Distribution network optimization method based on artificial intelligence

Abstract

The present invention relates to the field of distribution network power supply technology, and discloses a distribution network optimization method based on artificial intelligence. First, the attribute information of each power supply substation is periodically obtained to determine its first stability. Then, different power supply data sequences of each substation are collected and analyzed to generate first sample data. Subsequently, label data is extracted through historical operation data for subsequent model training. On this basis, the prediction model is trained using the first sample data, label data and first stability. Finally, the second sample data of each power supply substation collected in real time is input into the model to obtain the predicted stability of each substation. The method aims to improve the overall stability and reliability of the distribution network, achieve optimized scheduling, reduce energy losses, and improve flexibility by integrating data analysis and predictive modeling, so as to provide technical support for the construction of intelligent distribution networks.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution network, and in particular to a distribution network optimization method based on artificial intelligence. Background Art

[0002] In modern power systems, with the widespread access to new power sources and loads such as distributed power sources, energy storage systems and charging piles, distribution networks are facing increasingly complex challenges. Especially in substations, due to the uncertainty of single-phase source and load and the split-phase access method, the power source and load show significant inconsistency in temporal and spatial characteristics. This inconsistency further exacerbates the inherent problems of the distribution network, such as overload, underload and power quality issues.

[0003] The specific manifestations are:

[0004] 1. Heavy load and light load problem: When distributed power sources such as solar photovoltaic or wind power are connected to the distribution network, the power supply capacity of these sources and loads at different times and locations will fluctuate greatly due to changes in natural conditions. This may cause some areas to face heavy overloads, while others may be under light load, which poses a challenge to the stability and security of the power grid.

[0005] 2. Power quality issues: When electricity from different sources is connected to the grid, power quality issues such as waveform distortion, frequency fluctuation, and harmonic interference may occur. These issues not only affect the power consumption experience of end users, but may also damage equipment and cause economic losses.

[0006] 3. Demand for stable power supply: With the diversification and complexity of power demand, providing continuous and stable power supply has become the primary task that power companies must face. In addition, power companies must ensure that the power quality is within the qualified range to meet the ever-increasing power demand of various customers.

[0007] 4. Accommodation of distributed power sources: The design of the new distribution network must be able to maximize the absorption of connected distributed power sources to achieve efficient use of renewable energy and ensure energy security, which puts higher requirements on the flexibility and capacity of the power grid.

[0008] In summary, there are problems such as low demand matching accuracy, high energy loss, low flexibility, low reliability and low stability in the distribution network. Summary of the invention

[0009] In view of the shortcomings of the prior art, the present invention provides a distribution network optimization method based on artificial intelligence. The distribution network optimization method based on artificial intelligence is developed around two aspects: factors affecting the power supply quality in distribution station areas and intelligent early warning of power quality in distribution station areas. It effectively solves the contradiction between load supply and energy consumption after heavy overload, low voltage and faults in distribution networks, provides technical support for the operation and control of new distribution networks, and is of great significance to achieving demand matching, low-consumption economy, flexible and reliable high-quality power supply.

[0010] The technical solution adopted by the present invention is:

[0011] The distribution network optimization method based on artificial intelligence includes the following steps:

[0012] Step 1, periodically obtain attribute information of each power supply area;

[0013] Step 2, determining the first stability of each power supply area according to the attribute information;

[0014] Step 3, periodically obtaining multiple power supply data sequences corresponding to each power supply area, each power supply data sequence includes multiple data points, and each data point corresponds to a time point;

[0015] Step 4, determining first sample data according to the attribute information and the power supply data sequence;

[0016] Step 5, determining label data according to the historical operation data of each power supply area;

[0017] Step 6, training a prediction model according to the first sample data, the label data and the first stability;

[0018] Step 7: input the second sample data of each power supply area collected in real time into the prediction model to obtain the prediction stability of each power supply area.

[0019] Furthermore, in step 1, each power supply area corresponds to different area types. According to different load types, the areas are divided into four categories: urban residential areas, commercial areas, industrial areas and agricultural areas;

[0020] The attribute information of the power supply area includes: area type and value attribute;

[0021] The types of zones include: residential, commercial, industrial, and agricultural;

[0022] Numerical attributes include: output power, output current, output voltage, supply area, and supply population.

[0023] Further, in step 2, determining the first stability of each power supply area according to the attribute information includes:

[0024] Step 2.1, determining a first fluctuation value by determining the output power collected periodically;

[0025] The method for determining the first fluctuation value satisfies the following relationship:

[0026]

[0027] Step 2.2, determining a second fluctuation value by determining the output current collected periodically;

[0028] The method for determining the second fluctuation value satisfies the following relationship:

[0029]

[0030] Step 2.3, determining a third fluctuation value by determining the periodically collected output voltage;

[0031] The method for determining the third fluctuation value satisfies the following relationship:

[0032]

[0033] Step 2.4, respectively normalizing the supply area and the supply population corresponding to each power supply area to obtain the normalized supply area and the normalized supply population corresponding to each power supply area; determining the normalized supply area as the first weight, and determining the normalized supply population as the second weight; determining the average of the first weight and the second weight corresponding to each power supply area as the first confidence corresponding to each power supply area;

[0034] Step 2.5: for each power supply area, determine the corresponding first stability according to the first fluctuation value, the second fluctuation value, the third fluctuation value and the first confidence.

