Resource scheduling method based on big data

Abstract

The invention relates to the technical field of power resource scheduling, and discloses a resource scheduling method based on big data, comprising the following steps: collecting historical power consumption data of all plots and historical environmental data of the location of a wind generating set; inputting the environmental characteristics into a generating capacity prediction model; inputting the comprehensive characteristics into a power dispatching strategy network, and outputting a power dispatching strategy; distributing coal-fired power generation to each land parcel according to the output power dispatching strategy; according to the method, the generating capacity generated by the wind generating set in the future is predicted through the generating capacity prediction model, the missing power supply capacity of each land parcel of the power supply area is obtained based on the predicted generating capacity, and then the power dispatching strategy is generated through the power dispatching strategy network to make up the power generated by insufficient wind power generation in the vacuum period in advance. The time of a vacuum period is effectively shortened, and the influence of unstable wind power generation and power supply on a power supply area is reduced.

Description

Technical Field

[0001] The present invention relates to the field of power resource scheduling, and more specifically, to a resource scheduling method based on big data. Background Art

[0002] At present, the power supply in some areas is provided by a combination of wind power generation and coal-fired power generation. When wind power generation and coal-fired power generation form a power supply network to provide a stable power supply for the entire region, due to the intermittent nature of wind turbines, when the wind speed is moderate or strong, wind power generation will continue to be high-yielding, but when the wind speed is weak or even stagnant, the wind power supply will be reduced or even interrupted, resulting in the need for dynamic scheduling of coal-fired power generation to make up for the insufficient supply of wind power generation or reduce its own power supply. However, dynamic scheduling can only be carried out after the wind power generation is insufficient or excessive, resulting in this result, and then the power supply of coal-fired power generation can be adjusted. This will easily cause a vacuum period between the insufficient wind power generation and the scheduling, resulting in unstable power supply in the area during the vacuum period, affecting the life of the area. Summary of the invention

[0003] The present invention provides a resource scheduling method based on big data to solve the technical problems in the above-mentioned background technology.

[0004] The present invention provides a resource scheduling method based on big data, comprising the following steps:

[0005] Step 101, gridding the power supply area into multiple plots, collecting historical power consumption data of all plots and historical environmental data of the locations where the wind turbines are located; the power consumption data includes peak power consumption, valley power consumption and average power consumption, and the historical environmental data of the locations where the wind turbines are located includes wind speed, wind direction, temperature, humidity and air pressure data of the locations where the wind turbines are located;

[0006] Step 102, encoding the historical environmental data of the location of the wind turbine generator set into environmental features, inputting the environmental features into a power generation prediction model, and the power generation prediction model outputs the wind power generation provided by the wind turbine generator set for the power supply area in the next 6 hours;

[0007] Step 103, generating the electricity consumption characteristics of each plot through the electricity consumption data of the plot, dividing the wind power generation according to the number of plots to generate the wind power generation, splicing the wind power generation with the electricity consumption characteristics of each plot to form a comprehensive characteristic, inputting the comprehensive characteristic into the power dispatching strategy network, and outputting the power dispatching strategy, wherein one power dispatching strategy includes the corresponding coal-fired power generation output by the coal-fired power plant to all plots;

[0008] Step 104, allocating coal-fired power generation to each plot according to the output power dispatching strategy.

[0009] In a preferred embodiment, the power generation prediction model includes an input layer, an LSTM layer and a fully connected layer. The input layer inputs environmental features to the LSTM layer, the LSTM layer outputs the hidden state of the current time step to the fully connected layer, and the fully connected layer outputs the wind power generation provided by the wind turbine to the power supply area in the next 6 hours.

