Air conditioner load demand response potential evaluation method based on time sequence matching strategy

Abstract

The invention discloses an air conditioner load demand response potential evaluation method based on a time sequence matching strategy, and the method comprises the following steps: (1) constructing an air conditioner load prediction model based on a classical GRU recurrent neural network architecture to capture time dynamic characteristics in an air conditioner load time sequence, and initializing model parameters based on user historical load data; (2) a time sequence distribution drift phenomenon is considered, a time sequence distribution similarity quantification and time sequence distribution matching algorithm module is introduced, and a self-adaptive RNN air conditioner load prediction model is constructed based on the time sequence distribution similarity quantification and time sequence distribution matching algorithm module; (3) in combination with the adaptive RNN load prediction model and the user demand response optimal decision simulation model based on the household energy management system, evaluating a load difference value before and after the air conditioner user participates in the demand response, namely demand response potential; the built adaptive RNN model improves the accuracy of air conditioner load prediction and demand response potential evaluation, and can provide effective reference for a demand response scheduling center.

Description

Technical Field

[0001] The invention relates to the field of air conditioning load management, and in particular to an air conditioning load demand response potential evaluation method based on a time sequence matching strategy. Background Art

[0002] In recent years, the proportion of renewable energy in the power system has gradually increased, which has brought tremendous pressure on the balance of supply and demand of the power grid. In this context, demand response as an effective means can mobilize the flexible load resources on the user side as an alternative to the operation and standby of traditional power generation units in the power grid to participate in peak load and frequency regulation, thereby greatly enhancing the regulation capacity of the power grid. According to relevant statistical data, the proportion of air-conditioning power load in the basic cycle of peak load in three different countries, China, Spain and the United States, is more than 45%, which fully demonstrates the importance of air-conditioning load as a typical flexible load resource. Therefore, it is necessary to conduct in-depth research on how to fully tap the demand response potential of air-conditioning load resources. Therefore, it is crucial for aggregators and power markets to accurately evaluate the load adjustable margin of air-conditioning users (i.e., demand response potential). Summary of the invention

[0003] The technical problem to be solved by the present invention is that, considering the air conditioning load demand response potential evaluation method based on the time series matching strategy, the time series distribution drift problem in the air conditioning load forecasting is effectively handled, and the accuracy of the load forecasting is improved. At the same time, combined with the user demand response optimal decision simulation model, the accurate evaluation of the air conditioning user demand response potential is realized, which provides valuable information for aggregators and power markets, and helps to improve the operating efficiency and reliability of the power system.

[0004] In order to solve the above technical problems, the present invention adopts the following technical means:

[0005] A method for evaluating the potential of air conditioning load demand response based on a time series matching strategy comprises the following steps:

[0006] S1, build an air conditioning load prediction model based on the classic GRU recurrent neural network architecture to capture the temporal dynamic characteristics of the air conditioning load time series, and initialize the model parameters based on the user's historical load data;

[0007] S2, considering the time series distribution drift phenomenon of air-conditioning load sequence, introduces the time series distribution similarity quantification and time series distribution matching algorithm modules, and builds an adaptive RNN air-conditioning load prediction model based on this;

[0008] S3, combining the improved adaptive RNN load forecasting model and the user demand response optimal decision simulation model based on the home energy management system, evaluates the load difference before and after the air-conditioning users participate in the demand response, that is, the demand response potential.

[0009] The air conditioning load prediction model described in step S1 includes an input layer, several GRU hidden layers and an output layer, wherein the input layer receives the air conditioning load data and historical load data of the current time step, the GRU hidden layer is responsible for extracting the characteristics of the time series and controlling the flow of information through a gating mechanism to avoid the gradient vanishing problem, and the output layer predicts the future air conditioning load;

[0010] The GRU model is trained using historical load data input into the input layer. During the training process, the back propagation algorithm and the Adam optimizer are used to optimize the model parameters, and the model weight parameters are initialized by minimizing the mean square error between the true value and the model predicted value.