[0035] Further, determining the corresponding first stability according to the first fluctuation value, the second fluctuation value, the third fluctuation value and the first confidence level includes:

[0036]

[0037] in, Represents the first stability corresponding to the i-th power supply area;

[0038] Among them, the higher the fluctuation degree of the output power, the higher the first stability; the higher the fluctuation degree of the output current, the higher the first stability; the higher the fluctuation degree of the output voltage, the higher the first stability; the first confidence is used to characterize the credibility of the first stability.

[0039] Furthermore, in step 3, power supply data such as voltage and power are collected periodically; the power supply quality of each substation is predicted based on the power supply data sequence of each substation in each period;

[0040] Each power supply data sequence includes multiple data points, each data point corresponds to a time point, forming a variety of real-time power supply data; the multiple real-time power supply data include:

[0041] The frequency deviation index V corresponding to each time point is expressed as:

[0042]

[0043] In the formula, is the actual frequency, is the rated frequency;

[0044] Voltage deviation index corresponding to each time point , the expression is:

[0045]

[0046] In the formula, Actual voltage, is the rated voltage;

[0047] The voltage fluctuation index d corresponding to each time point is expressed as:

[0048]

[0049] In the formula, The maximum value of the RMS voltage at the common point where the equipment is connected. It is the minimum value of the RMS voltage at the common point where the equipment is connected;

[0050] Voltage flicker index corresponding to each time point , the expression is:

[0051]

[0052] In the formula, is the short-time flicker value, is the number of short-time flicker values ​​in a long time, j is the index;

[0053] The three-phase imbalance index corresponding to each time point , the expression is:

[0054]

[0055] In the formula, is the positive sequence voltage; is the negative sequence voltage;

[0056] Harmonic indicators corresponding to each time point , the expression is:

[0057]

[0058] In the formula, is the harmonic order; is the hth harmonic voltage; is the total fundamental voltage;

[0059] Voltage sag index corresponding to each time point , the expression is:

[0060]

[0061] In the formula, is the voltage sag value; The duration of the temporary drop.

[0062] Furthermore, in step 4, for each power supply area, first sample data of each power supply area is generated by concatenating the attribute information and the power supply data sequence;

[0063] The first sample data is in the form of a vector: [b,c,d',e,f,g,h,t']; wherein b represents the type of the power supply station, c represents the output power of the power supply station, d' represents the output current of the power supply station, e represents the output voltage of the power supply station, f represents the supply area of ​​the power supply station, g represents the supply population of the power supply station, h represents the power supply data sequence of the power supply station, and t' represents the timestamp.

[0064] Further, in step 5, the label data of each power supply area is determined according to the number of outages k of the equipment in the power supply area, the historical power outage frequency o of the power supply area, the average power outage duration l of the power supply area, the average power outage duration q of all users in the power supply area, and the system average power supply availability r of the power supply area; the combination of the above data forms a label vector [k, o, l, q, r];

[0065] Combining historical operation data and stability data, the stability data C is calculated by normalizing each dimension of the historical data.

[0066] Further, determining the stability data corresponding to each power supply area according to the normalized dimensional data includes:

[0067]

[0068] in, Represents the label data of the i-th power supply area; a represents other normalized dimensional data in the historical operation data except the average power supply availability; i represents the index of the power supply area; j represents the index of the dimensional data; n represents the number of other dimensions in the dimensional data except the average power supply availability.

[0069] Further, in step 6, training the prediction model includes the following steps:

[0070] Select the long short-term memory model as the initial prediction model, randomly set the initial state and initial label, the format is the same as their respective data, and randomly initialize the weights and biases, including the weights corresponding to the previous data , the deviation corresponding to the previous data , weight corresponding to the input data , the deviation corresponding to the input data , the weight corresponding to the hidden output data , the deviation corresponding to the hidden output data The value of

[0071] Step 6.1, input the first sample data into the pre-built initial prediction model; obtain the prediction label corresponding to each power supply area;

[0072] Input the first sample data at time t To the neuron corresponding to time t; according to the calculation formula, process , initial label and initial state, and get the predicted label at time t , the status data at time t ; Input the first sample data at time t+1 To the neuron corresponding to time t+1; according to the calculation formula, process , as well as , get the predicted label at time t+1 , the status data at time t+1 ; Repeat this cycle until the corresponding prediction data for each moment is obtained;

[0073] The predicted label at time t The calculation formula is:

[0074]

[0075] In the formula, is the hidden output data at time t; is the hyperbolic tangent activation function; Represents the state data at time t;

[0076] Forgotten data The calculation formula is:

[0077]

[0078] In the formula, is the Sigmoid activation function; and are the weights and biases corresponding to previous data; is the predicted label at time t-1; is the first sample data at time t;

[0079] Input Data The calculation formula is:

[0080]

[0081] In the formula, is the Sigmoid activation function; and are the weights and biases corresponding to the input data; is the predicted label at time t-1; is the first sample data at time t;

[0082] The initial state value of the neuron at time t The calculation formula is:

[0083]

[0084] In the formula, is the hyperbolic tangent activation function; and are the weights and biases corresponding to the initial state values; is the predicted label at time t-1; is the first sample data at time t;

[0085] Hidden output data at time t The calculation formula is:

[0086]

[0087] In the formula, is the Sigmoid activation function; and are the weights and biases corresponding to the hidden output data; is the predicted label at time t-1; is the first sample data at time t;

[0088] The state value of the neuron at time t The calculation formula is:

[0089]

[0090] In the formula, To forget the data; Represents the state value of the neuron at time t-1; For input data; is the initial state value of the neuron at time t;

[0091] Step 6.2, determine the loss value of the initial prediction model based on the first stability, the predicted label and the label data; the expression is:

[0092]

[0093] Wherein, Loss represents the loss value of the initial prediction model; represents the first stability of the i-th power supply area; i represents the index of the power supply area, m represents the number of power supply areas; t represents the index of the time corresponding to the first sample data at the moment, and n represents the total number of moments; is the predicted label data of the i-th power supply area at time t; The predicted data of the ith power supply area at time t;

[0094] Step 6.3: When the loss value is greater than the preset termination threshold, the parameters in the initial prediction model are updated based on the back propagation algorithm, including the initial state, initial label, and weights corresponding to previous data. , the deviation corresponding to the previous data , weight corresponding to the input data , the deviation corresponding to the input data , the weight corresponding to the hidden output data , the deviation corresponding to the hidden output data ; After updating the parameters, repeat the above process again until the loss value is less than or equal to the termination threshold, and obtain the prediction model trained to the convergence state.

[0095] Further, in step 7, the process of processing the second sample data using the prediction model includes:

[0096] Input the second sample data X at any time point in the future that needs to be predicted to the first neuron in the prediction model;

[0097] According to the calculation formulas in step 6, , initial label and initial state, and get the predicted label at that moment , the status data at that moment ;

[0098] Enter the next moment, that is, the second sample data of the second moment to the second neuron;

[0099] According to the calculation formulas in step 6, , as well as , get the predicted label at the second moment , the state data at the second moment ;

[0100] This cycle continues until the predicted label corresponding to each moment is obtained.

[0101] The beneficial effects of the present invention are:

[0102] Improve accuracy: By periodically acquiring power supply area attribute information and power supply data, the AI ​​model can accurately capture dynamic changes, thereby improving the accuracy of demand matching.

[0103] Reduce energy loss: The stability and power supply demand predicted by the model are more forward-looking, which can allocate resources more effectively, reduce unnecessary energy loss, and improve overall efficiency.

[0104] Enhanced flexibility: AI methods enable the distribution network to respond to load changes in real time and quickly adjust power supply strategies to ensure that each substation has good responsiveness under different load conditions.

[0105] Improve reliability: Accurate prediction and real-time monitoring can significantly reduce the occurrence of power failures, enhance the system's response capabilities, and ensure high reliability of the distribution network.

[0106] Enhanced stability: Through data-driven analysis, potential stability issues can be identified in advance, and preventive measures can be taken to ensure the safe operation of the distribution network.

[0107] Optimize decision-making: The predictive ability of AI models provides strong support for operational decisions, helping managers to better plan and schedule and improve management efficiency.

[0108] Overall, this method can effectively improve the operational efficiency of the distribution network, reduce costs, and lay the foundation for the development of future smart grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] In order to more clearly illustrate the embodiments of the present application 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 recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0110] Figure 1 A flow chart of a distribution network optimization method based on artificial intelligence used in the present invention;

[0111] Figure 2 A structural diagram of each neuron in the prediction model of the present invention;

[0112] Figure 3 This is a model structure diagram of the prediction model of the present invention. DETAILED DESCRIPTION

[0113] To make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be described clearly and completely in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0114] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely represents some embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should be noted that the embodiments of the present invention and the features and technical solutions in the embodiments can be combined with each other without conflict. It should be noted that similar numbers and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0115] In order to solve the problems of low demand matching accuracy, high energy loss, low flexibility, low reliability and low stability of the distribution network, the present embodiment provides a distribution network optimization method based on artificial intelligence; the distribution network optimization method based on artificial intelligence first analyzes the factors affecting the power supply quality of the distribution station area, including: analysis of the power supply and consumption characteristics of the distribution station area; analysis of the laws affecting the power supply quality of the distribution station area; research on intelligent early warning technology for power quality in the distribution station area; secondly, analysis of layered and phased expansion of the distribution network with complex source and load access is carried out, including: analysis of the characteristics of layered sharing and expansion demand; layered and phased expansion and load supply guarantee topology design; and layered and phased expansion technology of the substation area.

[0116] like Figure 1 As shown, the specific steps include:

[0117] Step 1: Periodically obtain the attribute information of each power supply area.

[0118] The power supply area is also called the transformer power supply area, which refers to the power supply range or area covered by a transformer. It is a basic unit in the power system, responsible for converting high-voltage electric energy into low-voltage electric energy and distributing it to power users in the area. The multiple areas covered by the distribution network correspond to different power supply areas. Among them, each power supply area corresponds to different types of areas. For example, according to the different load types, the areas can be divided into four categories: urban residential areas, commercial areas, industrial areas, and agricultural areas. These different types of areas differ in terms of power consumption characteristics, load distribution, and power supply needs. In addition, different areas correspond to different types of renewable energy, so different areas correspond to different properties of renewable energy.

[0119] In this embodiment, the attribute information of the power supply area includes:

[0120] District types: residential, commercial, industrial, agricultural.

[0121] Numerical attributes: output power, output current, output voltage, supply area, supply population.

[0122] Step 2: determine the first stability of each power supply area according to the attribute information.

[0123] Different substations have different corresponding attribute information, resulting in different power generation stability in different substations.

[0124] The determining of the first stability of each power supply area according to the attribute information comprises:

[0125] Step 2.1, determining a first fluctuation value by determining the output power collected periodically;

[0126] Specifically, the method for determining the first fluctuation value satisfies the following relationship:

[0127]

[0128] Step 2.2, determining a second fluctuation value by determining the output current collected periodically;

[0129] Specifically, the method for determining the second fluctuation value satisfies the following relationship:

[0130]

[0131] Step 2.3, determining a third fluctuation value by determining the periodically collected output voltage;

[0132] Specifically, the method for determining the third fluctuation value satisfies the following relationship:

[0133]

[0134] Step 2.4, respectively normalizing the supply area and the supply population corresponding to each power supply area to obtain the normalized supply area and the normalized supply population corresponding to each power supply area; determining the normalized supply area as the first weight, and determining the normalized supply population as the second weight; determining the average of the first weight and the second weight corresponding to each power supply area as the first confidence corresponding to each power supply area;

[0135] Step 2.5: for each power supply area, determine the corresponding first stability according to the first fluctuation value, the second fluctuation value, the third fluctuation value and the first confidence.

[0136] Specifically, determining the corresponding first stability according to the first fluctuation value, the second fluctuation value, the third fluctuation value, and the first confidence level includes:

[0137]

[0138] in, Represents the first stability corresponding to the i-th power supply area.

[0139] The higher the fluctuation degree of the output power, the higher the first stability; the higher the fluctuation degree of the output current, the higher the first stability; the higher the fluctuation degree of the output voltage, the higher the first stability. The first confidence level is used to characterize the credibility of the first stability.

[0140] Step 3: periodically obtain multiple power supply data sequences corresponding to each power supply area, each power supply data sequence includes multiple data points, and each data point corresponds to a time point.

[0141] In this embodiment, in order to analyze the power supply quality of the power supply area, power supply data such as voltage and power can be periodically collected; the power supply quality of each area is predicted based on the power supply data sequence of each area in each period.

[0142] In this embodiment, the various real-time power supply data include:

[0143] (1) The frequency deviation index V corresponding to each time point is defined as the difference between the actual frequency and the rated frequency;

[0144] The expression of V is:

[0145]

[0146] In the formula, is the actual frequency, is the rated frequency, The unit is Hz.

[0147] (2) Voltage deviation index corresponding to each time point , which is defined as the ratio of the difference between the actual voltage and the rated voltage to the rated voltage;

[0148] The expression is:

[0149]

[0150] In the formula, Actual voltage, is the rated voltage.

[0151] (3) The voltage fluctuation index d corresponding to each time point is defined as the maximum value of the RMS voltage at the common point where the equipment is connected and minimum value The ratio of the difference to the rated voltage of the grid;

[0152] The expression of d is:

[0153]

[0154] (4) Voltage flicker index corresponding to each time point , also known as long-term flicker value; voltage flicker can be expressed as the degree of harm caused by voltage fluctuations;

[0155] The expression is:

[0156]

[0157] In the formula, is the short-time flicker value, is the number of short-time flicker values ​​in a long period of time, and j is the index.

[0158] (5) Three-phase imbalance index corresponding to each time point , the connection of three-phase asymmetric load will cause the three-phase power supply voltage to be unbalanced, resulting in negative sequence components, which will cause harm to the motor and transformer;

[0159] The expression is:

[0160]

[0161] In the formula, is the positive sequence voltage; is the negative sequence voltage.

[0162] (6) Harmonic indicators corresponding to each time point , which is also called harmonic distortion rate; nonlinear load connected to the power supply system will induce fundamental voltage and current distortion, generate a large number of harmonics, and affect safe operation;

[0163] The expression is:

[0164]

[0165] In the formula, is the harmonic order; is the hth harmonic voltage; is the total fundamental voltage.

[0166] (7) Voltage sag index corresponding to each time point , which is defined as the degree of voltage sag caused by a sudden drop in voltage within a short period of time;

[0167] The expression is:

[0168]

[0169] In the formula, is the voltage sag value, which indicates the ratio of voltage during sag; It is the duration of voltage sag, which indicates the duration of voltage sag.

[0170] Step 4, determining first sample data according to the attribute information and the power supply data sequence;

[0171] In this embodiment, for each power supply area, the attribute information and the power supply data sequence are spliced ​​in a preset order to obtain the first sample data corresponding to each power supply area. Exemplarily, the first sample data may be [area type, output power, output current, output voltage, supply area, supply population, power supply data sequence].

[0172] For any power supply station, the first sample data of the power supply station can be in the form of a vector, for example: [b, c, d', e, f, g, h, t']; wherein b represents the station type of the power supply station, c represents the output power of the power supply station, d' represents the output current of the power supply station, e represents the output voltage of the power supply station, f represents the supply area of ​​the power supply station, g represents the supply population of the power supply station, h represents the power supply data sequence of the power supply station, and t' represents the timestamp.

[0173] Step 5: determine the label data according to the historical operation data of each power supply area.

[0174] In this embodiment, the label data of each power supply area can be determined according to the following data; wherein the label data is used to quantitatively characterize the stability of each power supply area during operation. Exemplarily, the historical operation data includes: the number of outages of equipment in the power supply area, the historical power outage frequency, the average power outage duration, the average power outage duration of all users in each power supply area, and the average power supply availability of the system.

[0175] Among them, the higher the number of equipment outages in the power supply area, the lower the stability of the power supply area; the higher the historical power outage frequency in the power supply area, the lower the stability of the power supply area; the higher the average power outage duration, the lower the stability; the longer the average power outage duration of all users, the lower the stability; the higher the average power supply availability, the higher the stability.

[0176] Subsequently, the stability of the power supply station operation can be evaluated based on the stability predicted at any time. When the predicted stability needs to be adjusted, the closest stability data is checked, and then the corresponding power supply sequence data is determined, and then the nearest normal power supply sequence data is determined as the confirmed power supply sequence data.

[0177] For any power supply area, the label data of the power supply area can be in the form of a vector, for example: [k, o, l, q, r]. Among them, k represents the number of outages of equipment in the power supply area, o represents the historical power outage frequency of the power supply area, l represents the average power outage duration of the power supply area, q represents the average power outage duration of all users in the power supply area, and r represents the average power supply availability of the system in the power supply area.

[0178] In this embodiment, 80% of the training data may be used as a training set, and 20% of the training data may be used as a training set.

[0179] Another method for determining label data is to determine the stability data of the power supply area according to the historical operation data of the power supply area, and use the historical operation data and the stability data as label data. Determining the stability data according to the historical operation data of each power supply area includes: determining multiple dimensional data in the historical operation data corresponding to each power supply area; normalizing each dimensional data respectively to obtain normalized dimensional data;

[0180] Determining stability data corresponding to each power supply area according to the normalized dimensional data includes:

[0181]

[0182] in, Represents the label data of the i-th power supply area; a represents other normalized dimensional data in the historical operation data except the average power supply availability; i represents the index of the power supply area; j represents the index of the dimensional data; n represents the number of other dimensions in the dimensional data except the average power supply availability.

[0183] In this case, for any power supply station, the label data of the power supply station can be in the form of a vector, for example: [k, o, l, q, r, C]; where C represents the stability data of the power supply station determined according to the normalized dimensional data.

[0184] Step 6: training a prediction model based on the first sample data, the label data, and the first stability.

[0185] In this embodiment, training the prediction model includes: inputting first sample data into a pre-constructed initial prediction model; obtaining a prediction label corresponding to each power supply station area; determining a loss value of the initial prediction model based on the first stability, the prediction label and the label data; when the loss value is greater than a preset termination threshold, updating the prediction generation model based on a back propagation algorithm until the loss value is less than or equal to the termination threshold, thereby obtaining a prediction model trained to a convergence state.

[0186] The initial prediction model can be a long short-term memory model, which includes a plurality of neurons connected in series, and the structure of each neuron is as follows: Figure 2 As shown, the model structure is as Figure 3 As shown;

[0187] in, is the first sample data input at time t;

[0188] Represents the predicted label at time t, and Jointly obtain the predicted label at time t The expression is:

[0189]

[0190] In the formula, is the hidden output data at time t; is the hyperbolic tangent activation function; Represents the state value of the neuron at time t.

[0191] is the forgotten data, with a value range of 0 to 1. and The operation is performed and the sigmoid function is used to obtain a value between 0 and 1. The closer the forgotten data is to 0, the more predicted labels predicted by the first sample data at the previous moment need to be forgotten. The closer the forgotten data is to 1, the fewer predicted labels predicted by the first sample data at the previous moment need to be forgotten. The expression is:

[0192]

[0193] In the formula, is the Sigmoid activation function; and are the weights and biases corresponding to previous data; is the predicted label at time t-1; is the first sample data at time t.

[0194] is the input data, through and The operation is performed and processed by the sigmoid function to obtain a value between 0 and 1; input data The expression is:

[0195]

[0196] In the formula, is the Sigmoid activation function; and are the weights and biases corresponding to the input data; is the predicted label at time t-1; is the first sample data at time t.

[0197] is the initial state value of the neuron at time t, expressed as:

[0198]

[0199] In the formula, is the hyperbolic tangent activation function; and are the weights and biases corresponding to the initial state values; is the predicted label at time t-1; is the first sample data at time t.

[0200] is the hidden output data at time t, through and , and processed by the sigmoid function to obtain a value between 0 and 1; hide the output data The expression is:

[0201]

[0202] In the formula, is the Sigmoid activation function; and are the weights and biases corresponding to the hidden output data; is the predicted label at time t-1; is the first sample data at time t.

[0203] Represents the state value of the neuron at time t, expressed as:

[0204]

[0205] In the formula, To forget the data; Represents the state value of the neuron at time t-1; For input data; is the initial state value of the neuron at time t.

[0206] In this embodiment, the loss value of the initial prediction model is determined according to the first stability, the prediction label and the label data, and the expression is:

[0207]

[0208] Wherein, Loss represents the loss value of the initial prediction model; represents the first stability of the i-th power supply area; i represents the index of the power supply area, m represents the number of power supply areas; t represents the index of the time corresponding to the first sample data at the moment, and n represents the total number of moments; is the predicted label data of the i-th power supply area at time t; The predicted data of the i-th power supply area at time t.

[0209] Furthermore, the specific process of training the initial prediction model is as follows:

[0210] Preset the initial state and initial label, where the format of the initial state is the same as the state data and the value is random, and the format of the initial label is the same as the predicted label and the value is random;

[0211] Pre-initialize and randomly set the weights corresponding to previous data , the deviation corresponding to the previous data , weight corresponding to the input data , the deviation corresponding to the input data , the weight corresponding to the hidden output data , the deviation corresponding to the hidden output data The value of

[0212] Input the first sample data at time t The neuron corresponding to time t;

[0213] According to the above calculation formulas, , initial label and initial state, and get the predicted label at time t , the status data at time t ;

[0214] Input the first sample data at time t+1 The neuron corresponding to time t+1;

[0215] According to the above calculation formulas, , as well as , get the predicted label at time t+1 , the status data at time t+1 ;

[0216] This cycle continues until the corresponding prediction data for each moment is obtained.

[0217] The loss value of the initial prediction model is calculated according to the above loss function. When the loss value is greater than the termination threshold, it indicates that the difference between the predicted label and the label data predicted by the initial prediction model is large, indicating that the performance of the initial prediction model is poor and the accuracy of the predicted label is low. Therefore, the back propagation algorithm can be used to update the parameters in the initial prediction model, including the initial state, initial label, and weights corresponding to previous data. , the deviation corresponding to the previous data , weight corresponding to the input data , the deviation corresponding to the input data , the weight corresponding to the hidden output data , the deviation corresponding to the hidden output data .

[0218] After updating the parameters, the above process is repeated again until the loss value is less than or equal to the termination threshold, and the update is stopped to obtain a prediction model trained to a converged state.

[0219] Step 7: input the second sample data of each power supply area collected in real time into the prediction model to obtain the prediction stability of each power supply area.

[0220] For any power supply area, the second sample data of the power supply area is in the same format as the first sample data, and can be in the form of a vector, for example: [b, c, d', e, f, g, h, t']. b represents the type of the power supply area, c represents the output power of the power supply area, d' represents the output current of the power supply area, e represents the output voltage of the power supply area, f represents the supply area of ​​the power supply area, g represents the supply population of the power supply area, h represents the power supply data sequence of the power supply area, and t' represents the timestamp.

[0221] The process of processing the second sample data using the prediction model includes:

[0222] Input a second sample data X at any moment, that is, at any moment in the future that needs to be predicted, to the first neuron in the prediction model;

[0223] According to the calculation formulas in step 6, , initial label and initial state, and get the predicted label at that moment , the status data at that moment ;

[0224] Enter the next moment, that is, the second sample data of the second moment to the second neuron;

[0225] According to the calculation formulas in step 6, , as well as , get the predicted label at the second moment , the state data at the second moment ;

[0226] This cycle continues until the predicted label corresponding to each moment is obtained.

[0227] At the same time, it should be pointed out that the above description is only a typical embodiment of the present invention and does not limit the present invention in any form. Any changes made to the above embodiments based on the ideas and technical essence of the present invention are within the protection scope of the present invention.

Claims

1. A distribution network optimization method based on artificial intelligence, characterized in that: The following steps are involved: Step 1, periodically obtain attribute information of each power supply area; Step 2, determining the first stability of each power supply area according to the attribute information; Step 3, periodically obtaining multiple power supply data sequences corresponding to each power supply area, each power supply data sequence includes multiple data points, and each data point corresponds to a time point; Step 4, determining first sample data according to the attribute information and the power supply data sequence; Step 5, determining label data according to the historical operation data of each power supply area; Step 6, training a prediction model according to the first sample data, the label data and the first stability; Step 7, inputting the second sample data of each power supply area collected in real time into the prediction model to obtain the prediction stability of each power supply area; In step 1, each power supply area corresponds to different area types. According to different load types, the areas are divided into four categories: urban residential areas, commercial areas, industrial areas and agricultural areas; The attribute information of the power supply area includes: area type and value attribute; The types of zones include: residential, commercial, industrial, and agricultural; Numerical attributes include: output power, output current, output voltage, supply area, supply population; In step 2, determining the first stability of each power supply area according to the attribute information includes: Step 2.1, determining a first fluctuation value by determining the output power collected periodically; The method for determining the first fluctuation value satisfies the following relationship: Step 2.2, determining a second fluctuation value by determining the output current collected periodically; The method for determining the second fluctuation value satisfies the following relationship: Step 2.3, determining a third fluctuation value by determining the periodically collected output voltage; The method for determining the third fluctuation value satisfies the following relationship: Step 2.4, respectively normalizing the supply area and the supply population corresponding to each power supply area to obtain the normalized supply area and the normalized supply population corresponding to each power supply area; determining the normalized supply area as the first weight, and determining the normalized supply population as the second weight; determining the average of the first weight and the second weight corresponding to each power supply area as the first confidence corresponding to each power supply area; Step 2.5, for each power supply area, determining a corresponding first stability according to the first fluctuation value, the second fluctuation value, the third fluctuation value and the first confidence level; Determining the corresponding first stability according to the first fluctuation value, the second fluctuation value, the third fluctuation value and the first confidence level includes: Among them, A i Represents the first stability corresponding to the i-th power supply area; Among them, the higher the fluctuation degree of the output power, the higher the first stability; the higher the fluctuation degree of the output current, the higher the first stability; the higher the fluctuation degree of the output voltage, the higher the first stability; the first confidence is used to characterize the credibility of the first stability.

2. The distribution network optimization method based on artificial intelligence according to claim 1 is characterized in that: In step 3, power supply data voltage and power data are collected periodically; the power supply quality of each substation is predicted based on the power supply data sequence of each substation in each period; Each power supply data sequence includes multiple data points, each data point corresponds to a time point, forming a variety of real-time power supply data; A variety of real-time power supply data including: The frequency deviation index V corresponding to each time point is expressed as: V=Δf=|f-f0| In the formula, f is the actual frequency and f0 is the rated frequency; The voltage deviation index δU corresponding to each time point is expressed as: Where, U is the actual voltage, U is the N is the rated voltage; The voltage fluctuation index d corresponding to each time point is expressed as: Where U max The maximum value of the RMS voltage at the common point where the equipment is connected, U min It is the minimum value of the RMS voltage at the common point where the equipment is connected; The voltage flicker index P corresponding to each time point is expressed as: Where P S is the short-time flicker value, N is the number of short-time flicker values ​​in a long time, and j is the index; The three-phase unbalance index ε corresponding to each time point is expressed as: Where U + is the positive sequence voltage; U - is the negative sequence voltage; The harmonic index D corresponding to each time point TH , the expression is: Where M is the harmonic order; U h is the hth harmonic voltage; U1 is the total fundamental voltage; The voltage sag index E corresponding to each time point is expressed as: E=(1-V sag ) 2 T Where V sag is the voltage sag value; T is the sag duration.

3. The distribution network optimization method based on artificial intelligence according to claim 1 is characterized in that: In step 4, for each power supply area, first sample data of each power supply area is generated by concatenating the attribute information and the power supply data sequence; The first sample data is in the form of a vector: [b, c, d, e, f, g, h, t']; wherein b represents the type of the power supply station, c represents the output power of the power supply station, d represents the output current of the power supply station, e represents the output voltage of the power supply station, f represents the supply area of ​​the power supply station, g represents the supply population of the power supply station, h represents the power supply data sequence of the power supply station, and t' represents the timestamp.

4. The distribution network optimization method based on artificial intelligence according to claim 1 is characterized in that: In step 5, the label data of each power supply area is determined according to the number of outages k of the equipment in the power supply area, the historical power outage frequency o of the power supply area, the average power outage duration p of the power supply area, the average power outage duration q of all users in the power supply area, and the system average power supply availability e of the power supply area; the combination of the above data forms a label vector [k, o, p, q, e]; Combining historical operation data and stability data, the stability data C is calculated by normalizing each dimension of the historical data.

