Method, device and equipment for processing heavy overload in power distribution area and storage medium

Abstract

The invention relates to the technical field of load prediction in a novel power system, in particular to a power distribution area heavy overload processing method, device and equipment and a storage medium. The method comprises the steps of obtaining an original load; using a VMD algorithm to decompose and normalize the original load to obtain a normalized component, using a Pearson correlation coefficient to calculate the correlation degree of the original load and the characteristic variables, and selecting the characteristic variables of which the correlation degree is greater than a preset value in the characteristic variables as key characteristics; the normalized component and the key feature are input into a prediction model, a prediction result output by the prediction model is obtained, and the prediction model is a machine learning model based on an LSTM-CNN combined neural network; searching a corresponding early warning level in a preset early warning level table according to the prediction result; issuing early warning information according to the early warning level; and based on the early warning information and a preset demand response model of the intelligent power distribution network, solving the demand response model of the intelligent power distribution network by adopting a particle swarm algorithm to obtain a treatment strategy. According to the scheme provided by the invention, timely graded early warning can be carried out on the heavy overload phenomenon caused by load fluctuation and external environment factors in the power distribution area; an intelligent power distribution network demand response model considering time-of-use electricity price and direct load control can also be established, the model is solved by adopting a particle swarm algorithm, and finally an intelligent power distribution network load management optimization strategy is obtained.

Description

Technical Field

[0001] The present application relates to the technical field of load prediction in a novel power system, and in particular to a method, device, equipment and storage medium for processing a heavy overload in a distribution station area. Background Art

[0002] Power equipment is an important part of the power system, responsible for providing energy and electricity for humans. Its operating status directly affects the normal operation and safety of the power grid, so it is necessary to monitor and warn the operating status of power grid equipment so as to take timely measures and perform maintenance.

[0003] The continuous advancement of new power systems and the extensive use of distributed photovoltaics, electric vehicles, and refrigeration / heating equipment have made the load volatility and randomness of distribution substations more obvious, and the probability of heavy overload in distribution substations has greatly increased. If the distribution substation is in a heavy overload state for a long time, it will aggravate the abnormal loss of distribution substation equipment, reduce the service life of the equipment, and even cause safety accidents such as transformer burning and explosion, bringing hidden faults and operational risks to the power grid. However, the distribution substation does not belong to the category of precision equipment, and the cost is relatively low. The load itself has a large randomness and unpredictability. Therefore, the current research and development on online monitoring of distribution substation equipment is relatively slow, and it mainly focuses on online monitoring of transformers. There is still a lack of heavy overload warning technology for distribution substations, so it is necessary to develop reliable distribution substation warning technology and develop effective management strategies for heavy overload in distribution substations. Summary of the invention

[0004] In order to solve or partially solve the problems existing in the related technology, the present application provides a method, device, equipment and storage medium for handling heavy overload in distribution station areas, which can timely issue early warnings for heavy overload phenomena in distribution station areas caused by load fluctuations and external environmental factors, and can effectively control the heavy overload phenomena in distribution station areas.

[0005] The first aspect of the present application provides a method for handling a heavy overload in a distribution station area, comprising:

[0006] Get the original load;

[0007] The original load is decomposed and normalized by using VMD algorithm to obtain normalized components, the correlation between the original load and the characteristic variables is calculated by using Pearson correlation coefficient, and the characteristic variables with correlation greater than a preset value among the characteristic variables are selected as key features;

[0008] Inputting the normalized component and the key feature into a prediction model to obtain a prediction result output by the prediction model; wherein the prediction model is a machine learning model based on an LSTM-CNN combined neural network;

[0009] According to the prediction result, searching for the corresponding warning level in the preset warning level table;

[0010] Issue warning information according to the warning level;

[0011] Preferably, the processing method also includes solving the smart distribution network demand response model based on the early warning information and the preset smart distribution network demand response model using a particle swarm algorithm to obtain a governance strategy; wherein the preset smart distribution network demand response model is a model that takes time-of-use electricity prices and direct load control into consideration.