[0011] Step S2 is specifically as follows:

[0012] S21, based on the maximum entropy principle, introduces a time series distribution similarity quantification algorithm module, and uses a greedy algorithm to solve the following optimization problem to obtain the sub-segment division with the largest distribution difference between the original training data:

[0013]

[0014]

[0015] in, represents two basic time periods, Δ1 and Δ2 are predefined parameters to avoid trivial solutions, K0 is a hyperparameter to avoid over-segmentation, and d(D i ,D j ) represents the Euclidean distance between these two basic cycles;

[0016] The goal of solving the above optimization problem is to maximize the average load sub-segment distribution distance by searching the number of load sub-segments K and the corresponding load sub-segment divisions, so that the distribution of each sub-segment is as diverse as possible, so that the learned prediction model has better generalization ability;

[0017] The specific solution strategy for the load sub-segment distribution distance is as follows: first, the original training load sequence is evenly divided into N = 10 parts, where each part is the smallest unit basic cycle that cannot be further divided. Then, the value of K is randomly searched in {2, 3, 5, 7, 10}. Given K, the length n of each basic cycle is selected according to the greedy strategy. j , A and B represent the starting point and end point of the time series respectively. First, consider K = 2, and maximize the distribution distance d(S AC , S CB ) Select one segmentation point from the nine candidate segmentation points and record it as C. After determining C, consider K=3 and use the same strategy to select another point D. Apply similar strategies to different K values ​​and so on;

[0018] S22, based on the obtained optimal original training load sequence data sub-segment division, introduces the time series distribution matching algorithm module to learn the common domain knowledge between different sub-segments by matching their distributions. The specific prediction loss function term and time series matching loss function term are as follows:

[0019]

[0020]

[0021] The final objective function of time series distribution matching is as follows:

[0022]

[0023] In the final objective function of time series distribution matching, λ is a trade-off hyperparameter, formula Represents the average value of the distribution distance of all paired basic cycles; a small batch of training data segments Di and Dj is used to perform the forward operation of the RNN layer and connect all hidden features. Finally, time series distribution matching can be achieved by minimizing the following formula during the training process;

[0024] S23, combining time series similarity quantification and distribution matching algorithm, constructs an adaptive RNN air conditioning load prediction model. During the training process, the parameters of the GRU model can be dynamically adjusted according to the distribution changes of the time series to improve the accuracy of the prediction.

[0025] Step S3 is specifically as follows:

[0026] S31, define the air conditioning demand response potential value DR(t);

[0027] S32, using the optimal decision-making simulation model of air-conditioning user demand response based on the home energy management system, simulating the load changes of air-conditioning users under different demand response strategies, thereby constructing a load data set after the target air-conditioning users respond;

[0028] S33, combined with the improved adaptive RNN load forecasting model, uses the historical load data of the target user to train and predict its baseline load before participating in demand response. The post-response load data set constructed in S2 is used to train the model and predict the air-conditioning load of the target user after participating in demand response. The difference in load and electricity consumption before and after the user participates in demand response is calculated as an evaluation indicator of demand response potential.

[0029] Step S31, defining the air conditioning demand response potential value DR(t) as:

[0030] The ability of air-conditioning users to reduce peak load or increase valley load before and after participating in demand response is expressed in the following mathematical expressions:

[0031]

[0032] Where P AC (t) is the load of air-conditioning users before participating in demand response, Load of air-conditioning users after participating in demand response.

[0033] In step S32, an optimal decision simulation model for air-conditioning user demand response based on a home energy management system is used, and the optimal decision simulation model includes: objective function and air-conditioning load constraint modeling.

[0034] The objective function specifically includes: In the incentive-based demand response, the user's goal is to minimize his total daily electricity cost, as shown below:

[0035]

[0036] Where λ is the daily fixed electricity price, s(μ) is a logical variable used to determine whether the electricity consumption reaches the contract level, so as to know whether the customer needs to pay a fine, and is expressed as:

[0037]

[0038] Δd represents the reduction in electricity consumption during a DR event, i.e., the user’s DR potential. The specific expression is as follows:

[0039]

[0040] The definitions of the remaining variables in the objective function are as follows, where d cont represents the contractual electricity reduction between individual customers and load aggregators, Inc is the monetary reward per unit electricity reduction, and Pen is the monetary penalty per unit electricity reduction when the reduction of individual customers does not reach the pre-agreed contract level;

[0041] The following equation is used to determine whether the customer will pay less compared to the situation without DR;

[0042]

[0043] The power balance constraints are as follows:

[0044]

[0045] The left part is the total household power consumption, and the right part is the transferable load Air conditioning load and other inelastic loads composition.

