CN117878979B - Power balance and dynamic compensation system based on electric energy storage - Google Patents
Power balance and dynamic compensation system based on electric energy storage Download PDFInfo
- Publication number
- CN117878979B CN117878979B CN202410286867.1A CN202410286867A CN117878979B CN 117878979 B CN117878979 B CN 117878979B CN 202410286867 A CN202410286867 A CN 202410286867A CN 117878979 B CN117878979 B CN 117878979B
- Authority
- CN
- China
- Prior art keywords
- power
- data
- model
- training
- time sequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004146 energy storage Methods 0.000 title claims abstract description 65
- 238000012549 training Methods 0.000 claims abstract description 93
- 238000012163 sequencing technique Methods 0.000 claims abstract description 7
- 238000000034 method Methods 0.000 claims description 60
- 239000013598 vector Substances 0.000 claims description 42
- 230000006870 function Effects 0.000 claims description 34
- 238000004422 calculation algorithm Methods 0.000 claims description 19
- 238000012706 support-vector machine Methods 0.000 claims description 19
- 230000002457 bidirectional effect Effects 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 18
- 238000013507 mapping Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 12
- 238000009826 distribution Methods 0.000 claims description 9
- 238000000926 separation method Methods 0.000 claims description 9
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 7
- 238000010586 diagram Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 238000009792 diffusion process Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 5
- 230000005520 electrodynamics Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000008014 freezing Effects 0.000 claims description 3
- 238000007710 freezing Methods 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 230000006698 induction Effects 0.000 claims description 3
- 238000012821 model calculation Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000000630 rising effect Effects 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 claims description 3
- 238000002759 z-score normalization Methods 0.000 claims description 3
- 238000009434 installation Methods 0.000 claims description 2
- 230000005611 electricity Effects 0.000 abstract description 8
- 238000012423 maintenance Methods 0.000 abstract description 4
- 230000033228 biological regulation Effects 0.000 abstract description 2
- 238000012545 processing Methods 0.000 abstract description 2
- 230000007613 environmental effect Effects 0.000 description 18
- 239000000523 sample Substances 0.000 description 13
- 238000005286 illumination Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000011217 control strategy Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000002035 prolonged effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000013068 control sample Substances 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000007599 discharging Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000004148 unit process Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention belongs to the technical field of power grid regulation and control, and discloses a power balance and dynamic compensation system based on electric energy storage; the data acquisition module is used for acquiring the comprehensive data of the historical system and processing the comprehensive data to obtain a pre-training comprehensive data set; the model training module is used for training a system power time sequence predictor model; training a power balance classification sub-model; the model simultaneous model is used for constructing a system power balance prediction model; the threshold value acquisition module inputs the real-time system comprehensive data into a power balance robust prediction model, and predicts to obtain a power balance state; if the power balance state is unbalanced, constructing an electric quantity dynamics model for acquiring a first electric quantity threshold value and a second electric quantity threshold value; the dynamic adjustment module is used for sequencing the importance priorities of the n load devices; and the operation and maintenance cost of the power grid system is reduced, and the important electricity utilization tasks are ensured to be completed smoothly.
Description
Technical Field
The invention relates to the technical field of power grid regulation, in particular to a power balance and dynamic compensation system based on electric energy storage.
Background
The patent with the publication number of CN115719979A discloses a source load storage coordination control method for off-grid operation of a new energy micro-grid, wherein a control object comprises an energy storage system, a new energy unit and load switches of each station area, and a control framework comprises an equipment on-site control layer, a micro-grid coordination control layer and a distribution network main station control layer; the on-site control layer of the equipment adopts a power balance strategy based on energy storage dynamic compensation, and the power difference between a new energy unit and a load is dynamically compensated through an energy storage converter, so that instantaneous power balance is realized; the micro-grid coordination control layer adopts a short-term electric quantity balance strategy based on source storage cooperative control, and adjusts the output of a new energy unit through a micro-grid controller to realize short-term electric quantity balance; the control layer of the distribution network main station adopts a long-time electric quantity balance strategy based on load adjustment, and the load is adjusted through the distribution main station so as to realize long-time electric quantity balance. The method and the system are beneficial to realizing off-grid stable operation of the micro-grid.
With the continuous improvement of the social electrification level, a power grid system faces increasingly complex and high-dimensional operation management challenges; under the superposition of various uncertain factors, the system is extremely easy to generate unexpected events such as unbalance of supply and demand, electric quantity drop and the like; the existing network Internet of things sensing system and information processing means are rough, and are in face of massive heterogeneous data, and the overall analysis and judgment capability and the processing response speed of the system have obvious weaknesses; this directly results in a difficulty in quickly and accurately judging the risk status of the system and implementing an orderly hierarchical load control strategy when a significant power supply pressure or critical node energy source is urgent. And the active and dynamic state supplement intervention of the electric quantity cannot be performed. The critical tasks and production guarantee capability of the system are often affected, and great uncertainty is added to the production order and the social development. At the same time, frequent high-intensity operation also aggravates the cost load and operation and maintenance management pressure of the power grid.
In view of the above, the present invention proposes a power balance and dynamic compensation system based on electric energy storage to solve the above-mentioned problems.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides the following technical scheme for achieving the purposes: a power balance and dynamic compensation system based on electrical energy storage, comprising: the data acquisition module is used for acquiring the comprehensive data of the historical system and preprocessing the comprehensive data of the historical system to obtain a pre-training comprehensive data set;
The model training module is used for training a system power time sequence predictor model for predicting the system power value according to the pre-training comprehensive data set; training a power balance classification sub-model for outputting a current system power classification result according to the predicted system power value;
The model simultaneous module is used for integrating the system power time sequence prediction sub-model and the power balance classification sub-model to construct a system power balance prediction model for predicting the power balance state; the power balance state is balanced or unbalanced;
the threshold value acquisition module is used for inputting the real-time system comprehensive data into the power balance robust prediction model, and predicting to obtain a power balance state; if the power balance state is unbalanced, constructing an electric quantity dynamics model for acquiring a first electric quantity threshold value and a second electric quantity threshold value;
The dynamic adjustment module is used for sequencing the importance priorities of the n load devices; when the electric quantity of the electric energy storage device is smaller than a first electric quantity threshold value, load equipment with low importance priority is sequentially cut off until the electric quantity is larger than or equal to the first electric quantity threshold value; and when the electric quantity of the electric energy storage device is greater than or equal to the second electric quantity threshold value, sequentially recovering the cut load equipment.
Further, the historical system comprehensive data comprises load power data, generator set power data, power grid access point power data, battery energy storage system charge and discharge power data and environmental data coefficients;
The method comprises the steps that load power is obtained through each load equipment installation power sensor, and corresponding time sequence data are formed according to fixed time periods and recorded according to sequence;
the power data of the generator set is obtained by recording and reading data values according to a fixed time period through a power monitoring sensor arranged on the generator set, and forming corresponding time sequence data according to the sequence;
the power data of the power grid access point is obtained by installing a power distribution power measurement device at the power grid access point, recording and reading data values according to a fixed time period, and forming corresponding time sequence data according to a sequence;
The method for acquiring the charge and discharge power data of the battery energy storage system comprises the steps of installing power measurement devices for power distribution at two ends of a direct current bus of the battery energy storage system, measuring charge and discharge current and terminal voltage, and calculating to obtain charge and discharge power through a power formula; forming corresponding time sequence data according to the fixed time period and the sequence record;
Environmental data coefficient ；
In the method, in the process of the invention,For the illumination intensity,/>Is the temperature,/>Is weather humidity,/>Is wind speed,/>Is electricity price/>AndIs a reference coefficient;
the illumination intensity is acquired through a light intensity sensor; the air temperature is obtained by a temperature sensor; the weather humidity is obtained through a humidity sensor; the wind speed is obtained through a wind speed sensor; electricity prices are obtained from official websites of grid companies.
