CN112348380A - Demand response schedulable capacity probability prediction method and device and electronic equipment - Google Patents

Abstract

The invention provides a demand response schedulable capacity probability prediction method, a demand response schedulable capacity probability prediction device and electronic equipment, wherein the demand response schedulable capacity probability prediction method comprises the following steps: acquiring demand response data of each user subordinate to the load aggregation provider on a historical demand response day; calculating time-sharing aggregation response capacity according to the demand response data; performing feature extraction on the time-sharing polymerization response capacity to obtain a feature parameter of the sample time-sharing polymerization response capacity; taking the sample time-sharing aggregation response capacity characteristic parameter as input, taking the time-sharing aggregation response capacity as output, and performing schedulable capacity estimation according to the characteristic parameter by using a meta-learning algorithm; and performing probability prediction of the demand response schedulable capacity of the load aggregator by using the nonparametric kernel density probability prediction model. And through a meta-learning algorithm and nonparametric kernel density estimation, the small sample probability prediction of the demand response schedulable capacity for the load aggregation provider is realized.

Description

Demand response dispatchable capacity probability prediction method, device and electronic device

technical field

The present invention relates to the field of data processing, and in particular, to a method, device and electronic equipment for probabilistic prediction of demand response schedulable capacity.

Background technique

Since 2012, the state has successively issued relevant notices and opinions on power demand-side management, emphasizing the improvement of the supply-demand balance guarantee level mainly based on demand-side management, and gradually forming a demand-side dynamic peak shaving capacity that accounts for about 3% of the maximum electricity load. Demand response uses price signals and incentives to guide users to cut or shift loads during peak loads, ease the tension of power supply resources, and promote flexibility in the operation of the power system. Demand response can urge users to actively participate in the load regulation of the power system, use electricity reasonably, and reduce the operating pressure of the system during peak power consumption, so as to ensure a safer, more stable and efficient operation of the power system.

For a single residential user, its load mobilization ability is small, and the electricity consumption law is uncertain, so it is difficult to directly conduct electricity transactions with the system operator to achieve load peak regulation of the power grid. As an intermediary between resident users and system operators, load aggregators undertake the key task of integrating demand-side resources, and can formulate bidding strategies with the goal of maximizing their own market-based transaction benefits. In order to maximize the benefits of market-based transactions, load aggregators need to accurately predict the aggregate demand response capacity of users. At the same time, due to the volatility of users' electricity consumption, if the strategy is formulated only by the aggregate demand response capacity prediction and determination value, the actual demand response of the user may not meet the policy requirements, and the load aggregator will be punished economically; or the actual demand response of the user may exceed the actual demand response. Policy requirements that will reduce the profit of load aggregators. Compared with point forecasting, probability forecasting can not only predict the expected value at a certain time in the future, but also obtain its probability distribution information, thus providing more comprehensive reference information for the decision-making of load aggregators. Load aggregators can reasonably formulate bidding strategies based on the probability prediction information, so as to maximize themselves not to exceed the requirements of the strategy and not be subject to economic penalties, thereby reducing decision-making risks.

Therefore, the probabilistic prediction of the schedulable capacity of demand response for load aggregators is more conducive to load aggregators to formulate optimal bidding strategies, reduce risks, and maximize the benefits of market-based transactions.

At present, there are few researches on the probability prediction of the schedulable capacity of demand response of load aggregators at home and abroad. At the same time, due to the fact that demand response has been implemented in fewer years and regions, and there are fewer demand response data samples, it is difficult for traditional machine learning algorithms to compare the samples. In the case of less, higher prediction accuracy can be obtained. Therefore, it is of great significance to find a small-sample probabilistic prediction method for the schedulable capacity of demand response for load aggregators.

SUMMARY OF THE INVENTION

The embodiments of the present invention provide a method for probabilistic prediction of demand response schedulable capacity, which can perform a small sample probability prediction of demand response schedulable capacity for load aggregators.

In a first aspect, an embodiment of the present invention provides a demand response schedulable capacity probability prediction method, including:

Obtain the demand response data of each user under the load aggregator on the historical demand response day;

Calculate the time-sharing aggregated response capacity according to the demand response data;

Perform feature extraction on the time-sharing aggregated response capacity to obtain the characteristic parameters of the sample time-sharing aggregated response capacity;

Taking the time-sharing aggregated response capacity characteristic parameter of the sample as an input, and the time-sharing aggregated response capacity as an output, using a meta-learning algorithm to estimate the schedulable capacity according to the characteristic parameter;

Probabilistic prediction of demand response schedulable capacity of load aggregators using a nonparametric kernel density probabilistic prediction model.

Optionally, the step of calculating the time-sharing aggregated response capacity according to the demand response data specifically includes:

Accumulate all user demand responses in different demand response time periods to obtain the time-sharing aggregated response capacity.

Optionally, the step of performing feature extraction on the time-sharing aggregated response capacity to obtain characteristic parameters of the sample time-sharing aggregated response capacity specifically includes:

performing feature extraction on the time-sharing aggregated response capacity;

The extracted features are screened according to the maximum information coefficient, and the characteristic parameters of the sample time-sharing aggregated response capacity are obtained.

Optionally, the extracted features include: daily features, hourly features, and cumulative benefit features.

Optionally, the step of using the meta-learning algorithm to estimate the schedulable capacity specifically includes:

classifying the sample time-sharing aggregated response capacity according to preset conditions;

grouping the sample time-sharing aggregated response capacity according to the training model type;

Parameter optimization using model-agnostic meta-learning MAML as a meta-learning algorithm.

Optionally, the step of using the model-independent meta-learning MAML as the meta-learning algorithm for parameter optimization specifically includes:

According to the initial parameters of the model, the internal learning rate, and the task of aggregating the response capacity based on the time-sharing of samples, the stochastic gradient descent method is used to optimize the target parameters.

Optionally, the specific steps of using the non-parametric kernel density probability prediction model to perform the probability prediction of the demand response schedulable capacity of the load aggregator include:

Obtain a probability density function according to the sample quantity, kernel function, demand response schedulable capacity prediction error and sample bandwidth of the sample time-sharing aggregated response capacity;

Based on the probability density function, a probability distribution function is obtained;

Probabilistic prediction of the demand response schedulable capacity of the load aggregator is performed through the probability distribution function.

In a second aspect, an embodiment of the present invention further provides a demand response schedulable capacity probability prediction device, the device comprising:

The acquisition module is used to acquire the demand response data of each user under the load aggregator on the historical demand response day;

a calculation module, configured to calculate the time-sharing aggregated response capacity according to the demand response data;

The feature extraction module is used to perform feature extraction on the time-sharing aggregated response capacity, and obtain the characteristic parameters of the sample time-sharing aggregated response capacity;

a processing module, configured to use the time-sharing aggregated response capacity characteristic parameter of the sample as an input, and the time-sharing aggregated response capacity as an output, and use a meta-learning algorithm to estimate the schedulable capacity according to the characteristic parameter;

The forecasting module is used for probabilistic forecasting of the demand response dispatchable capacity of the load aggregator by using the nonparametric kernel density probability forecasting model.

In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, when the processor executes the computer program The steps in the method for probabilistic prediction of demand response schedulable capacity provided by the embodiment of the present invention are implemented.

In the embodiment of the present invention, the demand response data of each user under the load aggregator on the historical demand response day is obtained; according to the demand response data, the time-sharing aggregated response capacity is calculated; the time-sharing aggregated response capacity is extracted by feature extraction to obtain a sample The characteristic parameter of the time-sharing aggregated response capacity; the time-sharing aggregated response capacity characteristic parameter of the sample is used as input, and the time-sharing aggregated response capacity is used as the output, and the schedulable capacity is estimated by using the meta-learning algorithm according to the characteristic parameter; using the non-parametric kernel density probability The forecasting model makes probabilistic forecasts of the demand response schedulable capacity of the load aggregator. Through meta-learning algorithm and non-parametric kernel density estimation, small-sample probability prediction of schedulable capacity of demand response for load aggregators is realized.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

1 is a flowchart of a method for probabilistic prediction of demand response schedulable capacity provided by an embodiment of the present invention;

FIG. 2 is a flowchart of a time-sharing aggregated response capacity feature parameter extraction and screening provided by an embodiment of the present invention;

3 is a model architecture diagram provided by an embodiment of the present invention;

4 is a schematic diagram of a training model and a testing model provided by an embodiment of the present invention;

FIG. 5 is a schematic diagram of a probabilistic prediction result of demand response schedulable capacity provided by an embodiment of the present invention;

6 is a schematic structural diagram of a demand response schedulable capacity probability prediction device provided by an embodiment of the present invention;

7 is a schematic structural diagram of a processing module provided by an embodiment of the present invention;

8 is a schematic structural diagram of a prediction module provided by an embodiment of the present invention;

FIG. 9 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Please refer to FIG. 1. FIG. 1 is a flowchart of a method for probabilistic prediction of demand response schedulable capacity provided by an embodiment of the present invention. As shown in FIG. 1, the method includes the following steps:

S1. Obtain the demand response data of each user subordinate to the load aggregator on the historical demand response day.

In the embodiment of the present invention, the above-mentioned demand response refers to that power users temporarily change their inherent habitual power consumption patterns according to dynamic changes in power prices and power policies, so as to reduce or shift the power consumption load for a certain period of time and respond to power supply, So as to ensure the stability of the power grid system. The above-mentioned demand response day refers to the demand response implementation day, and the above-mentioned historical demand response day refers to the demand response day extracted from the demand response data.

The above-mentioned demand response can be divided into price-based demand response and incentive-based demand response. The above-mentioned price-based demand response refers to that the user adjusts the electricity demand accordingly according to the received electricity price signal, which may be time-of-use electricity price response, real-time electricity price response, and peak electricity price response. The above incentive-based demand response refers to the fact that users actively reduce power demand when the system needs it to obtain compensation, which can be direct load control, interruptible load, demand-side bidding, and emergency power demand response.

Among them, the above-mentioned time-of-use electricity price response can be understood as the conversion of fixed electricity prices into different price mechanisms at different time periods.

The above-mentioned real-time electricity price response has a faster electricity price update cycle than the hourly electricity price. For example, the electricity price update cycle is one hour or less. Since the time-of-use electricity price response cannot cope with short-term capacity shortages, in the case of short-term capacity shortages, real-time electricity can be used. Electricity price response.

The above-mentioned peak electricity price response can pre-set the price during peak electricity consumption, and notify users in advance of a certain period of time, which can have the effect of resisting sudden electricity consumption peaks.

The above-mentioned direct load control can be remotely controlled by the load aggregator to turn on and off the user equipment, so as to avoid peak electricity consumption and notify in advance.

The above-mentioned interruptible load may be remotely controlled by the load aggregator to turn on and off the user equipment under the condition of obtaining the consent of the user, so as to avoid peak electricity consumption and notify in advance.

The above-mentioned demand-side bidding can be a way for load aggregators to actively participate in market competition in the form of bidding.

The above-mentioned emergency power demand response can be that when the stability of the power system is threatened, the load aggregator provides compensation for the user to reduce the load, and the user voluntarily chooses to participate or give up.

S2. Calculate the time-sharing aggregated response capacity according to the demand response data.

In the embodiment of the present invention, the above-mentioned demand response data includes the response capacity of the user, and the time-sharing aggregated response capacity refers to all the response capacity in a preset time period. Specifically, the above-mentioned time-sharing aggregate response capacity can be calculated by the following formula:

为T时间段聚合响应容量，上述f_i,k,T为第i个用户第k天T时间段内的需求响应容量。Among them, the above

is the aggregated response capacity in the T time period, and the above f _i,k,T is the demand response capacity of the i-th user in the k-th time period on the kth day.

S3. Perform feature extraction on the time-sharing aggregated response capacity to obtain characteristic parameters of the time-sharing aggregated response capacity of the sample.

In the embodiment of the present invention, the above-mentioned feature extraction is performed on the time-sharing aggregated response capacity, and the extracted features are features that affect user demand response. According to the characteristics that affect the user's demand response, the characteristic parameters of the sample time-sharing aggregated response capacity are obtained.

Specifically, the above-mentioned features that affect user demand response include daily features, hourly features, and cumulative benefit features. The above-mentioned characteristics can affect the user's demand response.

The above-mentioned daily features may include: predicted daily maximum temperature, minimum temperature, maximum humidity, minimum humidity, rainfall, season label, weekday label, weekend label, and incentive amount.

The above hourly features may include: actual temperature, temperature and humidity index, cold and humidity index, human comfort index, demand response time temperature, humidity, wind speed, demand response baseline load power consumption, and time label.

The above-mentioned cumulative benefit features may include: cumulative maximum temperature benefit and cumulative minimum temperature benefit.

Optionally, please refer to FIG. 2, the step S3 specifically includes the following steps:

S31. Perform feature extraction on the time-sharing aggregated response capacity.

In this embodiment of the present invention, the above-mentioned features may include daily features, hourly features, and cumulative benefit features. The above-mentioned characteristics can affect the user's demand response.

S32. Screen the extracted features according to the maximum information coefficient to obtain feature parameters of the sample time-sharing aggregated response capacity.

In the embodiment of the present invention, the extracted features can be screened by the calculation formula of the maximum information coefficient, and the calculation formula of the maximum information coefficient is as follows:

Among them, the above p(x, y) is the joint probability density of x and y, p(x) and p(y) are the marginal probability distribution densities of x and y, respectively, x and y are two variables with different characteristics, The number of division grids of a and b in the x and y directions, respectively.

Specifically, the calculation formula of the maximum information coefficient can be understood as gridding the features according to the preset grid conditions, and obtaining the maximum mutual information value, then normalizing the maximum mutual information value, and selecting different scales The maximum amount of mutual information is taken as the maximum information coefficient.

In the embodiment of the present invention, filtering the extracted features can reduce the features with small information coefficients, thereby reducing the data amount of the time-sharing aggregated response capacity feature parameters of the samples.

S4. Use the time-sharing aggregated response capacity characteristic parameter of the sample as an input, and the time-sharing aggregated response capacity as an output, and use a meta-learning algorithm to estimate the schedulable capacity according to the characteristic parameter.

并将其分为两组，分别为p组和q组。p组作为元学习的训练模型，即D_train，q组作为测试模型，即D_test。元学习模型中训练模型和测试模型均包含训练集和测试集。对于元学习中的训练模型，其训练集

包括p组中采样的N个类别。对于每个选定的类别，随机选择K个示例作为训练数据；其测试集

也是采样N个类别，每个类别随机选择K个示例作为测试数据。同理，对于元学习中的测试模型，其训练集和测试集包括q组中采样的N个类别，对于每个选定的类别，同样随机选择K个示例作为训练/测试数据。In the embodiment of the present invention, in order to adapt to the meta-learning framework, the time-sharing aggregated response capacity of the samples is classified according to the day, that is,

And they are divided into two groups, namely p group and q group. The p group is used as the training model of meta-learning, namely D _train , and the q group is used as the test model, namely D _test . Both the training model and the test model in the meta-learning model include training sets and test sets. For a trained model in meta-learning, its training set

Include N classes sampled in p groups. For each selected class, K examples are randomly selected as training data; its test set

It is also sampling N categories, and each category randomly selects K examples as test data. Similarly, for a test model in meta-learning, its training and test sets include N categories sampled in q groups, and for each selected category, K examples are also randomly selected as training/testing data.

