A heating load scheduling method based on data-physical-knowledge joint driving

Abstract

The application discloses a kind of based on data-physical-knowledge joint driving heating load scheduling method, for the problem that heating load randomness is big, not easy to control, first design load clustering method and load power prediction algorithm based on data, physics and knowledge model;Then design heating double-layer scheduling model, user layer deployment load control terminal, collect and report power data, scheduling layer includes prediction module, index set module and scheduling module, wherein prediction module classifies load and predicts heating power, index set module includes economy, comfort and fairness index, scheduling module is based on load prediction data, with economy, comfort and fairness maximization as goal, adopts particle swarm algorithm to solve scheduling power, and by load control terminal implementation, example analysis the feasibility of double-layer scheduling strategy.The application provides technical method for load prediction, scheduling under the transformation of heating, can effectively improve the fairness and effectiveness of heating.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of flexible load coordinated scheduling, in particular to a heating load scheduling method based on data-physical-knowledge joint driving. BACKGROUND

[0002] With the advancement of urbanization in China, the traditional heating method using coal and firewood is gradually replaced by electric heating technology. Compared with traditional heating technology, electric heating technology has less environmental pollution, higher energy utilization rate and higher environmental friendliness. In addition, using electric heating technology for heating in winter and low temperature reduces the degree of building reconstruction and is convenient and easy to use. However, with the increase of electric heating users, the power of electric heating users changes randomly and fluctuates greatly. After the temperature decreases, there will be a large number of heating equipment in the residential area connected to the power grid at the same time, which will cause a certain impact on the regional power grid. Therefore, it is necessary to schedule the electric heating users to meet the user demand and reduce the pressure on the power grid, and maximize the heating goal.

[0003] The conventional electric heating scheduling strategy mainly adopts direct load control strategy and round control strategy to directly control the load, which can meet the scheduling demand of the scheduling end to a certain extent. For example, the patent CN112488372A-multiple time scale electric heating load double-layer optimization scheduling method adopts an orderly heating strategy and a time-of-use electricity price strategy, which are direct control of the load and actual demand data of the load under real-time electricity price. The scheduling strategy has a certain delay effect. In the patent CN114169236A-phase change heat storage type electric heating load control system control method based on LSTM algorithm, the LSTM algorithm is used for power value prediction, but the prediction method is single and does not comprehensively consider the influence of other factors. SUMMARY

[0004] To solve the above technical problems, the present application provides an electric heating load scheduling method based on data-physical-knowledge joint driving, which fully considers the demand of the scheduling layer and the user layer from the scheduling end and the user end, designs an electric heating equipment power prediction model, constructs a power prediction algorithm based on multiple model fusion prediction according to different user categories, and improves the power prediction accuracy by fusing and correcting the prediction results of multiple prediction models. At the same time, based on the prediction results and the particle swarm algorithm, a scheduling scheme meeting the scheduling economy, fairness and comfort indicators is obtained, the stability of the power system is improved, the heating demand of the user is met, and a win-win effect is obtained.

[0005] The heating load scheduling method based on data-physical-knowledge joint driving provided by the present application comprises the following steps:

[0006] Step 1, analyze the existing electric heating load with warm habits and characteristics, classify the load, and then design a kind of warm load power prediction algorithm based on data-physical-knowledge model joint driving; Specifically, first, a data model based on historical data is established to predict power, then a physical model based on time-of-use electricity price and load demand elasticity coefficient is established, then a knowledge reasoning model based on incremental knowledge base and intelligent inference algorithm is established, and finally the prediction results of the physical model and the knowledge model are used to correct the prediction results of the data model by using the correction algorithm, and the future power data of the load is obtained;

[0007] Step 2, design a double-layer scheduling model for electric heating load, specifically: the double-layer scheduling model includes a user layer and a scheduling layer, a load control terminal is installed in the user layer to collect and report the heating power data of the user; The scheduling layer includes a prediction module, an index set module and a scheduling module, wherein the prediction module is used to implement the load classification and load power prediction algorithm in step 1; The index set module includes economic, comfort and fairness indexes, which are used to reflect the heating improvement of the user after implementing the scheduling strategy output by the scheduling module; The scheduling module outputs the scheduling strategy: using the particle swarm algorithm, taking the future power data of the load as the initial particle, taking the index set data, i.e. the maximum user heating improvement as the objective function, solving the algorithm, outputting the scheduling power, and the load control terminal in the user layer implements the scheduling strategy.