[0012] Preferably, the VMD algorithm is used to decompose and normalize the original load to obtain normalized components, the Pearson correlation coefficient is used to calculate the correlation between the original load and the characteristic variable, and the characteristic variable with a correlation greater than a preset value is selected as a key feature, including:

[0013] Decomposing the original load into K modal components with different characteristics by using VMD algorithm;

[0014] Normalizing the K modal components with different characteristics to obtain normalized components;

[0015] The Pearson correlation coefficient is used to calculate the correlation between the characteristic variables and the original load, and a correlation greater than a preset value is selected as a key feature, wherein the characteristic variables include external environmental variables such as temperature, humidity, and light radiation.

[0016] Preferably, the VMD algorithm is used to decompose the original load into K modal components with different characteristics, including:

[0017] Acquire load data through the power consumption information collection system, clean up the load data, and form the original load;

[0018] Decomposing the original load by using VMD algorithm;

[0019] The decomposition step comprises:

[0020] First, the original load is f, which is decomposed into the K modal components with different characteristics;

[0021]

[0022] Where: u k are the modal components; ω k is the center frequency of each mode; is the partial derivative with respect to time t; δ(t) is the impulse function; u k(t) is the kth modal component; f is the original signal; st represents the constraint condition;

[0023] Secondly, transform the constrained variational problem into an unconstrained problem:

[0024]

[0025] Where λ(t) is the Lagrange multiplier; α is the quadratic penalty factor; [·] represents the inner product;

[0026] Finally, iterate and solve:

[0027]

[0028] set up λ 1 The initial value of , let n = 0, N is the maximum number of iterations;

[0029]

[0030] Spend.

[0031] Preferably, the normalizing the K modal components having different characteristics comprises:

[0032]

[0033] Among them, X′ is the normalized component, X is the modal component, and X max is the maximum value among the modal components, X min is the minimum value among the modal components.

[0034] Preferably, the method of calculating the correlation between the original load and the characteristic variable using the Pearson correlation coefficient and selecting the characteristic variable with the correlation greater than a preset value as the key feature includes:

[0035] The Pearson correlation coefficient is calculated as follows:

[0036]

[0037] Where: characteristic variable A={a1,...,a n}, the original load B = {b1,...,b n}, R is the Pearson correlation coefficient; n is the dimension of the variable; and is the average value of the characteristic variable and the original load respectively; k∈[1,n]; n∈[-1,1]; positive and negative values ​​represent the positive correlation and negative correlation between the two, where the larger the absolute value, the greater the correlation.

[0038] Preferably, the normalized component and the key feature are input into a prediction model to obtain a prediction result output by the prediction model; wherein the prediction model is a machine learning model based on an LSTM-CNN combined neural network, comprising:

[0039] A LSTM-CNN combined neural network model is established, the normalized component is input, multiple LSTM networks learn multi-scale time feature representation in parallel, and the CNN convolutional neural network convolves the output time step of each LSTM to extract its output features:

[0040] The CNN convolutional neural network includes three parts: convolution layer, pooling layer and fully connected layer. The convolution layer extracts one-dimensional features. The specific extraction formula is:

[0041]

[0042] Where: is the feature map j of the lth layer; σ is the activation function ReLU; is the bias; F l is the number of feature maps on the l-th layer feature map f; Convolution is performed on the feature map f of the lth layer to create the feature map j in the l+1th layer;

[0043] The pooling layer receives the features extracted by the convolutional layer and reduces the dimension of the features to improve robustness. The specific pooling formula is:

[0044]

[0045] Where: f is the convolution kernel size; s0 is the step size; x is the input, p is the number of padding layers, and when p is infinite, the maximum value is taken in the pooling area, that is, the maximum pooling;

[0046] Considering the key features, multiple LSTM networks are established to learn and predict the impact of the key features on the load at the target time. The specific formula is:

[0047] f t =signoid(w f x t +b f )

[0048] i t =signoid(w i x t +b i )

[0049] c t =f t c t-1 +i t tanh(w cx t +w c h t-1 +b c )

[0050] o t =signoid(w o x t +b o )

[0051] h t =o t tanh(c t )

[0052] Where: f t is the forget gate state, i t is the input gate state, o t is the output gate state, c t is the temporary state of the memory unit at time t generated by the tanh layer, x t is the input information of the LSTM basic unit at time t, w f 、w i 、w c 、w o is the corresponding weight matrix, b f 、b i 、b c 、b o is the corresponding bias matrix, c t is the current state;

[0053] The normalized component is combined with the key features and the time module, and the prediction result of the final time series is obtained through the fully connected layer.

