CN102645580A - Intelligent detection method for forward direction active energy incremental data of ammeter - Google Patents

Abstract

The invention discloses an intelligent detection method for forward direction active energy incremental data of an ammeter and relates to the technical field of data detection of an electrical power system. The method comprises the following steps of: S1: inputting a time sequence x=[x0, x1 ,..., xn] of the forward direction active energy incremental data of the ammeter; calculating a differential vector deltax=[deltax1, deltax2 ,..., deltaxk,..., deltaxn] of the time sequence, wherein a differential value deltaxk=/xk-xk-1/ and n represents that the time sequence has n values; S2: clustering the differential vector deltax=[deltax1, deltax2, ..., deltaxk,..., deltaxn] by utilizing a self-organizing characteristic map neural network SOM (Self-Organizing Map) algorithm to obtain the clustering center of each type; and S3: defining data contained by each type of the clustering centers which exceed the threshold value into abnormal data according to the pre-set threshold value. The intelligent detection method disclosed by the invention clusters the differential value of the forward direction active energy incremental data of the ammeter to find and eliminate the abnormal data, so that the detection efficiency and the accuracy of data detection are improved.

Description

Be used for ammeter forward active energy incremental data intelligent detecting method

Technical field

The present invention relates to electric power system data detection technique field, particularly a kind of ammeter forward active energy incremental data intelligent detecting method that is used for.

Background technology

In the heart, forward active energy increment is one of important content of measuring of intelligent electric meter in present electric power system dispatching, also is one of data of being concerned about most of dispatching center.Because the existence of surveying instrument mistake or execution error exists abnormal data in the ammeter forward active energy incremental data unavoidably.Abnormal data has adverse influence to the operation of electric system, therefore should actively abnormal data found out and rejected.Present Data Detection method relies on the eye-observation error correction mostly, and the low and accuracy of efficient is difficult to be protected.Therefore, need a kind of Data Detection method of intelligence badly.

Summary of the invention

The technical matters that (one) will solve

The technical matters that the present invention will solve is: how ammeter forward active energy incremental data is detected, to improve detection efficiency and the accuracy that detects data.

(2) technical scheme

For solving the problems of the technologies described above, the invention provides a kind of ammeter forward active energy incremental data intelligent detecting method that is used for, may further comprise the steps:

S1: the time series x=[x of input ammeter forward active energy incremental data ₀, x ₁..., x _n], calculate said seasonal effect in time series difference vector Δ x=[Δ x ₁, Δ x ₂..., Δ x _k..., Δ x _n], difference value Δ x wherein _k=| x _k-x _K-1|, n value arranged in the n express time sequence;

S2: to said difference vector Δ x=[Δ x ₁, Δ x ₂..., Δ x _k..., Δ x _n], utilize self-organizing feature map neural network SOM algorithm to carry out cluster, and obtain all kinds of cluster centres;

S3: according to pre-set threshold, all kinds of data definitions that comprise that cluster centre exceeded said threshold value are abnormal data.

Wherein, said step S2 specifically comprises:

S2.1: confirm the SOM network topology structure, setting the input layer number is 1, and the neuron number of setting output layer is M;

S2.2: t=0 is set, the initialization weight w _j(0) (j=1,2 ..., M), and each w _j(0) different;

S2.3: for said SOM network provides input value Δ x _k

S2.4: calculate the distance between current input value and each weights, and the minimum pairing neuron of weights of chosen distance is the triumph neuron;

$q\left(t\right)=\underset{j}{\min }\left|\right|{\mathrm{\Δx}}_k-w_j\left(t\right)\left|\right|$

S2.5: neuronic weight vector is in the adjustment triumph neuron and the radius of neighbourhood thereof:

$w_j(t+1)=\left\{\begin{array}{c}{w}_j\left(t\right)+\mathrm{\η}\left(t\right)[{\mathrm{\Δx}}_k-w_j\left(t\right)],j\∈N_q\left(t\right)\\ w_j\left(t\right),j\∉N_q\left(t\right)\end{array}\right.$

Wherein, η (t) is the learning rate parameter, and scope is 0＜η (t)＜1, successively decreases N in time _q(t) be the radius of neighbourhood of triumph neuron q (t), N _q(t) also successively decrease in time;

S2.6: make t=t+1, judge whether input vector all offers network,, otherwise return step S2.3 if then change step S2.7 over to;

S2.7: upgrade the learning rate parameter and the radius of neighbourhood:

$\mathrm{\η}\left(t\right)=\mathrm{\η}\left(0\right)(1-\frac{t}{T})$

$N_q\left(t\right)=\mathrm{int}\left[N_q\left(0\right)(1-\frac{t}{T})\right]$

Wherein η (0) is the initial learn rate, N _q(0) be the initial neighborhood radius, T is predetermined total iterations, and int is integer translation operation symbol;

S2.8: then finish if total iterations reaches pre-determined number T, otherwise get back to step S2.3.