Further, the process of acquiring the pre-training comprehensive data set includes:
detecting and filtering invalid, abnormal and repeated data in the acquired data; filling missing values in the filtered data by using an average method, a median method, an adjacent value method and an interpolation method; smoothing the load power data, the power data of the generator set, the power data of the power grid access point and the time sequence data of the charge and discharge power data of the battery energy storage system by using a 3-point average filtering algorithm;
Extracting statistical characteristics of load power data, wherein the statistical characteristics comprise a mean value, a standard deviation and a power variation of a fixed time period; extracting engine characteristics of power data of the generator set, wherein the engine characteristics comprise rising time, falling time and fluctuation times of a fixed time period; extracting power grid characteristics of power data of a power grid access point, wherein the power grid characteristics comprise extremum, peak-valley difference and forward and reverse power ratio of a fixed time period; extracting battery characteristics of charge and discharge power data of a battery energy storage system, wherein the battery characteristics comprise power range and charge and discharge power ratio in a fixed time period; mapping statistical features, engine features, grid features and battery features to the same order of magnitude using 0-1 normalization or Z-score normalization methods;
obtaining m environmental data coefficients in a historical fixed time, taking the m environmental data coefficients as data coordinate points, fitting the data coordinate points into an environmental coefficient curve according to a time sequence, and integrating the environmental coefficient curve and the fixed time to obtain environmental characteristic parameters;
The statistical features, engine features, grid features, battery features, and environmental feature parameters constitute a pre-training comprehensive dataset.
Further, the obtaining mode of the system power time sequence predictor model includes:
Dividing the pre-training comprehensive data set into a comprehensive training set and a comprehensive verification set; extracting features in the pre-training comprehensive data set according to a time sequence, and constructing a high-dimensional time sequence feature vector serving as an input vector of a system power time sequence predictor model; collecting system power values corresponding to the extracted features according to time sequence; taking the corresponding system power as the output of a time sequence predictor model;
Constructing a basic structure of a system power time sequence predictor model based on a long-short-time memory network, wherein the basic structure comprises an input layer, a bidirectional LSTM encoder layer, a unidirectional LSTM decoder layer, an attention layer and an output layer, and adding residual connection between the bidirectional LSTM encoder and the unidirectional LSTM decoder layer;
the manner of adding a residual connection between the bi-directional LSTM encoder and the uni-directional LSTM decoder layers includes:
The bidirectional LSTM encoder is of a stacked structure of N bidirectional LSTM layers, and takes the output characteristic H of the last LSTM layer; splicing the output characteristic H of the bidirectional LSTM encoder with the original input characteristic X to obtain the input characteristic HX of the decoder; the unidirectional LSTM decoder is a stacked structure of M unidirectional LSTM layers; the input of the first LSTM layer of the unidirectional LSTM decoder is HX, and recursive modeling of time sequence characteristics is carried out; the layers are then connected according to standard LSTM; skipping a certain number of LSTM layers, and adding the output characteristic H of the bidirectional LSTM coder and the characteristic output by the middle layer of the decoder to realize depth residual error;
When a loss function of a system power time sequence predictor model is set, adopting a mean square error loss function;
Mean square error loss function ；
Wherein,For the number of samples,/>Representation of samples/>Is the true system power value of/>Representation of samples/>System power value obtained through model prediction,/>Representing model parameters;
Training a system power time sequence predictor model by adopting a random gradient descent method of an Adam self-adaptive optimization algorithm; in the process of training a model by using the comprehensive training set, gradually reducing the value of the loss function by calculating gradient information of the loss function and updating parameters in a system power time sequence predictor model according to an Adam algorithm; terminating training when the value of the loss function of the comprehensive verification set no longer drops; and obtaining a system power time sequence predictor model.
Further, the obtaining manner of the power balance classification submodel includes:
Constructing a power balance classification sub-model based on a support vector machine, wherein an input feature vector of the power balance classification sub-model comprises statistical features, engine features and output of a system power time sequence prediction sub-model, and the output is a power classification result;
Collecting the output of statistical features, engine features and a system power time sequence predictor model as input feature vector samples; marking the system power balance state corresponding to the collected input feature vector sample, and marking the input feature vector sample with a class label as balanced or unbalanced; the input feature vector samples marked with the categories form a vector sample set; dividing the vector sample set into a training set and a testing set;
Inputting the training set into a power balance classification sub-model, and training an SVM classifier to enable the SVM classifier to learn the corresponding relation between the input feature vector and the class of the power balance state of the system;
The learning process includes:
Representing input feature vector samples of a training set as Category labels are denoted/>; Using RBF kernel functionsMapping the input feature vector samples to a high-dimensional space;
; in the/> Is a core parameter in RBF core function,/>Represents the/>Training feature vectors; /(I)Represents the/>Training feature vectors;
obtaining separation hyperplane parameters by optimizing the classification loss function of an SVM classifier ；
Classification loss function；
In the method, in the process of the invention,Is a penalty coefficient; /(I)The number of samples is the training set; /(I)Indexing the training set samples; /(I)Represents the/>Loss function values of the individual training samples; /(I)Represents the/>A training class label;
by separating hyperplane parameters Obtaining bias term/>, of separation hyperplane; Wherein/>To separate hyperplane parameters/>Is a transpose of (2); /(I)For/>Mapping the individual samples to function values on the separation hyperplane;
setting classification precision of SVM classifier ; Wherein/>For the number of real cases,/>The number of true negative examples; /(I)The number of false positive examples; /(I)The number of the false negative cases;
when the classification precision of the SVM classifier on the test set reaches a preset threshold value, model training is completed; when the test precision is not up to the requirement, adjusting the parameters of the SVM classifier And repeating training until the classification precision of the SVM classifier on the test set reaches a preset threshold.
Further, the construction mode of the system power balance prediction model comprises the following steps:
Adopting the output of the power balance classification sub-model at the output end of the system power balance prediction model, and freezing the parameters of the system power time sequence prediction sub-model and the power balance classification sub-model; and the construction of the system power balance prediction model is completed.