[0008] Further, in step 1, the data model based on historical data for power prediction is established, and the steps are as follows:

[0009] 1) Use the load control terminal to collect historical power data, time data, indoor and outdoor temperature and humidity data to establish a training set for the data model;

[0010] 2) Input the training set data into the model training, and the training algorithm can be selected as neural network algorithm, decision tree algorithm, etc. After training, the model can output power prediction data based on the data model according to the real-time data collected by the load control terminal, denoted as P data .

[0011] Further, in step 1, in step 1, the physical model based on time-of-use electricity price and load demand elasticity coefficient is established, and the steps are as follows:

[0012] 1) First, select typical day data, and calculate the time-of-use electricity price by using formula (1), wherein C i,t represents the real-time electricity price of the i-th load at time t; P i,t represents the real-time power of the i-th load at time t, C1, C2, C3, C4 are the purchase electricity cost, transmission and distribution loss, transmission and distribution price and government fund respectively;

[0013]

[0014] 2) Utilize the time-of-use electricity price, and combine the price demand self-elasticity coefficient E (t,t) and the elasticity coefficient E (t,h) , calculate the power demand data based on the time-of-use electricity price and the load demand elasticity coefficient by using formula (2), denoted as P physical , wherein C i,h represents the electricity price of the i-th load at time h;

[0015]

[0016] Further, in step 1, the knowledge reasoning model based on the incremental knowledge base and the intelligent inference algorithm is established, specifically: 1) design a load trend calculation method, wherein T P (t, q) represents the trend of the electric heating load at time t, and the time span before and after is q, represents the load power sequence;

[0017]

[0018] 2) Construct an incremental knowledge base, first use the triple D = {(h, r, t) | h ∈ E, r ∈ R, t ∈ E} to establish a rule base, the rule base includes the heating rules between temperature and load, wherein h is the head entity, t is the tail entity, and r is the relationship between the two, E and R represent the entity set and the relationship set respectively; then use the collected historical data to establish a fact database to store the process information of the scheduling strategy and user changes, wherein T represents the outside temperature, C represents the change of dynamic electricity price in the scheduling strategy, and P represents the change of user heating power;

[0019]

[0020] Finally, the model algorithm library is established. Under the influence of the environment and its own factors on the electric heating load, the existing knowledge in the rule base and the fact database is used to reason the load heating trend, and heuristic knowledge and experience are used to optimize the model parameters.

[0021] 3) In the case that the knowledge base lacks existing knowledge and cannot be inferred, the intelligent inference algorithm is adopted, and the optional inference algorithm includes LSTM, SVM, etc. The algorithm input is time data, temperature data and load power, and the output is load heating trend prediction data; in combination with the above two steps, the power trend data of the user can be inferred, denoted as P know .

[0022] Further, in step 1, the prediction results of the physical model and the knowledge model are used to correct the prediction results of the data model, and the specific steps of obtaining the future load power data are as follows: 1) first use the data model to obtain the power prediction data P data(t), and then obtain the power demand data P physical (t) using the knowledge model know (t); 2) design a correction algorithm, calculate the future power data of the load using formula (5), denoted as P fuse (t). Wherein δ1, δ2 are random errors of the physical model and the knowledge model, m1, m2, m3 are the ratios of P data (t), P physical (t) and P know (t):

[0023]

[0024] Further, in step 2, the user layer load control terminal of the double-layer scheduling model is used to collect and report the heating power data of the user; the scheduling layer includes a prediction module, an index set module and a scheduling module, and the step of outputting the scheduling strategy is:

[0025] 1) the prediction module realizes the algorithm in step 1, and outputs the future power P fuse (t) of the load;

[0026] 2) using the heating power data, the index set module calculates the index set data;

[0027] 3) using the particle swarm algorithm, taking the future power of the load as the initial particle, designing the objective function according to the index set data, performing algorithm solving, and outputting the scheduling power.

[0028] Further, the index set module includes an economic index, a fairness index and a comfort index, specifically:

[0029] 1) the economic index, using the time-of-use electricity price and the heating equipment power before and after optimization to reflect the economy of scheduling, wherein E e,cost is the economic index; P e , P e,f (t) are the fixed electricity price before and after optimization and the time-of-use electricity price at t; L load (t), L load,f (t) are the powers consumed by the user except the heating equipment before and after optimization at t; L ev (t), L ev,f (t) are the heating equipment powers at t before and after optimization; i represents a single load of each type of load, m is the number of load classification, n is the number of loads, T takes 24 hours; Δt takes 60 minutes:

[0030]

[0031] 2) the comfort index, using the comfortable temperature Q i,j and the current temperature Ti,j The average absolute error percentage of (t) is measured, wherein E e,copy is the electricity comfort index.

[0032]

[0033] 3) The fairness index is characterized by the load non-heating time T control , the heating time T warm , and the optimal heating time T best , wherein E m (z) is the fairness index.