Wherein, among the said step S2.2, the said weight w of picked at random from difference vector Δ x _j(0) a M initial value.

Wherein, said pre-determined number T is no less than 200 times.

Wherein, said initial learn rate η (0) value is 0.8.

Wherein, said initial neighborhood radius N _q(0) value is 2～3.

Wherein, said M value is 5～7.

Wherein, the established standards of said threshold value is to guarantee to be no less than 95% data within said threshold value.

(3) beneficial effect

The present invention introduces electric system with the thought that clustering algorithm detects outlier; Proposed to be used for the intelligent detecting method of ammeter forward active energy incremental data; Through cluster to ammeter forward active energy incremental data difference value; Can realize discovery, thereby offer help for the rejecting abnormalities data, thereby improve the accuracy of detection efficiency with the detection data to abnormal data.Particularly, the present invention has following effect to electric system: find the abnormal data of ammeter forward active energy increment, the reliability and the security that improve Operation of Electric Systems.The intelligent detecting method that is used for ammeter forward active energy incremental data that the present invention proposes can be useful among the dispatch automated system of each provincial electric system of China and regional power system; Can improve the security and the reliability of system, produce great economic and social benefit.

Description of drawings

Fig. 1 is a kind of ammeter forward active energy incremental data intelligent detecting method process flow diagram that is used for of the embodiment of the invention.

Embodiment

Below in conjunction with accompanying drawing and embodiment, specific embodiments of the invention describes in further detail.Following examples are used to explain the present invention, but are not used for limiting scope of the present invention.

Method through cluster is told outlier, is that abnormal data detects one of method commonly used.The self-organizing feature map neural network algorithm is also existing very ripe application on clustering problem.Just under such background; The present invention proposes ammeter forward active energy incremental data intelligent detecting method based on self-organizing feature map neural network (SOM) cluster; Be used for ammeter forward active energy incremental data Intelligent Measurement, idiographic flow is as shown in Figure 1, comprising:

Step S101, the time series x=[x of input ammeter forward active energy incremental data ₀, x ₁..., x _n], calculate said seasonal effect in time series difference value Δ x=[Δ x ₁, Δ x ₂..., Δ x _k..., Δ x _n], Δ x wherein _k=| x _k-x _K-1|, so that in the detection time sequence adjacent two constantly ammeter forward active energy increments differ great sudden change value.Before and after being in the time series, the sudden change value differs very large value, through making the difference that difference obtains two values.Wherein, n value arranged in the n express time sequence.

Step S102 is to above-mentioned difference vector Δ x=[Δ x ₁, Δ x ₂..., Δ x _k..., Δ x _n], utilize self-organizing feature map neural network algorithm (SOM) to carry out cluster.Concrete cluster process is following:

Step 1, according to data scale, confirm the SOM network topology structure.Because ammeter forward active energy incremental difference score value is an one-dimension array, so the input layer number is 1.According to the requirement and the classifying quality of number of categories, set the neuron number M of output layer.Cluster centre can effectively be distinguished normal data and abnormal data after the choosing of output layer neuron number M should guarantee cluster; Should guarantee that promptly the cluster centre of abnormal data and the cluster centre of normal data have enough big discrimination, general M value 5～7 is comparatively suitable.

Step 2, t=0 is set, the initialization weight w _j(0) (j=1,2 ..., M), the initial value of picked at random weights from the available set of difference vector Δ x, unique here restriction is w _j(0) different.

Step 3, input value Δ x is provided for said SOM network _k

Step 4, calculate the distance between current input vector and each weights, and the pairing neuron of weights of chosen distance minimum is the triumph neuron,

$q\left(t\right)=\underset{j}{\min }\left|\right|{\mathrm{\Δx}}_k-w_j\left(t\right)\left|\right|.$

Neuronic weight vector is in step 5, adjustment triumph neuron and the radius of neighbourhood thereof:

$w_j(t+1)=\left\{\begin{array}{c}{w}_j\left(t\right)+\mathrm{\η}\left(t\right)[{\mathrm{\Δx}}_k-w_j\left(t\right)],j\∈N_q\left(t\right)\\ w_j\left(t\right),j\∉N_q\left(t\right)\end{array}\right.$

Wherein, η (t) is the learning rate parameter, and scope is 0＜η (t)＜1, successively decreases in time.N _q(t) be the radius of neighbourhood of triumph neuron q (t), N _q(t) also successively decrease in time.