Further, the formula of the electrodynamic model is that; Wherein,Time/>Is provided; /(I)Time/>Is provided; Time/> The system power value is obtained through a system power time sequence predictor model; /(I)Time/>A control signal to the load device;
setting state of charge constraints for an electrical energy storage device Wherein/>Is the minimum constraint; is the maximum constraint; constructing a control candidate set of control signals of load equipment, and iteratively searching for the optimal/>, wherein the control candidate set aims at meeting the constraint of the electric quantity state Control sequence/>；
Selecting a current time from a control sequenceIs substituted into dynamic model calculation, and solves/>, in real timeIs the optimal state trajectory of (1);
Recording the electric quantity of the corresponding electric energy storage device at each moment on the optimal state track; forming a time sequence pair; in the time sequence pair, sequentially iterating each time sequence point, and judging the electric quantity and the minimum constraint of the electric energy storage equipment Selecting a time sequence point with the smallest difference value, and extracting the electric quantity of the electric energy storage equipment corresponding to the time sequence point as a first electric quantity threshold;
in the time sequence pair, sequentially iterating each time sequence point, and judging the electric quantity and the maximum constraint of the electric energy storage equipment And selecting a time sequence point with the smallest difference value, and extracting the electric quantity of the electric energy storage equipment corresponding to the time sequence point as a second electric quantity threshold value.
Further, the control candidate set is constructed by defining a control variableFor controlling the proportion of the load equipment, the value range of the control proportion is [0.1,1]; the [0.1,1] section is divided into A sub-sections on average, and a control candidate set is constituted.
Further, the means for prioritizing the importance of the n loads includes:
Step 1, collecting attribute data of n load devices, wherein the attribute data comprise rated power, load rate, working environment temperature, historical failure rate and importance coefficients;
Step 2, setting weights corresponding to the attribute data respectively; rated power is weighted as The weight of the load factor isThe weight of the importance coefficient is/>The historical failure rate is weighted/>The weight of the importance coefficient is/>; And is also provided with；
Step 3, normalizing the attribute data of n load devices, and mapping the attribute data into a range of [0,1 ];
Step 4, synthesizing and calculating the comprehensive scores of each load device by using an exponential weighting method, and sequencing n load devices from large to small according to the magnitude of the comprehensive scores;
First, the Comprehensive scoring/>, of individual load devices；
；
Wherein,For/>Rated power of the individual load devices; /(I)For/>Load factors of the individual load devices; /(I)For/>The operating environment temperature of the individual load devices; /(I)For/>Historical failure rates of individual load devices; /(I)For/>Importance coefficients of the individual load devices.
Further, the obtaining manner of the attribute data includes:
Consult the data manual of the load equipment, record its rated active power index; measuring the maximum working current and the maximum working voltage of the load equipment by using an oscilloscope and an induction current clamp, and calculating the maximum power value according to the maximum working current and the maximum working voltage; comparing the rated active power index with the maximum power value, and taking the value with the larger value as the rated power;
Installing a power acquisition device at a device power access point, and reading the actual active power consumption of the device recorded in the acquisition device in real time; load ratio = actual active power consumption of the device/rated power;
Detecting the working environment temperature of the load equipment in real time by using a temperature and humidity sensor arranged around the equipment; counting the failure times and total failure time of the equipment in the past v years, and calculating the failure rate;
Collecting topological structure diagrams of load equipment and other subsystems of the system, and determining topological connection relations and dependency relations; setting a dependence propagation algorithm, and evaluating the importance coefficient of the numerical values of the load devices 1-10 according to the influence diffusion force and the dependence degree of the devices in the system topology network.
The power balance and dynamic compensation system based on electric energy storage has the technical effects and advantages that:
The initiative and the adaptability of the power state judgment and the electric quantity compensation of the power grid system are greatly improved; the accurate prediction, constraint and optimization control of the system power change and the electric quantity state under various complex conditions are realized; by establishing a perfect system state prediction and evaluation mechanism, risks can be perceived in advance, and a control strategy is adopted to dynamically adjust loads so as to ensure stable and reliable operation of the system; the depth fusion application of the multi-source heterogeneous data and the end-to-end depth learning model realize the accurate grasp of the current state and the development trend of the system; each module works cooperatively and supports each other, so that the overall sensing, analyzing and decision making capability of the system is enhanced; the running and maintenance cost of the power grid system is reduced, and the service life of load equipment is prolonged; ensuring that important electricity utilization tasks are successfully completed.
Drawings
FIG. 1 is a schematic diagram of an electric energy storage based power balance and dynamic compensation system according to the present invention;
fig. 2 is a schematic diagram of a power balancing and dynamic compensation method based on electric energy storage according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
Referring to fig. 1, the power balance and dynamic compensation system based on electric energy storage according to the present embodiment includes:
The data acquisition module is used for acquiring the comprehensive data of the historical system and preprocessing the comprehensive data of the historical system to obtain a pre-training comprehensive data set;
the model training module is used for training a system power time sequence predictor model for predicting the system power value according to the pre-training comprehensive data set; training a power balance classification sub-model for outputting the balance state of the current system power state according to the predicted system power value;
The model simultaneous module is used for integrating the system power time sequence prediction sub-model and the power balance classification sub-model to construct a system power balance prediction model for predicting the power balance state; the power balance state is balanced or unbalanced;
the threshold value acquisition module is used for inputting the real-time system comprehensive data into the power balance robust prediction model, and predicting to obtain a power balance state; if the power balance state is unbalanced, constructing an electric quantity dynamics model for acquiring a first electric quantity threshold value and a second electric quantity threshold value;
the dynamic adjustment module is used for sequencing the importance priorities of the n load devices; when the electric quantity of the electric energy storage device is smaller than a first electric quantity threshold value, load equipment with low importance priority is sequentially cut off until the electric quantity is larger than or equal to the first electric quantity threshold value; when the electric quantity of the electric energy storage device is larger than or equal to a second electric quantity threshold value, sequentially recovering the cut load equipment;
the historical system comprehensive data comprise load power data, generator set power data, power grid access point power data, battery energy storage system charge and discharge power data and environmental data coefficients;
Installing power sensors on each load device to obtain load power, for example, a power sampling device or a bypass sampling resistor, and calculating and recording the active power of the load in real time by measuring voltage and current; forming corresponding time sequence data according to the fixed time period and the sequence record;
the power data of the generator set is obtained by recording and reading data values according to a fixed time period through a power monitoring sensor arranged on the generator set, and forming corresponding time sequence data according to the sequence;
the power data of the power grid access point is obtained by installing a power measurement device for power distribution on the power grid access point, such as an intelligent electric energy meter and the like, and measuring the forward/reverse active power and reactive power of the power grid access point; recording and reading data values according to a fixed time period, and forming corresponding time sequence data according to the sequence;
The method for acquiring the charge and discharge power data of the battery energy storage system comprises the steps of installing power measurement devices for power distribution at two ends of a direct current bus of the battery energy storage system, measuring charge and discharge current and terminal voltage, and calculating to obtain charge and discharge power through a power formula; forming corresponding time sequence data according to the fixed time period and the sequence record;
Environmental data coefficient ；
In the method, in the process of the invention,For the illumination intensity,/>Is the temperature,/>Is weather humidity,/>Is wind speed,/>Is electricity price/>AndIs a reference coefficient; the reference coefficient is obtained by fitting experimental data through regression analysis and a least square method; the experimental data are a plurality of groups of collected historical environmental parameters; the historical environmental parameters include historical illumination intensity, air temperature, weather humidity, wind speed and electricity price;
The illumination intensity is acquired through a light intensity sensor; the air temperature is obtained by a temperature sensor; the weather humidity is obtained through a humidity sensor; the wind speed is obtained through a wind speed sensor; the electricity price is obtained from an official website of the power grid company;
The acquisition process of the pre-training comprehensive data set comprises the following steps:
detecting and filtering invalid, abnormal and repeated data in the collected data, for example, setting a load power negative value, a value exceeding the rated power of equipment and the like as a missing value;
filling missing values in the filtered data by using an average method, a median method, an adjacent value method and an interpolation method; smoothing the load power data, the power data of the generator set, the power data of the power grid access point and the time sequence data of the charge and discharge power data of the battery energy storage system by using a 3-point average filtering algorithm, so as to reduce random noise;
Extracting statistical characteristics of load power data, wherein the statistical characteristics comprise a mean value, a standard deviation and a power variation of a fixed time period; extracting engine characteristics of power data of the generator set, wherein the engine characteristics comprise rising time, falling time and fluctuation times of a fixed time period; extracting power grid characteristics of power data of a power grid access point, wherein the power grid characteristics comprise extremum, peak-valley difference and forward and reverse power ratio of a fixed time period; extracting battery characteristics of charge and discharge power data of a battery energy storage system, wherein the battery characteristics comprise power range and charge and discharge power ratio in a fixed time period;
the method for acquiring the rise time includes the steps of reading power time series data of a generator set, calculating power variation between adjacent data points, counting the rise time when the variation is a positive value for the first time, and stopping counting when the variation is 0, wherein the elapsed time is the rise time;
The falling time is the time that the power value of the generator set falls from the maximum value to the minimum value in a fixed time period; the method for acquiring the rise time is the same as the method for acquiring the rise time;
The fluctuation frequency acquisition method is to read the power time series data of the generator set, calculate the difference between the data point and the average power value, and when the difference changes number, the fluctuation is once, and the statistic number change frequency is the fluctuation frequency.