[0010] In a preferred embodiment, the input layer:

[0011] The dimension of the input environment feature is (N, T, D), where N represents the number of samples, T represents the time step, and D represents the number of features;

[0012] LSTM Layer:

[0013] The calculation formula of the input gate is:

[0014] i t =σ(W i ·[h t-1 , x t ]+b i )

[0015] Among them, i t represents the input gate, σ represents the sigmoid activation function, W i represents the second weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b i represents the second bias vector;

[0016] The calculation formula of the forget gate is:

[0017] f t =σ(W f ·[h t-1 , x t ]+b f )

[0018] Among them, f t represents the forget gate, σ represents the sigmoid activation function, W f represents the third weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b f represents the third bias vector;

[0019] The calculation formula of the output gate is:

[0020] o t =σ(W o ·[h t-1 , x t ]+bo )

[0021] Among them, t represents the output gate, σ represents the sigmoid activation function, W o represents the fourth weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b o represents the fourth bias vector;

[0022] The update formula of the memory unit is:

[0023] C t =f t *C t-1 +i t *tanh(W C ·[h t-1 , x t ]+b C )

[0024] Among them, C t Represents the memory cell state at the current time step, C t-1 represents the state of the memory cell at the previous time step, f t represents the forget gate, i t represents the input gate, tanh represents the hyperbolic tangent activation function, and h t-1 represents the hidden state of the previous time step, x t Indicates the current input, W C represents the fifth weight matrix, b C represents the fifth bias vector;

[0025] The update formula of the hidden state is:

[0026] h t =o t *tanh(C t )

[0027] Among them, h t represents the hidden state of the current time step, o t represents the input gate, C t Represents the state of the memory unit at the current time step, and tanh represents the hyperbolic tangent activation function;

[0028] Fully connected layer:

[0029] The calculation formula of the fully connected layer is:

[0030] y=RELU(W y ·h t +b y )

[0031] Where y represents the predicted wind power generation provided by the wind turbine to the power supply area in the next 6 hours, W y represents the weight matrix of the fully connected layer, h t represents the hidden state of the current time step, b y Represents the bias term of the fully connected layer, and RELU represents the RELU function.

[0032] In a preferred embodiment, the power dispatching strategy network includes a feature splicing layer, a hidden layer and an output layer, wherein the feature splicing layer inputs power consumption characteristics and wind power generation, outputs comprehensive characteristics, inputs the comprehensive characteristics into the hidden layer, outputs hidden characteristics, and inputs the hidden characteristics into the output layer. The output layer outputs the Q value of each optional power dispatching strategy, and selects the power dispatching strategy with the largest Q value as the output.

[0033] In a preferred embodiment, the calculation formula of the feature splicing layer is:

[0034] Z=CONCAT(Y D , y*)

[0035] Among them, Z represents the comprehensive characteristics, Y D represents the electricity consumption characteristics, y* represents the averaged wind power generation vector, and CONCAT represents vector concatenation.

[0036] In a preferred embodiment, the calculation formula of the hidden layer is:

[0037] H=f(W H Z+b H )

[0038] Among them, H represents hidden features, Z represents comprehensive features, and W H represents the weight matrix of the hidden layer, b H represents the bias term of the hidden layer, and f represents the tanh activation function;

[0039] The calculation formula of the output layer is:

[0040] Q pred =σ(W o *H+b o )

[0041] Among them, Q pred represents the Q value of the power dispatch strategy corresponding to the power supply area, H represents the hidden feature, and W o represents the weight matrix of the output layer, b o represents the bias term of the output layer, and σ represents the sigmoid activation function.

[0042] In a preferred embodiment, the state space of the power dispatch strategy network includes environmental data and power consumption data;

[0043] Action space: The exportable coal-fired power generation of each plot is combined to generate the action space;

[0044] One action includes the amount of coal-fired generation added or subtracted across all plots;

[0045] The calculation formula of the reward function of the power dispatch strategy network is:

[0046] R=-α·S-β·D-γ·T+C

[0047] Among them, R represents the total reward, S represents the power supply stability reward value, D represents the power demand satisfaction reward value, T represents the processing efficiency reward value, α, β, γ represent the first, second, and third weight coefficients respectively, and C represents the bias term;

[0048] The calculation formula for power supply stability reward is:

[0049]

[0050] Among them, S represents the power supply stability reward value, N represents the number of plots, and P demand，i and P supply，i They represent the electricity demand and supply of the ith plot respectively;

[0051] The calculation formula for the bonus value of meeting electricity demand is:

[0052]

[0053] Where D represents the reward value for meeting electricity demand, N represents the number of plots, and P demand，i and P supply，i Represent the electricity demand and supply of the i-th plot respectively. When P demand，i <P supply，i When the power supply is insufficient, the reward is negative. demand，i >P supply，i When , the electricity demand is met and the reward is zero or positive;

[0054] The calculation formula for the processing efficiency bonus value is:

[0055]

[0056] Among them, T represents the processing efficiency bonus value, N represents the number of plots, and A t，i , A t-1，i They represent the adjustment of coal-fired power generation for the i-th plot at the current and previous moments respectively.

[0057] In a preferred embodiment, the training of the power dispatch strategy network includes the following steps:

[0058] Step 1: Initialize the experience pool, randomly initialize the power dispatch strategy network, and initialize the targetQ network, whose parameters are the same as those of the power dispatch strategy network;

[0059] Step 2: reset the environment and obtain the first state;

[0060] Step 3: Generate an action using the ∈-greedy strategy: with a probability of ∈, an action will be randomly selected, which is the exploration mode; otherwise, a t =max a Q(s t , a; θ), choose s t In the state The biggest action is the experience mode. Through the power dispatch strategy network calculation;

[0061] According to the interaction between the action and the environment, the feedback R is obtained t , the next state s t+1 and whether the termination condition done is triggered, the termination condition is the end of the current water fee settlement cycle;

[0062] Experience t , a t , Rt ， s t+1 , done is stored in the experience pool, an experience is a four-tuple s t , a t , R t ,s t+1 or t， a t , R t , done;

[0063] Step 4: Get an experience randomly from the experience pool;

[0064]

[0065] max a Q target (s t+1 , a; θ) represents the state s t+1 The maximum output of the targetQ network;

[0066] according to and Calculate the loss and use the gradient descent method to update the drip control strategy network;

[0067]

[0068] Indicates state st and action a t The output Q value of the power dispatch strategy network is as follows;

[0069] γ is a discount factor used to balance the weights of immediate rewards and future rewards, with a default value of 0.6;

[0070] Step 5: Update the targetQ network every fixed number of network updates to make its parameters the same as those of the current drip control strategy network;

[0071] Step 6: The step is terminated until the Q network converges or the number of updates reaches the set value.

[0072] A resource scheduling system based on big data, including the following modules:

[0073] The data collection module collects the historical electricity consumption data of all plots and the historical environmental data of the locations of wind turbines;

[0074] The power generation prediction module is used to output the wind power generation provided by the wind turbines to the power supply area in the next 6 hours;

[0075] A feature generation module, used to encode the historical environmental data of the location of the wind turbine generator set into environmental features and generate the power consumption features of each plot of land through the power consumption data of the plot of land;

[0076] The comprehensive feature splicing module splices the wind power generation with the electricity consumption characteristics of each plot to form a comprehensive feature;

[0077] The power dispatching strategy network module is used to input the comprehensive features into the power dispatching strategy network and output the power dispatching strategy;

[0078] The resource allocation execution module is used to execute the power dispatching strategy output by the power dispatching strategy network.

[0079] A storage medium stores non-temporary computer-readable instructions. When the non-temporary computer-readable instructions are executed by a computer, the steps in the above-mentioned resource scheduling method based on big data can be executed.

[0080] The beneficial effects of the present invention are as follows: the present invention first predicts the amount of power that a wind turbine can generate in the future through a power generation prediction model, obtains the missing power supply for each plot in the power supply area based on the predicted power generation, and then generates a power dispatching strategy through a power dispatching strategy network to pre-compensate for the amount of power during the vacuum period caused by insufficient wind power generation, thereby effectively shortening the time of the vacuum period and reducing the impact of unstable wind power supply on the power supply area. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1It is a flow chart of the resource scheduling method based on big data of the present invention.

[0082] Figure 2 It is a block diagram of the resource scheduling system based on big data of the present invention. DETAILED DESCRIPTION

[0083] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0084] like Figure 1 As shown, a resource scheduling method based on big data includes the following steps:

[0085] Step 101, gridding the power supply area into multiple plots, collecting historical power consumption data of all plots and historical environmental data of the locations where the wind turbines are located; the power consumption data includes peak power consumption, valley power consumption and average power consumption, and the historical environmental data of the locations where the wind turbines are located includes wind speed, wind direction, temperature, humidity and air pressure data of the locations where the wind turbines are located;

[0086] The acquisition of electricity consumption data for all plots is done by using the power company's recording system to collect electricity consumption data for each plot. The electricity consumption data includes peak, valley and average electricity consumption in different time periods in order to analyze electricity consumption patterns and predict demand;