Step 6, make t=t+1, judge that whether input vector all offers network, if then change step 7 over to, otherwise returns step 3.

Step 7, renewal learning rate parameter and the radius of neighbourhood:

$\mathrm{\η}\left(t\right)=\mathrm{\η}\left(0\right)(1-\frac{t}{T})$

$N_q\left(t\right)=\mathrm{int}\left[N_q\left(0\right)(1-\frac{t}{T})\right]$

Wherein η (0) is the initial learn rate, and value is 0.8, N _q(0) be the initial neighborhood radius, value is 2～3, and T is predetermined total iterations, and this number of times is not less than 200 times usually.

Step 8, if total iterations reaches pre-determined number T then to be finished, otherwise get back to step 3.

Step S103 according to SOM algorithm cluster situation, is provided with threshold value.The established standards of threshold value is to guarantee to be no less than 95% data within threshold value.For example: Δ x ₁..., Δ x _nBe divided into 5 types, cluster centre is respectively 1,2,3,5,10, and getting threshold value is 8, and the cluster centre that is that then surpasses threshold value is that the data of those types of 10 then are outlier.Find out the outlier that exceeds threshold value, it is defined as abnormal data, thereby accomplish the Intelligent Measurement of ammeter forward active energy incremental data.

Give an actual example as follows at present, get certain ammeter forward active energy incremental data, S101 calculates difference value set by step.Be recorded in table 1.

Table 1 forward active energy incremental data difference value (unit: MWh)

Δx 1	Δx 2	Δx 3	Δx 4	Δx 5	Δx 6	Δx 7	Δx 8	Δx 9	Δx 10
										0	0.176	0	0.176	0.176	0	0	0.176	0.352	0.352
Δx 11	Δx 12	Δx 13	Δx 14	Δx 15	Δx 16	Δx 17	Δx 18	Δx 19	Δx 20
										0.352	0.352	0.176	0.352	0.176	0.176	0.176	0.176	0.176	0.176
Δx 21	Δx 22	Δx 23	Δx 24	Δx 25	Δx 26	Δx 27	Δx 28	Δx 29	Δx 30
										0.176	0.528	0.352	0.352	0.352	0.352	0.352	0.88	1.232	0.352
Δx 31	Δx 32	Δx 33	Δx 34	Δx 35	Δx 36	Δx 37	Δx 38	Δx 39	Δx 40
										0.176	0.176	0.352	0.352	0.528	0.352	0.176	0.352	0.528	0.528
Δx 41	Δx 42	Δx 43	Δx 44	Δx 45	Δx 46	Δx 47	Δx 48	Δx 49	Δx 50
										0.176	0.352	0.352	0.352	0.352	0.88	0.528	0.528	0.528	0.704

Δx 51	Δx 52	Δx 53	Δx 54	Δx 55	Δx 56	Δx 57	Δx 58	Δx 59	Δx 60
										0.528	1.232	0.704	0.176	0.704	3.168	2.816	0.352	0.176	0.176
Δx 61	Δx 62	Δx 63	Δx 64	Δx 65	Δx 66	Δx 67	Δx 68	Δx 69	Δx 70
										0.176	0.176	0.352	0.352	0.352	0.352	0.176	0.352	0.352	0.704
Δx 71	Δx 72	Δx 73	Δx 74	Δx 75	Δx 76	Δx 77	Δx 78	Δx 79	Δx 80
										0.704	0.176	0.176	2.288	0.352	0.176001	0.176001	0.176001	0.176	0.352
Δx 81	Δx 82	Δx 83	Δx 84	Δx 85	Δx 86	Δx 87	Δx 88	Δx 89	Δx 90
										1.408	0.528	0.528	1.056	0.352	0.352	0.528	0.528	1.056	1.584
Δx 91	Δx 92	Δx 93	Δx 94	Δx 95	Δx 96	Δx 97	Δx 98	Δx 99	Δx 100
										1.936	1.408	3.872	3.52	2.112	1.232	3.344	4.224	1.232	0.528
Δx 101	Δx 102	Δx 103	Δx 104	Δx 105	Δx 106	Δx 107	Δx 108	Δx 109	Δx 110
										0.88	1.408	1.584	1.056	0.88	1.232	0.528	0.528	1.76	1.408
Δx 111	Δx 112	Δx 113	Δx 114	Δx 115	Δx 116	Δx 117	Δx 118	Δx 119	Δx 120
										1.584	1.76	1.584	0.704	1.056	0.704	0.528	1.408	1.584	0.528
Δx 121	Δx 122	Δx 123	Δx 124	Δx 125	Δx 126	Δx 127	Δx 128	Δx 129	Δx 130
										3.344	1.056	0.88	0.352	0.704	0.704001	0.352	0.88	0.528	0.704
Δx 131	Δx 132	Δx 133	Δx 134	Δx 135	Δx 136	Δx 137	Δx 138	Δx 139	Δx 140