The extremum is directly extracted from the power time sequence data of the power grid access point; the peak-valley difference is also extracted from the power time series data of the power grid access point, and the difference is calculated by the maximum and minimum values;
the forward-reverse power ratio is the ratio of the forward power provided by the grid access point to the grid to the reverse power absorbed from the grid within a fixed period of time; wherein, the forward power refers to the power provided by the grid access point to the grid; reverse power refers to the power absorbed by the grid access point from the grid; the absolute values thereof were measured and the ratio was calculated, respectively.
The power range obtaining method is to measure the charging power time sequence and the discharging power time sequence respectively, and calculate the difference after taking the maximum value respectively;
The charge-discharge power ratio is obtained by measuring the charge power time sequence and the discharge power time sequence respectively, taking average value respectively, and calculating the ratio.
Mapping statistical features, engine features, grid features and battery features to the same order of magnitude using 0-1 normalization or Z-score normalization methods;
obtaining m environmental data coefficients in a historical fixed time, taking the m environmental data coefficients as data coordinate points, fitting the data coordinate points into an environmental coefficient curve according to a time sequence, and integrating the environmental coefficient curve and the fixed time to obtain environmental characteristic parameters;
The statistical features, engine features, grid features, battery features, and environmental feature parameters constitute a pre-training comprehensive dataset.
Further, the obtaining manner of the system power time sequence predictor model includes:
Dividing the pre-training comprehensive data set into a comprehensive training set and a comprehensive verification set; the specific dividing ratio can be determined according to actual conditions; for example, an 80% comprehensive training set and a 20% comprehensive validation set;
Extracting features in the pre-training comprehensive data set according to a time sequence, and constructing a high-dimensional time sequence feature vector reflecting the comprehensive condition of the system as an input vector of a system power time sequence predictor model; collecting system power values corresponding to the extracted features according to time sequence; taking the corresponding system power as the output of a time sequence predictor model;
Constructing a basic structure of a system power time sequence predictor model based on a long short-time memory (LSTM) network, wherein the basic structure comprises an input layer, a bidirectional LSTM encoder layer, a unidirectional LSTM decoder layer, an attention layer and an output layer, and adding residual connection between the bidirectional LSTM encoder and the unidirectional LSTM decoder layer to enable the output characteristics of the encoder to be directly transmitted to the decoder so as to realize the propagation of deeper information, and enhancing the end-to-end training capacity of the model;
the manner of adding a residual connection between the bi-directional LSTM encoder and the uni-directional LSTM decoder layers includes:
The bidirectional LSTM encoder is of a stacked structure of N bidirectional LSTM layers, and takes the output characteristic H of the last LSTM layer; splicing the output characteristic H of the bidirectional LSTM encoder with the original input characteristic X to obtain the input characteristic HX of the decoder; the residual migration of the encoder characteristics is realized by splicing, and richer condition information is provided for a decoder;
the unidirectional LSTM decoder is a stacked structure of M unidirectional LSTM layers; the input of the first LSTM layer of the unidirectional LSTM decoder is HX, and recursive modeling of time sequence characteristics is carried out; the layers are then connected according to standard LSTM; skipping a certain number of LSTM layers, and adding the output characteristic H of the bidirectional LSTM coder and the characteristic output by the middle layer of the decoder to realize depth residual error;
It should be noted that, the standard LSTM connection refers to a standard connection manner between layers in the LSTM network, that is, an LSTM unit and a next LSTM unit are connected by expanding according to time steps, and each LSTM unit processes an output of a previous LSTM unit;
When a loss function of a system power time sequence predictor model is set, a mean square error loss function (MSE) is adopted, so that the difference between the model prediction output and a true value is effectively measured;
Loss function ；
Wherein,For the number of samples,/>Representation of samples/>Is the true system power value of/>Representation of samples/>System power value obtained through model prediction,/>Representing model parameters;
training a system power time sequence predictor model by adopting a random gradient descent method of an Adam self-adaptive optimization algorithm; the Adam algorithm combines the cumulative gradient in AdaGrad algorithm and the exponential decay average in RMSProp algorithm, so that the self-adaptive learning rate of each parameter can be effectively calculated;
In the process of training a model by using the comprehensive training set, calculating gradient information of a loss function, updating parameters in a system power time sequence predictor model according to an Adam algorithm, gradually reducing the value of the loss function, and enabling the predicted output of the system power time sequence predictor model to continuously approximate to the real system power so as to optimize the model;
Gradient clipping technology is adopted to limit the gradient range; terminating training when the value of the loss function of the comprehensive verification set no longer drops; obtaining a system power time sequence predictor model;
it should be noted that, the output power predicted value of the system power time sequence predicted sub-model is used as one of the input features of the classification sub-model for judging the balance of the current power state, so as to improve the classification accuracy; assisting parameter setting of the electrodynamic model; the predicted system power time sequence can reflect the system power change trend more accurately, and is favorable for the rationality of state constraint conditions and control variable setting in the electric quantity dynamics model, so that a control strategy which is more in line with reality is solved;
The power balance classification submodel acquisition mode comprises the following steps:
constructing a power balance classification sub-model based on a Support Vector Machine (SVM), wherein an input feature vector of the sub-model comprises statistical features, engine features and output of a system power time sequence prediction sub-model, and the output is a power classification result; the power classification result is that the current system power state is balanced or unbalanced;
Collecting the output of statistical features, engine features and a system power time sequence predictor model as input feature vector samples;
marking the system power balance state corresponding to the collected input feature vector sample, and marking the input feature vector sample with a class label as balanced or unbalanced;
The input feature vector samples marked with the categories form a vector sample set; dividing the vector sample set into a training set and a testing set;
Inputting the training set into a power balance classification sub-model, and training an SVM classifier to enable the SVM classifier to learn the corresponding relation between the input feature vector and the class of the power balance state of the system;
The learning process includes:
Representing input feature vector samples of a training set as Category labels are denoted/>; Using RBF kernel functionsMapping the input feature vector samples to a high-dimensional space;
; in the/> Is a core parameter in RBF core function,/>Represents the/>Training feature vectors; /(I)Represents the/>Training feature vectors;
obtaining separation hyperplane parameters by optimizing the classification loss function of an SVM classifier ；
Classification loss function；
In the method, in the process of the invention,Is a penalty coefficient; /(I)The number of samples is the training set; /(I)Indexing the training set samples; /(I)Represents the/>Loss function values of the individual training samples; /(I)Represents the/>A training class label;
by separating hyperplane parameters Obtaining bias term/>, of separation hyperplane; Wherein,To separate hyperplane parameters/>Is a transpose of (2); /(I)For/>Mapping the individual samples to function values on the separation hyperplane;
setting classification precision of SVM classifier ; Wherein/>For the number of real cases,/>The number of true negative examples; /(I)The number of false positive examples; /(I)The number of the false negative cases;
the true example refers to the number of positive samples that are correctly determined to be positive, for example, the actual positive samples, and the classifier predicts the positive samples as well.