[0054] Preferably, based on the early warning information and the preset smart distribution network demand response model, a particle swarm algorithm is used to solve the smart distribution network demand response model to obtain a governance strategy, including:

[0055] Considering the time-of-use electricity price and the direct load control, using the demand price elasticity matrix E, updating the time-of-use electricity price, and obtaining the peak and valley load after its implementation;

[0056]

[0057] Among them, L peak , L flat and L valley are the peak, flat and valley loads before the implementation of time-of-use electricity prices; L p ' eak , L′ flat and L v ' alleyare the peak, flat and valley loads affected by time-of-use electricity prices; p peak 、p flat and p valley is the previous time-of-use electricity price; Δp peak , Δp flat and Δp valley It is the difference in electricity prices before and after the implementation of time-of-use electricity prices;

[0058] Among them, the responsibility r and the actual interruption compensation rate are defined respectively for controlling the user's load;

[0059]

[0060] a=r0F

[0061] F = 0.35arctanhx + 1

[0062] where t off.max and t on.min is the maximum continuous control time and minimum continuous operation time of the user; F is a function used to calculate the differential compensation rate; where the objective function used is min(max(L)-min(L)), which is used to represent the optimization goal of minimizing the peak-valley load difference, L is the load, and the constraints include the total load constraint, the threshold constraint of the time-of-use electricity price, and the direct load control compensation price constraint;

[0063] The proposed variable price and interruption compensation price are optimized using the particle swarm algorithm and the solution of the governance strategy is obtained based on the particle swarm algorithm.

[0064] A second aspect of the present application provides a device for processing a heavy overload in a distribution station area, comprising:

[0065] An acquisition module, used to obtain the original load;

[0066] An extraction module, used for extracting the normalized component and the key feature;

[0067] A prediction module, used for inputting the normalized component and the key feature into a prediction model to obtain a prediction result output by the prediction model; wherein the prediction model is a machine learning model based on a LSTM-CNN combined neural network;

[0068] A search module, used to search for a corresponding warning level in a preset warning level table according to the prediction result;

[0069] An alarm module, used to issue warning information according to the warning level;

[0070] Preferably, the device also includes an optimization module, which is used to solve the smart distribution network demand response model based on the early warning information and the preset smart distribution network demand response model using a particle swarm algorithm to obtain a governance strategy; wherein the preset smart distribution network demand response model is a model that takes time-of-use electricity prices and direct load control into consideration.

[0071] Preferably, the acquisition module is used to obtain the original load.

[0072] Preferably, the extraction module uses a VMD algorithm to decompose the original load into K modal components with different characteristics; normalizes the K modal components with different characteristics to obtain normalized components; uses the Pearson correlation coefficient to calculate the correlation between the characteristic variables and the original load, and selects characteristic variables with a correlation greater than a preset value as key features, and the characteristic variables include external environmental variables such as temperature, humidity, and light radiation.

[0073] Preferably, the prediction module is used to input the normalized component and the key feature into the prediction model to obtain the prediction result output by the prediction model; wherein the prediction model is a machine learning model based on the LSTM-CNN combined neural network, including: establishing an LSTM-CNN combined neural network model, inputting the normalized component, multiple LSTM networks learning multi-scale time feature representation in parallel, and the CNN convolutional neural network convolving the output time step of each LSTM to extract its output features; the CNN convolutional neural network includes three parts: a convolutional layer, a pooling layer and a fully connected layer, the convolutional layer extracts one-dimensional features, the pooling layer receives the features extracted by the convolutional layer and reduces the dimension of the features to improve robustness; considering the key features, establishing multiple LSTM networks, learning to predict the impact of the key features on the load at the target time; combining the normalized component with the key features and the time module, and obtaining the prediction result of the final time series through the fully connected layer.

[0074] Preferably, the search module is used to search for the corresponding warning level in a preset warning level table according to the prediction result.