0.528001	0.704	0.704001	0.704	0.704	1.232	0.704	0.704	1.232	1.232
										Δx 141	Δx 142	Δx 143	Δx 144	Δx 145	Δx 146	Δx 147	Δx 148	Δx 149	Δx 150
0.704001	0.704	0.88	0.528001	0.88	0.528001	0.704	1.936	1.936	1.936
										Δx 151	Δx 152	Δx 153	Δx 154	Δx 155	Δx 156	Δx 157	Δx 158	Δx 159	Δx 160
1.936	0.88	1.76	1.232	0.88	2.288	1.76	2.112	0.176	0.176
										Δx 161	Δx 162	Δx 163	Δx 164	Δx 165	Δx 166	Δx 167	Δx 168	Δx 169	Δx 170
0.176	0.176	1.936	0.528	3.344	1.408	2.288	0.704	1.76	1.76
										Δx 171	Δx 172	Δx 173	Δx 174	Δx 175	Δx 176	Δx 177	Δx 178	Δx 179	Δx 180
1.584	7.216	7.392	0.352	0.352	0.176	0.176	0.176	0.176	0.352
										Δx 181	Δx 182	Δx 183	Δx 184	Δx 185	Δx 186	Δx 187	Δx 188	Δx 189	Δx 190
0.352	6.512	6.688	6.864	6.864	1.056	1.232	0.704	1.584	0.528001
										Δx 191	Δx 192	Δx 193	Δx 194	Δx 195	Δx 196	Δx 197	Δx 198	Δx 199	Δx 200
0.528001	0.88	0.704	0.528	1.408	1.936	2.112	1.056	1.056	0.704001
										Δx 201	Δx 202	Δx 203	Δx 204	Δx 205	Δx 206	Δx 207	Δx 208	Δx 209	Δx 210
0.704	0.704	0.88	1.056	0.528	0.528	0.528	0.528	0.528	0.88
										Δx 211	Δx 212	Δx 213	Δx 214	Δx 215	Δx 216	Δx 217	Δx 218	Δx 219	Δx 220
2.64	0.528	0.351999	0.704	1.584	0.88	0.528	0.528	0.528	0.528

Δx 221	Δx 222	Δx 223	Δx 224	Δx 225	Δx 226	Δx 227	Δx 228	Δx 229	Δx 230
										0.351999	0.528	0.351999	0.176	0.176	0.176001	0.176	0.176001	0.352	0.352
Δx 231	Δx 232	Δx 233	Δx 234	Δx 235	Δx 236	Δx 237	Δx 238	Δx 239	Δx 240
										0.352	0.352	0.352	0.176	0.176	0.176	0.176	0.352	0	0
Δx 241	Δx 242	Δx 243	Δx 244	Δx 245	Δx 246	Δx 247	Δx 248	Δx 249	Δx 250
										0.352	0.352	0.176	0.176	0.352	0.176	0.176	0	0.352	0.352

S102 carries out cluster to above-mentioned difference value set by step, cluster result such as table 2:

Table 2 cluster result

Classification	1	2	3	4	5	6	7
								Cluster centre	0.1561	0.3520	0.5984	1.1098	1.8513	3.3636	6.8288
Comprise data volume	56	50	62	40	27	9	6

Requirement among the S103 set by step, the choosing of threshold value should guarantee that 95% data are in threshold value.Thus, for above-mentioned 250 groups of data, should guarantee that 238 groups of data of 250 * 0.95 ≈ are in threshold value.Therefore, can think that cluster centre is that 6 groups of data of 6.8288 (classifications 7) are abnormal data, with its rejecting.According to the cluster feedback information, 6 abnormal datas are seen table 3.