The meta-learning algorithm is a general-purpose algorithm that can solve various types of tasks. Specifically, the meta-learning algorithm learns the rules for initializing parameters. The initialized parameter θ has the optimal parameter solution for each task in the parameter space. θ1,2,..n are highly sensitive (in fact, the gradient direction is vertical), so that it can quickly reach the optimal point along the gradient direction.

The above task can be understood as a group of samples (ie, p group or q group), this group of samples contains N categories, and each category randomly selects K examples as training/testing data. It is also possible to take K training samples and K' test samples for each type.

Specifically, the model-independent meta-learning MAML can be used as a meta-learning algorithm for parameter optimization. According to the initial parameters of the model, the internal learning rate, and the task of time-sharing aggregation response capacity based on samples, the target parameters are evaluated by the stochastic gradient descent method. parameter optimization. For the parameter model f _θ with parameter θ, the parameter update optimization formula is:

为第i个任务。Among them, θ is the initial parameter of the model, θ′ _i is the updated parameter, α is the internal learning rate,

for the i-th task.

Further, the updated parameter θ′ _i can be meta-optimized by the stochastic gradient descent method, and its formula is:

为第i个任务，β为元学习率。Among them, θ is the initial parameter of the model, θ′ _i is the updated parameter, α is the internal learning rate,

is the i-th task, and β is the meta learning rate.

S5. Use the non-parametric kernel density probability prediction model to perform the probability prediction of the demand response dispatchable capacity of the load aggregator.

In the embodiment of the present invention, a probability density function can be obtained according to the sample number, kernel function, demand response schedulable capacity prediction error and sample bandwidth of the sample time-sharing aggregated response capacity; based on the probability density function, a probability distribution can be obtained function; the probability distribution function is used to perform the probability prediction of the demand response schedulable capacity of the load aggregator.

Specifically, the probability density function obtained by using nonparametric kernel density estimation can be expressed as the following formula:

Among them, N is the number of samples; h is the sample bandwidth; K( ) is the kernel function; e _i is the demand response schedulable capacity prediction error of the ith sample.

The above-mentioned kernel functions K(·) mainly include Gaussian kernel, uniform kernel, triangular kernel and Epanechnikov function.

进行积分得到概率分布函数F(x)，任意给定γ(0＜γ＜1)，在置信度1-γ下，则需求响应可调度容量的预测区间为：Further, for the probability density function

Integrate to obtain the probability distribution function F(x), given any γ (0 < γ < 1), under the confidence of 1-γ, the prediction interval of the schedulable capacity of demand response is:

为聚合响应容量预测值；

F^-1(y)为F(x)的反函数P{x≤F^-1(y)}＝y，y分别为γ₁、γ₂。in,

is the predicted value for the aggregated response capacity;

F ^-1 (y) is the inverse function of F(x) P{x≤F ^-1 (y)}=y, and y is γ ₁ and γ ₂ , respectively.

In the embodiment of the present invention, the demand response data of each user under the load aggregator on the historical demand response day is obtained; according to the demand response data, the time-sharing aggregated response capacity is calculated; the time-sharing aggregated response capacity is extracted by feature extraction to obtain a sample The characteristic parameter of the time-sharing aggregated response capacity; the characteristic parameter of the time-sharing aggregated response capacity of the sample is used as input, and the time-sharing aggregated response capacity is used as the output, and the schedulable capacity is estimated by using the meta-learning algorithm according to the characteristic parameter; using the non-parametric kernel density The probabilistic forecasting model performs probabilistic forecasting of the demand response schedulable capacity of the load aggregator. Through meta-learning algorithm and non-parametric kernel density estimation, small-sample probability prediction of schedulable capacity of demand response for load aggregators is realized.

