Consider a string 'ABBAA'
Possible substrings with even number of characters are $4$
'ABBA' : Count of 'A' is even and 'B' is even
'AA' : Count of 'A' is even and 'B' is even - ($0$)
Similarly 'BB' and 'BBAA' substrings.
So a total of $4$ substrings are possible.
If given string is 'AABCCBBCAA'
substrings possible are "AA" "BB" "AA" "CC" "BCCB" "CCBB" "CBBC" "CBBCAA" "AABCCB"
So total $9$ substrings are possible
Is there any pattern or sequence or algorithm to find this?
Assuming that your alphabet has constant size, you can solve your problem in linear time in the length of the input string.
Let $\Sigma = \{a_1, a_2, \dots, a_m\}$ be your alphabet and $s = s_1 s_2 \dots s_n$ be your input string. For $i=0, \dots, n$, and $j=1,\dots,m$ let $n_j(i)$ be $0$ iff the number of occurrences of $a_j$ in $s_1 \dots s_i$ is even and $1$ otherwise. Let $\eta(i) \in \{0,1\}^{m}$ be the vector $(\eta_1(i), \dots, \eta_m(i))$.
Consider any substring $s_{hk} = s_{h+1},s_{h+2}, \dots, s_k$ of $s$, where $0 \le h < k \le n$. The number of occurrences of each $a_j$ in $s_{hk}$ is even if and only if $\eta(k)-\eta(h) = (0, \dots, 0)$, where the difference is taken modulo $2$, i.e., iff $\eta(k)=\eta(h)$.
Then your problem is equivalent to counting the number of unordered pairs of distinct indices $\{h,k\}$ for which $\eta(h)=\eta(k)$. This can be done as follows: given $x \in \{0,1\}^{m}$ let $\alpha(x)$ be the number of times of $x$ appears in $\eta(0), \dots, \eta(n)$, so that the number of unordered pairs $\{h,k\}$ for which $\eta(h) = \eta(k) = x$ is $\binom{\alpha(x)}{2} = \frac{\alpha(x) (\alpha(x)-1)}{2}$.
To find the find the overall number of substrings of interest it suffices to sum the above quantity over all choices of $x \in \{0,1\}^m$: $$ \sum_{x \in \{0,1\}^m} \binom{\alpha(x)}{2} = \frac{1}{2} \sum_{x \in \{0,1\}^m} \alpha(x) (\alpha(x)-1). $$
Since $m=O(1)$, all values of $\eta(i)$, for $i=0, \dots, n$ can be computed in $O(nm) = O(n)$ time. Moreover, all values of $\alpha(x)$ for $x \in \{0,1\}^m$ can also be found in $O(n m \log \min\{n, 2^m\}) = O(n)$ time by scanning $\eta(0), \dots, \eta(n)$ once and maintaining a counter for each $x$ (these counters can be stored in a dictionary with at most $\min\{n,2^m\}$ elements). Finally, the final sum can be computed using $O(\min\{n, 2^m\}) = O(1)$ arithmetic operations (as it only needs to extend to $x \in \{0,1\}$ for which $\alpha(x) > 0$).
An example implementation due to @orlp of the above in Python using a binary integer as the vector $x$ and storing all $a(x)$ in a hashmap counter is as follows:
def num_even_substrings(s):
a = {0: 1}
x = 0
for c in s:
x ^= 1 << ord(c)
a[x] = a.get(x, 0) + 1
return sum(ax*(ax-1)//2 for ax in a.values())