[0075] Preferably, the alarm module is used to issue warning information according to the warning level.

[0076] A third aspect of the present application provides an electronic device, including:

[0077] A processor; and a memory having executable codes stored thereon, which, when executed by the processor, cause the processor to execute the method as described above.

[0078] A fourth aspect of the present application provides a computer-readable storage medium having executable code stored thereon. When the executable code is executed by a processor of an electronic device, the processor is caused to execute the method as described above.

[0079] The technical solution provided in this application may include the following beneficial effects: it can provide timely graded warnings for heavy overload phenomena caused by load fluctuations and external environmental factors in distribution station areas; it can also establish a smart distribution network demand response model that takes into account time-of-use electricity prices and direct load control, and use a particle swarm algorithm to solve the model, ultimately obtaining a smart distribution network load management optimization strategy.

[0080] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The above and other objects, features and advantages of the present application will become more apparent by describing in more detail exemplary embodiments of the present application in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same components in the exemplary embodiments of the present application.

[0082] Figure 1 A flow chart of a method for handling severe overload in a distribution station area proposed by the present invention;

[0083] Figure 2 A flowchart of a specific embodiment of a method for processing a heavy overload in a distribution station area proposed by the present invention;

[0084] Figure 3 A module connection diagram of a specific embodiment of a heavy overload processing device for a distribution station area proposed by the present invention;

[0085] Figure 4 This is a structural diagram of an electronic device of a specific embodiment of a heavy overload handling device for a distribution station area proposed by the present invention. DETAILED DESCRIPTION

[0086] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0087] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0088] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0089] The embodiment of the present invention provides a method for handling heavy overload in a distribution station area, which can provide timely and graded early warning of heavy overload phenomena caused by load fluctuations and external environmental factors in the distribution station area; it can also establish an intelligent distribution network demand response model that considers time-of-use electricity prices and direct load control, and use a particle swarm algorithm to solve the model, and finally obtain an intelligent distribution network load management optimization strategy.

[0090] The technical solution of the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0091] See also Figure 1 , a flow chart of a method for handling a heavy overload in a distribution station area proposed by the present invention discloses a process of a method for handling a heavy overload in a distribution station area:

[0092] S1. Obtain the original load;

[0093] S2. Decompose and normalize the original load using the VMD algorithm to obtain normalized components, use the Pearson correlation coefficient to calculate the correlation between the original load and the characteristic variables, and select the characteristic variables whose correlation is greater than the preset value as key features;

[0094] S3, input the normalized component and the key features into the prediction model to obtain the prediction result output by the prediction model; wherein the prediction model is a machine learning model based on the LSTM-CNN combined neural network;

[0095] S4. According to the prediction result, find the corresponding warning level in the preset warning level table;

[0096] S5. Issue warning information according to the warning level;

[0097] S6. Based on the early warning information and the preset smart distribution network demand response model, a particle swarm algorithm is used to solve the smart distribution network demand response model to obtain a governance strategy.

[0098] The technical solution provided in this application can provide timely graded warnings for heavy overload phenomena caused by load fluctuations and external environmental factors in distribution station areas; it can also establish a smart distribution network demand response model that takes into account time-of-use electricity prices and direct load control, and use a particle swarm algorithm to solve the model, ultimately obtaining a smart distribution network load management optimization strategy.

[0099] See also Figure 2 , is a flow chart of a specific embodiment of a method for handling heavy overload in a distribution station area proposed by the present invention, and the specific steps are as follows:

[0100] S101. Obtain 96 load data points of the load in the substation through the power consumption information collection system, clean the load data, such as removing abnormal values ​​and missing values, to ensure data quality, and obtain the original load; decompose the original load to obtain K modal components with different characteristics;

[0101] S102, normalizing the K modal components with different characteristics to obtain normalized components;

[0102] S103. According to formula (8), the Pearson correlation coefficient (PCC) is used to calculate the correlation between the original load and characteristic variables such as temperature, humidity, and light radiation, and characteristic variables with correlations greater than a preset value are selected as key features.