[0034]

[0035] Further, the objective function of the scheduling module is specifically: the target function F1: the maximum of user heating economy and comfort; the target function F2: the maximum of the scheduling fairness index in the scheduling process.

[0036] F1=1 / E e,cost +E e,copy (9)

[0037] F2=1 / E m (z) (10)

[0038] Further, the algorithm solving of the scheduling module outputs the scheduling power, specifically: taking the future power of the load as the initial particle, wherein the velocity updating formula of the d-dimensional component of the particle i of the particle swarm algorithm is formula (11), wherein is the d-dimensional component of the flight velocity vector of the particle i in the kth iteration; is the d-dimensional component of the position vector of the particle i in the kth iteration; c1 and c2 are step lengths; r1 and r2 are random functions, and the value range is [0, 1]; w is the inertia weight:

[0039]

[0040] Further, the data transmission is communicated through the wireless module, and the system includes a terminal side and a master station side, specifically: installing a load control terminal on the user side to collect electric heating data; the scheduling layer performs load prediction, index set calculation, and scheduling strategy solving according to the collected power data.

[0041] The beneficial effects of the application are: the application comprehensively considers the influence of power demand and trend change on future load power, starts from the scheduling end and the user end, fully considers the demand of the scheduling layer and the user layer, first needs to design an electric heating equipment power prediction model, according to different user categories, respectively constructs a power prediction algorithm based on multiple model fusion prediction, the algorithm fully embodies the power prediction data, power demand data and power trend data, and the future power demand of the user can be more accurately obtained by using the correction algorithm; at the same time, starting from the user side, an index representing the user's heating consumption and comfort level is designed, starting from the scheduling layer, the fairness problem of multi-region and different heating user scheduling is considered, and then an index set is designed to represent the effectiveness of the scheduling strategy implementation, finally, the particle swarm algorithm is used to integrate the power prediction data and the index set to solve the user power scheduling strategy, the scheduling strategy is dynamically solved according to the index, the scheduling power more meeting the demand is obtained, the peak load is cut and the valley is filled, the stability of the power system is improved, the demand of the user for heating is met, and a win-win effect is obtained. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to make the content of the application more easily understood, the application will be further described in detail below according to specific embodiments and in conjunction with the drawings.

[0043] Figure 1 Flow chart of electric heating load scheduling strategy based on data-physical-knowledge joint driving;

[0044] Figure 2 Power prediction schematic diagram based on data-physical-knowledge fusion;

[0045] Figure 3 Triplet relationship diagram;

[0046] Figure 4 LSTM cell structure diagram;

[0047] Figure 5 Scheduling system structure diagram;

[0048] Figure 6 Electric heating load control system schematic diagram;

[0049] Figure 7 Load daily prediction result schematic diagram;

[0050] Figure 8 Scheduling strategy model solving effect diagram. DETAILED DESCRIPTION

[0051] As Figure 1As shown, the heating load scheduling method based on data-physical-knowledge joint driving provided by the present application comprises the following steps: 1. analyzing the existing electric heating load use habit and characteristics, and building a data-physical-knowledge joint driving based heating power prediction model, specifically, first, classifying the load according to the electric heating load use characteristics, then building a data-physical-knowledge joint driving based heating power prediction model, including a data model based on historical data, a physical model based on time-of-use electricity price and load demand elasticity coefficient, and a user heating trend prediction model based on knowledge reasoning, and respectively performing power prediction, using a correction algorithm to correct the power, and the functions and correction of each model are as follows Figure 2 .

[0052] Further, the electric heating load classification algorithm specifically classifies the electric heating load according to the load historical heating power data, heating time period, heating time length and other factors, so that the scheduling layer can more specifically schedule a certain type of load.

[0053] 2. Building a load prediction data model based on historical data. Specifically, first, build a historical data set of different types of loads, and then build a training set data according to the collected load historical power data, time data (quarter, month, day, hour), indoor and outdoor temperature data, then input the data into the power prediction model based on historical data, and the model algorithm can be selected from neural network algorithm, decision tree algorithm, support vector machine and other data training based prediction models. After the model training is completed, the model can output the power prediction data based on the data model according to the real-time data collected by the load control terminal, denoted as P data .