[0087] The environmental data of the wind turbine location is obtained by setting up a weather station at the wind turbine location or using an existing weather monitoring system to collect real-time and historical data such as wind speed, wind direction, temperature, humidity and air pressure;

[0088] It should be noted that the power supply area is divided into multiple smaller plots by using a geographic information system (GIS), and these plots can be square, rectangular or irregular shapes based on specific geographical features; the size of the plot can be determined according to the distribution of power infrastructure, geographical conditions and the location of data collection points; in the present invention, for the sake of simple description, the power supply area can be divided into rectangular plots in a longitude and latitude grid manner; and each plot is numbered in sequence;

[0089] It should be noted that before performing feature engineering on environmental data and electricity consumption data, it is necessary to preprocess the historical environmental data and electricity consumption data of all collected plots. The preprocessing methods include data cleaning, feature extraction, feature selection and data standardization.

[0090] Step 102, encoding the historical environmental data of the location of the wind turbine generator set into environmental features, inputting the environmental features into a power generation prediction model, and the power generation prediction model outputs the wind power generation provided by the wind turbine generator set for the power supply area in the next 6 hours;

[0091] Coding of environmental characteristics:

[0092] Wind speed, wind direction, temperature, humidity and air pressure. In order to encode these environmental data, each environmental data is first scaled to the same range, such as using the maximum and minimum scaling method; then each environmental data is organized into a time window, for example, using the data of the past 12 hours to predict the wind power generation in the next 6 hours;

[0093] It should be noted that the environmental feature is a matrix of shape (time_steps, features), where time_steps represents the length of the time window and features represents the dimension of the environmental data;

[0094] The power generation prediction model includes an input layer, an LSTM layer, and a fully connected layer. The input layer inputs environmental features to the LSTM layer, and the LSTM layer outputs the hidden state of the current time step to the fully connected layer. The fully connected layer outputs the wind power generation provided by the wind turbine for the power supply area in the next 6 hours.

[0095] Input Layer:

[0096] The dimension of the input environmental features is (N, T, D), where N represents the number of samples, T represents the time step, which in the scenario of the present invention represents the data of the past 12 hours, and D represents the number of features, which are wind speed, wind direction, temperature, humidity and air pressure, so D = 5;

[0097] LSTM Layer:

[0098] The calculation formula of the input gate is:

[0099] i t =σ(W i ·[h t-1 , x t ]+b i )

[0100] Among them, i t represents the input gate, σ represents the sigmoid activation function, W i represents the second weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b i represents the second bias vector;

[0101] The calculation formula of the forget gate is:

[0102] f t =σ(W f ·[h t-1 , X t ]+b f )

[0103] Among them, f t represents the forget gate, σ represents the sigmoid activation function, W f represents the third weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b f represents the third bias vector;

[0104] The calculation formula of the output gate is:

[0105] o t =σ(W o ·[h t-1 , x t ]+b o )

[0106] Among them, t represents the output gate, σ represents the sigmoid activation function, W o represents the fourth weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b o represents the fourth bias vector;

[0107] The update formula of the memory unit is:

[0108] C t =f t *C t-1 +i t *tanh(W C ·[h t-1 , x t ]+b C )

[0109] Among them, C t Represents the memory cell state at the current time step, C t-1 represents the state of the memory cell at the previous time step, f t represents the forget gate, i t represents the input gate, tanh represents the hyperbolic tangent activation function, and h t-1 represents the hidden state of the previous time step, x t Indicates the current input, W C represents the fifth weight matrix, b Crepresents the fifth bias vector;

[0110] The update formula of the hidden state is:

[0111] h t =o t *tanh(C t )

[0112] Among them, h t represents the hidden state of the current time step, o t represents the input gate, C t Represents the state of the memory unit at the current time step, and tanh represents the hyperbolic tangent activation function;

[0113] Fully connected layer:

[0114] The calculation formula of the fully connected layer is:

[0115] y=RELU(W y ·h t +b y )

[0116] Where y represents the predicted wind power generation provided by the wind turbine to the power supply area in the next 6 hours, W y represents the weight matrix of the fully connected layer, h t represents the hidden state of the current time step, b y Represents the bias term of the fully connected layer, and RELU represents the RELU function;