[0103]

[0104] Where: characteristic variable A={a1,...,a n ], the original load B = {b1,...,b n}, R is the Pearson correlation coefficient; n is the dimension of the variable; and is the average value of the characteristic variable and the original load; k∈[1,n]; n∈[-1,1]; positive and negative values ​​indicate positive and negative correlations between the two. The larger the absolute value, the greater the correlation, and the characteristic variable greater than the preset value is selected as the key feature;

[0105] S104. Establish an LSTM-CNN combined neural network model, input a normalized component, and multiple LSTM networks learn multi-scale time feature representation in parallel. The CNN convolutional neural network convolves the output time step of each LSTM to extract its output features.

[0106] CNN convolutional neural network consists of three parts: convolutional layer, pooling layer and fully connected layer:

[0107] The convolution layer extracts one-dimensional features. The specific extraction formula is:

[0108]

[0109] Where: is the feature map j of the lth layer; σ is the activation function ReLU; is the bias; F l is the number of feature maps on the l-th layer feature map f; Convolution is performed on the feature map f of the lth layer to create the feature map j in the l+1th layer.

[0110] The pooling layer receives the features extracted by the convolutional layer and reduces the dimension of the features to improve robustness. The specific pooling formula is:

[0111]

[0112] Where: f is the convolution kernel size; s0 is the step size; x is the input, p is the number of padding layers, and when p is infinite, the maximum value is taken in the pooling area, that is, maximum pooling.

[0113] S105. Considering the key features, multiple LSTM networks are established to learn and predict the impact of the key features on the load at the target time. The specific formula is:

[0114] f t =signoid(w f x t +b f )(11)

[0115] i t =signoid(w i x t +b i )(12)

[0116] c t =f t c t-1 +i t tanh(w c x t +w c h t-1 +b c )(13)

[0117] o t =signoid(w o x t +b o )c t =f t ct-1 +i t tanh(w c x t +w c h t-1 +b c )(14)

[0118] h t =o t tanh(c t )(15)

[0119] Where: f t is the forget gate state, i t is the input gate state, o t is the output gate state, c t is the temporary state of the memory unit at time t generated by the tanh layer, x t is the input information of the LSTM basic unit at time t, w f 、w i 、w c 、w o is the corresponding weight matrix, b f 、b i 、b c 、b o is the corresponding bias matrix, c t The current state.

[0120] S106. Combine the normalized component with the key features and the time module, and obtain the load forecasting result of the final time series through the fully connected layer.

[0121] S107. According to the load forecast result, the corresponding warning level is searched in the preset warning level table, and warning information is issued according to the warning level. The preset warning level table is as follows:

[0122] Warning level Load factor Display Color state Safety [0,0.3) blue Light load Orange warning [0.3,8) orange color Pro Overload Yellow Warning [0.8,1) yellow Overload Red Alert [1,∞) red Overload

[0123] Among them, load rate = actual load / equipment rated load.

[0124] S108. Establish a demand response model for a smart distribution network that takes into account time-of-use electricity prices and direct load control, and use a particle swarm algorithm to solve the model, and finally obtain an optimization strategy for load management of the smart distribution network. The specific steps are as follows:

[0125] Taking into account the time-of-use electricity price, the peak and valley times are divided into: peak-valley time is 20:00-2:00 (2h), flat-valley time is 7:00-20:00 and 22:00-24:00 (15h), and valley time is 0:00-7:00 (7h).

[0126] Using the demand price elasticity matrix E, the time-of-use electricity price is updated and the peak and valley loads after its implementation are obtained.

[0127]

[0128] Among them, L peak , L flat and L valley They are the peak, flat and valley loads before the implementation of time-of-use electricity prices. p ' eak , L′ flat and L v ' alley They are the peak, flat and valley loads affected by time-of-use electricity prices. peak 、p flat and p valley is the previous time-of-use electricity price. Δp peak ,Δp flat and Δp valley It is the difference in electricity prices before and after the implementation of time-of-use electricity prices.

[0129] Considering direct load control, there are three main strategies: (1) directly shutting down the user's electrical equipment; (2) turning on / off the loop control of the power supply equipment; (3) adjusting the set temperature of the user's thermostat to achieve an effect similar to interrupting the power supply. After the user participates in direct load control, the power supplier will incentivize the user to respond to demand in the form of interruption compensation (assuming that the basic interruption compensation rate is consistent). In order to clearly distinguish the contribution degree of users participating in direct load control and implement reasonable differential compensation for users with different contributions, the responsibility r and the actual interruption compensation rate a are defined for the user's load control.