Table 3 abnormal data

Sequence number	Δx 172	Δx 173	Δx 182	Δx 183	Δx 184	Δx 185
							Numerical value	7.216	7.392	6.512	6.688	6.864	6.864

Adopt the mode of above-mentioned cluster automatically the seasonal effect in time series difference value of ammeter forward active energy incremental data to be carried out cluster; And judge abnormal data and removal according to pre-set threshold, realized that ammeter forward active energy incremental data detects efficiently and accurately.

Above embodiment only is used to explain the present invention; And be not limitation of the present invention; The those of ordinary skill in relevant technologies field under the situation that does not break away from the spirit and scope of the present invention, can also be made various variations and modification; Therefore all technical schemes that are equal to also belong to category of the present invention, and scope of patent protection of the present invention should be defined by the claims.

Claims (8)

1. one kind is used for ammeter forward active energy incremental data intelligent detecting method, it is characterized in that, may further comprise the steps:

S1: the time series x=[x of input ammeter forward active energy incremental data ₀, x ₁..., x _n], calculate said seasonal effect in time series difference vector Δ x=[Δ x ₁, Δ x ₂..., Δ x _k..., Δ x _n], difference value Δ x wherein _k=| x _k-x _K-1|, n value arranged on the n express time sequence;

S2: to said difference vector Δ x=[Δ x ₁, Δ x ₂..., Δ x _k..., Δ x _n], utilize self-organizing feature map neural network SOM algorithm to carry out cluster, and obtain all kinds of cluster centres;

S3: according to pre-set threshold, all kinds of data definitions that comprise that cluster centre exceeded said threshold value are abnormal data.

2. the ammeter forward active energy incremental data intelligent detecting method that is used for as claimed in claim 1 is characterized in that said step S2 specifically comprises:

S2.1: confirm the SOM network topology structure, setting the input layer number is 1, and the neuron number of setting output layer is M;

S2.2: t=0 is set, the initialization weight w _j(0) (j=1,2 ..., M), and each w _j(0) different;

S2.3: for said SOM network provides input value Δ x _k

S2.4: calculate the distance between current input value and each weights, and the minimum pairing neuron of weights of chosen distance is the triumph neuron;

$q\left(t\right)=\underset{j}{\min }\left|\right|{\mathrm{\Δx}}_k-w_j\left(t\right)\left|\right|$

S2.5: neuronic weight vector is in the adjustment triumph neuron and the radius of neighbourhood thereof:

$w_j(t+1)=\left\{\begin{array}{c}{w}_j\left(t\right)+\mathrm{\η}\left(t\right)[{\mathrm{\Δx}}_k-w_j\left(t\right)],j\∈N_q\left(t\right)\\ w_j\left(t\right),j\∉N_q\left(t\right)\end{array}\right.$

Wherein, η (t) is the learning rate parameter, and scope is 0＜η (t)＜1, successively decreases N in time _q(t) be the radius of neighbourhood of triumph neuron q (t), N _q(t) also successively decrease in time;

S2.6: make t=t+1, judge whether input vector all offers network,, otherwise return step S2.3 if then change step S2.7 over to;

S2.7: upgrade the learning rate parameter and the radius of neighbourhood:

$\mathrm{\η}\left(t\right)=\mathrm{\η}\left(0\right)(1-\frac{t}{T})$

$N_q\left(t\right)=\mathrm{int}\left[N_q\left(0\right)(1-\frac{t}{T})\right]$

Wherein η (0) is the initial learn rate, N _q(0) be the initial neighborhood radius, T is predetermined total iterations, and int is integer translation operation symbol;

S2.8: then finish if total iterations reaches pre-determined number T, otherwise get back to step S2.3.

3. the ammeter forward active energy incremental data intelligent detecting method that is used for as claimed in claim 2 is characterized in that, among the said step S2.2, and the said weight w of picked at random from difference vector Δ x _j(0) a M initial value.

4. the ammeter forward active energy incremental data intelligent detecting method that is used for as claimed in claim 2 is characterized in that said pre-determined number T is no less than 200 times.

5. the ammeter forward active energy incremental data intelligent detecting method that is used for as claimed in claim 2 is characterized in that said initial learn rate η (0) value is 0.8.

6. the ammeter forward active energy incremental data intelligent detecting method that is used for as claimed in claim 2 is characterized in that said initial neighborhood radius N _q(0) value is 2～3.

7. the ammeter forward active energy incremental data intelligent detecting method that is used for as claimed in claim 2 is characterized in that said M value is 5～7.

8. the ammeter forward active energy incremental data intelligent detecting method that is used for as claimed in claim 1 is characterized in that, the established standards of said threshold value is to guarantee to be no less than 95% data within said threshold value.

Images