CN103781108A - Neural network-based wireless sensor network data prediction method - Google Patents
Neural network-based wireless sensor network data prediction method Download PDFInfo
- Publication number
- CN103781108A CN103781108A CN201210411715.7A CN201210411715A CN103781108A CN 103781108 A CN103781108 A CN 103781108A CN 201210411715 A CN201210411715 A CN 201210411715A CN 103781108 A CN103781108 A CN 103781108A
- Authority
- CN
- China
- Prior art keywords
- data
- wireless sensor
- sensor network
- network
- neuron
- 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.)
- Pending
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a neural network-based wireless sensor network data prediction method. According to the method, a neural network is led into a wireless sensor network, and each wireless sensor datum is described through neurons, and a neural network model can be built; the fusion and extraction of acquired data of the wireless sensor network can be realized through using the neural network model of the wireless sensor network; motion items are added, and multiple factors are effectively considered, and not only is the gradient descent direction of a certain moment adjusted, but also a gradient direction before the moment is adjusted, and therefore, a vibration range can be authentically reduced, and training speed can be improved; and main factors that influence data output results are selected through differences between various kinds of application types, and a prediction model is established. The neural network-based wireless sensor network data prediction method of the invention can perform self learning in a certain field so as to obtain influence factors therein and construct complex nonlinear relationships, and therefore, influence of human intervention on prediction results can be effectively reduced, and the results are of greater objectivity.
Description
Technical field
The present invention relates to wireless communication technology field, particularly wireless sensor network data Forecasting Methodology.
Background technology
The development of informationization technology and the communication technology, people improve day by day to obtaining with the transmission rate request of data of data, and traditional communication technology has been difficult to meet people's demand.Wireless sensor network is the product of realizing this goal and requiring, by wireless sensor network, and the Monitoring Data of obtaining various zoness of different that user can be real-time, it is one of important technology of next-generation communication network technology.Wireless sensor network has been applied in a lot of fields, and has obtained good achievement.
Wireless sensor network is to have the micro-node of big data quantity, can carry out the New Complex network of remote monitoring, Real time data acquisition along with what wireless technology and network technical development were got up.The product that wireless sensor network combines as calculating, communication and transducer three technology is a kind of brand-new Information acquisi-tion technology.Along with the development of transducer, microprocessor and wireless communication technology, wireless sensor network is widely used, and has very wide application prospect in fields such as ECOLOGICAL ENVIRONMENTAL MONITORING, infrastructure safety, advanced manufacture, logistics management, medical treatment & health, industrial sensor, intelligent transportation control, intelligent energy and military affairs.Wireless sensor network is towards the future development of self-organizing adaptive network, and meanwhile, the wireless sensor technology application of multidisciplinary blending is also just obtaining more and more people's concern, and neural net is wherein to apply more one.
Because nerve calculates in the uniqueness aspect information processing and using value, become one of focus of Intelligent Information Processing research and application at present.Artificial neural net is by processing unit and is called the undirected signalling channel interconnection of connection and the parallel distributed processor that forms.People's neural network force artwork is imitated biological nervous system, by accepting the stimulation of outside input, constantly obtains and accumulates knowledge, and then have certain judgement predictive ability.Artificial neural net has polytype, can be for domain feature in different applications, adopt different network models, and comparatively commonly BP network model, it is applicable to various common engineering fields.Meanwhile, be also applicable to wireless sensor network.This model is by learning training data (comprising input and output value), constantly changes and connects weights, builds the non-linear relationship of model, thereby according to the size of new input value prediction output valve.The research and development of artificial neural net has produced far-reaching influence to modern science and technology.
Summary of the invention
The present invention is a kind of wireless sensor network data Forecasting Methodology based on neural net.The present invention is the concrete feature based on wireless sensor network data completely, utilizes neural net to solve data prediction problem, selects to affect the principal element of data Output rusults, sets up a kind ofly can realize the model of data prediction fast and accurately.
Concrete steps of the present invention comprise:
Step 1: determine neuron.This step is classified to the data type of collecting, data bulk, according to the requirement of neural net, determines the neuron type of the data of obtaining, and comprises the neuron kind of input and the neuron kind of output.
Step 2: training sample set.For neural network model can rationally be predicted accurately, realize neural net is trained, the selection of training sample set should meet " versatility, generality " principle, focuses on objectivity.
Step 3: add momentum term.The weights adjustment amount that contains momentum term in neural net is: Δ W (t)=η δ X+ α Δ W (t-1), wherein, α is momentum term coefficient.W is the layer weight matrix of certain network layer, the input vector that X is this level, and η is learning rate.In the time that the weights of each network and threshold value change, need to add last change amount.
Step 4: calculate adaptive learning rate.
Adaptive learning rate formula is as follows:
η(k+1)＝β×η(k)E＜0 (1)
η (k+1)=γ × η other (2)
Wherein, E=E (K)-E (K-1), β and γ are respectively the constant that is greater than and is less than 1.η (k) is the k time learning rate, and E (K) is the quadratic sum of the k time neural net error.In the iterative process of network training, learning rate can carry out self adaptation adjustment according to this formula, connects different weight coefficients by different learning rates, and reasonable manner is approached to minimal point fast.