[0117] Step 103, generating the electricity consumption characteristics of each plot through the electricity consumption data of the plot, dividing the wind power generation according to the number of plots to generate the wind power generation, splicing the wind power generation with the electricity consumption characteristics of each plot to form a comprehensive characteristic, inputting the comprehensive characteristic into the power dispatching strategy network, and outputting the power dispatching strategy, wherein one power dispatching strategy includes the corresponding coal-fired power generation output by the coal-fired power plant to all plots;

[0118] The power dispatching strategy network includes a feature concatenation layer, a hidden layer and an output layer, wherein the feature concatenation layer inputs power consumption features and wind power generation, outputs comprehensive features, inputs the comprehensive features into the hidden layer, outputs hidden features, and inputs the hidden features into the output layer. The output layer outputs the Q value of each optional power dispatching strategy, and selects the power dispatching strategy with the largest Q value as the output;

[0119] The calculation formula of the feature concatenation layer is:

[0120] Z=CONCAT(Y D ,y * )

[0121] Among them, Z represents the comprehensive characteristics, Y Drepresents the electricity consumption characteristics, y* represents the averaged wind power generation vector, and CONCAT represents vector concatenation;

[0122] The calculation formula for the hidden layer is:

[0123] H=f(W H Z+b H )

[0124] Among them, H represents hidden features, Z represents comprehensive features, and W H represents the weight matrix of the hidden layer, b H represents the bias term of the hidden layer, and f represents the tanh activation function;

[0125] The calculation formula of the output layer is:

[0126] Q pred =σ(W o *H+b o )

[0127] Among them, Q pred represents the Q value of the power dispatch strategy corresponding to the power supply area, H represents the hidden feature, and W o represents the weight matrix of the output layer, b o represents the bias term of the output layer, σ represents the sigmoid activation function;

[0128] It should be noted that before each power dispatch, environmental data and power consumption data need to be updated to regenerate the power dispatch strategy;

[0129] The training of the power dispatch strategy network includes the following steps:

[0130] Step 1: Initialize the experience pool, randomly initialize the power dispatch strategy network, and initialize the targetQ network, whose parameters are the same as those of the power dispatch strategy network;

[0131] Step 2: reset the environment and obtain the first state;

[0132] Step 3: Generate an action using the ∈-greedy strategy: with a probability of ∈, an action will be randomly selected, which is the exploration mode; otherwise, a t =max a Q(s t , a; θ), choose s t In the state The biggest action is the experience mode. Through the power dispatch strategy network calculation;

[0133] According to the interaction between the action and the environment, the feedback R is obtained t , the next state s t+1and whether the termination condition done is triggered, the termination condition is the end of the current water fee settlement cycle;

[0134] Experience t , a t , R t ,s t+1 , done is stored in the experience pool, an experience is a four-tuple s t , a t , R t ,s t+1 or t , a t , R t , done;

[0135] Step 4: Get an experience randomly from the experience pool;

[0136]

[0137] max a Q target (s t+1 , a; θ) represents the state s t+1 The maximum output of the targetQ network;

[0138] according to and Calculate the loss and use the gradient descent method to update the drip control strategy network;

[0139]

[0140] Indicates state s t and action a t The output Q value of the power dispatch strategy network is as follows;

[0141] γ is the discount factor, which is a value between 0 and 1, used to balance the weights of immediate rewards and future rewards. The default value is 0.6;

[0142] Step 5: Update the targetQ network every fixed number of network updates to make its parameters the same as those of the current drip control strategy network;

[0143] Step 6: The step is terminated until the Q network converges or the number of updates reaches the set value.