[0130]

[0131] a=r0F(18)

[0132] F=0.35arctanhx+1(19)

[0133] where t off.max and t on.min are the maximum continuous control time and minimum continuous operation time of the user; F is the function used to calculate the differential compensation rate.

[0134] Taking into account time-of-use electricity prices and direct load control, the specific objective function and constraints are as follows:

[0135] min(max(L)-min(L))(20)

[0136] min(C) (21)

[0137] sum(L)=sum(L0)(22)

[0138] 0.3≤p valley ≤p flat ≤p peak ≤1.5 (23)

[0139] x<0.4(24)

[0140] Formula (20) is the main optimization objective of minimizing the peak-valley load difference. Considering the interests of electricity users, formula (21) is the minimum user cost, where L is the load, L0 is the initial load, C is the user cost, and x is the compensation cost. Formula (22) is the total load constraint. The ideal assumption is to eliminate the extreme case where users directly reduce electricity consumption to save electricity costs. In order to prevent users from overreacting or peak-valley reversal or drift of the load curve, considering the user's tolerance and power supply cost, formula (23) is the threshold constraint of time-of-use electricity price. In the optimization process, it is a reasonable threshold value of peak-valley electricity price. Formula (24) is the direct load control compensation price constraint, which is also to limit the load transfer amount of users, thereby avoiding peak-valley reversal in extreme cases.

[0141] The proposed variable price and interruption compensation price are optimized using the particle swarm algorithm. In the D-dimensional space, the position vector X of the i-th particle i =[x i1 ,x i2 ,...,x iD ], its flight speed V i =[v i1 ,v i2 ,...,v iD ], and its fitness in space is determined by the objective function. The particle with the best fitness in each iteration is recorded as p best , and the best fitness particle in the entire iteration is recorded as g best The particle continuously updates its position and velocity during the optimization process, and the update formula is as follows:

[0142]

[0143] Where k is the number of iterations; is the velocity of the particle in the kth iteration; c1 and c2 are acceleration constants, usually 2; ω is the inertia weight, the larger the global search capability, the better, and the smaller the local search capability, the better. In order to obtain the global optimal solution, an improved strategy of linearly decreasing inertia weight is adopted.

[0144]

[0145] Among them, ω max and ω minare the maximum and minimum inertia weights, here we take 0.9 and 0.4; t max is the maximum number of iterations, and t is the current number of iterations. In this way, the particle swarm algorithm first starts to explore a larger area and locates the approximate position of the optimal solution more quickly. ω gradually decreases, the particle speed slows down, and a fine local search begins. The particle swarm algorithm can better control the ability of global and local searches, speed up the convergence speed, and finally obtain the optimal solution, that is, the optimization strategy for load management of smart distribution networks.

[0146] See also Figure 3 , which is a module connection diagram of a specific embodiment of a heavy overload processing device for a distribution station area proposed by the present invention, specifically comprising:

[0147] Acquisition module 1, used to obtain the original load;

[0148] Extraction module 2, using variational mode decomposition VMD algorithm to decompose and normalize the original load to obtain normalized components, using Pearson correlation coefficient to calculate the correlation between the original load and the characteristic variable, and selecting the correlation greater than a preset value as the key feature, including: using VMD algorithm to decompose the original load into K modal components with different characteristics; normalizing the K modal components with different characteristics to obtain normalized components; using Pearson correlation coefficient to calculate the correlation between the characteristic variable and the original load, and selecting the correlation greater than a preset value as the key feature, the characteristic variables include temperature, humidity, light radiation and other external environmental variables;