Optionally, the characteristic parameters that affect the user's demand response are used as the input of the neural network model, and the demand response quantity is used as the output of the neural network model, and the parameters are updated through the model-independent meta-learning MAML, and then the non-parametric kernel density model is used to perform probability analysis. prediction, as shown in Figure 3.

It should be noted that both training data and test data include training samples and test samples, as shown in Figure 4.

In the embodiment of the present invention, in order to specifically illustrate the effect of the embodiment of the present invention, the collected data is used for verification. In the embodiment of the present invention, the feature that affects the user's demand response is used as the input of the neural network model, and the demand response amount is used as the output of the neural network model. The parameters are updated through the model-independent meta-learning MAML, and then the Epanechnikov kernel density model is used for probability prediction. . The predicted results are shown in Figure 5.

It can be seen from FIG. 5 that, after actual testing, the demand response schedulable capacity probability prediction effect provided by the embodiment of the present invention is good, and the 90% confidence interval covers all the real aggregate demand response capacity. This shows that the prediction effect of the method of the present invention is better.

It should be noted that the method for probabilistic prediction of demand response schedulable capacity provided by the embodiment of the present invention can be applied to devices such as mobile phones, computers, and servers that can perform probabilistic prediction of demand response schedulable capacity.

Please refer to FIG. 6. FIG. 6 is a schematic structural diagram of a demand response schedulable capacity probability prediction device provided by an embodiment of the present invention. As shown in FIG. 6, the device includes:

The obtaining module 601 is used to obtain the demand response data of each user subordinate to the load aggregator on the historical demand response day;

A calculation module 602, configured to calculate the time-sharing aggregated response capacity according to the demand response data;

A feature extraction module 603, configured to perform feature extraction on the time-sharing aggregated response capacity to obtain characteristic parameters of the sample time-sharing aggregated response capacity;

The processing module 604 is configured to use the time-sharing aggregated response capacity characteristic parameter of the sample as an input, and the time-sharing aggregated response capacity as an output, and use a meta-learning algorithm to estimate the schedulable capacity according to the characteristic parameter;

The prediction module 605 is configured to use the non-parametric kernel density probability prediction model to perform the probability prediction of the demand response schedulable capacity of the load aggregator.

Optionally, the calculation module 602 is further configured to accumulate all user demand responses in different demand response time periods to obtain a time-sharing aggregated response capacity.

Optionally, the computing module 602 specifically includes:

a feature extraction unit, configured to perform feature extraction on the time-sharing aggregated response capacity;

The screening unit is used to screen the extracted features according to the maximum information coefficient to obtain the feature parameters of the sample time-sharing aggregated response capacity.

Optionally, the extracted features include: daily features, hourly features, and cumulative benefit features.

Optionally, as shown in FIG. 7 , the processing module 604 specifically includes:

A classification unit 6041, configured to classify the sample time-sharing aggregated response capacity according to preset conditions; and

a grouping unit 6042, configured to group the sample time-sharing aggregated response capacity according to the training model type;

The parameter unit 6043 is configured to use the model-independent meta-learning MAML as a meta-learning algorithm to perform parameter optimization.

Optionally, the parameter unit 6043 is further configured to perform parameter optimization on the target parameters through the stochastic gradient descent method according to the initial parameters of the model, the internal learning rate, and the task of aggregating response capacity based on time-sharing of samples.

Optionally, as shown in FIG. 8 , the prediction module 605 specifically includes:

The first processing unit 6051 is configured to obtain a probability density function according to the sample number, kernel function, demand response schedulable capacity prediction error and sample bandwidth of the sample time-sharing aggregation response capacity;

a second processing unit 6052, configured to obtain a probability distribution function based on the probability density function;

The prediction unit 6053 is configured to perform probability prediction of the demand response schedulable capacity of the load aggregator through the probability distribution function.

It should be noted that the device for probabilistic prediction of demand response schedulable capacity provided by the embodiment of the present invention can be applied to devices such as mobile phones, computers, and servers that can perform probabilistic prediction of demand response schedulable capacity.

The demand response schedulable capacity probability prediction device provided in the embodiment of the present invention can implement each process implemented by the demand response schedulable capacity probability prediction method in the above method embodiments, and can achieve the same beneficial effects. In order to avoid repetition, details are not repeated here.

Referring to FIG. 9, FIG. 9 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. As shown in FIG. 9, it includes: a memory 902, a processor 901, and a memory 902 and a processor A computer program running on 901, which:

The processor 901 is configured to call the computer program stored in the memory 902, and perform the following steps:

Obtain the demand response data of each user under the load aggregator on the historical demand response day;

Calculate the time-sharing aggregated response capacity according to the demand response data;

Perform feature extraction on the time-sharing aggregated response capacity to obtain the characteristic parameters of the sample time-sharing aggregated response capacity;

The characteristic parameter of the time-sharing aggregated response capacity of the sample is used as input, and the time-sharing aggregated response capacity is used as output, and a meta-learning algorithm is used to estimate the schedulable capacity according to the characteristic parameter;

Probabilistic prediction of demand response schedulable capacity of load aggregators using a nonparametric kernel density probabilistic prediction model.

Optionally, the step of calculating the time-sharing aggregated response capacity according to the demand response data performed by the processor 901 specifically includes:

All user demand responses in different demand response time periods are accumulated to obtain the time-sharing aggregated response capacity.

Optionally, the step of performing feature extraction on the time-sharing aggregated response capacity performed by the processor 901 to obtain the sample time-sharing aggregated response capacity specifically includes:

performing feature extraction on the time-sharing aggregated response capacity;

The extracted features are screened according to the maximum information coefficient, and the characteristic parameters of the sample time-sharing aggregated response capacity are obtained.

Optionally, the extracted features include: daily features, hourly features, and cumulative benefit features.

Optionally, the step of using the meta-learning algorithm to estimate the schedulable capacity according to the characteristic parameter executed by the processor 901 specifically includes:

classifying the sample time-sharing aggregated response capacity according to preset conditions;

grouping the sample time-sharing aggregated response capacity according to the training model type;

Use model-agnostic meta-learning MAML as a meta-learning algorithm for parameter optimization.

Optionally, the step of using the model-independent meta-learning MAML as a meta-learning algorithm to perform parameter optimization performed by the processor 901 specifically includes:

According to the initial parameters of the model, the internal learning rate, and the task of aggregating the response capacity based on time-sharing of samples, the target parameters are optimized by the stochastic gradient descent method.

Optionally, the specific steps of using the non-parametric kernel density probabilistic prediction model to perform the probabilistic prediction of the demand response schedulable capacity of the load aggregator performed by the processor 901 include:

Obtain a probability density function according to the sample quantity, kernel function, demand response schedulable capacity prediction error and sample bandwidth of the sample time-sharing aggregated response capacity;

Based on the probability density function, a probability distribution function is obtained;

Probabilistic prediction of the demand response schedulable capacity of the load aggregator is performed through the probability distribution function.

It should be noted that the above-mentioned electronic devices may be mobile phones, computers, servers and other devices that can be applied to the probability prediction of demand response schedulable capacity.

The electronic device provided by the embodiments of the present invention can implement the various processes implemented by the demand response schedulable capacity probability prediction method in the above method embodiments, and can achieve the same beneficial effects. To avoid repetition, details are not described here.

Those of ordinary skill in the art can understand that the realization of all or part of the processes in the methods of the above embodiments can be accomplished by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. During execution, the processes of the embodiments of the above-mentioned methods may be included. The storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM for short).