[0046] The air conditioning load constraint modeling is as follows: the air conditioning load is an ideal flexible load resource for demand response, because when it actually participates in demand response, it does not need to be completely shut down. It only needs to adjust its set temperature appropriately within the user comfort range to flexibly adjust the air conditioning load; this section uses the simplified first-order ETP model mentioned in Chapter 2 to model the air conditioning load unit; when the air conditioner is turned on, the following constraints will be executed:

[0047]

[0048] When the air conditioner is off, there are:

[0049]

[0050] Where τ is the time step of 1 min, and the duration of the air conditioner in the on and off states is as follows:

[0051]

[0052]

[0053] Q i =αηp i air

[0054]

[0055] Here, θ1 is set to θ max , θ2 is set to θ min , when the air conditioner is in cooling mode in summer, α is set to -1.

[0056] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that the memory stores the computer program, and when the processor executes the computer program, it implements any of the aforementioned methods for evaluating the potential of air conditioning load demand response based on a timing matching strategy.

[0057] A computer-readable storage medium having a computer program stored thereon, wherein the storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute any of the aforementioned methods for evaluating the potential for air conditioning load demand response based on a timing matching strategy.

[0058] The adaptive RNN model constructed in the present invention improves the accuracy of air-conditioning load prediction and demand response potential assessment, and can provide an effective reference for the demand response dispatch center. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the method flow of the present invention.

[0060] Figure 2 It is a schematic diagram of the single-step load baseline prediction result curve of the present invention.

[0061] Figure 3 It is a schematic diagram of the multi-step load baseline prediction result curve of the present invention.

[0062] Figure 4 This is a schematic diagram of the prediction results of the first demand response day of the present invention.

[0063] Figure 5 This is a schematic diagram of the prediction results of the third demand response day of the present invention.

[0064] Figure 6 This is a schematic diagram of the prediction results of the fifth demand response day of the present invention.

[0065] Figure 7 This is a comparison chart of the demand response potential prediction results for 7 demand response days calculated by the present invention.

[0066] Figure 8 This is a schematic diagram of the air conditioning load time series distribution drift phenomenon introduced by the present invention.

[0067] Fig. 9 Schematic diagram of the principle of the time series distribution similarity quantization algorithm of the present invention. DETAILED DESCRIPTION

[0068] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0069] like Figure 1 As shown, the present invention relates to an air conditioning load demand response potential evaluation method based on a time series matching strategy, comprising the following steps:

[0070] (1) An air conditioning load prediction model is constructed based on the classic GRU recurrent neural network architecture to capture the temporal dynamic characteristics of the air conditioning load time series, and the model parameters are initialized based on the user's historical load data;

[0071] (2) Considering the time series distribution drift phenomenon of air-conditioning load sequence, the time series distribution similarity quantification and time series distribution matching algorithm modules are introduced, and based on this, an adaptive RNN air-conditioning load prediction model is constructed;

[0072] (3) The improved adaptive RNN load forecasting model and the user demand response optimal decision simulation model based on the home energy management system are combined to evaluate the load difference before and after air-conditioning users participate in demand response, that is, the demand response potential.

[0073] The technical solution of the present invention is described in detail below:

[0074] Step 1: Build an air conditioning load prediction model based on the classic GRU recurrent neural network architecture to capture the temporal dynamic characteristics of the air conditioning load time series, and initialize the model parameters based on the user's historical load data. The main implementation steps are as follows:

[0075] (1) The air conditioning load prediction model is constructed using the gated recurrent unit (GRU) recurrent neural network architecture. GRU is an improved RNN structure that can effectively capture long-term dependencies in time series. The model structure includes an input layer, several GRU hidden layers, and an output layer.

[0076] (2) The input layer receives the air conditioning load data and historical load data of the current time step. The GRU hidden layer is responsible for extracting the features of the time series and controlling the flow of information through the gating mechanism to avoid the gradient vanishing problem. The output layer predicts the future air conditioning load.

[0077] (3) The GRU model is trained using historical load data. During the training process, the back propagation algorithm and Adam optimizer are used to optimize the model parameters, and the model weight parameters are initialized by minimizing the mean square error between the true value and the model predicted value.

[0078] S2: Considering the time series distribution drift phenomenon of air-conditioning load sequence, the time series distribution similarity quantification and time series distribution matching algorithm modules are introduced, and based on this, an adaptive RNN air-conditioning load prediction model is constructed. The main implementation steps are as follows:

[0079] (1) Analyze the time series drift phenomenon in air conditioning load forecasting, which is specifically defined as follows: Figure 8 As shown in the figure, the data distribution P(x) of different intervals A, B and C are different (x and y represent training samples and predicted values, respectively). Especially for the test data that have not been seen during the training process, its distribution is also quite different from the training data. However, the conditional distribution P(y|x) is usually considered to remain unchanged in this scenario, which is also reasonable in actual load forecasting applications. For example, in air-conditioning load forecasting, the volatility of meteorological factors such as indoor and outdoor temperature causes the load curve (P(x)) to change, but the basic operation mode and control logic law of the air conditioner (P(y|x)) do not change due to changes in external factors. The traditional GRU model assumes that the historical load and the future load to be predicted follow the ideal assumption of independent and identical distribution. Therefore, the prediction accuracy in actual scenarios is low, and new algorithm modules need to be further introduced.

[0080] (2) Based on the maximum entropy principle, a time series distribution similarity quantification algorithm module is introduced. The greedy algorithm is used to solve the following optimization problem to obtain the sub-segment division with the largest distribution difference between the original training data. This is also considered to be the worst case of time series distribution drift:

[0081]

[0082]

[0083] Where d() is the distance metric, Δ1 and Δ2 are predefined parameters to avoid trivial solutions (e.g., very small or very large values ​​may not capture the distribution information), and K0 is a hyperparameter to avoid over-segmentation. The metric d(·, ·) in the equation can be any distance function, for example, Euclidean distance or edit distance, or some distribution-based distance / divergence such as maximum mean difference and KL-divergence.

[0084] (3) Based on the obtained optimal original training load sequence data sub-segment division, the time series distribution matching algorithm module is introduced to learn the common domain knowledge between different sub-segments by matching their distributions. Therefore, compared with the method that only relies on local or statistical information, it is expected that the learned model M can have better generalization performance on test data with unknown distribution. The specific prediction loss function term and time series matching loss function term are as follows:

[0085]

[0086]

[0087] The final objective function of time series distribution matching is as follows, where λ is a trade-off hyperparameter, and the second term represents the average of the distribution distances of all paired basic cycles. Regarding the specific calculation process, the present invention uses a small batch of training data segments Di and Dj to perform the forward operation of the RNN layer and connect all hidden features. Finally, time series distribution matching can be achieved by minimizing the following formula during the training process.

[0088]

[0089] (4) Combine time series similarity quantification and distribution matching algorithm to build an adaptive RNN air conditioning load prediction model. During the training process, the parameters of the GRU model can be dynamically adjusted according to the distribution changes of the time series to improve the accuracy of the prediction.

[0090] Step 3: Combine the improved adaptive RNN load forecasting model and the user demand response optimal decision simulation model based on the home energy management system to evaluate the load difference before and after the air-conditioning user participates in the demand response, that is, the demand response potential. The specific implementation steps are as follows:

[0091] (31) The air conditioning demand response potential is defined as the ability of air conditioning users to reduce peak load or increase valley load before and after participating in demand response. The specific mathematical expression is as follows: AC (t) is the load of air-conditioning users before participating in demand response, is the load of the air-conditioning user after participating in demand response, and the difference between the two is the demand response potential value of the air-conditioning user.

