074]

[0075] Among them, in step 102, considering the impact of response uncertainty on the response performance of adjustable resources, a response performance analysis model is constructed, specifically as follows:

[0076] 1) Establish as Figure 2 shown in the relationship between the response performance of adjustable resources and the incentive intensity under the influence of uncertainty factors. This model describes the relationship between the multi-dimensional performance parameters of users' participation in demand response and the incentive intensity as a piecewise function. Among them, A is the dead zone inflection point of the incentive intensity, and its abscissa is r 1 ; B is the saturation zone inflection point of the incentive intensity, and its abscissa is r 2 , and the ordinate is r 3 .

[0077] 2) The response performance parameters of adjustable resources can be obtained by Equation (4):

[0078]

[0079] In the formula: η represents the multi-dimensional performance parameters of the response of adjustable resources; r 4 The parameter is a random parameter considering the influence of uncertainty. Through the random variation of r 4 , the random characteristics of the process of adjustable resources participating in the response can be characterized; δ is the incentive intensity.

[0080] 3) The uncertainty parameter model of the response performance parameters of adjustable resources can be obtained according to Equation (5):

[0081]

[0082] Among them, r 1 , r 2 , r 3 are known deterministic model parameters; r 4 is a normal distribution satisfying certain rules, where the mean of this distribution is and the standard deviation is The parameters μ 0 and σ 0 are estimated values obtained by point estimation based on the historical response data set of r 4 . These estimated values are regarded as the normal distribution parameters satisfied by r 4 .

[0083] Among them, in step 103, based on the response performance analysis model, the historical time series data of relevant parameters in the model are decomposed and features are extracted through data analysis methods, specifically as follows:

[0084] 1) The daily power consumption X collected kThe data is segmented at a single time node to obtain the time series data x of the electricity consumption power k,t , where k is the adjustable resource number and t represents the selected time node;

[0085] 2) Use the improved Ensemble Empirical Mode Decomposition (EEMD) to decompose the time series data x of the electricity consumption power k,t into a series of Intrinsic Mode Function (IMF) components and a residual component, specifically:

[0086] ① Add white noise with a normal distribution to the initial electricity consumption time series x according to formula (6) k,t , where: n is the normalized white noise, σ x is the signal standard deviation, a is the proportionality coefficient, and the signal-to-noise mixed sequence x′ k,t is obtained;

[0087] x′ k,t = x k,t + a·σ x ·n (6)

[0088] ② Take x′ k,t as the initial decomposition time series, and then perform EMD decomposition to obtain each IMF component;

[0089] ③ Obtain the remaining component r k,i from formula (7). If i = 1, then r k,0 = x′ k,t ;

[0090] r k,i = r k,i-1 - imf k,i (7)

[0091] ④ Take the decomposition mode component h k,j-1 = r k,i as the initial sequence for obtaining the i-th mode component, and extract the local extreme values;

[0092] ⑤ Use spline interpolation to form the upper and lower envelope lines, and calculate the mean m k,j-1 of the upper and lower envelope lines;

[0093] ⑥ Subtract the decomposition mode component h k,j-1 from the mean m k,j-1 of the upper and lower envelope lines according to formula (8) to obtain the decomposition mode component h k,j in the next iteration process:

[0094] h k,j = h k,j-1 - m k,j-1(8)

[0095] ⑦ Judge whether the extraction of the i-th modal component meets the iteration stop condition. If it meets, output h k,j as imf k,i , if it does not meet, repeat the above process until the iteration stop condition is met, and output the i-th IMF component imf k,i ;

[0096] ⑧ Add a new white noise sequence of normal distribution to obtain a new IMF component, and take the integrated average of the obtained IMF as the final result.

[0097] ⑨ On the basis of the above EEMD algorithm, screen each IMF component through the maximum mutual information coefficient (MIC), extract the components with greater correlation with each influencing factor, and calculate the MIC value of each influencing factor and the IMF component according to formula (9):

[0098]

[0099] where X and Y represent the divided regions corresponding to x and y respectively, and the size of B is usually set to about the 0.6th power of the data volume.

[0100] ⑩ Set the MIC value threshold, screen out the IMF components that meet the MIC value, and combine the screened IMF components to form a matrix Y i =[imf 1,i ; imf 2,i ; …; imf n,i ;

[0101] 3) Combine the decomposed IMF component data and the historical response data set to form an initial input matrix Y i and D i ;

[0102] 4) Use the Stacked Auto-Encoder (SAE) to extract features from the extracted and screened IMF components and form a feature parameter set, specifically:[[]]

[0103] ① Given a cluster with n adjustable resources, the input of the SAE network is the screened IMF index data set Y i =[imf 1,i ; imf 2,i ; …; imf n,i , then the encoding process is to transform the input into the hidden layer state h i , which can be expressed by formula (10);

[0104] h i =f(w i ·Yi +b i ) (10)

[0105] Among them, w i and b i are respectively the weight matrix and the bias vector between the input layer and the hidden layer for the feature extraction of the i-th eigencomponent at a single time node.

[0106] ② The decoding process is to reconstruct the hidden layer state h i into Y′ i through the decoding function g, which can be expressed by formula (11);

[0107] Y′ i = g(w′ i ·h i +b′ i ) (11)

[0108] Among them, w′ i and b′ i respectively represent the weight matrix and the bias vector between the hidden layer and the output layer for the feature extraction of the i-th eigencomponent.

[0109] ③ Set the goal of network training to minimize the reconstruction error, and through continuous self-iteration of the network encoding process and the decoding process, finally make the data Y i input in the encoding process as equal as possible to the output Y′ i after decoding, which can be expressed by formula (12);

[0110] Θ AE (w i ,b i ,w′ i ,b′ i ) = argminL(Y i ,Y′ i ) (12)

[0111] Among them, Θ AE is the value of w i , b i , w′ i and b′ i , and L is the error function between Y i and Y′ i .

[0112] ④ Output the feature parameter set h i extracted by the hidden layer of the trained neural network for parameter calculation of the subsequent adjustable resource response performance analysis model.

[0113] Step 104: According to the extracted feature set, obtain the parameters of the adjustable resource response performance analysis model at multiple time nodes, specifically:

[0114] 1) Based on the extracted set of characteristic parameters h i , the deterministic parameters of the adjustable resource response potential model are mined by least squares fitting, and the following relational expressions are obtained:

[0115]

[0116] In the formula: b 1j (t), b 2j (t), b 3j (t) are the characteristic coefficients of the key parameters at a single time node t, U j represents the principal component of the jth adjustable resource electricity consumption characteristic index extracted by the principal component analysis method, a ij (t) is the coefficient of the ith characteristic index in the composition of the jth principal component, h i is the value of the ith characteristic index.

[0117] 2) Perform point estimation on the response historical data set to obtain the estimated values at each time node, and regard this estimated value as the normal distribution parameter satisfied by r 4 (t);

[0118]

[0119] Among them, the mean reference value μ(t), the variance reference value σ 2 (t), the mean fuzzy value μ e (t) and the variance fuzzy value σ e (t) are the uncertainty parameters describing the response potential model of the adjustable resource at a specific time node.

[0120] 3) By performing rolling analysis on the time series data of each time node of the adjustable resource cluster, obtain the probability distribution of the multi-dimensional response performance parameters of the cluster at different times;

[0121] Step 105: According to the analyzed adjustable resource response performance, determine the objective function of the joint resource optimization strategy for the VPP to participate in peak shaving and frequency modulation, specifically:

[0122] 1) The optimization operation objective for the VPP to participate in peak shaving and frequency modulation services is to maximize the net profit W;

[0123]

[0124] Among them, W U,t is the response control cost of the VPP to the adjustable resource, W ES,t is the call cost of the battery in the VPP; W t pun is the penalty cost for the VPP not meeting the response demand when participating in the response; W tAGC and W t reg respectively represent the benefits of VPP participating in peak shaving and frequency modulation assistance, mainly including the settlement benefits of day-ahead response and intraday response, specifically:

[0125]

[0126]

[0127] Among them, c AC,t and c rc,t respectively represent the clearing prices of frequency modulation and peak shaving in the day-ahead market; c e,t is the predicted value of the intraday call power price; and are respectively the winning bid frequency modulation and peak shaving capacities of VPP participating in frequency modulation and peak shaving; and are respectively the expected frequency modulation and peak shaving powers of VPP to be called intraday.