The above disclosures are only preferred embodiments of the present invention, and of course, the scope of the rights of the present invention cannot be limited by this. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims (9)

1. A demand response schedulable capacity probability prediction method, comprising the steps of:

acquiring demand response data of each user subordinate to the load aggregation provider on a historical demand response day;

calculating time-sharing aggregation response capacity according to the demand response data;

performing feature extraction on the time-sharing polymerization response capacity to obtain a feature parameter of the sample time-sharing polymerization response capacity;

taking the sample time-sharing aggregation response capacity characteristic parameter as input, taking the time-sharing aggregation response capacity as output, and performing schedulable capacity estimation according to the characteristic parameter by using a meta-learning algorithm;

and performing probability prediction of the demand response schedulable capacity of the load aggregator by using the nonparametric kernel density probability prediction model.

2. The demand response schedulable capacity probability prediction method of claim 1, wherein the step of calculating a time-share aggregated response capacity according to the demand response data specifically comprises:

and accumulating all the user demand responses in different demand response time periods to obtain time-sharing aggregate response capacity.

3. The demand response schedulable capacity probability prediction method of claim 1, wherein the step of performing the feature extraction on the time-sharing aggregate response capacity to obtain the feature parameter of the sample time-sharing aggregate response capacity specifically comprises:

performing feature extraction on the time-sharing polymerization response capacity;

and screening the extracted features according to the maximum information coefficient to obtain the feature parameters of the sample time-sharing polymerization response capacity.

4. The demand-response dispatchable capacity probability prediction method as set forth in claim 1, wherein the extracted characteristic parameters include: day-by-day characteristics, time-by-time characteristics, cumulative benefit characteristics.

5. The demand response schedulable capacity probability prediction method of claim 1, wherein the step of using a meta learning algorithm to estimate schedulable capacity according to the characteristic parameters specifically comprises:

classifying the sample time-sharing polymerization response capacity according to a preset condition;

grouping the sample time-sharing aggregation response capacity according to the type of a training model;

and using the model-independent meta-learning MAML as a meta-learning algorithm to carry out parameter optimization.

6. The demand-response schedulable capacity probability prediction method of claim 5, wherein the step of using model-independent meta-learning MAML as a meta-learning algorithm for parameter optimization specifically comprises:

and performing parameter optimization on the target parameter by a random gradient descent method according to the initial parameters of the model, the internal learning rate and the task based on the sample time-sharing aggregation response capacity.

7. The demand response schedulable capacity probability prediction method of claim 1, wherein the step of performing the probability prediction of the demand response schedulable capacity of the load aggregator using the non-parametric kernel density probability prediction model comprises:

obtaining a probability density function according to the sample quantity of the sample time-sharing aggregation response capacity, the kernel function, the demand response schedulable capacity prediction error and the sample bandwidth;

obtaining a probability distribution function based on the probability density function;

and performing probability prediction of the demand response schedulable capacity of the load aggregator through the probability distribution function.

8. A demand response schedulable capacity probability prediction apparatus, the apparatus comprising:

the acquisition module is used for acquiring the demand response data of each user subordinate to the load aggregation provider on the historical demand response day;

the calculation module is used for calculating time-sharing aggregation response capacity according to the demand response data;

the characteristic extraction module is used for carrying out characteristic extraction on the time-sharing polymerization response capacity to obtain a characteristic parameter of the sample time-sharing polymerization response capacity;

the processing module is used for taking the sample time-sharing aggregation response capacity characteristic parameter as input, taking the time-sharing aggregation response capacity as output, and carrying out schedulable capacity estimation according to the characteristic parameter by utilizing a meta-learning algorithm;

and the prediction module is used for performing probability prediction on the demand response schedulable capacity of the load aggregator by using the nonparametric kernel density probability prediction model.

9. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, the processor implementing the steps in the demand response schedulable capacity probability prediction method of any of the claims 1 to 7 when executing the computer program.

Images