Step 5: weights correction improves.Merge following momentum gradient algorithm and carry out the correction of network layer weights, formula is as follows:
Wherein, α is factor of momentum.In the time that α is set to zero, weights correction is only relevant to current negative gradient; In the time that α is set to 1, the correction of weights determines the negative gradient value by last iteration.Adopt this pattern to carry out neural metwork training, can reduce the shock range in learning process, with respect to having added damping term.
Step 6: data prediction.Based on the neural network model of above generation, the wireless sensor data that utilization is obtained, by determining neuron, the data type etc. in different application field, the wireless sensor network data forecast model of employing based on improved neural net can carry out the data prediction of various different field, different application.
This method can improve the performance of system, improves the degree of convergence, prevents that algorithm is trapped in the middle of local extremum, thereby has improved the overall accuracy of neural net, is more suitable for data prediction to calculate.
Embodiment
Embodiment mono-
In embodiments of the invention one, data source is collected for being positioned at Zhengjiang City's area sensor network, adopts method of the present invention to carry out data prediction, predicts the possibility of this region breaking out of fire.Concrete steps are as follows:
Step 1: Data Collection and data preliminary treatment.Consider geographical position and the surrounding environment feature in certain region, choosing the input of 6 factors of influence as this neural net herein, is respectively illumination (S1), temperature (S2), humidity (S3), wind-force wind (S4), landform (S5) and flammable type (S6).
Step 2: determine input neuron.Choose above-mentioned six features as input neuron, the intensity of illumination of this certain time point of region has been described in illumination, temperature is the current Celsius temperature in this region, humidity is the current humidity in this region, wind direction has represented monsoon intensity size and the tendency in this region, landform has reflected the geologic feature in this region, and flammable type also can exert an influence to the fire complexity in this region.
Step 3: determine output neuron.The complexity of definition output is divided into: high fire generation area, middle fire generation area, low fire generation area, without fire generation area.Fire by 4 kinds of different stages is described, and can offer the effective prediction mode of user, for whether can breaking out of fire making prediction scheme in certain region.Output neuron adopts the pattern of percent M to represent, 100 represent that this output neuron is the perfect forecast of certain prediction of output factor, is expressed as: (100 therefore the fire of 4 kinds of different stages is described, 0,0,0), (0,100,0,0), (0,0,100,0), (0,0,0,100).
Step 4: neural metwork training.Waveform of real-time data take certain region by wireless sensor network is basis, get 10 weeks to the observation data in this region as sample, carry out training study, the number of times of maximum training is 2,000,000 times, network convergence error sum of squares index is 0.5 × 10
-6, first learning rate is 0.01, momentum constant is 0.9.
Carry out the training of recursive neural network repeatedly according to above setting, until convergence meets when realizing the network setting and connecting weights and threshold value, deconditioning process.
Step 5: data prediction
After sample training by above process implementation neural network model, just can carry out the fire occurrence degree prediction of following period of this region, to the fire prediction result input neurons of following three weeks with output neuron as shown: 1, as shown in table 2.
Table 1 input neuron
Table 2 output neuron
Find by above embodiment mono-result and True Data comparative analysis, wireless sensor network data forecast model based on neural net can be analyzed the wireless sensor data of Real-time Obtaining, and make correct prediction, its result and actual conditions are coincide.
Therefore, the present invention utilizes wireless sensor network to obtain real time data and predicts it is a kind of effective and feasible mode.
In addition to the implementation, the present invention can also have other execution modes.All employings are equal to the technical scheme of replacement or equivalent transformation formation, all drop on the protection range of requirement of the present invention.
Claims (4)
1. the wireless sensor network data Forecasting Methodology based on neural net, is characterized in that, described method comprises step:
S1, obtains data from different wireless sensor networks.By this network, can be real-time obtain long-range sensor network data, Data Collection, to data center, is concentrated and is processed.
S2, data preliminary treatment.The data that wireless sensor network is obtained, carry out preliminary treatment according to the difference of application, to obtain the data prediction object that can meet different application.
S3, determines neuron.This step is classified to the data type of collecting, data bulk, according to the requirement of neural net, determines the neuron type of the data of obtaining, and comprises the neuron kind of input and the neuron kind of output.
S4, training sample set.For neural network model can rationally be predicted accurately, realize neural net is trained, the selection of training sample set should meet " versatility, generality " principle, focuses on objectivity.
S5, data prediction.Based on the neural network model of above generation, the wireless sensor data that utilization is obtained, by determining neuron, the data type etc. in different application field, the wireless sensor network data forecast model of employing based on improved neural net can carry out the data prediction of various different field, different application.
2. method according to claim 1, is characterized in that, in step S2, this process adopts the technology such as data normalization, data scrubbing, data stipulations.
3. method according to claim 1, is characterized in that, in step S3, neuron models are classified according to prediction data kind.
4. method according to claim 1, is characterized in that, in step S4, the weights adjustment amount that contains momentum term in neural net is: Δ W (t)=η δ X+ α Δ W (t-1), wherein, α is momentum term coefficient.W is the layer weight matrix of certain network layer, the input vector that X is this level, and η is learning rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210411715.7A CN103781108A (en) | 2012-10-23 | 2012-10-23 | Neural network-based wireless sensor network data prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210411715.7A CN103781108A (en) | 2012-10-23 | 2012-10-23 | Neural network-based wireless sensor network data prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103781108A true CN103781108A (en) | 2014-05-07 |
Family
ID=50572819
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210411715.7A Pending CN103781108A (en) | 2012-10-23 | 2012-10-23 | Neural network-based wireless sensor network data prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103781108A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2571541C1 (en) * | 2014-06-25 | 2015-12-20 | Станислав Станиславович Махров | Method for neural network clusterisation of wireless sensor network |
| CN106961656A (en) * | 2017-02-23 | 2017-07-18 | 南京邮电大学 | A kind of wireless sensor network data Forecasting Methodology |
| CN108388213A (en) * | 2018-02-05 | 2018-08-10 | 浙江天悟智能技术有限公司 | Direct-spinning of PET Fiber process control method based on local plasticity echo state network |
| CN111489006A (en) * | 2019-01-29 | 2020-08-04 | 深圳富桂精密工业有限公司 | Fire development situation prediction method and device and computer-readable storage medium |
| CN113169887A (en) * | 2018-09-28 | 2021-07-23 | 诺基亚技术有限公司 | Radio network self-optimization based on data from radio network and spatio-temporal sensors |
-
2012
- 2012-10-23 CN CN201210411715.7A patent/CN103781108A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| RU2571541C1 (en) * | 2014-06-25 | 2015-12-20 | Станислав Станиславович Махров | Method for neural network clusterisation of wireless sensor network |
| CN106961656A (en) * | 2017-02-23 | 2017-07-18 | 南京邮电大学 | A kind of wireless sensor network data Forecasting Methodology |
| CN106961656B (en) * | 2017-02-23 | 2020-04-07 | 南京邮电大学 | Wireless sensor network data prediction method |
| CN108388213A (en) * | 2018-02-05 | 2018-08-10 | 浙江天悟智能技术有限公司 | Direct-spinning of PET Fiber process control method based on local plasticity echo state network |
| CN108388213B (en) * | 2018-02-05 | 2019-11-08 | 浙江天悟智能技术有限公司 | Direct-spinning of PET Fiber process control method based on local plasticity echo state network |
| CN113169887A (en) * | 2018-09-28 | 2021-07-23 | 诺基亚技术有限公司 | Radio network self-optimization based on data from radio network and spatio-temporal sensors |
| CN113169887B (en) * | 2018-09-28 | 2023-07-25 | 诺基亚技术有限公司 | Radio network self-optimization based on data from radio networks and spatio-temporal sensors |
| CN111489006A (en) * | 2019-01-29 | 2020-08-04 | 深圳富桂精密工业有限公司 | Fire development situation prediction method and device and computer-readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liang et al. | A deep reinforcement learning network for traffic light cycle control | |
| CN107168402B (en) | Environment of chicken house temperature intelligent monitoring system based on CAN fieldbus | |
| CN103781108A (en) | Neural network-based wireless sensor network data prediction method | |
| Niu et al. | Model turbine heat rate by fast learning network with tuning based on ameliorated krill herd algorithm | |
| CN107808212A (en) | Solar energy collecting power forecasting method based on grey neural network | |
| Saxena et al. | A review study of weather forecasting using artificial neural network approach | |
| CN109726802B (en) | Machine learning prediction method for wind speed in railway and wind farm environment | |
| Senthil Kumar et al. | Feature selection used for wind speed forecasting with data driven approaches | |
| CN105631532A (en) | Power system load prediction method using fuzzy decision-based neural network model | |
| CN104036328A (en) | Self-adaptive wind power prediction system and prediction method | |
| CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
| CN104050547A (en) | Non-linear optimization decision-making method of planning schemes for oilfield development | |
| CN107945534A (en) | A kind of special bus method for predicting based on GMDH neutral nets | |
| Li et al. | An improved gray model for aquaculture water quality prediction | |
| CN102663493A (en) | Delaying nerve network used for time sequence prediction | |
| CN107301478A (en) | A kind of cable run short-term load forecasting method | |
| Ballı et al. | An application of artificial neural networks for prediction and comparison with statistical methods | |
| Zhengxiang et al. | Transport volume forecast based on GRNN network | |
| CN107563573A (en) | A kind of Forecasting Methodology of the solar power generation amount based on adaptive learning mixed model | |
| Guoqiang et al. | Study of RBF neural network based on PSO algorithm in nonlinear system identification | |
| Hu et al. | Prediction of PM2. 5 based on Elman neural network with chaos theory | |
| He et al. | Greenhouse air temperature and humidity prediction based on improved BP neural network and genetic algorithm | |
| Furze et al. | Mathematical methods to quantify and characterise the primary elements of trophic systems | |
| Getanda et al. | Data grouping techniques’ performance analysis in GM (1, 1)’s prediction accuracy improvement for forecasting traffic parameters | |
| Cai et al. | Application of improved wavelet neural network in MBR flux prediction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140507 |