[0054] 3. Building a physical model based on time-of-use electricity price and load demand elasticity coefficient, and the model establishment steps are as follows: 1) selecting typical daily load data for a certain type, and establishing a real-time electricity price model according to the load proportion; 2) calculating the load change and new elastic electricity price according to the elastic coefficient, at this time the electric heating load change is the correction amount based on the physical model. First, select 24h load data of a typical day, and calculate the time-of-use electricity price by using formula (1), wherein C i,t represents the real-time electricity price of the i-th load at time t; P i,t represents the real-time power of the i-th load at time t, C1, C2, C3 and C4 are the electricity purchase cost, power transmission and distribution loss, power transmission and distribution price and government fund respectively;

[0055]

[0056] The electricity price of this type of load in a typical day can be calculated by formula (1). In order to represent the relationship between the power demand of the load and the real-time electricity price, the price elasticity coefficient E reflecting the load demand is introduced, and the calculation formula of the price elasticity coefficient is as follows:

[0057]

[0058] P(t) = P(t) + P(t) * E(t,t) * (P(t) - P(t)) + P(t) * E(t,h) * (P(t) - P(t-h)) i,t＝0 P(t) represents the power of the i-th load at the initial time (t = 0).

[0059] According to the influence of price fluctuation at time t on the electricity demand at time t and time h, the price elasticity coefficient can be divided into self-elasticity coefficient and mutual-elasticity coefficient

[17] , represented by E(t,t) and E(t,h) respectively, and the calculation formula is as follows:

[0060]

[0061] Further, based on the time-of-use electricity price, the self-elasticity coefficient E (t,t) and the mutual-elasticity coefficient E (t,h) of price demand, a physical model for electric heating load forecasting based on price elasticity response can be constructed, and the formula is as follows, and the output result is denoted as P physical , where C i,h represents the electricity price of the i-th load at time h.

[0062]

[0063] 4. Build a user heating trend prediction based on knowledge reasoning, which includes load trend calculation, knowledge base construction and intelligent heating trend prediction algorithm.

[0064] 1) Design a load trend calculation method, where T P (t,q) represents the trend of the electric heating load at time t, and the time span before and after is q, represents the load power sequence

[0065]

[0066] Then build an incremental knowledge base to describe the relationship between heating trend and external factors, including rule base, fact database and model algorithm base, so as to find the potential relationship between electric heating load historical data and user behavior, and form a knowledge system. The rule base includes premise and conclusion, which can be represented by the concept of triple. Assuming that there is a knowledge graph containing n entities and m relationships, where the facts are stored as triples D = {(h, r, t) | h E, r R, t E}, each triple consists of a head entity h E, a tail entity t E and the relationship r R between them, where E represents the entity set, and R represents the relationship set. According to the relationship between real-time temperature T i , load power d i , real-time electricity price P(h) and load trend P1(t), the following triple is defined to build incremental knowledge base, such as Figure 3The fact database stores the dispatching strategy and the process information of user change using the collected historical data, and the formula is as follows:

[0067]

[0068] In the formula, T represents the external temperature, C represents the change of dynamic price in the dispatching strategy, and P represents the change of user heating power. The model algorithm library includes an inference algorithm, that is, under the influence of the environment and self factors on the electric heating load, the existing knowledge in the rule library and the fact database can be used to infer the heating trend of the load, and in the inference process, the heuristic knowledge and experience can be used to optimize the internal parameters of the model.

[0069] If the input external factors have related knowledge in the knowledge base, the change trend of user heating is inferred based on the established rules, including the following types: 1) consistent matching: the data to be inferred and the head entity of the knowledge are accurately matched, and the content includes quarter, month, hour, temperature, and average temperature; 2) domain value range matching: when the input quarter, month, hour, temperature, and average temperature data are matched within a certain domain value range, it is converted to the input value defined by the rule within the set range, and for the to-be-inferred data with large fluctuations, there are different matching objects and domain value ranges.

[0070] Finally, an intelligent heating trend prediction algorithm is constructed. When consistent matching and threshold matching cannot be performed, the application designs a knowledge inference model based on a long short-term memory neural network (LSTM), and the model input is time data, temperature data, and load power, and the output is load heating trend prediction data.

[0071] The LSTM model adopts a cell structure as shown in Figure 4 , which includes three controls of a forgetting gate, an input gate, and an output gate, which correspond to parts A, B, and C in the figure. In the gate structure, the load information is selectively passed forward through a neural layer using a sigmoid function and a point-by-point multiplication operation. At time t, the cell uses the forgetting gate to control the update and discard of the input power information and the past power information, where h t-1 t-1 represents the output information of the previous cell, and x t is the input of the current cell.

[0072] f t =σ(W f ·[h t-1 ,x t ]+b f ) (14)

[0073] The input gate is used to determine the load information to be added to the cell state, in the B box, first, the sigmoid function is used to determine the update information, and then the tanh function is used to generate the update content vector:

[0074]

[0075] The output gate is Figure 3 In the C box part, first, the sigmoid function is used to determine the part of the cell state that needs to be output, and then the tanh function is used to process the cell state C t And output the load trend result.

[0076]

[0077] 5. The specific steps for correcting the prediction result of the data model using the prediction results of the physical model, the knowledge model and the correction algorithm are as follows: 1) first, the power prediction data P data (t) at time t is obtained using the data model, then the power demand data P physical (t) is obtained using the physical model; the power trend data P know (t) is obtained using the knowledge model; 2) a correction algorithm is designed, and the load future power data is calculated using formula (5), denoted as P fuse (t). Wherein δ1, δ2 are random errors of the physical model and the knowledge model, m1, m2, m3 are the ratios of P data (t), P physical (t) and P know (t).

[0078]

[0079] 6. A kind of electric heating load double-layer scheduling model based on data-physical-knowledge joint driving and particle swarm algorithm, it is characterized in that, the scheduling model includes two layers of mechanism of scheduling layer and user layer, scheduling layer includes prediction module, index set module and scheduling module, the load control terminal of user layer is used to gather, report the heating power data of user.Scheduling model as Figure 5 Shown, the scheduling platform of provincial dispatching cloud executes the function of scheduling module, wisdom energy platform executes the function of prediction module and index set module, completes load prediction and index set calculation, the load control terminal of user layer reports, issues power and scheduling strategy, the physical map of overall deployment as Figure 6 Shown.

[0080] Further, the load fusion prediction module predicts the load future power using the load power fusion prediction algorithm based on data-physical-knowledge built in steps 2-5;

[0081] Further, the scheduling index set calculation module includes an economy index, a fairness index, and a comfort index, and specifically includes:

[0082] 1) The economy index uses the electricity price before and after optimization and the heating device power before and after optimization to reflect the scheduling economy;

[0083]

[0084] In the formula, E e,cost is the economy index; P e , P e,f (t) are the fixed electricity price before and after optimization and the time-of-use electricity price at time t, respectively; L load (t), L load,f (t) are the power consumed by the user other than the heating device before and after optimization at time t; L ev (t), L ev,f (t) are the heating device power at time t before and after optimization; i represents a single load of each type of load, m is the number of load classifications, n is the number of loads, T is 24 hours, and Δt is 60 minutes.

[0085] 2) The comfort index uses the average absolute error percentage of the comfortable temperature Q i,j and the current temperature T i,j (t) to measure, wherein E e,copy is the electricity comfort index.

[0086]

[0087] After implementing the double-layer scheduling strategy, when the user does not change the electricity consumption in each period, the user comfort is the largest, and the value is 1; the greater the change in the user's electricity consumption mode, i.e., the electricity consumption in each period, the lower the comfort index; in the extreme case where the user does not use electricity at all, the user's comfort is 0.

[0088] 3) The fairness index uses the non-heating time T control , the heating time T warm , and the optimal heating time T best to represent the heating fairness of the user, wherein E m (z) is the fairness index. The smaller the value of E m (z), the more fair the entire load rotation control process.

[0089]

[0090] Further, the scheduling module uses a particle swarm algorithm, takes the future load power as the initial particle, designs a target function according to the index set data, performs algorithm solving, and outputs the scheduling power. Specifically, the target function, the constraint condition, and the particle swarm algorithm solving include:

[0091] 1) Objective function

[0092] Objective function 1: maximize user's warm economy and comfort:

[0093] F1 = 1 / E e,cost + E e,copy (9)

[0094] Objective function 2: maximize scheduling fairness index in scheduling process:

[0095] F2 = 1 / E m (z) (10)

[0096] 2) Constraint conditions, including economy index, comfort index and fairness index constraints:

[0097] (1) Economy index requirement

[0098] The economy index reflects the consumption ratio after optimization and before optimization, so the index range needs to be controlled in 0.5-1.5 to comprehensively guarantee the economic benefits of the power grid layer and the user layer.

[0099] (2) Fairness index requirement

[0100] The fairness index is used to constrain the scheduling layer to implement the rotation strategy, avoiding the long heating-off time of the heating use of large demand, so the fairness index is used to comprehensively consider the heating demand of different loads, and the index should be in the range of 0-1.

[0101] (3) Comfort index requirement

[0102] After designing the time-of-use electricity price, it needs to be adjusted according to the comfort of the user to avoid the comfort of the user being too low after the implementation of the calculated electricity price C.

[0103]

[0104] (4) Peak load shifting requirement

[0105] According to the load power prediction data output by the load classification module and the load fusion prediction module, the results solved by the particle swarm algorithm are limited, and the total power after scheduling should be consistent with the power demand, i.e. the sum is consistent.

[0106] Further, the particle swarm algorithm is used to solve the model and obtain the scheduling result of the particle swarm. First, the objective function is integrated into the following formula during solving:

[0107] F = 1 / (F1+F2) (18)

[0108] 3) Particle swarm algorithm solution

[0109] The daily 24-hour load power is taken as 24 particles of the particle swarm, wherein a velocity updating formula of the particle i of the particle swarm algorithm in the d dimension is:

[0110]

[0111] A position updating formula of the particle i in the d dimension is:

[0112]

[0113] In the formula, A d dimension component of a flight velocity vector of the particle i in the kth iteration; A d dimension component of a position vector of the particle i in the kth iteration; c1, c2: acceleration constant, adjusting a maximum step length of learning; r1, r2: two random functions, taking a value range [0, 1], so as to increase randomness; w: inertia weight, non-negative number, adjusting a search range of a solution space.

[0114] Further, the user layer load control terminal receives scheduling data from the scheduling layer and issues to the user load responsible for the user layer load control terminal, so that the cloud-edge collaborative control of scheduling is realized.

[0115] 7. The delivery of the command is all carried out through wireless communication of the Bluetooth module, scheduling and data instructions are delivered between the terminal side and the main station side, the load control terminal is installed on the electric heating user side, electric heating data collection is realized, and is uploaded to the scheduling layer, that is, the negative control terminal and the intelligent circuit breaker are arranged on the user load side, the negative control terminal and the intelligent circuit breaker communicate through the 485 communication mode, the negative control terminal collects load data, and uploads to the load prediction module of the scheduling layer through 3G / 4G and other wireless communication modes, and simultaneously controls the user load by using the intelligent circuit breaker according to the power scheduling data of the scheduling layer. The scheduling layer: collects electric heating operation data, and uploads the electric heating area operation data to the main station side scheduling control system, decomposes the control target forwarded from the main station side control system according to the predetermined strategy, and issues to the load control terminal for execution.

[0116] Example:

[0117] The present application takes the electric heating load of a certain state region as an experimental object to carry out experimental test, and by March 2020, a total of 352.5 million electric energy meters are installed in the electric heating of the certain state, wherein 30.67 million electric energy meters are installed in the coal-to-electricity project, 4.58 million electric energy meters are installed in the scattered industry expansion, and various loads such as civil and industrial heating are included.

[0118] Firstly, the electric heating load is classified according to the load history, such as heating power data, heating period, heating time length and other factors, and is divided into residential heating, industrial park heating, campus heating and other loads. Then, the historical data set of different types of loads is constructed, and the training set data is established according to the collected load historical power data, time data (quarter, month, day, hour), indoor and outdoor temperature data.

[0119] Then, the load prediction model based on historical data is established, the neural network algorithm is selected as the prediction algorithm, the training set data is input into the neural network model, and the load power prediction result based on historical data is output; then, the demand response physical model based on price elasticity coefficient is established, the physical model is established according to the real-time electricity price calculation formula, the price elasticity coefficient calculation formula, and the power calculation formula based on the price demand self-elasticity coefficient and the power calculation formula based on the price elasticity coefficient, and the load power prediction data based on the price elasticity coefficient is calculated; in addition, the user heating trend speculation based on knowledge reasoning is established, firstly, the heating trend of the load in different periods is calculated by using the calculation method based on the sliding time window, then the incremental knowledge base is constructed to describe the relationship between the heating trend and the external factors, and finally the heating trend knowledge reasoning algorithm is constructed, if the input external factors have related knowledge in the knowledge base, the change trend of the user heating is inferred according to the established rules, when consistent matching and threshold matching cannot be performed, the knowledge reasoning model based on long short-term memory neural network (LSTM) is used for reasoning, so as to obtain the user heating trend speculation data based on knowledge reasoning.

[0120] Further, the above prediction data is fused by using the data-physical-knowledge load prediction power fusion method, so as to build the load power prediction algorithm based on the joint driving of data-physical-knowledge, and the prediction result is as shown in Figure 7 The prediction standard deviation of the data model is 4.948, the prediction standard deviation of the data-physical fusion is 0.859, the prediction standard deviation of the data-physical-knowledge reasoning model is 0.867, the fusion prediction algorithm has higher accuracy, and the prediction accuracy of the knowledge reasoning model after correction is not much different from that of the physical model correction, but the heating trend inference of the load is more accurate, and the change of the load trend between 17:00-21:00 can be better reflected.

[0121] Further, the strategy design and the deployment of the load side equipment are carried out at the dispatching layer and the load layer, as shown in Figure 5The upper layer is a scheduling layer, including a prediction module, an index set module and a scheduling module. In the scheduling process, first, the load is classified, data from the load control terminal is received, the real-time power of the load is predicted by using the fusion prediction algorithm, the current index set data is calculated by using the reported data, then the particle swarm algorithm is used to solve the initial particle by using the power prediction data and the current index set data as the initial particle, the scheduling data is obtained by using the objective function and the constraint condition, and the load heating power is sent to the load control terminal. The load control terminal and the intelligent circuit breaker are arranged on the load side, the load control terminal implements the scheduling strategy according to the scheduling data, collects the electric heating load telemetry, the circuit breaker state and other information through the RS-485 communication mode, and sends the circuit breaker control information, when the heating reaches the scheduling power, the intelligent circuit breaker is disconnected to stop heating. In the scheduling process, the load control terminal uploads the real-time heating power of various loads to the scheduling layer, and receives the heating power sent by the scheduling layer, which is used for load switching control.

[0122] The PSO algorithm is used to solve the scheduling model, and the solving result is as shown in Figure 8 The red and blue curves are the heating power data before and after optimization respectively, it can be seen that after using the scheduling strategy, the load power fluctuation is reduced, and the power mutation is less, which is beneficial to the stable operation of the power grid. In addition, the data of the 7th day is selected to calculate the index set, wherein the user power before optimization is 4.67 degrees, and the user power after optimization is 3.035 degrees, the economic index value of the user is greater than 1, which indicates that the heating consumption expenditure of the user has decreased; the comfort index is 0.65, which indicates that the heating behavior of the user has changed by 45% after the scheduling strategy; before participating in the scheduling, the user uses heating as needed, and the heating satisfaction is 1, and after the scheduling, the satisfaction is 0.75, and finally the W value is calculated as 5.185, which is greater than 1, indicating that the scheduling strategy is effective.

[0123] The above description is only the preferred embodiment of the application and the explanation of the applied technical principles. It should be understood by those skilled in the art that the scope of the application involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and also covers other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above-mentioned features are replaced with the technical features disclosed in the present application (but not limited to) having similar functions to form a technical solution.

[0124] In addition to the technical features described in the specification, the remaining technical features are known to those skilled in the art, and in order to highlight the innovative features of the present application, the remaining technical features will not be described here.

Claims

1. A heating load scheduling method based on data-physics-knowledge joint driving, characterized in that, The steps of the method are as follows: Step 1: Analyze the heating habits and characteristics of existing electric heating loads, classify the loads, and then design a power prediction algorithm for heating loads based on a data-physical-knowledge model. Specifically, first, establish a data model for power prediction based on historical data; then, establish a physical model based on time-of-use electricity pricing and load demand elasticity coefficient; next, establish a knowledge reasoning model based on an incremental knowledge base and intelligent inference algorithm; finally, use the prediction results of the physical model, the knowledge reasoning model, and the correction algorithm to correct the prediction results of the data model to obtain future load power data. Step 2: Design a two-layer scheduling model for electric heating loads. Specifically, the two-layer scheduling model includes a user layer and a scheduling layer. A load control terminal is installed at the user layer to collect and report user heating power data. The scheduling layer includes a prediction module, an index set module, and a scheduling module. The prediction module is used to implement the load classification and load power prediction algorithms in Step 1. The index set module includes economic, comfort, and fairness indicators to reflect the improvement in user heating after implementing the scheduling strategy output by the scheduling module. The scheduling module outputs a scheduling strategy: using a particle swarm optimization algorithm, the future load power data is used as the initial particles, and the objective function is to maximize the improvement in user heating based on the index set data. The algorithm is then solved to output the scheduling power, and the load control terminal at the user layer implements the scheduling strategy. The indicator set module includes economic indicators, fairness indicators, and comfort indicators, specifically: 1) Economic indicators: The economic efficiency of dispatch is reflected by using time-of-use electricity prices before and after optimization and the power consumption of heating equipment, where E e,cost For economic indicators; P e P e,f (t) represents the fixed electricity price before and after optimization, and the time-of-use electricity price at time t, respectively; L load (t), L load,f (t) represents the power consumed by users other than heating equipment at time t before and after optimization; L ev (t), L ev,f (t) represents the power of the heating equipment at time t before and after optimization; i represents a single load of each type, m is the number of load categories, n is the number of loads, T is 24 hours, and Δt is 60 minutes. 2) Comfort index, using the comfort temperature Q i,j With the current temperature T i,j The mean absolute error percentage of (t) is used to measure this, where E e,copy For electrical comfort indicators: 3) Fairness index, using the time T during which the load is not heated. control Heating time T warm Optimal heating time T best To characterize the fairness of heating for users, where E m (z) is the fairness indicator: Objective function F1: Maximize user heating economy and comfort; Objective function F2: Maximize scheduling fairness indicators during the scheduling process. F1=1 / E e,cost +E e,copy (9) F2=1 / E m (of) (10) The algorithm solution of the scheduling module outputs the scheduling power, specifically: the future load power is used as the initial particle, where the d-th dimension velocity update formula of particle i in the particle swarm algorithm is given by equation (11), where Let d be the d-th component of the velocity vector of particle i in the k-th iteration; Let be the d-th dimension component of the position vector of particle i in the k-th iteration; c1 and c2 are the step sizes; r1 and r2 are random functions with values ​​in the range [0,1]; w is the inertia weight.

2. The heating load scheduling method based on data-physics-knowledge joint driving as described in claim 1, characterized in that, In step 1, the steps for establishing a data model for power prediction based on historical data are as follows: 1) Collect historical load power data, time data, and indoor and outdoor temperature and humidity data using load control terminals to establish a training set for the data model; 2) Input the training set data into the model for training. After training, the model outputs power prediction data based on the data model according to the real-time data collected by the load control terminal, denoted as P. data .

3. The heating load scheduling method based on data-physics-knowledge joint driving as described in claim 1, characterized in that, In step 1, the physical model based on time-of-use pricing and load demand elasticity coefficient is established. The steps are as follows: 1) First, select typical daily data and calculate the time-of-use pricing using equation (1), where C i,t P represents the real-time electricity price at time t for the i-th load; i,t Let C1 represent the real-time power of the i-th load at time t, and C2, C3, and C4 represent the electricity purchase cost, transmission and distribution loss, transmission and distribution price, and government funds, respectively. 2) Utilize time-of-use pricing, combined with the price demand elasticity coefficient E. (t,t) And the elastic coefficient E (t,h) The power demand data based on time-of-use pricing and load demand elasticity coefficient is calculated using equation (2), and denoted as P. physical C i,h Let represent the electricity price at time h for the i-th load.

4. The heating load scheduling method based on data-physics-knowledge joint driving as described in claim 1, characterized in that, In step 1, the knowledge reasoning model based on the incremental knowledge base and intelligent inference algorithm is established, specifically: 1) Design a load trend calculation method, where T P (t,q) represents the trend of electric heating load at time t, with a time span of q. Represents the load power sequence; 2) Construct an incremental knowledge base. First, a rule base is established using the triple D = {(h,r,t)|h∈E,r∈R,t∈E}. The rule base includes heating rules between temperature and load, where h is the head entity, t is the tail entity, r is the relationship between the two, and E and R represent the entity set and relation set, respectively. Then, a fact database is established using collected historical data to store process information on scheduling strategies and user changes. Here, T represents the external temperature, C represents the change in dynamic electricity price in the scheduling strategy, and P represents the load trend. Finally, a model algorithm library is established. Under the influence of environmental and internal factors on electric heating load, the existing knowledge in the rule base and fact database is used to infer the heating trend of the load, and heuristic knowledge and experience are used to optimize the model parameters. 3) To address situations where the knowledge base lacks existing knowledge and reasoning is impossible, an intelligent inference algorithm is employed. The algorithm takes time data, temperature data, and load power as input and outputs load heating trend prediction data. Combining the above steps, the user's power trend data is inferred, denoted as P. know .

5. A heating load scheduling method based on data-physics-knowledge joint driving as described in claim 1, characterized in that, In step 1, the specific steps for obtaining future load power data by using the prediction results of the physical model, the knowledge model, and the correction algorithm to correct the prediction results of the data model are as follows: 1) First, use the data model to obtain the power prediction data P at time t. data (t), and then use the physical model to obtain the power demand data P. physical (t); Power trend data P obtained using a knowledge model know (t); 2) Design a correction algorithm and use equation (5) to calculate the future load power data, denoted as P. fuse (t); where δ1 and δ2 are the random errors of the physical model and the knowledge model, respectively, and m1, m2, and m3 are P data (t), P physical (t) and P know The ratio of (t):

6. The heating load scheduling method based on data-physics-knowledge joint driving as described in claim 1, characterized in that, In step 2, the user-layer load control terminal of the two-layer scheduling model is used to collect and report users' heating power consumption data; the scheduling layer includes a prediction module, an index set module, and a scheduling module, and the steps for outputting the scheduling strategy are as follows: 1) Utilize the prediction module to output the future load power P fuse (t); 2) Using heating power data, the index set module calculates index set data; 3) Using the particle swarm optimization algorithm, with the future power of the load as the initial particle, design the objective function based on the index set data, solve the algorithm, and output the scheduling power.

Images