True negatives refer to the number of negative samples that are correctly determined to be negative, e.g., actually negative, and the classifier predicts as negative as well.
The false positive example refers to the number of negative samples that are erroneously determined to be positive, for example, actually negative samples, and the classifier prediction is erroneously determined to be positive samples.
The false negative example refers to the number of positive samples that are erroneously determined to be negative, for example, actually positive samples, and the classifier prediction is erroneously determined to be negative.
When the classification precision of the SVM classifier on the test set reaches a preset threshold value, model training is completed; when the test precision is not up to the requirement, adjusting the parameters of the SVM classifierRepeating training until the classification precision of the SVM classifier on the test set reaches a preset threshold; the setting of the preset threshold value can be determined according to the actual training process;
in a preferred embodiment, a dynamic training sample selection mechanism based on an information quantity algorithm is designed, so that a system power balance prediction model pays more attention to a training sample with wrong classification in the training process, and the classification performance is enhanced;
Integrating a system power time sequence predictor model and a power balance classification model; adopting the output of the power balance classification sub-model at the output end of the system power balance prediction model, and freezing the parameters of the system power time sequence prediction sub-model and the power balance classification sub-model; the construction of a system power balance prediction model is completed;
the formula of the electrodynamic model is ; Wherein/>Time/>Is provided; /(I)Time/>Is provided; /(I)Time/>The system power value is obtained through a system power time sequence predictor model; /(I)Time/>A control signal to the load device;
setting state of charge constraints for an electrical energy storage device Wherein/>Is the minimum constraint; is the maximum constraint; constructing a control candidate set of control signals of load equipment, and iteratively searching for the optimal/>, wherein the control candidate set aims at meeting the constraint of the electric quantity state Control sequence/>；
Selecting a current time from a control sequenceIs substituted into dynamic model calculation, and solves/>, in real timeIs the optimal state trajectory of (1);
Recording the electric quantity of the corresponding electric energy storage device at each moment on the optimal state track; forming a time sequence pair; in the time sequence pair, sequentially iterating each time sequence point, and judging the electric quantity and the minimum constraint of the electric energy storage equipment Selecting a time sequence point with the smallest difference value, and extracting the electric quantity of the electric energy storage equipment corresponding to the time sequence point as a first electric quantity threshold;
in the time sequence pair, sequentially iterating each time sequence point, and judging the electric quantity and the maximum constraint of the electric energy storage equipment Selecting a time sequence point with the smallest difference value, and extracting the electric quantity of the electric energy storage equipment corresponding to the time sequence point as a second electric quantity threshold value;
The construction method of the control candidate set includes:
Definition of control variables For controlling the proportion of the load equipment, the value range is [0.1,1]; thus/>The whole control sample space of the device is a section [0.1,1] which can meet all possible control proportion values of the normal operation of the device; dividing the [0.1,1] interval into a plurality of subintervals on average, wherein each subinterval takes a representative value as a discrete sample to form a control candidate set;
further, the means for prioritizing importance of the n loads includes:
Step 1, collecting attribute data of n load devices, wherein the attribute data comprise rated power, load rate, working environment temperature, historical failure rate and importance coefficients;
Step 2, setting weights corresponding to the attribute data respectively; rated power is weighted as The weight of the load factor isThe weight of the importance coefficient is/>The historical failure rate is weighted/>The weight of the importance coefficient is/>; And is also provided with; The obtaining mode of the weight corresponding to each attribute data comprises the following steps:
Collecting attribute data of a sufficient number of historical load devices; the attribute data of the historical load equipment and the running state thereof at the time form a training sample, and the running state is marked as a sample label; constructing models such as random forests or SVMs, and predicting the running state by using attribute data of each load device as prediction features; evaluating the contribution degree of each attribute to the model prediction result through a feature importance analysis algorithm, and taking the contribution degree as the weight of attribute data;
step 3, normalizing the attribute data of n load devices, and mapping the attribute data into a range of [0,1 ];
Step 4, synthesizing and calculating the comprehensive scores of each load device by using an exponential weighting method, and sequencing n load devices from large to small according to the magnitude of the comprehensive scores;
First, the Comprehensive scoring/>, of individual load devices；
；
Wherein,For/>Rated power of the individual load devices; /(I)For/>Load factors of the individual load devices; /(I)For/>The operating environment temperature of the individual load devices; /(I)For/>Historical failure rates of individual load devices; /(I)For/>Importance coefficients of the individual load devices;
The acquisition mode of the attribute data comprises the following steps:
Consult the data manual of the load equipment, record its rated active power index; measuring the maximum working current and the maximum working voltage of the load equipment by using an oscilloscope and an induction current clamp, and calculating the maximum power value according to the maximum working current and the maximum working voltage; comparing the rated active power index with the maximum power value, and taking the larger of the rated active power index and the maximum power value as rated power;
Installing a power acquisition device, such as an intelligent ammeter, at a device power access point; the method comprises the steps of reading the actual active power consumption of equipment recorded in an acquisition device in real time; load ratio = actual active power consumption of the device/rated power;
Detecting the working environment temperature of the load equipment in real time by using a temperature and humidity sensor arranged around the equipment; counting the failure times and total failure time of the equipment in the past v years, and calculating the failure rate;
Collecting topological structure diagrams of load equipment and other subsystems of the system, and determining topological connection relations and dependency relations; setting a dependence propagation algorithm, and evaluating importance coefficients of the numerical values of the load devices 1-10 according to the influence diffusion force and the dependence degree of the devices in a system topology network;
the dependence propagation algorithm is based on a topological connection relation and a dependence relation, and a matrix is established to represent the dependence relation strength among devices; defining a propagation operator, and performing iterative computation to obtain an influence diffusion result;
according to the range and intensity of the device to influence diffusion, evaluating importance coefficients of 1-10 by using a clustering and discriminant analysis algorithm;
According to the embodiment, the initiative and the adaptability of the power state judgment and the electric quantity compensation of the power grid system are greatly improved; the accurate prediction, constraint and optimization control of the system power change and the electric quantity state under various complex conditions are realized; by establishing a perfect system state prediction and evaluation mechanism, risks can be perceived in advance, and a control strategy is adopted to dynamically adjust loads so as to ensure stable and reliable operation of the system; the depth fusion application of the multi-source heterogeneous data and the end-to-end depth learning model realize the accurate grasp of the current state and the development trend of the system; each module works cooperatively and supports each other, so that the overall sensing, analyzing and decision making capability of the system is enhanced; the running and maintenance cost of the power grid system is reduced, and the service life of load equipment is prolonged; ensuring that important electricity utilization tasks are successfully completed.
