Large-area power consumption prediction method based on distributed calculation

Abstract

The invention belongs to the technical field of regional power consumption prediction, and aims to solve the problem that detailed power consumption data of each region can be directly exposed by an existing regional power consumption prediction method. The distributed calculation-based large-area power consumption prediction method comprises the following steps of: dividing a large area of which the power consumption is to be predicted into a plurality of small areas, and performing secondary division on the divided small areas into a plurality of areas; constructing a power consumption prediction model of each district, and updating the power consumption prediction model of the core district by using the encryption loss data and gradient data of the other districts as the power consumption prediction model of the small district to which the core district belongs; the core small area receives encryption model parameters of the other small areas and generates model gradients corresponding to the small areas; and constructing a power consumption prediction model of the whole large area through encrypted data transmission and model updating iteration between the core small area and the other small areas. According to the method, the leakage problem in data transmission can be avoided, and the prediction accuracy and safety are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of regional power consumption prediction, and in particular relates to a large-area power consumption prediction method based on distributed computing. Background Art

[0002] At present, there have been many studies on power consumption prediction, but most of them involve the power consumption prediction of individual equipment such as electric vehicles and air conditioners, rather than the power consumption prediction under the integration of multiple business formats and multiple regions, such as the Chinese patents "A data-driven regional building air conditioner short-term power consumption prediction method", "New energy vehicle charging reminder analysis system based on big data", and "A regional power consumption side electric vehicle load time-sharing dynamic prediction method and system".

[0003] In addition, the Chinese patent "A planning method for a regional integrated energy system" mainly focuses on the planning method and does not provide a detailed description of the prediction method. Moreover, judging from the disclosure documents, the prediction method in this document mainly relies on long and short time series and has nothing to do with distributed algorithms.

[0004] The existing regional electricity consumption forecasting technology requires each region to aggregate detailed electricity consumption data into one system, then train an overall model in the system, and then use the model to form forecast data. This method will directly expose the detailed electricity consumption data of each region, increasing data security risks. Summary of the invention

[0005] In order to solve at least one of the above technical problems existing in the prior art, the present invention provides a large-area power consumption prediction method based on distributed computing.

[0006] The present invention is implemented by the following technical solution: a large area power consumption prediction method based on distributed computing, comprising the following steps:

[0007] S1: Divide the large area of ​​power consumption to be predicted into several small areas, and define one of the small areas as the core small area;

[0008] S2: Perform secondary division on the divided small areas, divide each small area into several areas, and randomly select one of the areas as the core area;

[0009] S3: Obtain first characteristic data affecting power consumption in each area, construct a power consumption prediction model for each area, and perform encryption training on the power consumption prediction model for the area based on the first characteristic data;

[0010] S4: In any small area, the core area uses the encrypted loss data and gradient data of the other areas to update its own power consumption prediction model. After the model converges, the power consumption prediction model of the core area is used as the power consumption prediction model of the small area to which it belongs;

[0011] S5: obtaining second characteristic data affecting power consumption in each small area, and performing secondary encryption training on a power consumption prediction model of the corresponding small area based on the second characteristic data;

[0012] S6: The core small area receives the encrypted model parameters of the other small areas and generates the model gradient of the corresponding small area, and the other small areas update their own power consumption prediction models based on the model gradient;

[0013] S7: Through encrypted data transmission and model update iteration between the core small area and other small areas, a power consumption prediction model for the entire large area is constructed, and power consumption prediction is performed.

[0014] Preferably, in any small area, the remaining areas except the core area are divided into two parts, one part encrypts the gradient data and sends it to the core area, and the other part encrypts the loss data and sends it to the core area.

[0015] Preferably, in any small area, the core patch and the remaining patches only exchange the first feature data under different features during the process of exchanging loss data and gradient data.

[0016] Preferably, the core area sends encryption keys to other areas in advance, and when other areas transmit gradients and losses to the core area, the encryption algorithms used are all asymmetric encryption RSA algorithms;

[0017] The core small area sends the encryption key to the other small areas in advance. When the other small areas transmit model parameters to the core small area, the encryption algorithm used is the asymmetric encryption RSA algorithm.

[0018] Preferably, the criteria for dividing a large area into several small areas are: the administrative divisions and business types at the next level of the large area;

[0019] Small areas are divided into several areas based on their functions, including but not limited to commercial areas, residential areas, industrial areas, educational areas and green park areas.

[0020] Preferably, the first characteristic data affecting power consumption in different areas are sorted in order of date from far to near, and used as input of the power consumption prediction model of the corresponding area;

[0021] The first characteristic data corresponding to the commercial area includes but is not limited to time, temperature, number of office buildings, business hours, and whether there are commercial activities;

[0022] The first characteristic data corresponding to the residential area includes but is not limited to time, temperature, occupancy rate, household appliance penetration rate, and residential area;

[0023] The first characteristic data corresponding to the industrial zone includes but is not limited to time, temperature, machine working hours, factory type, and factory area;

[0024] The first characteristic data corresponding to the education zone includes, but is not limited to, time, temperature, number of students in school, class duration, and school type;

[0025] The first characteristic data corresponding to the green park area includes but is not limited to time, temperature, park area, number of facilities, and visitor flow.

[0026] Preferably, the second characteristic data corresponding to any small area includes but is not limited to time, temperature, daily average power consumption, number of power consumption units, and total population.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention is based on distributed computing and can predict the overall power consumption of a large area composed of multiple cities without revealing the power consumption of a single city. At the same time, due to the complexity of the business forms of power consumption in each city, it is usually necessary to establish power consumption prediction models for different business forms in order to accurately predict. This method can also be used to integrate the power consumption prediction models of multiple business forms in a single area, so as to improve the accuracy of the prediction model of a single area.

[0029] The present invention adopts a local training model in each area, uploads the model gradient and loss to form a small area power consumption prediction model, and then transmits the small area model parameters to the core area, and obtains the final large area power consumption prediction model through secondary training to ensure the data security of power consumption details in each area. The establishment of the prediction model in this article requires the collection and management of a large amount of data, and it is necessary to improve the accuracy and generalization ability of the model for the power consumption scenarios in a certain area, so that it has better performance in energy consumption prediction in different regions, and provides a basis for the future application of the model in large area power consumption prediction perception.

[0030] The present invention avoids the problem of confidentiality leakage during data transmission. In the first training process, the power consumption data of a certain place is not directly transmitted, but only the feature intermediate result obtained by model training using the local data is transmitted. The feature intermediate result includes gradient data and loss data, and the feature intermediate result needs to be encrypted before being sent, thereby ensuring the data security of the local power consumption data. In the second training process, the local power consumption data is not directly transmitted, but the model parameters are transmitted to prevent data leakage while ensuring the accuracy of model training.

[0031] The present invention applies a distributed algorithm to the problem of large-area power consumption prediction. In the first model training phase, each area is independently calculated, and the small area only needs to obtain gradient and loss information to update the model accordingly. In the second model training phase, each small area trains the model independently, while the large area is responsible for integrating and updating the model parameters. In this way, the present invention can effectively reduce the computational complexity in both training processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 It is a large-area power consumption prediction flow chart of the present invention. DETAILED DESCRIPTION

[0034] In conjunction with the drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0035] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportional relationship or adjustment of size should fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention. It should be noted that in this specification, relational terms such as first and second are only used to distinguish one entity from several other entities, and do not necessarily require or imply any actual relationship or order between these entities.

[0036] The present invention provides an embodiment:

[0037] like Figure 1 As shown, a large area power consumption prediction method based on distributed computing includes the following steps:

[0038] S1: Divide the large area of ​​power consumption to be predicted into several small areas M1, M2, M3, ..., M n, and define one of the small areas M1 as the core small area; the standard for dividing a large area into several small areas is: the administrative division and business type of the next level of the large area;

[0039] S2: Perform secondary division on the divided small areas, and divide each small area into several areas. The divided areas are specifically: commercial area A, residential area B, industrial area C, educational area E and green park area F, and randomly select one of the areas as the core area; the small area is divided into several areas including but not limited to commercial area, residential area, industrial area, educational area and green park area based on the area function;

[0040] S3: Obtain first characteristic data affecting power consumption in each area, construct a power consumption prediction model for each area, and perform encryption training on the power consumption prediction model for the area based on the first characteristic data;

[0041] In this embodiment, different first characteristic data affecting power consumption in each area are collected through various methods such as system docking or manual reporting. The first characteristic data corresponding to the commercial area include but are not limited to time, temperature, number of office buildings, business hours, and whether there are commercial activities; the first characteristic data corresponding to the residential area include but are not limited to time, temperature, occupancy rate, home appliance penetration rate, and residential area; the first characteristic data corresponding to the industrial area include but are not limited to time, temperature, machine working hours, factory type, and factory area; the first characteristic data corresponding to the educational area include but are not limited to time, temperature, number of students in school, class time, and school type; the first characteristic data corresponding to the green park area include but are not limited to time, temperature, park area, number of facilities, and tourist flow.

[0042] The process of building and training the power consumption prediction model for each area is as follows:

[0043] Taking the core small area M1 as an example, the specific process is as follows:

[0044] (1) Assuming that commercial area A is the core area, commercial area A sends the public key to other areas;

[0045] (2) Calculate the intermediate results of gradient and loss calculation for each area. The power consumption prediction model for each area is Θ i∈{A,B,C,E,F} , assuming that the learning rate is η, the regularization parameter is λ, and the dataset Model parameter Θ A , Θ B , Θ C , Θ E , Θ F , corresponding to The feature space of

[0046] The training objectives are:

[0047]

[0048] make

[0049] The encryption loss is:

[0050]

[0051] Where [[·]] is the RSA asymmetric encryption algorithm, in,

[0052] [[L]]=[[L A ]]+[[L B ]]+[[L C ]]+[[L E ]]+[[L F ]]

[0053] The gradient is:

[0054]

[0055] (3) To ensure the generalization ability, only the gradients of the half of the slices except the core slice are used to update the power consumption prediction model. The loss of the half of the slices except the core slice is used to update the power consumption prediction model. Taking E and F sending model parameters and B and C sending gradients as an example, the slice exchange process is as follows:

[0056]

[0057]

[0058] S4: In any small area, the core area uses the encrypted loss data and gradient data of the other areas to update its own power consumption prediction model, and in order to reduce the computational complexity, the same features, such as time and temperature, are not exchanged during the exchange of parameters and gradients among the areas. After the model converges, the power consumption prediction model of the core area is used as the power consumption prediction model of the small area to which it belongs;

[0059] S5: Obtain the second characteristic data affecting power consumption in each small area (the second characteristic data corresponding to any small area includes but is not limited to time x1, temperature x2, daily average power consumption x3, number of power consumption units x4, and total population x5), and perform secondary encryption training on the power consumption prediction model of the small area based on the second characteristic data;

[0060] S6: The core small area receives the encrypted model parameters of the other small areas and generates the model gradient of the corresponding small area, and the other small areas update their own power consumption prediction models based on the model gradient;

[0061] The power consumption prediction model Θ1, Θ2, Θ3, ..., Θn , the specific process is as follows:

[0062] (1) The core small area M1 creates an encryption key pair and sends the public key to other small areas;

[0063] (2) Each small area uses its own data to train the power consumption prediction model, and the loss function is as follows:

[0064]

[0065] (3) After training is completed, each small area encrypts the model parameters and sends them to the core small area M1. After receiving the model parameters, the core area M1 decrypts the model parameters and calculates the model gradients.

[0066] (4) After the core small area M1 calculates the gradient, it encrypts and sends it to each of the remaining small areas. After receiving the gradient, the remaining small areas update their respective models.

[0067] (5) According to the above steps 1-4, the power consumption prediction model of each small area is repeatedly trained until convergence.

[0068] S7: Through encrypted data transmission and model update iteration between the core small area and the remaining small areas, the power consumption prediction model of the entire large area is constructed, and power consumption prediction is performed to obtain the large area power consumption prediction data for the corresponding date. Specifically, when the large area power consumption prediction model is used to predict the power consumption of a large area, the input features of the model are: time, temperature, daily average power consumption, number of power consumption units, and total population.

[0069] In this embodiment, the core area sends encryption keys to the remaining areas in advance, and when the remaining areas transmit gradients and losses to the core area, the encryption algorithms used are all asymmetric encryption RSA algorithms; the core small area sends encryption keys to the remaining small areas in advance, and when the remaining small areas transmit model parameters to the core small area, the encryption algorithms used are all asymmetric encryption RSA algorithms.