[0144] The reward function is calculated as:

[0145] R=-α·S-β·D-γ·T+C

[0146] Among them, R represents the total reward, S represents the power supply stability reward value, D represents the power demand satisfaction reward value, T represents the processing efficiency reward value, α, β, γ represent the first, second, and third weight coefficients respectively, and C represents the bias term;

[0147] The calculation formula for power supply stability reward is:

[0148]

[0149] Among them, S represents the power supply stability reward value, N represents the number of plots, and P demand，i and P supply，i They represent the electricity demand and supply of the ith plot respectively;

[0150] The calculation formula for the bonus value of meeting electricity demand is:

[0151]

[0152] Where D represents the reward value for meeting electricity demand, N represents the number of plots, and P demand，i and P supply，i Represent the electricity demand and supply of the i-th plot respectively. When P demand，i <P supply，i When the power supply is insufficient, the reward is negative. demand，i >P supply，i When , the electricity demand is met and the reward is zero or positive;

[0153] The calculation formula for the processing efficiency bonus value is:

[0154]

[0155] Where T represents the processing efficiency bonus value, N represents the number of plots, At i , A t-1，i They represent the adjustment of coal-fired power generation for the i-th plot at the current and previous moments respectively;

[0156] It should be noted that the purpose of designing the reward in this way is to ensure that the power dispatch strategy not only meets the immediate power demand, but also strikes a balance between economic benefits, environmental impact and system reliability.

[0157] The state space includes environmental data and electricity consumption data;

[0158] One action includes the amount of coal-fired generation added or subtracted across all plots;

[0159] Action space: The exportable coal-fired power generation of each plot is combined to generate the action space;

[0160] Step 104, allocating coal-fired power generation to each plot according to the output power dispatching strategy.

[0161] like Figure 2 As shown, a resource scheduling system based on big data includes the following modules:

[0162] The data collection module collects the historical electricity consumption data of all plots and the historical environmental data of the locations of wind turbines;

[0163] Function: Responsible for collecting historical electricity consumption data and environmental data of wind turbines from smart meters and other sensors in various power supply areas;

[0164] Input: original electricity consumption data and environmental data;

[0165] Output: pre-processed land electricity consumption data and environmental characteristic data;

[0166] The power generation prediction module is used to output the wind power generation provided by the wind turbines to the power supply area in the next 6 hours;

[0167] Function: Use machine learning models to predict wind power generation;

[0168] Input: environmental characteristic data (wind speed, wind direction, temperature, humidity, air pressure);

[0169] Output: 6-hour wind power generation forecast for the power supply area in the future;

[0170] A feature generation module, used to encode the historical environmental data of the location of the wind turbine generator set into environmental features and generate the power consumption features of each plot of land through the power consumption data of the plot of land;

[0171] Function: Generate plot electricity consumption characteristics and wind power generation characteristics based on historical data;

[0172] Input: preprocessed electricity consumption data and power generation forecast data;

[0173] Output: electricity consumption characteristics of each block and average wind power generation characteristics;

[0174] The comprehensive feature splicing module splices the wind power generation with the electricity consumption characteristics of each plot to form a comprehensive feature;

[0175] Function: Combine electricity consumption characteristics and wind power generation characteristics into comprehensive characteristics;

[0176] Input: electricity consumption characteristics and wind power generation characteristics;

[0177] Output: comprehensive feature dataset;

[0178] The power dispatching strategy network module is used to input the comprehensive features into the power dispatching strategy network and output the power dispatching strategy;

[0179] Function: Use reinforcement learning algorithm to determine the optimal power dispatch strategy;

[0180] Input: comprehensive feature dataset;

[0181] Output: Optimal power dispatch strategy, including allocation of coal-fired power generation;

[0182] A resource allocation execution module, used for executing the power dispatch strategy output by the power dispatch strategy network;

[0183] Function: Execute the allocation of coal-fired power generation according to the output power dispatch strategy;

[0184] Input: Power dispatch strategy;

[0185] Output: Actual coal-fired electricity generation allocated to each plot.

[0186] A storage medium stores non-temporary computer-readable instructions. When the non-temporary computer-readable instructions are executed by a computer, the steps in the above-mentioned resource scheduling method based on big data can be executed.

[0187] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A resource scheduling method based on big data, characterized in that: The following steps are involved: Step 101, gridding the power supply area into multiple plots, collecting historical power consumption data of all plots and historical environmental data of the locations where the wind turbines are located; the power consumption data includes peak power consumption, valley power consumption and average power consumption, and the historical environmental data of the locations where the wind turbines are located includes wind speed, wind direction, temperature, humidity and air pressure data of the locations where the wind turbines are located; Step 102, encoding the historical environmental data of the location of the wind turbine generator set into environmental features, inputting the environmental features into a power generation prediction model, and the power generation prediction model outputs the wind power generation provided by the wind turbine generator set for the power supply area in the next 6 hours; Step 103, generating the electricity consumption characteristics of each plot through the electricity consumption data of the plot, dividing the wind power generation according to the number of plots to generate the wind power generation, splicing the wind power generation with the electricity consumption characteristics of each plot to form a comprehensive characteristic, inputting the comprehensive characteristic into the power dispatching strategy network, and outputting the power dispatching strategy, wherein one power dispatching strategy includes the corresponding coal-fired power generation output by the coal-fired power plant to all plots; Step 104, allocating coal-fired power generation to each plot according to the output power dispatching strategy.

2. The resource scheduling method based on big data according to claim 1, characterized in that: The power generation prediction model includes an input layer, an LSTM layer and a fully connected layer. The input layer inputs environmental features to the LSTM layer, the LSTM layer outputs the hidden state of the current time step to the fully connected layer, and the fully connected layer outputs the wind power generation provided by the wind turbine to the power supply area in the next 6 hours.

3. The resource scheduling method based on big data according to claim 2, characterized in that: Input Layer: The dimension of the input environment feature is (N, T, D), where N represents the number of samples, T represents the time step, and D represents the number of features; LSTM Layer: The calculation formula of the input gate is: i t =σ(W i ·[h t-1 ,x t ]+b i ) Among them, i t represents the input gate, σ represents the sigmoid activation function, W i represents the second weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b i represents the second bias vector; The calculation formula of the forget gate is: f t =σ(W f ·[h t-1 ,x t ]+b f ) Among them, f t represents the forget gate, σ represents the sigmoid activation function, W f represents the third weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b f represents the third bias vector; The calculation formula of the output gate is: the t =σ(W o ·[h t-1 ,x t ]+b o ) Among them, t represents the output gate, σ represents the sigmoid activation function, W o represents the fourth weight matrix, h t-1 represents the hidden state of the previous time step, x t Indicates the current input, b o represents the fourth bias vector; The update formula of the memory unit is: C t =f t *C t-1 +i t *tanh(W C ·[h t-1 ,x t ]+b C ) Among them, C t Represents the memory cell state at the current time step, C t-1 represents the state of the memory cell at the previous time step, f t represents the forget gate, i t represents the input gate, tanh represents the hyperbolic tangent activation function, and h t-1 represents the hidden state of the previous time step, x t Indicates the current input, W C represents the fifth weight matrix, b C represents the fifth bias vector; The update formula of the hidden state is: h t =o t *tanh(C t ) Among them, h t represents the hidden state of the current time step, o t represents the input gate, C t Represents the state of the memory unit at the current time step, and tanh represents the hyperbolic tangent activation function; Fully connected layer: The calculation formula of the fully connected layer is: y=RELU(IN y ·h t +b y ) Where y represents the predicted wind power generation provided by the wind turbine to the power supply area in the next 6 hours, W y represents the weight matrix of the fully connected layer, h t represents the hidden state of the current time step, b y Represents the bias term of the fully connected layer, and RELU represents the RELU function.

4. The resource scheduling method based on big data according to claim 3 is characterized in that: The power dispatching strategy network includes a feature splicing layer, a hidden layer and an output layer, wherein the feature splicing layer inputs power consumption characteristics and wind power generation, outputs comprehensive characteristics, inputs the comprehensive characteristics into the hidden layer, outputs hidden characteristics, and inputs the hidden characteristics into the output layer. The output layer outputs the Q value of each optional power dispatching strategy, and selects the power dispatching strategy with the largest Q value as the output.

5. The resource scheduling method based on big data according to claim 4 is characterized in that: The calculation formula of the feature concatenation layer is: Z=CONCAT(AND D ,and * ) Among them, Z represents the comprehensive characteristics, Y D Represents the electricity consumption characteristics, y * represents the wind power generation vector after equalization, and CONCAT represents vector concatenation.

6. The resource scheduling method based on big data according to claim 5, characterized in that: The calculation formula for the hidden layer is: H=f(W H Z+b H ) Among them, H represents hidden features, Z represents comprehensive features, and W H represents the weight matrix of the hidden layer, b H represents the bias term of the hidden layer, and f represents the tanh activation function; The calculation formula of the output layer is: Q pred =σ(W o *H+b o ) Among them, Q pred represents the Q value of the power dispatch strategy corresponding to the power supply area, H represents the hidden feature, and W o represents the weight matrix of the output layer, b o represents the bias term of the output layer, and σ represents the sigmoid activation function.