[0128] 2) The winning bid frequency modulation and peak shaving capacities at each time node and depend on the capacity of VPP participating in the bid and the corresponding winning bid probability;

[0129]

[0130]

[0131] Among them, and respectively represent the bid capacities of VPP participating in the two services of frequency modulation and peak shaving; ρ 1 and ρ 2 respectively represent the winning bid probabilities of VPP participating in frequency modulation and peak shaving in the ancillary service market;

[0132] 3) The bidding probabilities ρ 1 and ρ 2 of VPP participating in frequency modulation and peak shaving are mainly related to the response performance of its internal adjustable resources.

[0133]

[0134]

[0135] Among them, and are respectively the comprehensive response performance parameters of the adjustable resource clusters for peak shaving and frequency modulation within VPP; k AGC and k reg are the winning bid probability and response performance coefficients of VPP, and this value can be obtained according to the historical response winning bid numbers of adjustable resources.

[0136]

[0137]

[0138] Among them, S AGC and S reg are respectively the historical winning bid times for frequency regulation and peak shaving, and are respectively the historical bidding times for frequency regulation and peak shaving, and are the mean values of the comprehensive response performance parameters of the adjustable resource cluster within the VPP;

[0139]

[0140]

[0141] Among them, N is the number of days of historical response data of the VPP adjustable resources obtained, and T is the number of data acquisitions within a day, usually taking T = 96;

[0142] 4) The response cost W of the VPP participating in frequency regulation and peak shaving U,t and W ES,t can be obtained according to formulas (26) and (27);

[0143]

[0144]

[0145] Among them, and are respectively the response power of the i-th adjustable resource cluster participating in frequency regulation and the response power of the j-th adjustable resource cluster participating in peak shaving; c ES represents the unit depreciation and maintenance cost of the battery; P ch,ESz,t and P dis,ES,t are respectively the charge and discharge power of the z-th battery at the t time node; N ES represents the number of battery units; and respectively represent the incentive prices of the adjustable resource cluster U.

[0146] 5) The penalty cost W for the VPP not meeting the response demand when participating in the response t pun can be obtained according to formula (28):

[0147]

[0148] Among them, and respectively represent the actual completion amounts of the day-ahead response tasks, The penalty price for responding to the task completion deviation.

[0149] Step 106: Determine the constraint conditions for the joint optimization control strategy of the adjustable resources for peak shaving and frequency modulation of the VPP, specifically including: the internal power balance constraint of the VPP, the adjustment constraint of the adjustable resource cluster, and the charge and discharge constraint of the energy storage.

[0150] 1) The internal power balance constraint of the VPP.

[0151] 1. The frequency modulation capacity power balance constraint:

[0152] 2. The peak shaving capacity power balance constraint:

[0153] 3. The power balance constraint of the battery:

[0154] Among them, is the bidding power of the z-th battery participating in frequency modulation and peak shaving auxiliary services at the t-th time node.

[0155] 2) The adjustment constraint of the adjustable resource cluster.

[0156]

[0157]

[0158] 3) The charge and discharge constraint of the energy storage.

[0159]

[0160]

[0161] The state of charge SOC of the battery at each moment is represented as S t , and is calculated by the following formula:

[0162] S t =(1 - ε)S t-1 +P ch Δtμ c -P dis Δt / μ d (33)

[0163] In the formula: ε is the self-discharge rate of the battery; μ c is the charging efficiency; μ d is the discharging efficiency. Usually, the total battery power also needs to be maintained within a certain limit:

[0164] S min ≤S t ≤S max (34)

[0165] S init = S T (35)

[0166] Among them, S min , S max and S T respectively represent the minimum state of charge, the maximum state of charge and the initial state of charge of the storage battery. And, since the storage battery cannot be charged and discharged simultaneously, there is:

[0167] P ch,ESz,t P dis,ESz,t = 0 (36)

[0168] Formula (36) introduces a non-linear factor into the VPP joint resource optimization, and introduces the variable α z to transform formula (36) into two sets of inequality relations of formulas (37) and (38).

[0169]

[0170]

[0171] When α z = 1, the storage battery z may be in the charging state, but must not be in the discharging state; when α z = 0, the storage battery z may be in the discharging state, but must not be in the charging state.

[0172] Embodiment 2

[0173] The following further introduces the solution in Embodiment 1 in combination with specific experimental data, as described in detail below:

[0174] Obtained through the data acquisition and monitoring control system (SCADA) or the advanced metering infrastructure (AMI). In the power system, AMI provides business services based on the cloud platform through data acquisition, data management and application of smart meters, meets the access capabilities of tens of millions of meters and millions of DCUs, and realizes multiple services such as prepaid and postpaid for users, electricity consumption information analysis, and large user billing support. At the same time, relying on the energy consumption data accumulated by automated systems such as marketing, fee control, and collection, data analysis is carried out to build a user power load prediction model. Using the historical data of enterprise electricity consumption, price and fee, and electricity load data, energy consumption analysis and bill optimization work are provided for users, and the electricity price structure, load curve, etc. of users are optimized.

[0175] Taking the adjustable resource cluster electricity load data from May 10, 2019 to February 25, 2020 obtained by simulation as the experimental data and inputting it into the EEMD model, the electricity price adopts Figure 6The real-time purchase price curve of the spot market shown, with the data collection frequency being 15 min / point. Select the power consumption time series data at 12:00 d, and decompose the original load sequence through EEMD, as Figure 7 shown.

[0176] For the convenience of data processing, the daily type data can be digitized (using 1 and 0 to represent holidays and non-holidays respectively, 1 to 12 represent January to December in sequence, 1 to 7 represent Monday to Sunday in sequence, and 1 to 24 represent 1 to 24 hours in a day);

[0177] In step S22 of this embodiment, the power consumption load data of the adjustable resource cluster from May 10, 2019 to February 25, 2020 obtained by simulation is used as the experimental data set. Since it is difficult to obtain temperature data at 15 min / point, the daily maximum temperature and the daily minimum temperature are used in the example to analyze the influence of temperature on the power consumption characteristics of each time series component of the adjustable resources. The probability distribution of the response potential of the adjustable resource cluster at different time nodes is finally obtained as Figure 8 shown.

[0178] According to the requirements of different response services for different dimensional indicators of adjustable resources in Table 1, the effective response potential of adjustable resources in different service scenarios is further obtained. Taking four different types of adjustable resource clusters simulated in this embodiment as examples, the load type composition and simulation parameters of each cluster are shown in Tables 2 and 3. The effective response load reduction power distribution of the adjustable resource cluster is finally obtained as Figure 10 shown.

[0179] To verify the superiority of the response performance analysis method proposed in the present invention, the performance analysis results of each time node in the example are compared. The average value of the simulated response performance parameters is used as the true value, and the average error of the potential performance parameter results of each time node within a day is used as the comparison index. Table 4 gives the comparison of the calculation results of several response potential performance analysis algorithms. From this result, it can be seen that although the analysis error of directly using the SAE and FCL methods for supervised label training is small, the calculation efficiency of this method is low, and it is difficult to obtain a large amount of historical data of the response potential of each time node in practice. At the same time, due to the complexity of the power consumption time series data of the adjustable resource cluster, the feature extraction effect of the PCA algorithm is poor, resulting in a large final analysis error. Among the power time series data decomposition algorithms, although the improved EEMD algorithm increases a small amount of calculation time, it can effectively reduce the analysis error compared with the EMD algorithm.

[0180] To verify the superiority of the optimization algorithm proposed in the present invention, 3 batteries ES1, ES2, and ES3 with the parameters shown in Table 3 are simulated. The rated capacities S N of ES1 to ES3 are all 1500 kW·h, the initial capacity is 1000 kW·h, and the minimum allowable capacity is 0.2SN , the maximum allowable capacity is 0.9S N , the maximum charge-discharge power and efficiency are the same. The intraday call price of adjustable resources, the clearing prices for frequency regulation and peak shaving, and the response penalty price are as Figure 11 shown. To ensure the reliability of the results, 40 groups of possible response performance parameters of Cluster II are selected by randomly sampling, and the average response power of Cluster II is used as the actual response power. The rated power of the adjustable resource response is used as the optimization result of the constraint (Case2), and is compared with the optimization method considering the cluster response performance proposed in the present invention (Case1). From Figure 12 the experimental simulation results shown, the method proposed in the present invention can effectively reduce the penalty cost of adjustable resources and improve the market revenue of virtual power plants participating in ancillary services.

[0181] Table 1

[0182]

[0183] Table 2

[0184]

[0185] Table 3

[0186]

[0187]

[0188] Table 4

[0189]

[0190] Table 5

[0191]

[0192] In a second aspect, a joint optimization control device for adjustable resources of a virtual power plant is provided. The structural block diagram of the device is as Figure 3 shown, and it is mainly used to execute the optimization method provided in the first aspect, specifically including a data processing unit, a data decomposition unit, a data processing unit, a feature extraction unit based on SAE, and a resource joint optimization control unit.

[0193] The data processing unit is used for collecting historical power consumption data and response data, identifying and correcting abnormal points;

[0194] The data decomposition unit is used for decomposing historical power consumption data to obtain each IMF component;

[0195] The feature extraction unit based on SAE is used for obtaining the characteristic relationship data h of the IMF component and the adjustable resource response data i。

[0196] The resource joint optimization control unit is used to obtain the response performance parameters of adjustable resources at each time node and solve the optimization control strategy algorithm proposed by the present invention.

[0197] Its further technical solution is that the data preprocessing unit includes a data acquisition module, an abnormal data point identification module, and an abnormal data point correction module;

[0198] The data acquisition module is used to obtain the historical power time series data x k,t and historical response data D of the evaluation target from the database;

[0199] The abnormal data point identification module is used to identify abnormal data from the acquired data;

[0200] The abnormal data point correction module is used to correct the identified abnormal data.

[0201] Its further technical solution is that the data decomposition unit includes a white noise mixing module, a data decomposition module, an IMF mean calculation module, and a MIC value acquisition module;

[0202] The white noise mixing module is used to add white noise components to the historical power time series data x k,t to obtain the time series data x' to be decomposed k,t ;

[0203] The data decomposition module is used to perform EMD decomposition on the time series data with different white noise components added to obtain the IMF components under different white noise components;

[0204] The IMF mean calculation module is used to calculate the mean of the IMF under different components obtained as the finally output IMF component imf;

[0205] The MIC value acquisition module is used to obtain the maximum mutual trust coefficient MIC value between each obtained IMF and the historical response data;

[0206] Its further technical solution is that the data processing unit includes an IMF component screening module, a data combination module, and a data normalization module;

[0207] The IMF component screening module is used to screen out the IMF components imf with a relatively large correlation with the historical response data i for further feature extraction;

[0208] The data combination module is used to combine the data of each screened IMF component and the historical response data set to form matrices Y i and D i ;

[0209] The data normalization module is used to eliminate the dimensions of different factors and limit the data range for facilitating data processing;

[0210] Its further technical solution is that the SAE-based feature extraction unit includes a data encoding module, a data decoding module, a data training module, and a feature output module;

[0211] The data encoding module is used to encode the network input data and retain the data feature information;

[0212] The data decoding module is used to decode the encoded data to obtain a set of data containing feature information again;

[0213] The data training module is used to compare the data feature data before encoding and the data feature data after decoding, and reduce the feature difference between the two sets of data through iterative training;

[0214] The feature output module is used to output the hidden layer feature parameter set h of the trained neural network i .

[0215] Its further technical solution is that the resource joint optimization control unit includes a model parameter calculation module, a response performance analysis module, and an optimization algorithm calculation module;

[0216] The model parameter calculation module is used to calculate each parameter r in the response performance analysis model according to the feature parameter set h i and obtain r 1 , r 2 , r 3 and r 4 ;

[0217] The response performance analysis module is used to calculate the probability distribution η of the adjustable resource response performance parameter and the comprehensive response performance parameter and

[0218] The optimization algorithm calculation module is used to solve the constructed adjustable resource joint optimization control algorithm.

[0219] In the embodiments of the present invention, except for those with special descriptions for the models of each device, the models of other devices are not limited, and any device that can complete the above functions can be used.

[0220] Those skilled in the art can understand that the drawings are only schematic diagrams of a preferred embodiment, and the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0221] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A joint optimization control method for large-scale adjustable resources, characterized in that, the method includes: Considering the cluster response dynamic process, constructing multi-dimensional response performance parameters of adjustable resources, and solving the comprehensive response performance parameters of adjustable resources; Considering the impact of response uncertainty on the response performance of adjustable resources, constructing a response performance analysis model; based on the response performance analysis model, decomposing and extracting features from the historical time series data of relevant parameters in the model through data analysis; According to the extracted feature set, obtaining the parameters of the adjustable resource response performance analysis model at multiple time nodes; Considering the difference in adjustable resource response performance, determining the objective function of the joint optimization control of adjustable resources for VPP to participate in peak shaving and frequency modulation; determining the constraint conditions for the joint optimization control of adjustable resources for VPP to participate in peak shaving and frequency modulation; Substituting the relevant parameters of the model, quickly solving the above linear optimization problem through Cplex software, and adjusting the electricity price structure and electricity load of users based on the solution results to reduce the transmission risk of transmission lines; Among them, the constraint conditions for determining the joint optimization control of adjustable resources for VPP to participate in peak shaving and frequency modulation are: 1) VPP internal power balance constraint; Frequency modulation capacity power balance constraint: Among them, N ES represents the number of battery units; Peak shaving capacity power balance constraint: Among them, and respectively represent the bidding capacities of VPP participating in two services of frequency regulation and peak shaving; Power balance constraint of the battery: Among them, is the bidding power of the z-th battery for frequency regulation and peak shaving ancillary services at the t time node; 2) Adjustable resource cluster regulation constraint; 3) Energy storage charge and discharge constraint; The state of charge SOC of the battery at each moment is represented as S t , and is calculated by the following formula: S t = (1 - ε)S t-1 + P ch Δtμ c - P dis Δt / μ d where ε is the self-discharge rate of the battery; μ c is the charging efficiency; μ d is the discharging efficiency, and are the response powers of the i-th adjustable resource cluster participating in frequency regulation and the j-th adjustable resource cluster participating in peak regulation, respectively; P ch,ESz,t and Pdis,ESz,t are the charging and discharging powers of the z-th battery at the t time node, respectively.

2. A joint optimization control method for large-scale adjustable resources according to claim 1, characterized in that, the response performance analysis model describes the relationship between the multi-dimensional performance parameters of users' participation in demand response and the incentive intensity as a piecewise function, The adjustable resource response performance parameter is: where: η represents the multi-dimensional performance parameter of the adjustable resource response; r 4 The parameter is a random parameter considering the influence of uncertainty, and the random characteristics of the adjustable resource participating in the response process can be characterized by the random variation of r 4 ; δ is the excitation intensity; The uncertainty parameter model of the adjustable resource response performance parameter is: where r 1 , r 2 , r 3 are known deterministic model parameters; r 4 is a normal distribution satisfying certain rules, where the mean of the distribution is and the standard deviation is The parameters μ 0 and σ 0 are estimated values obtained by point estimation based on the historical response data set of r 4 . Regarding these estimated values as the normal distribution parameters satisfied by r 4 , N is the number of days of historical response data of VPP adjustable resources obtained.

3. A joint optimization control device for adjustable resources of a virtual power plant, characterized in that, the device is used to execute the method described in any one of claims 1-2, and the device includes: a data processing unit, a data decomposition unit, a data processing unit, a feature extraction unit based on SAE, and a resource joint optimization control unit, The data processing unit is used for collecting historical electricity consumption power data and response data, identifying and correcting abnormal points; The data decomposition unit is used for decomposing historical electricity consumption power data to obtain each IMF component; The SAE-based feature extraction unit is used to obtain the feature relationship data h between the IMF components and the adjustable resource response data i ; The resource joint optimization control unit is used for obtaining the response performance parameters of adjustable resources at each time node and solving the optimal control strategy.

Images