[0149] Prediction module 3, inputs the normalized component and the key features into the prediction model to obtain the prediction results output by the prediction model; wherein the prediction model is a machine learning model based on the LSTM-CNN combined neural network, including: establishing an LSTM-CNN combined neural network model, inputting the normalized component, multiple LSTM networks learning multi-scale time feature representation in parallel, and the CNN convolutional neural network convolving the output time step of each LSTM to extract its output features; the CNN convolutional neural network includes three parts: a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer extracts one-dimensional features, and the pooling layer receives the features extracted by the convolutional layer and reduces the dimension of the features to improve robustness; considering the key features, establishing multiple LSTM networks, and learning to predict the impact of the key features on the load at the target time; combining the normalized component with the key features and the time module, and obtaining the prediction results of the final time series through the fully connected layer;

[0150] Search module 4, used to search for the corresponding warning level in the preset warning level table according to the prediction result;

[0151] Alarm module 5, used to issue warning information according to the warning level;

[0152] The optimization module 6 is used to obtain the load management optimization strategy of the smart distribution network considering the time-of-use electricity price.

[0153] See also Figure 4 , is a structural diagram of an electronic device of a specific embodiment of a heavy overload processing device for a distribution station area proposed by the present invention, which specifically includes:

[0154] The processor 1020 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0155] The memory 1010 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. Among them, ROM can store static data or instructions required by the processor 1020 or other modules of the computer. The permanent storage device may be a readable and writable storage device. The permanent storage device may be a non-volatile storage device that does not lose the stored instructions and data even after the computer is powered off. In some embodiments, the permanent storage device uses a large-capacity storage device (such as a magnetic or optical disk, flash memory) as a permanent storage device. In some other embodiments, the permanent storage device may be a removable storage device (such as a floppy disk, optical drive). The system memory may be a readable and writable storage device or a volatile readable and writable storage device, such as a dynamic random access memory. The system memory may store some or all instructions and data required by the processor at run time. In addition, the memory 1010 may include any combination of computer-readable storage media, including various types of semiconductor storage chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and disks and / or optical disks may also be used. In some embodiments, the memory 1010 may include a readable and / or writable removable storage device, such as a laser disc (CD), a read-only digital versatile disc (such as a DVD-ROM, a double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a mini SD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. The computer-readable storage medium does not include carrier waves and transient electronic signals transmitted wirelessly or wired.

[0156] The memory 1010 stores executable codes, and when the executable codes are processed by the processor 1020 , the processor 1020 can execute part or all of the methods described above.

[0157] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps in the above method of the present application.

[0158] Alternatively, the present application can also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium) on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or server, etc.), the processor executes part or all of the steps of the above-mentioned method according to the present application.

[0159] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for handling heavy overload in a distribution station area, characterized in that: include: Get the original load; Decomposing and normalizing the original load by using the variational mode decomposition (VMD) algorithm to obtain normalized components, calculating the correlation between the original load and the characteristic variable by using the Pearson correlation coefficient, and selecting the characteristic variable with a correlation greater than a preset value as a key feature; Inputting the normalized component and the key feature into a prediction model to obtain a prediction result output by the prediction model; wherein the prediction model is a machine learning model based on a long short-term memory (LSTM)-convolutional neural network (CNN) combined neural network; According to the prediction result, searching for the corresponding warning level in the preset warning level table; The warning information is issued according to the warning level.

2. The method according to claim 1, characterized in that Also includes: Based on the early warning information and the preset smart distribution network demand response model, a particle swarm algorithm is used to solve the smart distribution network demand response model to obtain a governance strategy; Among them, the preset smart distribution network demand response model is a model that takes time-of-use electricity prices and direct load control into consideration.

3. The method according to claim 1 or 2, characterized in that: The method of using the variational mode decomposition (VMD) algorithm to decompose and normalize the original load to obtain normalized components, using the Pearson correlation coefficient to calculate the correlation between the original load and the characteristic variable, and selecting the characteristic variable with a correlation greater than a preset value as a key feature includes: Decomposing the original load into K modal components with different characteristics by using VMD algorithm; Normalizing the K modal components with different characteristics to obtain normalized components; The Pearson correlation coefficient is used to calculate the correlation between the characteristic variables and the original load, and a correlation greater than a preset value is selected as a key feature, wherein the characteristic variables include external environmental variables such as temperature, humidity, and light radiation.