5. The distribution network optimization method based on artificial intelligence according to claim 4 is characterized in that: Determining stability data corresponding to each power supply area according to the normalized dimensional data includes: Among them, C i Represents the label data of the i-th power supply area; a represents other normalized dimensional data in the historical operation data except the average power supply availability; i represents the index of the power supply area; j represents the index of the dimensional data; n represents the number of other dimensions in the dimensional data except the average power supply availability.

6. The distribution network optimization method based on artificial intelligence according to claim 1, characterized in that: In step 6, training the prediction model includes the following steps: Select the long short-term memory model as the initial prediction model, randomly set the initial state and initial label, the format is the same as their respective data, and randomly initialize the weights and biases, including the weights W corresponding to the previous data f , the deviation b corresponding to the previous data f , the weight W corresponding to the input data i , the deviation b corresponding to the input data i , the weight W corresponding to the hidden output data o , the deviation b corresponding to the hidden output data o The value of Step 6.1, input the first sample data into the pre-built initial prediction model; obtain the prediction label corresponding to each power supply area; Input the first sample data X at time t t To the neuron corresponding to time t; according to the calculation formula, process X t , initial label and initial state, and get the predicted label h at time t t , the state data C at time t t ; Input the first sample data X at time t+1 t+1 To the neuron corresponding to time t+1; according to the calculation formula, process X t+1 、h t and C t , get the predicted label h at time t+1 t+1 , the state data C at time t+1 t+1 ; Repeat this cycle until the corresponding prediction data for each moment is obtained; The predicted label h at time t t The calculation formula is: h t =o t *tanh(C t ) In the formula, o t is the hidden output data at time t; tanh is the hyperbolic tangent activation function; C t Represents the state data at time t; Forget data f t The calculation formula is: f t =σ(W f ·[h t-1 ,X t ]+b f ) Where σ is the Sigmoid activation function; W f and b f are the weights and biases corresponding to previous data; h t-1 is the predicted label at time t-1; X t is the first sample data at time t; Input data i t The calculation formula is: i t =σ(W i ·[h t-1 ,X t ]+b i ) Where σ is the Sigmoid activation function; W i and b i are the weights and biases corresponding to the input data; h t-1 is the predicted label at time t-1; X t is the first sample data at time t; The initial state value of the neuron at time t is C t The calculation formula is: C t =tanh(W c ·[h t-1 ,X t ]+b j ) Where tanh is the hyperbolic tangent activation function; W c and b j are the weights and biases corresponding to the initial state values; h t-1 is the predicted label at time t-1; X t is the first sample data at time t; Hidden output data o at time t t The calculation formula is: the t =σ(W o ·[h t-1 ,X t ]+b o ) Where σ is the Sigmoid activation function; W o and b o are the weights and biases corresponding to the hidden output data; h t-1 is the predicted label at time t-1; X t is the first sample data at time t; The state value of the neuron at time t is C t The calculation formula is: C t =f t *C t-1 +i t *C t In the formula, f t Forgotten data; C t-1 represents the state value of the neuron at time t-1; i t is the input data; C t is the initial state value of the neuron at time t; Step 6.2, determine the loss value of the initial prediction model based on the first stability, the predicted label and the label data; the expression is: Wherein, Loss represents the loss value of the initial prediction model; A i represents the first stability of the i-th power supply area; i represents the index of the power supply area, m represents the number of power supply areas; t represents the index of the time corresponding to the first sample data at the moment, and n represents the total number of moments; label data i,t is the predicted label data of the ith power supply area at time t; predicted label i,t The predicted data of the ith power supply area at time t; Step 6.3: When the loss value is greater than the preset termination threshold, the parameters in the initial prediction model are updated based on the back propagation algorithm, including the initial state, initial label, and weight W corresponding to the previous data. f , the deviation b corresponding to the previous data f , the weight W corresponding to the input data i , the deviation b corresponding to the input data i , the weight W corresponding to the hidden output data o , the deviation b corresponding to the hidden output data o ; After updating the parameters, repeat the above process again until the loss value is less than or equal to the termination threshold, and obtain the prediction model trained to the convergence state.

7. The distribution network optimization method based on artificial intelligence according to claim 6 is characterized in that: In step 7, the process of processing the second sample data using the prediction model includes: Input the second sample data X at any time point in the future that needs to be predicted to the first neuron in the prediction model; According to the calculation formulas in step 6, process X1, the initial label and the initial state to obtain the predicted label h1 and the state data C1 at that moment; Input the second sample data X2 at the next moment, i.e., the second moment, to the second neuron; According to the calculation formulas in step 6, X2, h1 and C1 are processed to obtain the predicted label h2 at the second moment and the state data C2 at the second moment; This cycle continues until the predicted label corresponding to each moment is obtained.

Images