Example 2
Referring to fig. 2, in a power balance and dynamic compensation process of a power grid system, a power balance and dynamic compensation method based on electric energy storage implemented by a power balance and dynamic compensation system based on electric energy storage includes:
S1, collecting comprehensive data of a historical system, and preprocessing the comprehensive data of the historical system to obtain a pre-training comprehensive data set;
S2, training a system power time sequence predictor model for predicting the system power value according to the pre-training comprehensive data set; training a power balance classification sub-model for outputting a current system power classification result according to the predicted system power value;
S3, integrating the system power time sequence prediction sub-model and the power balance classification sub-model to construct a system power balance prediction model for predicting a power balance state; the power balance state is balanced or unbalanced;
S4, inputting real-time system comprehensive data into a power balance robust prediction model, and predicting to obtain a power balance state; if the power balance state is unbalanced, constructing an electric quantity dynamics model for acquiring a first electric quantity threshold value and a second electric quantity threshold value;
s5, carrying out importance priority ranking on the n load devices; when the electric quantity of the electric energy storage device is smaller than a first electric quantity threshold value, load equipment with low importance priority is sequentially cut off until the electric quantity is larger than or equal to the first electric quantity threshold value; and when the electric quantity of the electric energy storage device is greater than or equal to the second electric quantity threshold value, sequentially recovering the cut load equipment.
In addition, the process schematically depicted by the power balancing and dynamic compensation method based on electric energy storage may be implemented as a computer software program according to embodiments of the present application. For example, the present application provides a non-transitory machine-readable storage medium storing machine-readable instructions executable by a processor to perform instructions corresponding to the method steps provided by the present application, of course, the architecture shown in the power balancing and dynamic compensation method schematic diagram based on electric energy storage is merely exemplary, and when implementing different devices, adaptive selection or adjustment may be made according to actual needs.
The above formulas are all formulas with dimensionality removed and numerical calculation, the formulas are formulas with the latest real situation obtained by software simulation through collecting a large amount of data, and preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.
The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to those skilled in the art without departing from the principles of the present invention are intended to be comprehended within the scope of the present invention.
Claims (9)
1. The power balance and dynamic compensation system based on electric energy storage is characterized by comprising: the data acquisition module is used for acquiring the comprehensive data of the historical system and preprocessing the comprehensive data of the historical system to obtain a pre-training comprehensive data set;
The model training module is used for training a system power time sequence predictor model for predicting the system power value according to the pre-training comprehensive data set; training a power balance classification sub-model for outputting a current system power classification result according to the predicted system power value;
The model simultaneous module is used for integrating the system power time sequence prediction sub-model and the power balance classification sub-model to construct a system power balance prediction model for predicting the power balance state; the power balance state is balanced or unbalanced;
the threshold value acquisition module is used for inputting the real-time system comprehensive data into the power balance robust prediction model, and predicting to obtain a power balance state; if the power balance state is unbalanced, constructing an electric quantity dynamics model for acquiring a first electric quantity threshold value and a second electric quantity threshold value;
the dynamic adjustment module is used for sequencing the importance priorities of the n load devices; when the electric quantity of the electric energy storage device is smaller than a first electric quantity threshold value, load equipment with low importance priority is sequentially cut off until the electric quantity is larger than or equal to the first electric quantity threshold value; when the electric quantity of the electric energy storage device is larger than or equal to a second electric quantity threshold value, sequentially recovering the cut load equipment;
the historical system comprehensive data comprise load power data, generator set power data, power grid access point power data and battery energy storage system charge and discharge power data;
The method comprises the steps that load power is obtained through each load equipment installation power sensor, and corresponding time sequence data are formed according to fixed time periods and recorded according to sequence;
the power data of the generator set is obtained by recording and reading data values according to a fixed time period through a power monitoring sensor arranged on the generator set, and forming corresponding time sequence data according to the sequence;
the power data of the power grid access point is obtained by installing a power distribution power measurement device at the power grid access point, recording and reading data values according to a fixed time period, and forming corresponding time sequence data according to a sequence;
The method for acquiring the charge and discharge power data of the battery energy storage system comprises the steps of installing power measurement devices for power distribution at two ends of a direct current bus of the battery energy storage system, measuring charge and discharge current and terminal voltage, and calculating to obtain charge and discharge power through a power formula; and forming corresponding time series data according to the fixed time period and the sequence record.
2. The electric energy storage based power balance and dynamic compensation system of claim 1, wherein the acquisition process of the pre-training integrated data set comprises:
detecting and filtering invalid, abnormal and repeated data in the acquired data; filling missing values in the filtered data by using an average method, a median method, an adjacent value method and an interpolation method; smoothing the load power data, the power data of the generator set, the power data of the power grid access point and the time sequence data of the charge and discharge power data of the battery energy storage system by using a 3-point average filtering algorithm;
Extracting statistical characteristics of load power data, wherein the statistical characteristics comprise a mean value, a standard deviation and a power variation of a fixed time period; extracting engine characteristics of power data of the generator set, wherein the engine characteristics comprise rising time, falling time and fluctuation times of a fixed time period; extracting power grid characteristics of power data of a power grid access point, wherein the power grid characteristics comprise extremum, peak-valley difference and forward and reverse power ratio of a fixed time period; extracting battery characteristics of charge and discharge power data of a battery energy storage system, wherein the battery characteristics comprise power range and charge and discharge power ratio in a fixed time period; mapping statistical features, engine features, grid features and battery features to the same order of magnitude using 0-1 normalization or Z-score normalization methods;
the statistical features, engine features, grid features, and battery features form a pre-training comprehensive dataset.