4. The method according to claim 3, characterized in that The VMD algorithm is used to decompose the original load into K modal components with different characteristics, including: Acquire load data through the power consumption information collection system, clean up the load data, and form the original load; Decomposing the original load by using VMD algorithm; The decomposition step comprises: First, the original load is f, which is decomposed into the K modal components with different characteristics; Where: u k are the modal components; ω k is the center frequency of each mode; is the partial derivative with respect to time t; δ(t) is the impulse function; u k (t) is the kth modal component; f is the original signal; st represents the constraint condition; Secondly, transform the constrained variational problem into an unconstrained problem: Where λ(t) is the Lagrange multiplier; α is the quadratic penalty factor; [·] represents the inner product; Finally, iterate and solve: set up λ 1 The initial value of , let n = 0, N is the maximum number of iterations; Spend.

5. The method according to claim 3, characterized in that: The normalizing the K modal components having different characteristics comprises: Among them, X′ is the normalized component, X is the modal component, and X max is the maximum value among the modal components, X min is the minimum value among the modal components.

6. The method according to claim 3, characterized in that The method of calculating the correlation between the original load and the characteristic variable using the Pearson correlation coefficient and selecting the characteristic variable with a correlation greater than a preset value as a key feature includes: The Pearson correlation coefficient is calculated as follows: Where: characteristic variable A={a1,...,a n }, the original load B = {b1,...,b n }, R is the Pearson correlation coefficient; n is the dimension of the variable; and is the average value of the characteristic variable and the original load respectively; k∈[1,n]; n∈[-1,1]; positive and negative values ​​represent the positive correlation and negative correlation between the two, where the larger the absolute value, the greater the correlation.

7. The method according to claim 1, characterized in that The normalized component and the key feature are input into a prediction model to obtain a prediction result output by the prediction model; wherein the prediction model is a machine learning model based on an LSTM-CNN combined neural network, including: A LSTM-CNN combined neural network model is established, the normalized component is input, multiple LSTM networks learn multi-scale time feature representation in parallel, and the CNN convolutional neural network convolves the output time step of each LSTM to extract its output features: The CNN convolutional neural network includes three parts: convolution layer, pooling layer and fully connected layer. The convolution layer extracts one-dimensional features. The specific extraction formula is: Where: is the feature map j of the lth layer; σ is the activation function ReLU; is the bias; F l is the number of feature maps on the l-th layer feature map f; Convolution is performed on the feature map f of the lth layer to create the feature map j in the l+1th layer; The pooling layer receives the features extracted by the convolutional layer and reduces the dimension of the features to improve robustness. The specific pooling formula is: Where: f is the convolution kernel size; s0 is the step size; x is the input, p is the number of padding layers, and when p is infinite, the maximum value is taken in the pooling area, that is, the maximum pooling; Considering the key features, multiple LSTM networks are established to learn and predict the impact of the key features on the load at the target time. The specific formula is: f t =signoid(w f x t +b f ) i t =signoid(w i x t +b i ) c t =f t c t-1 +i t tanh(w c x t +w c h t-1 +b c ) o t =signoid(w o x t +b o ) h t =o t fishy t ) Where: f t is the forget gate state, i t is the input gate state, o t is the output gate state, c t is the temporary state of the memory unit at time t generated by the tanh layer, x t is the input information of the LSTM basic unit at time t, w f 、w i 、w c 、w o is the corresponding weight matrix, b f 、b i 、b c 、b o is the corresponding bias matrix, c t is the current state; The normalized component is combined with the key features and the time module, and the prediction result of the final time series is obtained through the fully connected layer.

8. A device for handling heavy overload in a distribution station area, characterized in that: include: An acquisition module, used to obtain the original load; An extraction module, used for extracting the normalized component and the key feature; A prediction module, used for inputting the normalized component and the key feature into a prediction model to obtain a prediction result output by the prediction model; wherein the prediction model is a machine learning model based on a LSTM-CNN combined neural network; A search module, used to search for a corresponding warning level in a preset warning level table according to the prediction result; The alarm module is used to issue warning information according to the warning level.

9. An electronic device, characterized in that: include: processor; and a memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable codes stored thereon, which, when executed by a processor of an electronic device, causes the processor to execute the method according to any one of claims 1 to 7.

Images