7. The resource scheduling method based on big data according to claim 6 is characterized in that: The state space of the power dispatch strategy network includes environmental data and power consumption data; Action space: The exportable coal-fired power generation of each plot is combined to generate the action space; One action includes the amount of coal-fired generation added or subtracted across all plots; The calculation formula of the reward function of the power dispatch strategy network is: R=-α·S-β·D-γ·T+C Among them, R represents the total reward, S represents the power supply stability reward value, D represents the power demand satisfaction reward value, T represents the processing efficiency reward value, α, β, γ represent the first, second, and third weight coefficients respectively, and C represents the bias term; The calculation formula for power supply stability reward is: Among them, S represents the power supply stability reward value, N represents the number of plots, P demand，i and P supply，i They represent the electricity demand and supply of the i-th plot respectively; The calculation formula for the bonus value of meeting electricity demand is: Where D represents the reward value for meeting electricity demand, N represents the number of plots, and P demand，i and P supply，i Represent the electricity demand and supply of the i-th plot respectively. P demand，i <P supply，i When the power supply is insufficient, the reward is negative. demand，i >P supply，i When , the electricity demand is met and the reward is zero or positive; The calculation formula for the processing efficiency bonus value is: Among them, T represents the processing efficiency bonus value, N represents the number of plots, and A t，i , A t-1，i They represent the adjustment of coal-fired power generation for the i-th plot at the current and previous moments respectively.

8. The resource scheduling method based on big data according to claim 7, characterized in that: The training of the power dispatch strategy network includes the following steps: Step 1: Initialize the experience pool, randomly initialize the power dispatch strategy network, and initialize the targetQ network, whose parameters are the same as those of the power dispatch strategy network; Step 2: reset the environment and obtain the first state; Step 3: Generate an action using the ∈-greedy strategy: with a probability of ∈, an action will be randomly selected, which is the exploration mode; otherwise, a t =max a Q(s t ,a;θ), choose t In the state The biggest action is the experience mode. Through the power dispatch strategy network calculation; According to the interaction between the action and the environment, the feedback R is obtained t , the next state s t+1 and whether the termination condition done is triggered, the termination condition is the end of the current water fee settlement cycle; Experience t ,a t ,R t ,s t+1 , done is stored in the experience pool, an experience is a four-tuple s t ,a t ,R t ,s t+1 or t ,a t ,R t ,done; Step 4: Get an experience randomly from the experience pool; max a Q target (s t+1 ,a;θ) represents the state s t+1 The maximum output of the targetQ network; according to and Calculate the loss and use the gradient descent method to update the drip control strategy network; Indicates state s t and action a t The output Q value of the power dispatch strategy network is as follows; γ is a discount factor used to balance the weights of immediate rewards and future rewards, with a default value of 0.6; Step 5: Update the targetQ network every fixed number of network updates to make its parameters the same as those of the current drip control strategy network; Step 6: The step is terminated until the Q network converges or the number of updates reaches the set value.

9. A resource scheduling system based on big data, characterized in that: Includes the following modules: The data collection module collects the historical electricity consumption data of all plots and the historical environmental data of the locations of wind turbines; The power generation prediction module is used to output the wind power generation provided by the wind turbines to the power supply area in the next 6 hours; A feature generation module, used to encode the historical environmental data of the location of the wind turbine generator set into environmental features and generate the power consumption features of each plot of land through the power consumption data of the plot of land; The comprehensive feature splicing module splices the wind power generation with the electricity consumption characteristics of each plot to form a comprehensive feature; The power dispatching strategy network module is used to input the comprehensive features into the power dispatching strategy network and output the power dispatching strategy; The resource allocation execution module is used to execute the power dispatching strategy output by the power dispatching strategy network.

10. A storage medium storing non-transitory computer-readable instructions, characterized in that: When the non-transitory computer-readable instructions are executed by a computer, the steps in the resource scheduling method based on big data as claimed in claim 8 can be executed.

Images