3. The system for power balance and dynamic compensation based on electric energy storage according to claim 2, wherein the system power time series predictor model is obtained by the following steps:
Dividing the pre-training comprehensive data set into a comprehensive training set and a comprehensive verification set; extracting features in the pre-training comprehensive data set according to a time sequence, and constructing a high-dimensional time sequence feature vector serving as an input vector of a system power time sequence predictor model; collecting system power values corresponding to the extracted features according to time sequence; taking the corresponding system power as the output of a time sequence predictor model;
Constructing a basic structure of a system power time sequence predictor model based on a long-short-time memory network, wherein the basic structure comprises an input layer, a bidirectional LSTM encoder layer, a unidirectional LSTM decoder layer, an attention layer and an output layer, and adding residual connection between the bidirectional LSTM encoder and the unidirectional LSTM decoder layer;
the manner of adding a residual connection between the bi-directional LSTM encoder and the uni-directional LSTM decoder layers includes:
The bidirectional LSTM encoder is of a stacked structure of N bidirectional LSTM layers, and takes the output characteristic H of the last LSTM layer; splicing the output characteristic H of the bidirectional LSTM encoder with the original input characteristic X to obtain the input characteristic HX of the decoder; the unidirectional LSTM decoder is a stacked structure of M unidirectional LSTM layers; the input of the first LSTM layer of the unidirectional LSTM decoder is HX, and recursive modeling of time sequence characteristics is carried out; the layers are then connected according to standard LSTM; skipping a certain number of LSTM layers, and adding the output characteristic H of the bidirectional LSTM coder and the characteristic output by the middle layer of the decoder to realize depth residual error;
When a loss function of a system power time sequence predictor model is set, adopting a mean square error loss function;
Mean square error loss function ；
Wherein,For the number of samples,/>Representation of samples/>Is the true system power value of/>Representation of samples/>System power value obtained through model prediction,/>Representing model parameters;
Training a system power time sequence predictor model by adopting a random gradient descent method of an Adam self-adaptive optimization algorithm; in the process of training a model by using the comprehensive training set, gradually reducing the value of the loss function by calculating gradient information of the loss function and updating parameters in a system power time sequence predictor model according to an Adam algorithm; terminating training when the value of the loss function of the comprehensive verification set no longer drops; and obtaining a system power time sequence predictor model.
4. The system for power balance and dynamic compensation based on electric energy storage according to claim 3, wherein the power balance classification sub-model is obtained by the following steps:
Constructing a power balance classification sub-model based on a support vector machine, wherein an input feature vector of the power balance classification sub-model comprises statistical features, engine features and output of a system power time sequence prediction sub-model, and the output is a power classification result;
Collecting the output of statistical features, engine features and a system power time sequence predictor model as input feature vector samples; marking the system power balance state corresponding to the collected input feature vector sample, and marking the input feature vector sample with a class label as balanced or unbalanced; the input feature vector samples marked with the categories form a vector sample set; dividing the vector sample set into a training set and a testing set;
Inputting the training set into a power balance classification sub-model, and training an SVM classifier to enable the SVM classifier to learn the corresponding relation between the input feature vector and the class of the power balance state of the system;
The learning process includes:
Representing input feature vector samples of a training set as Category labels are denoted/>; Using RBF kernel functions/>Mapping the input feature vector samples to a high-dimensional space;
; in the/> Is a core parameter in RBF core function,/>Represents the/>Training feature vectors; /(I)Represents the/>Training feature vectors;
obtaining separation hyperplane parameters by optimizing the classification loss function of an SVM classifier ；
Classification loss function；
In the method, in the process of the invention,Is a penalty coefficient; /(I)The number of samples is the training set; /(I)Indexing the training set samples; /(I)Represents the/>Loss function values of the individual training samples; /(I)Represents the/>A training class label;
by separating hyperplane parameters Obtaining bias term/>, of separation hyperplane; Wherein/>To separate hyperplane parameters/>Is a transpose of (2); /(I)For/>Mapping the individual samples to function values on the separation hyperplane;
setting classification precision of SVM classifier ; Wherein/>For the number of real examples,The number of true negative examples; /(I)The number of false positive examples; /(I)The number of the false negative cases;
when the classification precision of the SVM classifier on the test set reaches a preset threshold value, model training is completed; when the test precision is not up to the requirement, adjusting the parameters of the SVM classifier And repeating training until the classification precision of the SVM classifier on the test set reaches a preset threshold.
5. The electric energy storage-based power balance and dynamic compensation system according to claim 4, wherein the system power balance prediction model is constructed in a manner comprising:
Adopting the output of the power balance classification sub-model at the output end of the system power balance prediction model, and freezing the parameters of the system power time sequence prediction sub-model and the power balance classification sub-model; and the construction of the system power balance prediction model is completed.
6. The electric energy storage based power balance and dynamic compensation system of claim 5, wherein the formula of the electrodynamic model is; Wherein/>Time/>Is provided; /(I)Time/>Is provided; /(I)Time/>The system power value is obtained through a system power time sequence predictor model; /(I)Time/>A control signal to the load device;
setting state of charge constraints for an electrical energy storage device Wherein/>Is the minimum constraint; /(I)Is the maximum constraint; constructing a control candidate set of control signals of load equipment, and iteratively searching for the optimal/>, wherein the control candidate set aims at meeting the constraint of the electric quantity stateControl sequence/>；
Selecting a current time from a control sequenceIs substituted into dynamic model calculation, and solves/>, in real timeIs the optimal state trajectory of (1);
Recording the electric quantity of the corresponding electric energy storage device at each moment on the optimal state track; forming a time sequence pair; in the time sequence pair, sequentially iterating each time sequence point, and judging the electric quantity and the minimum constraint of the electric energy storage equipment Selecting a time sequence point with the smallest difference value, and extracting the electric quantity of the electric energy storage equipment corresponding to the time sequence point as a first electric quantity threshold;
in the time sequence pair, sequentially iterating each time sequence point, and judging the electric quantity and the maximum constraint of the electric energy storage equipment And selecting a time sequence point with the smallest difference value, and extracting the electric quantity of the electric energy storage equipment corresponding to the time sequence point as a second electric quantity threshold value.
7. The electric energy storage based power balance and dynamic compensation system of claim 6 wherein said control candidate set is constructed in a manner defining a control variableFor controlling the proportion of the load equipment, the value range of the control proportion is [0.1,1]; the [0.1,1] section is divided into A sub-sections on average, and a control candidate set is constituted.
8. The electric storage based power balancing and dynamic compensation system of claim 7, wherein the means for prioritizing the importance of the n loads comprises:
Step 1, collecting attribute data of n load devices, wherein the attribute data comprise rated power, load rate, working environment temperature, historical failure rate and importance coefficients;
Step 2, setting weights corresponding to the attribute data respectively; rated power is weighted as The weight of the load factor is/>The weight of the importance coefficient is/>The historical failure rate is weighted/>The weight of the importance coefficient is/>; And is also provided with；
Step 3, normalizing the attribute data of n load devices, and mapping the attribute data into a range of [0,1 ];
Step 4, synthesizing and calculating the comprehensive scores of each load device by using an exponential weighting method, and sequencing n load devices from large to small according to the magnitude of the comprehensive scores;
First, the Comprehensive scoring/>, of individual load devices；
；
Wherein,For/>Rated power of the individual load devices; /(I)For/>Load factors of the individual load devices; /(I)Is the firstThe operating environment temperature of the individual load devices; /(I)For/>Historical failure rates of individual load devices; /(I)For/>Importance coefficients of the individual load devices.