[0070] The present invention avoids the problem of confidentiality leakage during data transmission. During the two model training processes, the power consumption data of a certain place is not directly transmitted. Only the feature intermediate results or model parameters obtained by model training using the local data are transmitted. The feature intermediate results include gradient data and loss data, and are all sent in encrypted form, thereby ensuring the data security of the local power consumption data.

[0071] The present invention applies a distributed algorithm to the problem of large-area power consumption prediction. In the first model training phase, each area is independently calculated, and the small area only needs to obtain gradient and loss information to update the model accordingly. In the second model training phase, each small area trains the model independently, while the large area is responsible for integrating and updating the model parameters. In this way, the present invention effectively reduces the computational complexity in both training processes.

[0072] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for predicting power consumption in a large area based on distributed computing, characterized in that: The following steps are involved: S1: Divide the large area of ​​power consumption to be predicted into several small areas, and define one of the small areas as the core small area; S2: Perform secondary division on the divided small areas, divide each small area into several areas, and randomly select one of the areas as the core area; S3: Obtain first characteristic data affecting power consumption in each area, construct a power consumption prediction model for each area, and perform encryption training on the power consumption prediction model for the area based on the first characteristic data; S4: In any small area, the core area uses the encrypted loss data and gradient data of the other areas to update its own power consumption prediction model. After the model converges, the power consumption prediction model of the core area is used as the power consumption prediction model of the small area to which it belongs; S5: obtaining second characteristic data affecting power consumption in each small area, and performing secondary encryption training on a power consumption prediction model of the corresponding small area based on the second characteristic data; S6: The core small area receives the encrypted model parameters of the other small areas and generates the model gradient of the corresponding small area, and the other small areas update their own power consumption prediction models based on the model gradient; S7: Through encrypted data transmission and model update iteration between the core small area and other small areas, a power consumption prediction model for the entire large area is constructed, and power consumption prediction is performed.

2. The method for predicting power consumption in a large area based on distributed computing according to claim 1, characterized in that: In any small area, the remaining areas except the core area are divided into two parts, one of which encrypts the gradient data and sends it to the core area, and the other part encrypts the loss data and sends it to the core area.

3. The method for predicting power consumption in a large area based on distributed computing according to claim 2 is characterized in that: In any small area, the core patch and the remaining patches only exchange the first feature data under different features during the process of exchanging loss data and gradient data.

4. The method for predicting power consumption in a large area based on distributed computing according to claim 1 is characterized in that: The core area sends the encryption key to other areas in advance. When other areas transmit gradients and losses to the core area, the encryption algorithm used is the asymmetric encryption RSA algorithm. The core small area sends the encryption key to the other small areas in advance. When the other small areas transmit model parameters to the core small area, the encryption algorithm used is the asymmetric encryption RSA algorithm.

5. The method for predicting power consumption in a large area based on distributed computing according to claim 1 is characterized by: The criteria for dividing a large area into several small areas are: the administrative divisions and business types at the next level of the large area; Small areas are divided into several areas based on their functions, including but not limited to commercial areas, residential areas, industrial areas, educational areas and green park areas.

6. A method for predicting power consumption in a large area based on distributed computing according to claim 5, characterized in that: The first characteristic data affecting power consumption in different areas are sorted in order of date from far to near, and used as the input of the power consumption prediction model of the corresponding area; The first characteristic data corresponding to the commercial area includes but is not limited to time, temperature, number of office buildings, business hours, and whether there are commercial activities; The first characteristic data corresponding to the residential area includes but is not limited to time, temperature, occupancy rate, household appliance penetration rate, and residential area; The first characteristic data corresponding to the industrial zone includes but is not limited to time, temperature, machine working hours, factory type, and factory area; The first characteristic data corresponding to the education zone includes, but is not limited to, time, temperature, number of students in school, class duration, and school type; The first characteristic data corresponding to the green park area includes but is not limited to time, temperature, park area, number of facilities, and visitor flow.

7. A method for predicting power consumption in a large area based on distributed computing according to claim 6, characterized in that: The second characteristic data corresponding to any small area includes but is not limited to time, temperature, daily average power consumption, number of power consumption units, and total population.

Images