[0092]

[0093] (32) Using the optimal decision-making simulation model for air-conditioning user demand response based on the home energy management system, the load changes of air-conditioning users under different demand response strategies are simulated, thereby constructing a load data set after the target air-conditioning user responds. The objective function and air-conditioning load constraint modeling of the optimal decision-making simulation model for air-conditioning user demand response are as follows:

[0094] 1) Objective function: In the incentive-based demand response, the user's goal is to minimize his total daily electricity cost, as shown below:

[0095]

[0096] Where λ is the daily fixed electricity price, s(μ) is a logical variable used to determine whether the electricity consumption reaches the contract level, so as to know whether the customer needs to pay a fine, and is expressed as:

[0097]

[0098] For user i on the kth day of DR, Δd represents the amount of electricity consumption reduction during the DR event (i.e., the user’s DR potential), and the specific expression is as follows:

[0099]

[0100] The definitions of the remaining variables in the objective function are as follows, where d cont It represents the contractual electricity reduction between individual customers and load aggregators, Inc is the monetary reward per unit electricity reduction, and Pen is the monetary penalty per unit electricity reduction when the reduction amount of individual customers does not reach the pre-agreed contract level.

[0101] The following equation is used to determine whether the customer will pay less than if there were no DR

[0102]

[0103] The power balance constraints are as follows:

[0104]

[0105] The left part is the total household power consumption, and the right part is the transferable load Air conditioning load and other inelastic loads composition.

[0106] 2) Air conditioning load constraint modeling: Air conditioning load is an ideal flexible load resource for demand response, because when it actually participates in demand response, it does not need to be completely shut down. It only needs to adjust its set temperature appropriately within the user comfort range to flexibly adjust the air conditioning load. This section uses the simplified first-order ETP model mentioned in Chapter 2 to model the air conditioning load unit. When the air conditioner is turned on, the following constraints will be executed:

[0107]

[0108] When the air conditioner is off, there are:

[0109]

[0110] Where τ is the time step of 1 min. For the sake of simplicity, the optimal decision-making optimization model of user demand response in this paper only considers fixed-frequency air conditioners, whose working state can only switch between rated power and zero power. Therefore, the indoor temperature of the building provided by the air conditioner will be in [θ mn, θ max ](θ min =θ set -θ ab ,θ min =θ set +θ db ) range and changes periodically. Therefore, the duration of the air conditioner in the on and off states can be calculated as follows:

[0111]

[0112]

[0113]

[0114]

[0115] Here, θ1 is set to θ max , θ2 is set to θ min , when the air conditioner is in cooling mode in summer, α is set to -1. (This article takes the air conditioner in cooling mode as an example)

[0116] (33) Combined with the improved adaptive RNN load forecasting model, the historical load data of the target user is used to train and predict the baseline load before participating in demand response. The post-response load dataset constructed by S2 is used to train the model and predict the air-conditioning load of the target user after participating in demand response. The difference in load and electricity consumption before and after the user participates in demand response is calculated as an evaluation indicator of demand response potential.

[0117] (34) The real data set of the Pecan Street Energy Community Project in the United States was used to test and verify the effect of the present invention. The data of a household from July 1, 2016 to August 31, 2016 were selected for case analysis, which included air conditioning load data as labels and the corresponding indoor temperature, outdoor temperature, building area and air conditioning set temperature as input feature data, with a time resolution of one minute. Since the users of the original data set have never participated in demand response, this chapter artificially sets 7 demand response days. The demand response signal of each demand response day is mainly related to the start time, duration and incentive subsidy of the demand response, as shown in Table 1.

[0118] Table 1

[0119]

[0120] As can be seen from Table 1, there are three different types of incentive subsidies and six different demand response starts and durations. The incentive subsidies and demand response times on different demand response days are randomly selected, thereby simulating the randomness of demand response signals on different demand response days in real situations.

[0121] Specifically, the present invention was tested and verified from two aspects: load baseline prediction effect and demand response potential prediction effect. Figure 2 This is a comparison chart of the load baseline prediction curves of the adaptive RNN model, the GRU model with the same architecture, and the classic MLP model in the single-step prediction scenario. Figure 3 This is a comparison chart of the load baseline prediction curves of the adaptive RNN model, the GRU model with the same architecture, and the classic MLP model in the multi-step prediction scenario. Figure 4 , Figure 5 , Figure 6 These are the load forecast result diagrams before and after the demand response on the 1st, 3rd and 5th days among the 7 demand response days. Without loss of generality, the simulated demand response time periods of these three days are different, namely 12:00-14:00, 13:00-15:00, and 17:00-20:00, and their temperature setting values ​​are 24.5℃, 25.5℃, and 26.5℃ respectively. Figure 7 This is a comparison chart of the demand response potential prediction results for 7 demand response days.

Claims

1. A method for evaluating the potential of air conditioning load demand response based on a time series matching strategy, characterized in that: The steps include: S1, build an air conditioning load prediction model based on the classic GRU recurrent neural network architecture to capture the temporal dynamic characteristics of the air conditioning load time series, and initialize the model parameters based on the user's historical load data; S2, considering the time series distribution drift phenomenon of air-conditioning load sequence, introduces the time series distribution similarity quantification and time series distribution matching algorithm modules, and builds an adaptive RNN air-conditioning load prediction model based on this; S3, combining the improved adaptive RNN load forecasting model and the user demand response optimal decision simulation model based on the home energy management system, evaluates the load difference before and after the air-conditioning users participate in the demand response, that is, the demand response potential.

2. The method for evaluating the air conditioning load demand response potential based on the time series matching strategy according to claim 1, characterized in that: The air conditioning load prediction model described in step S1 includes an input layer, several GRU hidden layers and an output layer, wherein the input layer receives the air conditioning load data and historical load data of the current time step, the GRU hidden layer is responsible for extracting the characteristics of the time series and controlling the flow of information through a gating mechanism to avoid the gradient vanishing problem, and the output layer predicts the future air conditioning load; The GRU model is trained using historical load data input into the input layer. During the training process, the back propagation algorithm and the Adam optimizer are used to optimize the model parameters, and the model weight parameters are initialized by minimizing the mean square error between the true value and the model predicted value.

3. The method for evaluating the air conditioning load demand response potential based on the time series matching strategy according to claim 1, characterized in that: Step S2 is specifically as follows: S21, based on the maximum entropy principle, introduces a time series distribution similarity quantification algorithm module, and uses a greedy algorithm to solve the following optimization problem to obtain the sub-segment division with the largest distribution difference between the original training data: in, represents two basic time periods, Δ1 and Δ2 are predefined parameters to avoid trivial solutions, K0 is a hyperparameter to avoid over-segmentation, and d(D i ,D j ) represents the Euclidean distance between these two basic cycles; The goal of solving the above optimization problem is to maximize the average load sub-segment distribution distance by searching the number of load sub-segments K and the corresponding load sub-segment divisions, so that the distribution of each sub-segment is as diverse as possible, so that the learned prediction model has better generalization ability; The specific solution strategy for the load sub-segment distribution distance is as follows: first, the original training load sequence is evenly divided into N = 10 parts, where each part is the smallest unit basic cycle that cannot be further divided. Then, the value of K is randomly searched in {2, 3, 5, 7, 10}. Given K, the length n of each basic cycle is selected according to the greedy strategy. j , A and B represent the starting point and end point of the time series respectively. First, consider K = 2, and maximize the distribution distance d(S AC , S CB ) Select one segmentation point from the nine candidate segmentation points and record it as C. After determining C, consider K=3 and use the same strategy to select another point D. Apply similar strategies to different K values ​​and so on; S22, based on the obtained optimal original training load sequence data sub-segment division, introduces the time series distribution matching algorithm module to learn the common domain knowledge between different sub-segments by matching their distributions. The specific prediction loss function term and time series matching loss function term are as follows: The final objective function of time series distribution matching is as follows: In the final objective function of time series distribution matching, λ is a trade-off hyperparameter, formula Represents the average value of the distribution distance of all paired basic cycles; a small batch of training data segments Di and Dj is used to perform the forward operation of the RNN layer and connect all hidden features. Finally, time series distribution matching can be achieved by minimizing the following formula during the training process; S23, combining time series similarity quantification and distribution matching algorithm, constructs an adaptive RNN air conditioning load prediction model. During the training process, the parameters of the GRU model can be dynamically adjusted according to the distribution changes of the time series to improve the accuracy of the prediction.

4. The method for evaluating the air conditioning load demand response potential based on a time series matching strategy according to claim 1, characterized in that: Step S3 is specifically as follows: S31, define the air conditioning demand response potential value DR(t); S32, using the optimal decision-making simulation model of air-conditioning user demand response based on the home energy management system, simulating the load changes of air-conditioning users under different demand response strategies, thereby constructing a load data set after the target air-conditioning users respond; S33, combined with the improved adaptive RNN load forecasting model, uses the historical load data of the target user to train and predict its baseline load before participating in demand response. The post-response load data set constructed in S2 is used to train the model and predict the air-conditioning load of the target user after participating in demand response. The difference in load and electricity consumption before and after the user participates in demand response is calculated as an evaluation indicator of demand response potential.

5. The method for evaluating the air conditioning load demand response potential based on the time series matching strategy according to claim 4, characterized in that: In step S31, the air conditioning demand response potential value DR(t) is defined as: The ability of air-conditioning users to reduce peak load or increase valley load before and after participating in demand response is expressed in the following mathematical expressions: Where P AC (t) is the load of air-conditioning users before participating in demand response, Load of air-conditioning users after participating in demand response.

6. The method for evaluating the air conditioning load demand response potential based on a time series matching strategy according to claim 4, characterized in that: In step S32, an optimal decision simulation model for air-conditioning user demand response based on a home energy management system is used, and the optimal decision simulation model includes an objective function and air-conditioning load constraint modeling.

7. The method for evaluating the air conditioning load demand response potential based on the time series matching strategy according to claim 6, characterized in that: The objective function specifically includes: In the incentive-based demand response, the user's goal is to minimize his total daily electricity cost, as shown below: Where λ is the daily fixed electricity price, s(μ) is a logical variable used to determine whether the electricity consumption reaches the contract level, so as to know whether the customer needs to pay a fine, and is expressed as: Δd represents the reduction in electricity consumption during a DR event, i.e., the user’s DR potential. The specific expression is as follows: The definitions of the remaining variables in the objective function are as follows, where d cont represents the contractual electricity reduction between individual customers and load aggregators, Inc is the monetary reward per unit electricity reduction, and Pen is the monetary penalty per unit electricity reduction when the reduction of individual customers does not reach the pre-agreed contract level; The following equation is used to determine whether the customer will pay less compared to the situation without DR; The power balance constraints are as follows: The left part is the total household power consumption, and the right part is the transferable load Air conditioning load and other inelastic loads composition.

8. The method for evaluating the air conditioning load demand response potential based on a time series matching strategy according to claim 6, characterized in that: The air conditioning load constraint modeling is as follows: the air conditioning load is an ideal flexible load resource for demand response, because when it actually participates in demand response, it does not need to be completely shut down. It only needs to adjust its set temperature appropriately within the user comfort range to flexibly adjust the air conditioning load; this section uses the simplified first-order ETP model mentioned in Chapter 2 to model the air conditioning load unit; when the air conditioner is turned on, the following constraints will be executed: When the air conditioner is off, there are: Where τ is the time step of 1 min, and the duration of the air conditioner in the on and off states is as follows: Q i =αηp i air Here, θ1 is set to θ max , θ2 is set to θ min , when the air conditioner is in cooling mode in summer, α is set to -1.

9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that the memory stores a computer program, and when the processor executes the computer program, it implements an air conditioning load demand response potential assessment method based on a timing matching strategy as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute an air conditioning load demand response potential assessment method based on a timing matching strategy according to any one of claims 1 to 4.

Images