9. The system of claim 8, wherein the means for obtaining the attribute data comprises:
Consult the data manual of the load equipment, record its rated active power index; measuring the maximum working current and the maximum working voltage of the load equipment by using an oscilloscope and an induction current clamp, and calculating the maximum power value according to the maximum working current and the maximum working voltage; comparing the rated active power index with the maximum power value, and taking the value with the larger value as the rated power;
Installing a power acquisition device at a device power access point, and reading the actual active power consumption of the device recorded in the acquisition device in real time; load ratio = actual active power consumption of the device/rated power;
Detecting the working environment temperature of the load equipment in real time by using a temperature and humidity sensor arranged around the equipment; counting the failure times and total failure time of the equipment in the past v years, and calculating the failure rate;
Collecting a topological structure diagram of the load equipment and a system subsystem, and determining a topological connection relation and a dependency relation; setting a dependence propagation algorithm, and evaluating the importance coefficient of the numerical values of the load devices 1-10 according to the influence diffusion force and the dependence degree of the devices in the system topology network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410286867.1A CN117878979B (en) | 2024-03-13 | 2024-03-13 | Power balance and dynamic compensation system based on electric energy storage |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410286867.1A CN117878979B (en) | 2024-03-13 | 2024-03-13 | Power balance and dynamic compensation system based on electric energy storage |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117878979A CN117878979A (en) | 2024-04-12 |
| CN117878979B true CN117878979B (en) | 2024-05-17 |
Family
ID=90590528
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410286867.1A Active CN117878979B (en) | 2024-03-13 | 2024-03-13 | Power balance and dynamic compensation system based on electric energy storage |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117878979B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101833310A (en) * | 2009-03-09 | 2010-09-15 | 纵横网路资讯股份有限公司 | Power consumption management and control system and method |
| WO2013120347A1 (en) * | 2012-02-17 | 2013-08-22 | 中国电力科学研究院 | Control method for unifying self-balancing and self-smoothing of micro-grid |
| CN104659819A (en) * | 2015-03-19 | 2015-05-27 | 武汉大学 | Wind power adsorption estimation method involving wind power prediction error |
| CN110365052A (en) * | 2019-08-08 | 2019-10-22 | 东北大学 | Microgrid energy-storage system state consistency control method based on power optimization scheduling |
| CN112510701A (en) * | 2020-11-27 | 2021-03-16 | 江苏科技大学 | Multi-energy ship energy management control device and method |
| CN115133577A (en) * | 2022-07-26 | 2022-09-30 | 广东电网有限责任公司 | Wave power generator and coordination control method and device thereof |
| CN116706962A (en) * | 2023-04-25 | 2023-09-05 | 华能青海发电有限公司新能源分公司 | Energy storage system call control method and system |
| CN117485177A (en) * | 2023-12-01 | 2024-02-02 | 上海速普顿新能源科技有限公司 | Intelligent energy-saving control system and method for charging pile |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200379417A1 (en) * | 2019-05-29 | 2020-12-03 | University Of Washington | Techniques for using machine learning for control and predictive maintenance of buildings |
| EP4338022A2 (en) * | 2021-05-06 | 2024-03-20 | Strong Force IoT Portfolio 2016, LLC | Quantum, biological, computer vision, and neural network systems for industrial internet of things |
| US20230170697A1 (en) * | 2021-11-30 | 2023-06-01 | Elexity, Inc. | Techniques for predictive control of energy use in a structure |
-
2024
- 2024-03-13 CN CN202410286867.1A patent/CN117878979B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101833310A (en) * | 2009-03-09 | 2010-09-15 | 纵横网路资讯股份有限公司 | Power consumption management and control system and method |
| WO2013120347A1 (en) * | 2012-02-17 | 2013-08-22 | 中国电力科学研究院 | Control method for unifying self-balancing and self-smoothing of micro-grid |
| CN104659819A (en) * | 2015-03-19 | 2015-05-27 | 武汉大学 | Wind power adsorption estimation method involving wind power prediction error |
| CN110365052A (en) * | 2019-08-08 | 2019-10-22 | 东北大学 | Microgrid energy-storage system state consistency control method based on power optimization scheduling |
| CN112510701A (en) * | 2020-11-27 | 2021-03-16 | 江苏科技大学 | Multi-energy ship energy management control device and method |
| CN115133577A (en) * | 2022-07-26 | 2022-09-30 | 广东电网有限责任公司 | Wave power generator and coordination control method and device thereof |
| CN116706962A (en) * | 2023-04-25 | 2023-09-05 | 华能青海发电有限公司新能源分公司 | Energy storage system call control method and system |
| CN117485177A (en) * | 2023-12-01 | 2024-02-02 | 上海速普顿新能源科技有限公司 | Intelligent energy-saving control system and method for charging pile |
Non-Patent Citations (2)
| Title |
|---|
| Short-Term Three-Phase Load Prediction With Advanced Metering Infrastructure Data in Smart Solar Microgrid Based Convolution Neural Network Bidirectional Gated Recurrent Unit;PHUONG NGUYEN THANH 等;IEEE Access;20220731;第10卷;68686-68699 * |
| 施烨 等.缺乏可靠功率预测条件下的微电网实时调度策略.《电力信息与通信技术》.第19卷(第3期),80-91. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117878979A (en) | 2024-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110097297A (en) | A kind of various dimensions stealing situation Intellisense method, system, equipment and medium | |
| CN110782153A (en) | Modeling method and system for comprehensive energy efficiency assessment system of enterprise park | |
| CN117277958B (en) | Intelligent operation and maintenance management method and system for photovoltaic power station based on big data | |
| CN110969306B (en) | Power distribution low-voltage transformer area load prediction method and device based on deep learning | |
| CN109325607A (en) | A kind of short-term wind power forecast method and system | |
| CN112149873A (en) | Low-voltage transformer area line loss reasonable interval prediction method based on deep learning | |
| CN116167527B (en) | Pure data-driven power system static safety operation risk online assessment method | |
| CN112990587A (en) | Method, system, equipment and medium for accurately predicting power consumption of transformer area | |
| CN116148753A (en) | Intelligent electric energy meter operation error monitoring system | |
| CN110968703B (en) | Method and system for constructing abnormal metering point knowledge base based on LSTM end-to-end extraction algorithm | |
| CN113516271A (en) | Wind power cluster power day-ahead prediction method based on space-time neural network | |
| CN112288157A (en) | Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning | |
| CN108256693A (en) | A kind of photovoltaic power generation power prediction method, apparatus and system | |
| CN111754029A (en) | Community load prediction system | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN117878979B (en) | Power balance and dynamic compensation system based on electric energy storage | |
| CN115907228A (en) | Short-term power load prediction analysis method based on PSO-LSSVM | |
| CN111177278A (en) | Grid user short-term load prediction real-time processing tool | |
| CN116305741A (en) | Updating method and device for digital twin model of power equipment | |
| CN113034018B (en) | Power grid supply and demand balance adjustment method based on power moment analysis | |
| CN112818592A (en) | Photovoltaic cell parameter and power generation capacity prediction method and system based on data mining | |
| CN110837932A (en) | Thermal power prediction method of solar heat collection system based on DBN-GA model | |
| CN105975736B (en) | A kind of renewable energy output power modelling method of probabilistic based on orthogonal series | |
| CN117791626B (en) | Intelligent comprehensive power box power supply optimization method | |
| CN115114983B (en) | Method for acquiring and analyzing electric quantity data based on big data equipment and computer system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |